Home

development of a wearable mobility monitoring system

1. Taking elevator to 1 1 1 1 1 1 0 1 1 14 10 71 496 2 floor Standing waiting 1 1 1 1 1 1 1 1 1 NA 1 1 1 1 1 14 14 100 0 Walking to get out of elevator and 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 15 100 0 keep walking on level ground Standing waiting for elevator 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 15 100 095 q xipueddy urojs amp g Suuojuo NA K1 Iqo A 9 q91e9AA Jo juouido oAo T 6S1 Walking to get in ihoelovator 0 0 0 0 0 0 1 0 1 0 1 1 Taking elevator to NOP 1 floor 1 1 1 0 0 0 1 IC 0 0 1 0 1 1 1 14 8 57 1 Walking to get out of elevator and 1 1 0 1 1 1 1 1 1 0 1 1 1 1 1 15 13 86 7 keep walking on level ground NOP NOP NOP NOP NOP OP NOP NOP NOP NOP NOP Walk pepe Me 0 i Ic ic IC ED ic ic te I6 ic Walking on stair intermediate NOP NOP NOP NOP NOP Nor Noe NOP NOP landing level IC 1 IC IC IC IC IC IC 1 1 1 ground for approx 1 5 meter 7 NOP NOP NOP NOP NOP p YA NOP Walking up stairs 0 1 1 IC IC IC IC IC IC 0 1 1 Walking on level 4 NOPI 0 NOP 1 NOP NOP NOP NOP ground C IC IC IC IC IC Walking down Stairs 0 0 0 0 0 0 0 Walking on stair intermediate landing level NOP 0 0 0 NOP 0 0 IC IC ground for approx 1 5 meter Walking down NOP o stais IC 0 0 0 0 0 0 0
2. Subject 1 Subject 2 Subject 3 Subject 4 S PA Change of State uo a a identifying 1 A 5 context Walking on level 1 1 1 93 395 ground Stand to sit transition 0 1 1 92 9 Sitting 0 1 1 93 3 Sit to stand 1 1 1 100 0 Walking on level o ground 9 1 ird Standing waiting 1 1 1 85 7 for elevator Walking to get in o the elevator 9 d 9 9 KA Taking elevator to 2 floor 78 6 q xipueddy urojs amp s SuuojmuoJA Ad IqoJA ALII AA JO juoeuido oAo q SSI Walking to get out of elevator and keep walking on level ground Standing waiting for elevator Walking to get in the elevator Taking elevator to 1 floor Walking to get out of elevator and keep walking on level ground Walking up stairs Walking on stair intermediate landing level ground for approx 1 5 meter Walking up stairs Walking on level ground Walking down stairs Walking on stair intermediate landing level ground for approx 1 5 meter NOP IC NOP IC IC IC NOP NOP NOP IC NOP NOP NOP IC NOP NOP NOP IC IC IC 1 1 1 1 1 1 NOP NOP NOP IC IC IC 0 0 0 NOP IC 0 0 NOP IC 15 100 0 93 3 7 1 85 7 86 7 100 0 100 0 100 0 75 0 0 0 0 0 q xipueddy
3. 2 2 3 Technologies for Biomechanical Measurements The following describes commonly used instruments to quantify different biomechanical parameters in laboratory settings This includes gait and foot pressure analyses Some of the following instruments have the advantage of being very accurate but are limited by space requirements setup time and cost Development of a Wearable Mobility Monitoring System 13 Literature Review 2 2 3 1 Visual Motion Tracking System Visual motion tracking systems can be either a marker or marker free system based on whether they need markers to be affixed to body parts Motion tracking systems can be integrated with force plates and electromyography EMG Figure 2 3 Vicon Motion System 62 systems in a laboratory setting In marker based tracking systems cameras record the motion of light reflecting or light producing markers attached to the human body An example is the Vicon Motion System 62 Figure 2 3 These video based systems often represent the gold standard in human motion analysis 63 In a marker free system human motion is analyzed with computer vision techniques and algorithms 64 For both marker and marker free systems the number of cameras used to capture three dimensional 3D data will vary depending on the laboratory needs size and configuration Major drawbacks include the time for setup camera calibration and marker placement 2 2 3 2 Non Visual Mot
4. 74 C Tudor Locke J E Williams J P Reis and D Pluto Utility of pedometers for assessing physical activity Convergent validity Sports Medicine vol 32 pp 795 808 2002 75 J T Cavanaugh K L Coleman J M Gaines L Laing and M C Morey Using step activity monitoring to characterize ambulatory activity in community dwelling older adults Journal of the American Geriatrics Society vol 55 pp 120 124 2007 76 D Giansanti V Macellari and G Maccioni Telemonitoring and telerehabilitation of patients with Parkinson s disease Health technology assessment of a novel wearable step counter Telemedicine and e Health vol 14 pp 76 83 2008 77 A Godfrey R Conway D Meagher and G Laighin Direct measurement of human movement by accelerometry Medical Engineering and Physics vol 30 pp 1364 1386 2008 78 Stayhealthy Inc R73 Research Activity Monitor Stayhealthy Online Available http www stayhealthy com page view3789 html id products rt3 Accessed 11 Nov 2009 79 PAL Technologies Ltd ActivPAL PALTechnologies Limited Online Available http www paltechnologies com Accessed 11 Nov 2009 80 C V C Bouten K T M Koekkoek M Verduin R Kodde and J D Janssen A triaxial accelerometer and portable data processing unit for the assessment of daily physical activity IEEE Transactions on Biomedical Engineering vol 44 pp 136 147 1997 81 M J Mathie A C
5. 109 K Aminian and B Najafi Capturing human motion using body fixed sensors Outdoor measurement and clinical applications Computer Animation and Virtual Worlds vol 15 pp 79 94 2004 110 K M Culhane M O Connor D Lyons and G M Lyons Accelerometers in rehabilitation medicine for older adults Age and Ageing vol 34 pp 556 560 2005 111 J J Kavanagh and H B Menz Accelerometry A technique for quantifying movement patterns during walking Gait and Posture vol 28 pp 1 15 2008 112 J F Knight H W Bristow S Anastopoulou C Baber A Schwirtz and T N Arvanitis Uses of accelerometer data collected from a wearable system Personal and Ubiquitous Computing vol 11 pp 117 132 2007 113 H J Luinge and P H Veltink Measuring orientation of human body segments using miniature gyroscopes and accelerometers Medical and Biological Engineering and Computing vol 43 pp 273 282 2005 114 H Lau and K Tong The reliability of using accelerometer and gyroscope for gait event identification on persons with dropped foot Gait and Posture vol 27 pp 248 257 2008 115 J Favre B M Jolles R Aissaoui and K Aminian Ambulatory measurement of 3D knee joint angle Journal of Biomechanics vol 41 pp 1029 1035 2008 116 M N Nyan F E H Tay and E Murugasu A wearable system for pre impact fall detection Journal of Biomechanics vol 41 pp 3475 3481 2008 117 D
6. 13 G H Jin S B Lee and T S Lee Context awareness of human motion states using accelerometer Journal of Medical Systems vol 32 pp 93 100 2008 14 T Choudhury G Borriello S Consolvo D Haehnel B Harrison B Hemingway J Hightower P Klasnja K Koscher A LaMarca J A Landay L LeGrand J Lester A Rahimi A Rea and D Wyatt The mobile sensing platform An embedded activity recognition system JEEE Pervasive Computing vol 7 pp 32 41 2008 15 U Maurer A Rowe A Smailagic and D Siewiorek Location and activity recognition using eWatch A wearable sensor platform in Ambient Intelligence in Everyday Life 2006 pp 86 102 16 S E Lord K McPherson H K McNaughton L Rochester and M Weatherall Community ambulation after stroke How important and obtainable is it and what measures appear predictive Archives of Physical Medicine and Rehabilitation vol 85 pp 234 239 2004 17 J S Frank and A E Patla Balance and mobility challenges in older adults Implications for preserving community mobility American Journal of Preventive Medicine vol 25 pp 157 163 2003 18 World Health Organisation WHO International Classification of Functioning Disability and Health ICF Geneva Switzerland World Health Organisation 2001 19 World Health Organisation International Classification of Functioning Disability and Health ICF World Health Organization 2009 Online Av
7. 194 J Stokes and J Lindsay Major caues of death and hospitalization in Canadian seniors Chronic Diseases in Canada vol 17 pp 63 73 1996 Development of a Wearable Mobility Monitoring System 145 Appendix A Appendix A Final schematics of the external board used for the WMMS Development of a Wearable Mobility Monitoring System 146 Appendix A D ETI 600272 JSquie oN T og aequis waung anz S0809 r Bises reupz zo rps ecupel OF apex 2017 v zeredos aq ues zy os DIOS 91 ZEZEXVIN neono za noH Dino NOT Bio NISL INO Ho NT z9 A A qaog bnqeq S940011 s0904 a amo 0909 919I0S 1985442t z3 430V3H Lol ENEJ ant 0809 19 xu zeesu Z XL zezsy z 147 Development of a Wearable Mobility Monitoring System Appendix A T zr z T Y z p z eg 8002 Te j9qum oN Apu sig I 2 L a lt g gt s zagwny iueumzo aa Ijonuogoo Hiveessiy SN 0909 ZI SNO 0909 fine 29 FE now Kumogsmog 13990 He 10 0m01079 0e z0 cue beiz 220 2 dN 0904 TIO 70 owezsuy vexpabei sp 59 27 1eze4oieTeccy Twdopzdo SNO 0909 ad Kumogismog 13904 X isas 13904 sGeuoA 100 TaD IW lt 1193138 7399 SNG 4doz lt 0193738 7390v 01 0809 ud 001 46 NNOD sisa ASN WIKIO 1d SSA MI 0904 umogie
8. Hardware Design and Evaluation arannana Ces ep lt Power and eee Rechargeable Circuit Light Sensor Temperature and Humidity Sensor Accelerometer Debugand Figure 6 3 Image of the board with all the sensors identified Table 6 1 Summary of specifications for main component of the external sensors board Device Type Manufacturer Part Summary of Specifications Number Microcontrol Cypress CY8C27443 MBC Processor Speeds to 24MHz ler 188 Semiconductor 248XI 8x8 Multiply 32 Bit Accumulate Corporation Low Power at High Speed 3 0V to 5 25V Operating Voltage 12 Rail to Rail Analog PSoC blocks 8 Digital PSoC Blocks Programmable Clocking 16K Flash Program Storage 256 Bytes SRAM Data Storage Watchdog and Sleep Timers Physical size LxWxH mm 18 1x7 6x0 1 Weight 0 85grams Development of a Wearable Mobility Monitoring System 71 Bluetooth Module 189 Accelerometer 166 Light Sensor 190 Digital Humidity and Temperature Sensor 191 Free2Move ST Microelectronics Avago Technologies Limited Sensirion AG F2M03GLA LIS344ALH APDS 9005 SHT71 Hardware Design and Evaluation Fully qualified end product with Bluetooth v2 0 EDR CE FCC and IC Low Power consumption Nominal transmit power 6dBm Nominal sensitivity 83dBm Frequency 2 4GHz ISM band Range up to 250m line of sight Integrated high output antenna 8Mbit Flash for complete system solu
9. no peak with increase in intensity would occur during riding in a car This false state was detected with the car s stop and go motion at a stop sign Since an increase in intensity should happen when the person is moving another threshold to verify that the person was in a certain active state was added to the algorithm The algorithm verified that the standard deviation was above 0 1g in order to detect the state no peak with increase in intensity SMA of x y and z axis acceleration signals and standard deviation of y axis STDY SMA gt High Threshold AND STDY gt Active Threshold State No Peak with State Peak p State Previous state EUER increase in intensity RENTEN State No peak normal intensity Figure 7 8 Flowchart of the SMA algorithm 7 3 Light The light sensor on the external board measured light intensity of the ambient environment Light intensity level detected indoor and outdoor states during the day During preliminary hardware testing the light sensor was calibrated with different light conditions Table 6 2 in Section 6 2 7 From those results it was estimated that a high threshold of 1000 mV and a Development of a Wearable Mobility Monitoring System 92 Development of the Prototype WMMS low threshold of 300 mV would differentiate outdoor from indoor states during the day The same DT algorithm as the one applied to the standard deviation was applied to the light
10. 1 1 EE 1 0 0 1 EXE 1 1 10 5 66 67 Walking on level ground 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 0 100 00 Stand to lie transition 1 1 1 1 1 ERES 1 1 1 1 eT 1 1 15 0 100 00 Lying 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 0 100 0096 Lie to Stand transition 1 1 1 1 1 Rage 1 1 1 1 EXP 1 1 15 0 100 0096 Walking on level ground 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 0 100 00 Walking on ramp 1 1 1 0 0 lolo 1 0 1 1 kale 0 0 6 9 40 00 Walking on level ground 1 1 0 0 0 0 0 1 1 0 1 1 0 1 0 7 8 46 67 1 1 7 8 46 67 Transition indoor outdoor and keep 0 0 0 1 1 9 xipueddy uiojs S Surio1uoJA NTIQON 9 qe1eoAA Jo juouido oAo q EST Heee ic 0 SAAS ean A ip pO S0 eg pss Del 28 Stand to sit transition to get in the car 1 1 1 1 1 REY 1 1 1 1 PEE 1 1 15 0 100 00 Sitting in the car 1 1 1 1 1 elai 1 1 1 1 ERES 1 1 15 0 100 0096 Starts of car ride 0 1 0 0 1 1 1 1 1 0 0 1 1 1 1 10 5 66 67 Stop of car ride 0 1 0 0 NA PX E 1 1 0 0 PUTES 1 1 9 5 64 29 Sit to stand transition 1 1 1 1 1 1 1 1 1 1 1 1 0 0 1 13 2 86 67 Walking on level ground 0 1 1 1 1 ae 1 1 1 1 rae lea 1 1 14 1 93 33 Transition outdoor indoor and keep 0 0 0 0 0 1 0 0 0 1 1 1 0 0 0 4 11 26 6796 walking on level ground Standing peda fafatafafafafata mala fa fafa a 0 100 00 9 xipueddy urojs amp s SuuojmuoJA i IqoJA ALII AA JO juouido oAo q YSI Appendix D Table D 1 Picture evaluation results from evaluator 1
11. 46 7 20 0 100 0 100 0 66 7 64 3 86 7 93 3 26 7 100 0 1 7 Technical and Mobility Evaluation of the Prototype WMMS 8 2 5 BlackBerry Image Evaluation Results Two evaluators evaluated each picture taken for true positive changes of state Table 8 5 gives the percentage of pictures where each evaluator identified the context successfully An overall percentage of 74 3 1 9 was obtained The results from each evaluator for each of the trials are given in Appendix D Some contexts were frequently identified from the images gt 95 Most successful image categorizations happened during good light condition and when fewer details had to be identified in the image i e indoor Walking while in the Rehab Technology Lab which was a darker room had 53 3 success rate for walking on level ground after getting up from the bed and 42 996 for walking on level ground after walking on the ramp Also in the lab the ramp was not well identified at 16 7 Walking in the elevator had low results as well 15 4 for the elevator going up and 21 4 the elevator going down However standing in the elevator obtained 75 0 for going up and 71 496 for going down For walking up stairs the stairs could be identified from seven images out of eight for one evaluator and all eight images for the other evaluator However the WMMS pictures were unable to identify stairs descent For images taken while sitt
12. Stand to sit transition to get in the car 1 86 7 80 0 83 3 4 7 5 15 93 3 80 0 86 7 9 4 100 0 100 0 100 0 0 0 HStopofcarride 88 9 80 0 84 4 6 3 92 3 69 2 80 8 16 3 92 9 73 3 83 1 13 8 meer eee keep walking on level ground 4 25 096 50 0 37 5 17 7 100 0 100 0 100 0 0 0 Total Percentage of Successfully Identifying Context 75 7 73 0 74 3 1 9 N Development of a Wearable Mobility Monitoring System 120 Technical and Mobility Evaluation of the Prototype WMMS 8 3 Mobility Task Discussion As previously emphasized by the ICF model 18 and the Dimensions of Mobility from Patla and Shumway Cook 1 accounting for the environmental factors during mobility assessment is important Our results suggest that BlackBerry smartphones have great potential for community mobility monitoring The integrated camera can capture information on the context environment in which mobility events take place Additionally the BlackBerry had the necessary processing power to process and log data run algorithms collect GPS data and take pictures all without data loss 8 3 1 Use of Images in WMMS Our approach of taking a photograph when a change of state occurred demonstrated that mobility tasks such as taking an elevator or going up stairs could be identified from the images For the photographs taken when the subjects took the elevator the elevator context was identified from the images at 75 0 and 71 4
13. Support Vector Machine Ultra wideband Wireless Body Area Networks Wireless Body Sensor Networks Wireless Local Area Network Wearable Mobility Monitoring System Development of a Wearable Mobility Monitoring System xi Acknowledgment I would like to thank Dr Lemaire and Dr Baddour for their support and guidance that helped me complete my research I would never be able to thank enough my fianc Keith Heggie for his help support understanding patience and encouragement throughout that journey Thanks to Keith Heggie for designing and providing the external sensor board Thanks to Research In Motion RIM for their technical and financial support The Ontario Graduate Scholarships in Science and Technology program and the Ontario Centers of Excellence are also acknowledged for financial support Thanks to all the people at the Ottawa Hospital Rehabilitation Center especially to Cindy Kendell for all your help and friendship Development of a Wearable Mobility Monitoring System xii Introduction Chapter 1 Introduction Mobility can be defined as the ability to move independently from one point to another 1 and is essential for maintaining independence Mobility is required to perform many activities of daily life such as cooking dressing shopping and visiting friends According to Statistics Canada mobility problems are one of the issues that affect the greatest number of adults 2 The number of people wit
14. but is now also used in research settings patient care and general population surveys 45 The HAQ disability dimension consists of a self report of 20 questions that covers eight areas dressing and grooming arising eating walking hygiene reaching gripping and outdoor activities The score on each question is averaged to create a global Functional Disability Index score 59 2 2 2 4 Environmental Analysis of Mobility Questionnaire The Environmental Analysis of Mobility Questionnaire EAMQ was developed by Shumway Cook et al 60 as a self report questionnaire EAMQ collects information on 24 features of the physical environment grouped within eight dimensions Section 2 1 2 Subjects were asked to report the frequency of encounters or avoidance using a five point ordinal scale never rarely sometimes often always for each of the features Preliminary results indicated that mobility disability is characterized by a reduction in the number and type of environmental challenges A reduction of encounters could lead to a reduction in movement for an individual which could potentially lead to further deterioration in physical status and social interactions The questionnaire was suggested to be a valid method for determining environmentally specific mobility disability 61 EAMQ was validated using video camera and direct observation Further research with a larger sample was still necessary to verify the findings from this study 61
15. uiojsAs Suriojuo A Ki rqoJA QLI AA Jo juouido oAo T 9ST Walking on level 0 0 Walking down NOP NOPI NOP 0 0 0 0 NOP NOP 0 s stairs IC C IC IC IC 1 1 1 1 1 0 o 11 ov ground Stand to lie 100 0 transition Lying Lie to Stand transition Walking on level ground on level ground NOP NOP NOP NOP NOP NOP NOP NOP NOP Walking on ramp 0 1 0 IC IC IC IC 1 IC 0 0 IC IC IC IC 8 E 33 396 Walking on level NOP NOP NOP NOP NOP NOP NOP NOP x ground 1 0 cl 1G 1 4e de de efie ie d IC 3 421996 Transition indoor outdoor ira jc cs 1 1 1 1 pa p 1 1 1 Me Me us 7 7 100 0 and keep walking on level ground Transition outdoor indoor T od 1 1 0 1 e o 1 0 1 p o jud 7 5 71 496 and keep walking on level ground Gn UE ade NOP NOPI Nop NOP nop NOP NOP NOP NOP NOP NOP NOP A and keep walking IC C IC IC IC IC IC IC IC IC IC IC VUE o q xipueddy uiojsAs Surioduo A Ki rqoJA QLI AA Jo juouido oAo T LST Identifying Context 86 796 Stand to sit transition to get in 0 1 1 1 1 1 0 13 the car Sitting in the car 0 1 1 1 1 1 1 14 93 396 NOP NOP NOP h Starts of car ride IC 1 IC IC 1 1 1 10 100 096 NOP NOP NOP 2
16. 1 chest gyroscopes accelerometer Detection success Visual detection errors 20 some cases Correlations r 0 77 and 0 89 for IMA and EE44 8890 spontaneous 96 standard video to monitor 9590 posture 67 ambulation Visual detection 99 postural transition gt 90 lie walk Motivation and activity recognition Physical Activity PA static dynamic activities stand sit lying supine walking cycling ascending descending stairs speed of activity PA bench test of device correlation of activities of daily living dressing walk lie desk work etc in respiration chamber to monitor output Psychophysiological study in the young static dynamic activities 40 activity protocols sit lie stand walk variations etc Electrocardiogram ECG Ambulatory monitoring retests 9 postures lab ref of sit lie walk stairs etc recording in real world vs observer speech activity and heart rate PA 11 postures e g lying left right supine and prone Postures posture transitions gyroscope walking periods Signal processing and algorithm Threshold mean values standard deviation signal morphology correlations cycle times Time integrals from separate measurement direction IM Av versus energy expenditure due to physical activity EE act chamber mean std deviation FFTs Threshold video analysis 1 second resolutions psychophysiological effect of benzodia
17. 10 pp 144 151 2004 Development of a Wearable Mobility Monitoring System 140 References 141 J Baek G Lee W Park and B J Yun Accelerometer signal processing for user activity detection in Lecture Notes in Computer Science Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics 2004 pp 573 580 142 A K Bourke J V O Brien and G M Lyons Evaluation of a threshold based tri axial accelerometer fall detection algorithm Gait and Posture vol 26 pp 194 199 2007 143 T Yoshida F Mizuno T Hayasaka K Tsubota Y Imai T Ishikawa and T Yamaguchi Development of a wearable surveillance system using gait analysis Telemedicine and e Health vol 13 pp 703 713 2007 144 J B J Bussmann L Damen and H J Stam Analysis and decomposition of signals obtained by thigh fixed uni axial accelerometry during normal walking Medical and Biological Engineering and Computing vol 38 pp 632 638 2000 145 K Aminian K Rezakhanlou E De Andres C Fritsch P F Leyvraz and P Robert Temporal feature estimation during walking using miniature accelerometers An analysis of gait improvement after hip arthroplasty Medical and Biological Engineering and Computing vol 37 pp 686 691 1999 146 R LeMoyne C Coroian and T Mastroianni Quantification of Parkinson s disease characteristics using wireless accelerometers in CME International Conference
18. 123 S E Wiehe A E Carroll G C Liu K L Haberkorn S C Hoch J S Wilson and J D Dennis Using GPS enabled cell phones to track the travel patterns of adolescents International Journal of Health Geographics vol 7 2008 124 G MacLellan and L Baillie Development of a location and movement monitoring system to quantify physical activity in Proceeding for the Conference on Human Factors in Computing Systems 2008 pp 2889 2894 125 Y Michael E M McGregor J Allen and S Fickas Observing outdoor activity using global positioning system enabled cell phones in Lecture Notes in Computer Science Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics 2008 pp 177 184 126 N Ueda Y Nakanishi S Matsukawa and M Motoe Developing a GIS using a mobile phone equipped with a camera and a GPS and its exhibitions in Proceedings of the 24th International Conference on Distributed Computing Systems Workshops 2004 pp 414 417 127 A Le Faucheur P Abraham V Jaquinandi P Bouy J L Saumet and B Noury Desvaux Study of human outdoor walking with a low cost GPS and simple spreadsheet analysis Medicine and Science in Sports and Exercise vol 39 pp 1570 1578 2007 128 H Yamazoe A Utsumi K Hosaka and M Yachida A body mounted camera system for head pose estimation and user view image synthesis Image and Vision Computing vol 25 pp 1848 1855
19. 2007 129 Microsoft Corporation Introduction to SenseCam Microsoft Research 2007 Online Available http research microsoft com en us um cambridge projects sensecam Accessed 13 Oct 2009 130 D Byrne B Lavelle A Doherty G Jones and A F Smeaton Using Bluetooth and GPS metadata to measure event similarity in SenseCam images in IMAT 07 5th Development of a Wearable Mobility Monitoring System 139 References International Conference on Intelligent Multimedia and Ambient Intelligence 2007 pp 1454 1460 131 E Berry N Kapur L Williams S Hodges P Watson G Smyth J Srinivasan R Smith B Wilson and K Wood The use of a wearable camera SenseCam as a pictorial diary to improve autobiographical memory in a patient with limbic encephalitis A preliminary report Neuropsychological Rehabilitation vol 17 pp 582 601 2007 132 A J Sellen A Fogg M Aitken S Hodges C Rother and K Wood Do life logging technologies support memory for the past an experimental study using sensecam in Proceedings of the SIGCHI Conference on Human Factors in Computing Systems 2007 133 E L Berry A Hampshire J Rowe S Hodges N Kapur P Watson G Browne G Smyth K Wood and A M Owen The neural basis of effective memory therapy in a patient with limbic encephalitis British Medical Journal 2009 134 A K Dey and G D Abowd Towards a Better Understanding of Context and Context
20. Figure 6 1 was used as the platform or central node of the WMMS As shown in Chapter 5 BlackBerry smartphones are appropriate for a WMMS and the BlackBerry 9000 met the design criteria as outlined in Section 4 1 BlackBerry is a commercially available technology reliable and user friendly The device is also small and lightweight and does not interfere with movement when worn on the waist Potential issues with power capacity and memory could be resolved by upgrading to a Figure 6 1 Front side and back view of larger size battery and memory card BlackBerry Bold 181 Other important features of BlackBerry smartphones are the built in industry leading security features that come with the use of the Blackberry Enterprise Solution Additionally newer BlackBerry smartphone models provide access to accelerometer raw data that could enable the design of an all in one WMMS A mature Java environment and many secure APT s are also available with the BlackBerry devices 6 1 1 BlackBerry Bold Specifications and Features The following summarizes the BlackBerry 9000 specifications and features 187 e Built in GPS e 2 0 Mega Pixel Camera with flash and 3x digital zoom e Video Recording Development of a Wearable Mobility Monitoring System 68 6 2 Hardware Design and Evaluation Web browser Corporate Data Access Phone SMS MMS Wi Fi support 802 11a b g enabled Bluetooth v2 0 Serial Port Profile supported Devic
21. Termination Standing Development of a Wearable Mobility Monitoring System 106 17 18 19 20 2 22 23 24 25 26 27 28 29 30 31 32 Technical and Mobility Evaluation of the Prototype WMMS Standing waiting for elevator a Initiation Standing b Termination Start of forward walking progression to get inside the elevator Get in the elevator a Initiation Start of forward walking progression to get inside the elevator b Termination Standing inside the elevator Take the elevator to the first floor a Initiation Standing inside the elevator b Termination Start of forward walking progression to get outside the elevator Walk 50 meters towards the stairwell a Initiation Start of forward walking progression get outside the elevator b Termination Start pushing on the door of the stairwell Open door and enter stairwell a Initiation Start pushing on the door of the stairwell b Termination Lead leg contacts a stair Walk up stairs 13 steps a Initiation Lead leg contacts a stair b Termination Trail leg off of last stair Walk on stair intermediate landing level ground for approx 1 5 meter a Initiation Trail leg off of last stair b Termination Lead leg contacts a stair Walk up stairs 13 steps a Initiation Lead leg contacts a stair b Termination Trail leg off of last stair Open door and turn right a Initiation Trail leg off of last stair b Termination Exit stairw
22. The Netherlands Xsens Technologies B V 2008 186 Xsens Technologies B V XM B Technical Documentation Document XMO101P Revision D The Netherlands Xsens Technologies B V 2008 Development of a Wearable Mobility Monitoring System 144 References 187 Research In Motion Limited BlackBerry Bold BlackBerry Online Available http na blackberry com eng devices blackberrybold Accessed 17 Sep 2009 188 Cypress Semiconductor Corporation PSoC Mixed Signal Array Final Data Sheet Datasheet for CY8C27143 CY8C27243 CY8C27443 CY8C27543 and CY8C27643 Document No 38 12012 Rev L San Jose CA Cypress Semiconductor Corporation 2009 189 Free2Move AB Low Power Bluetooth Module with Antenna FZM03GLA Datasheet Rev C Sweden Free2move AB 2006 190 Avago Technologies Miniature Surface Mount Ambient Light Photo Sensor ADPS 9005 Datasheet San Jose CA Avago Technologies 2007 191 Sensirion The sensor Company Temperature and Humidity Sensor Datasheet SH7x Version 4 2 Switzerland Sensirion 2009 192 Wikipedia Low Pass Filter Wikipedia The Free Encyclopedia 2006 Online Available http en wikipedia org wiki Low pass filter Accessed 27 Nov 2009 193 Research In Motion Limited Sun Microsystems and Nokia Corporation BlackBerry JDE API Reference 4 6 1 Release Online Available http docs blackberry com en developers deliverables 6022 package summary html Accessed 30 Oct 2009
23. Wr f la 2 2 13 t t 0 t 0 t 0 yi Since the amplitude and duration of the acceleration signal vary depending on the type of activity between subjects and even for the same subject and activity calculating SMA is a good way to capture both amplitude and duration effects 7 Bourke et al 142 studied fall detections from a triaxial accelerometer worn at the chest The resultant or root sum of square RSS of the accelerometer signal was calculated Equation 2 14 and compared to a threshold to detect falling with 100 success for 240 falls RSS Ja ta a 2 14 Bourke et al 173 also examined vertical velocity for pre impact detection of fall The vertical velocity was calculated from the integration of the vertical acceleration during static and dynamic periods Bourke et al s method was able to detect pre impact of falls before trunk and knee touch the ground with an average lead time of 323ms The next sub category as identified by Preece et al 171 is time domain features which are typically statistic features For example Veltink et al 147 calculated the standard deviation of an accelerometer signal to differentiate between static and dynamic movement To distinguish between different dynamic activities Veltink et al also examined the signal morphology correlations mean standard deviation and cycle time Other statistic features are skewness kurtosis and eccentricity of the accelerometer signal which have
24. been used in WBAN for health care monitoring 84 94 and in context awareness applications 14 95 96 Mobile phone and smartphones e g a mobile phone with advanced functionality 97 have been used to compile information on a person s location and health status 98 as well as wireless platforms to monitor mobility and fall incidents for elderly people 99 Multiple sensors have been integrated in mobile phones allowing monitoring to happen at only one location on the body 87 This makes it easier to use and less obtrusive to the user With the constant increase in processing power allowing for sophisticated real time data processing smartphones are a great choice as a central node of WBSN They also take advantage of the user s acquaintance with the mobile device 98 Other advantages are that smartphones and handheld devices are often already integrated with sensors such as accelerometers camera and global positioning system GPS which makes them attractive for a fully integrated Development of a Wearable Mobility Monitoring System 20 Literature Review wearable mobility monitoring system In addition these devices come with a programming development platform for mobile devices usually based on Java ME Java Platform Micro Edition The portability of Java has made Java ME an attractive platform in mobile medical application 94 98 100 However Java ME may not be as portable as advertised 101 As mentioned by Xiaowe
25. for going to the second floor and first floor respectively When entering the elevator subjects usually stood and faced the door A good image was usually obtained when the door was just starting to close However if an image was taken before the subject was facing the door or if the door was already closed the image was dark and not clear These low quality images could be due to the BlackBerry camera not performing well under low light conditions A flash could have help but the camera flash was not accessible through the Java API version 4 6 For stair ascent the stairs context was identified in seven out of eight cases for one evaluator and all eight cases for the other evaluator On the other hand stairs could not be identified from images taken when walking down stairs Since the camera was pointing forward from the pelvis the WMMS did not provide the downward angle that would be required for viewing the stairs during downstairs walking Using a wide angle camera or a sphere camera could improve context identification by providing a larger view of the current environment Having a short video of a few seconds or being able to take multiple pictures of the same context could potentially help in identifying the context However from our BlackBerry camera test Chapter 6 a picture could only be taken every 1 5 seconds and that is with only Development of a Wearable Mobility Monitoring System 121 Technical and Mobility Evaluation o
26. intensity feature Figure 7 3 However during preliminary testing while driving many false changes of state were recorded due to the light intensity changes To remove those false changes of state the DT algorithm was only applied to the light intensity feature when the state was not detected as riding in a vehicle Light intensity versus Time 1600 Outdoor 1400 1 1200 Indoor High Threshold Light Intensity mV o e eo LowThreshold 4 0 0 pa AAA U H M se AAMA 200 0 1 2 3 4 5 6 7 8 Time minutes Figure 7 9 Ezample of the light intensity feature signal while performing mobility tasks indoors and outdoors 7 4 GPS GPS data have been used in mobility monitoring to complement motion data improve activity recognition and provide contextual data Section 2 3 4 5 Therefore the GPS location coordinates and speed were collected and added to the WMMS output file Development of a Wearable Mobility Monitoring System 93 Development of the Prototype WMMS GPS data were provided by the BlackBerry Bold Both data were extracted every 9 seconds using the Java locationListener interface The speed value was based on the Doppler Effect as explain in Section 2 3 4 5 For this WMMS prototype only the speed was considered for the change of state detection algorithm The speed feature was added for its potentia
27. piezoelectric accelerometers should not be used to calculate tilt or inclination angle since the gravitational force cannot be measured However many human motion applications use piezoresistive accelerometers or variable capacitance accelerometers 81 These two types are capable of detecting both static and dynamic motion Another advantage of having a DC response is that the accelerometer can be calibrated with the body segment by rotating the segment around the gravitational axis However the DC response adds an offset in the output signal that should be corrected to avoid over or under estimate of the measured acceleration 80 Variable capacitance accelerometers are typically made of a differential capacitor with their two central plates attached to the moving mass and external fixed plates Acceleration applied to the mass modifies the distance between the capacitor s plates resulting in an output voltage change The accelerometer output voltage is proportional to the applied acceleration When using accelerometers to assess movement their main limitation is that they give no indication of a body segment s initial conditions and they are sensitive to gravity Therefore additional information regarding segment orientation is needed to measure acceleration accurately 112 Other limitations include the relative movement of the accelerometer against the body and signal drift over time 80 112 Calibration methods should be consider
28. 1 Sensitivity 95 walking stick specificity gt 95 CDT 82 ADT 86 NN 82 1 wrist 1 chest Six methods for walking periods Remote sensor for home care sit stand lie walk Stroke patients Motor tasks at home assessment of mobility assistive devices cane accelerometers gyroscopes Lie row cycling sit stand run Nordic walk walk includes heat rate ECG SaO2 skin temperature skin resistance light intensity compass audio GPS and altitude sensors Video analysis thresholds applied not to short time Fourier transform StFTz StFT 77 discrete wavelet transform DWT DWT continuous wavelet transform CWT CWT less coefficients Means thresholds SMS message on GSM network Dominant frequencies energy aspects cross correlations auto covariance s Neural Network NN threshold wireless transmission Mean variance median skewness kurtosis percentiles spectral centroid spread peak frequencies power power in frequency bands custom decision trees CDT automatically generated decision tree ADT and neural network NN 2006 Karantonis et al 9 Overall 90 896 posture 94 1 walk 83 3 possible falls 95 1 waist Ambulatory monitoring activity 12 tasks rest posture walking falls estimation of metabolic energy FFT normalised signal magnitude area SMA signal magnitude vector SMV threshold MOITADY INICIO WT Literature Review 2
29. 10 0 0 096 Walking on level Stand to lie 1 1 1 1 1 1 1 1 1 1 15 15 100 0 transition q xipueddy uiojs S Surio1uoJA NTIQON 9 qe1eoAA Jo juouido oAo q 091 15 13 Lying 86 796 Lie to Stand 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 15 100 0 transition Waring omevel 0 0 0 0 0 1 1 1 1 0 0 0 1 1 1 15 7 46 7 ground j NOP NOP NOP NOP NOP NOP NOP NOP NOP Walking on ramp 0 0 0 IC IC IC IC 0 IC 0 0 IC IC IC IC 6 0 0 0 Walking on level NOP NOP NOP NOP NOP NOP NOP NOP 2 ground 1 9 IC ic tc it tc AA o fie ET NES e 7 42 973 Transition indoor outdoor NIE du e EE MEN drea E a Rao E S 71 4 and keep walking on level ground Transition outdoor indoor NA MA pu o o o 0 Nee prd 1 1 1 bd pd i 7 3 and keep walking 42 9 on level ground Transition NOP NOPI NOP NOP NOP NOP NOP NOP NOP NOP NOP NOP indoor outdoor IC C IC 1 IC IC 1 IC IC IC IC 1 IC IC IC 3 3 100 0 and keep walking on level ground Stand to sit transition to getin 0 1 0 1 1 1 1 1 1 0 1 1 1 1 1 15 12 80 0 the car q xipueddy urojs amp s SuuojmuoJA AITIQOJ AQLI AA JO juoeuido oAo q I9I Sitting in the car 0 1 0 1 1 1 1 1 1 0 1 1 1 1 1 15 12 80 0 NOP NOP NOP NOP NOP Starts of car ride IC 1 IC IC 1 1 1 1 1 IC IC 1 1 1 1 10 10 100 0 A NOP NOP NOP NOP NOP S
30. 2 12 0 70 would have be measured Very good algorithm performance was obtained for detecting changes of state produced by start stop motions sensitivity of 97 4 5 3 Furthermore as opposed to walking up stairs walking down stairs was detected at 100 0 However the stair intermediate landing was not always detected causing the lower stair section to have a sensitivity of 66 7 However the lower section would still be considered as stair descent since the state would not have changed from the upper stair section A sensitivity of 97 8 4 7 was found for the change of state caused by postural change i e stand to sit sitting sit to stand lying etc The start and stop of the car ride was detected at 66 7 and 64 3 respectively These results depended on the BlackBerry detecting the GPS satellites Development of a Wearable Mobility Monitoring System 115 Technical and Mobility Evaluation of the Prototype WMMS Table 8 3 Summary performance results for the each subject Subject Sensitivity Average Specificity Yo Average Standard deviation Standard deviation 1 75 4 4 0 96 4 0 4 2 79 7 2 9 93 3 0 7 3 75 4 1 5 96 7 1 1 4 80 9 8 6 96 1 1 6 5 77 2 1 5 99 5 0 5 Overall 77 7 X 2 5 96 4 1 2 2 Table 8 4 Performance results for each of the mobility tasks Change of State True Positive False Negative Walking on level ground 15 0 Stand to sit transition 14 1 Sit to
31. 50 Hz and a window size of 1 02 seconds It can be observed that the elapsed time is approximately 1 second when no picture is taken and an extra second is added after a picture is taken Another observation is that the second window of data after a picture is taken is smaller but by the third window the timing is back to normal Therefore it was decided to wait at least 2 windows or 2 04 seconds before taking another picture i e 3 seconds later Table 7 2 Section of a WMMS output file to demonstrate timing of the picture taken Time Frame Elapsed Time Image Name State of the s from previous or 0 if no User window s image taken 0 0 0 100000 0 978 0 978 0 100000 2 057 1 079 0 100000 3 053 0 996 IMAGE9 10100000 Picture taken 5 09 2 037 0 10100000 f 5 269 0 179 0 10100000 6 078 0 809 0 10100000 Ready to take 7 055 0 977 0 10100000 picture again 8 093 1 038 0 10100000 9 111 1 018 0 10100000 10 129 1 018 0 10100000 11 147 1 018 IMAGE10 10100001 Picture taken 12 964 1 817 0 100010 13 183 0 219 i 0 10 14 18 0 997 IMAGE11 0 Ready to take 15 987 1 807 0 0 picture again and 16 207 0 22 0 0 picture taken 17 185 0 978 0 0 Ready to take 18 232 1 047 0 0 picture again 19 23 0 998 0 0 20 248 1 018 0 0 21 207 0 959 0 0 22 204 0 997 0 0 23 262 1 058 0 0 Development of a Wearable Mobility Moni
32. 51 samples were used for the window size Therefore the window size is 1 02 seconds instead of 1 second During preliminary testing peaks occurred during transition when the person sat down rose from a chair or lay down Figure 7 7 For this reason SMA was added to the algorithm to determine the current state Another reason to add SMA is to help identify activity intensity changes which could mean a change of state Therefore three thresholds were used and three states were determined no peak with normal intensity no peak with increased in intensity or a peak The low threshold value was 0 100g and the high threshold value was 0 190g The threshold for the peak was set to 0 320g A DT algorithm was used to determine increases in intensity and peak detection illustrates the DT algorithm flowchart applied to the SMA feature When a peak was detected the next data window was not classified as a peak again until the signal went below the low threshold This avoided inappropriately switching from state peak to state no peak with increased in intensity and then to no peak with normal intensity since each Development of a Wearable Mobility Monitoring System 90 Development of the Prototype WMMS windows is independently analysed However if the transition was slow and a change happens across windows it was possible to detect the state no peak with increase in intensity just before detecting the state peak Th
33. F Coster N H Lovell and B G Celler Accelerometry Providing an integrated practical method for long term ambulatory monitoring of human movement Physiological Measurement vol 25 2004 82 J A Levine Measurement of energy expenditure Public Health Nutrition vol 8 pp 1123 2005 83 P Bonato Advances in wearable technology and applications in physical medicine and rehabilitation Journal of NeuroEngineering and Rehabilitation vol 2 2005 Development of a Wearable Mobility Monitoring System 135 References 84 E Jovanov A Milenkovic C Otto and P C De Groen A wireless body area network of intelligent motion sensors for computer assisted physical rehabilitation Journal of NeuroEngineering and Rehabilitation vol 2 2005 85 Y Hao and R Foster Wireless body sensor networks for health monitoring applications Physiological Measurement vol 29 pp R27 R56 2008 86 E Stuart M Moh and T S Moh Privacy and security in biomedical applications of wireless sensor networks in st International Symposium on Applied Sciences in Biomedical and Communication Technologies 2008 87 J Lester T Choudhury and G Borriello A practical approach to recognizing physical activities in 4th International Conference on Pervasive Computing 2006 pp 1 16 88 H Chen W Wu and J Lee A WBAN based real time electroencephalogram monitoring system Design and implementation Journal of M
34. Ninomiya and W M Caldwell A wearable posture behavior and activity recording system in Proceedings of the 22th Annual International Conference of the IEEE Engineering in Medicine and Biology 2000 pp 1278 154 B Najafi K Aminian F Loew Y Blanc and P Robert An ambulatory system for physical activity monitoring in elderly in Proceedings of the 1st Annual International Conference on Microtechnologies in Medicine and Biology 2000 155 B Najafi K Aminian A Paraschiv Ionescu F Loew C J B la and P Robert Ambulatory system for human motion analysis using a kinematic sensor Monitoring of daily physical activity in the elderly IEEE Transactions on Biomedical Engineering vol 50 pp 711 723 2003 156 L Bao and S S Intille Activity recognition from user annotated acceleration data Lecture Notes in Computer Science Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics vol 3001 pp 1 17 2004 157 H J Luinge and P H Veltink Inclination Measurement of Human Movement Using a 3 D Accelerometer with Autocalibration JEEE Transactions on Neural Systems and Rehabilitation Engineering vol 12 pp 112 121 2004 158 P Barralon N Noury and N Vuillerme Classification of daily physical activities from a single kinematic sensor in Proceedings of the 27th Annual International Conference of the IEEE Engineering in Medicine and Biology 2005 pp 2447 24
35. Stop of car ride IC 1 IC IC NA 1 1 8 88 996 Sit to stand 5 transition 0 3 1 1 1 1 1 12 92 3 Walking outside NOP 4 1 1 1 1 1 1 1 1 1 1 1 0 1 14 13 92 9 on level ground IC Transition outdoor indoor NOP NOPI NOP NOP NOP 0 NOP NOP 0 1 NOP NOP 1 25 0 and keep walking IC C IC IC IC IC IC IC IC a on level ground Standing 1 1 1 1 1 1 1 1 1 1 NA 1 1 1 1 14 14 100 0 Total Number of Pictures 27 30 29 29 30 31 29 29 28 26 30 34 29 30 29 440 Total Number of SUA 16 25 23 23 23 22 23 23 23 15 22 28 22 24 21 333 Total Yo of 59 3 79 3 79 3 76 7 71 0 79 3 79 3 82 1 57 7 73 3 82 4 75 9 80 0 724 Successfully 83 3 96 96 96 75 7 q xipueddy urojs amp g Suuojuo A Ki Iqo A 9 q91e9AA JO juouido oAo T 8ST Table D 2 Picture evaluation results from evaluator 2 Change of State Subject 1 Subject 2 Subject 3 Subject 4 Subject 5 d Hg 96 of Success in identifying uu UA EEE za S 2 context Walking on level DE ANITA AA Bie 0 1 MERECE ECTETUR AE EEUU Eo 86 7 sana doe 1 1 is esie Pete facien Jaga A ose T Fa TM eee si gie oue faa a 100 0 transition IC Sitting 1 1 EAE NEWE NESNEBDERMESEZEREEBXEE 100 0 Sit to stand 1 1 ae ae a Th ite ll ae WS ae ae sae ea I arg ae lhe Bie lill ais 100 0 Walking on level 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 15 14 93 3 ground for elevator Walking to get in 0 the elevator
36. The nodes of the tree are where questions are asked and the nodes are connected to other nodes through links branches Mathie et al 175 developed a generic framework Figure 2 12 using a binary tree structure to classify movements from a single triaxial waist accelerometer The advantage of Mathie et al s framework was its flexibility to allow nodes to be added removed and reordered without affecting the rest of the tree When applied to the classification of specific movements e g upright lying sit to stand stand to sit transitions walking and fall performed in a controlled laboratory setting this generic classification framework demonstrated an average Development of a Wearable Mobility Monitoring System 46 Literature Review classification performance of 97 7 for sensitivity and 98 7 for specificity 175 This classification was well suited for real time applications because it did not require a large amount of computational power This was shown by Karantonis et al 9 who implemented a simpler version of Mathie s algorithm to create a real time human movement classification system A similar approach to hierarchical classification is the decision tree The difference is that decision trees are automatically generated Automatic generation of decision trees can be done using popular algorithms such as CART classification and regression tree ID3 iterative dichotomiser 3 and C4 5 77 These techniques require train
37. a person was riding in a vehicle The user s state was determined for every data window of 1 02 seconds When a change of state was detected a picture was taken However due to the limitations of the BlackBerry camera Chapter 6 the WMMS had to wait at least 2 04 seconds before being able to take another a picture Therefore the current state was compared with the three previous states to determine if a change of state happened The prototype WMMS software application was developed using the Java Development Environment and API version 4 6 1 All WMMS output data were saved to the BlackBerry SD card Development of a Wearable Mobility Monitoring System 102 Technical and Mobility Evaluation of the Prototype WMMS Chapter8 Technical and Mobility Evaluation of the Prototype WMMS The WMMS evaluation was divided into two main parts the technical evaluation and the mobility evaluation The technical evaluation examined the BlackBerry battery and the data loss The purpose of the mobility evaluation was to evaluate the performance of the WMMS for detecting changes of state The mobility evaluation was also to evaluate the pictures taken by the WMMS for their usefulness in determining context associated with the mobility tasks The following describes the method for the WMMS evaluations 8 1 Technical Evaluation The battery life of the BlackBerry Bold while running the full application GPS data processing camera was evaluated usin
38. activities take place and to analyze mobility in the community has not yet been explored 2 4 Data Analysis Algorithms As previously mentioned accelerometers are the most used wearable sensor to detect activity and to measure mobility Many researchers have already explored algorithms and data analysis techniques to extract useful information from the raw acceleration data and to Development of a Wearable Mobility Monitoring System 31 Literature Review classify activities A review from Godfrey et al 77 highlighted laboratory and clinical studies using accelerometers Table 2 2 As mentioned by Mathie et al 81 the output of an accelerometer when worn on the body will vary depending on four factors 1 Position at which it is placed 2 Its orientation relative to the subject 3 The posture of the subject 4 The activity being performed by the subject The following sections review concepts and techniques applied to accelerometers to detect human body activity These sections focus on sensor placement and specifications data calibration filtering windowing feature extractions and classification algorithms 2 4 1 Accelerometer Placement An accelerometer s location and its orientation relative to the body will affect the way its output signal will vary Deciding on the accelerometer placement on the body is important in human motion measurement Normally the sensor is attached to the body part whose movement is be
39. activities using a tri axial Development of a Wearable Mobility Monitoring System 48 Literature Review accelerometer with a performance of 93 accuracy In a context awareness system Jin et al 13 used fuzzy logic to detect user motion states such as lying sitting walking and running with a recognition rate of 98 9 98 9 99 7 and 99 9 respectively Emergency situations such as falling while walking and falling while running were also recognized at a rate of 100 Markov chain is a random process where future states depend on the present state and is independent of the past states 181 The Hidden Markov model HMM is similar to Markov chain but the present state is unknown Once trained a classification algorithm using HMM can identify a sequence of activities from a sequence of measured features and the likelihood of a transition from previous activity 171 He et al 182 used the HMM for real time activity classification using data from three two axis accelerometers Data was collected from five subjects performing 11 different activity series stable states such as standing sitting lying and transition states such as standing to sitting sitting to lying sitting to standing lying to sitting and falling The activity detection accuracy was 95 8296 HMM can also be combined with other classifiers For example Lester et al 87 used HMM as a second classifier to differentiate a range of daily activities The outputs
40. and Rehabilitation vol 85 2004 33 K O Berg S L Wood Dauphinee J I Williams and B Maki Measuring balance in the elderly Validation of an instrument Canadian Journal of Public Health vol 83 pp S7 S11 1992 34 L Blum and N Korner Bitensky Usefulness of the Berg Balance Scale in stroke rehabilitation A systematic review Physical Therapy vol 88 pp 559 566 2008 Development of a Wearable Mobility Monitoring System 131 References 35 S Mathias U S L Nayak and B Isaacs Balance in elderly patients The get up and go test Archives of Physical Medicine and Rehabilitation vol 67 pp 387 389 1986 36 D Podsiadlo and S Richardson The timed Up and Go A test of basic functional mobility for frail elderly persons Journal of the American Geriatrics Society vol 39 pp 142 148 1991 37 R O Crapo R Casaburi A L Coates P L Enright N R MacIntyre R T McKay D Johnson J S Wanger R J Zeballos V Bittner and C Mottram ATS statement Guidelines for the six minute walk test American Journal of Respiratory and Critical Care Medicine vol 166 pp 111 117 2002 38 K Donovan S E Lord H K McNaughton and M Weatherall Mobility beyond the clinic The effect of environment on gait and its measurement in community ambulant stroke survivors Clinical Rehabilitation vol 22 pp 556 563 2008 39 M E Tinetti Performance orientated assessment of mobility p
41. and contextual factors An illustration of the ICF model is presented in Figure 2 1 demonstrating the interaction between the different components ICF environmental factors comprise the physical social and attitudinal environment in which people live and conduct their lives 18 Understanding the impact that the physical environment can have on community mobility is important because some environments may have barriers that could decrease a person s mobility or may also have facilitators that could increase mobility Development of a Wearable Mobility Monitoring System 6 Literature Review Health condition disorder or disease Body function 4 Activity Participation amp structure Environmental Personal factors factors Figure 2 1 Interaction between ICF components reproduced from 18 2 1 2 Dimensions of Mobility Framework The Dimensions of Mobility framework was developed by Patla and Shumway Cook 1 to define community ambulation with respect to the physical environment s impact on a person s mobility Figure 2 2 This framework consists of eight environmental factors called dimensions which determine the degree of complexity and difficulty of mobility The dimensions are minimum walking distance time constraints on locomotion ambient conditions terrain conditions physical load interaction attention demands postural transitions and density of traffic both vehicular and o
42. caused problems with the light sensor since the view could potentially be blocked by the user s clothing As seen in the images evaluation results pictures could be used to detect indoors outdoors GPS speed was used to detect if a person was in a vehicle For the trials where the GPS satellites were detected the change caused by being in a vehicle was well detected While the initiation of being in a vehicle can be identified using the camera images WMMS classification was delayed by the 9 second sampling interval for GPS speed and the 7 m s threshold The main problem with the BlackBerry GPS during evaluation was the time required to detect satellites and initiate GPS data acquisition Based on preliminary tests the BlackBerry Bold 9000 could take 30 minutes to detect GPS satellites depending on the exterior conditions The BlackBerry was set to autonomous mode to detect location which is slower but more precise than using cell site mapping For our WMMS GPS speed was required for the detection of vehicle riding Cell tower based location could be investigated since location estimation occurs faster and would work indoors and in cloudy weather although this method is of lower precision Development of a Wearable Mobility Monitoring System 125 Technical and Mobility Evaluation of the Prototype WMMS 8 3 3 Limitations Some limitations of the study were that the BlackBerry Bold 9000 did not have an internal accelerometer Since a smartph
43. determined that there was no need for re calibrating the accelerometer during trials These results were also expected since the external board used a low drift accelerometer that has a trimming circuit to reset the device trimming value during power up Therefore calibration of the accelerometer was performed once prior to the evaluation Development of a Wearable Mobility Monitoring System 76 Hardware Design and Evaluation Testing for Drift Mean DC Acceleration of X axis versus Time 0 075 0 07 4 cy 5 0 065 lt SA AA DE c E z 0 06 0 055 T T T T T T T 0 00 0 25 0 50 0 75 1 00 1 25 1 50 1 75 2 00 Time Hours Testing for Drift Mean DC Acceleration of Y Axis versus Time 0 005 0 01 4 o0 E 0 015 lt a g 002 0 025 gt t t 0 00 0 25 0 50 0 75 1 00 1 25 1 50 1 75 2 00 Time Hours Testing for Drift Mean DC Acceleration of Z axis versus Time 1 005 1 01 2 S E x 1 015 S e E 1 02 1 025 0 00 0 25 0 50 0 75 1 00 1 25 1 50 1 75 2 00 Time hours Figure 6 4 Examples of the drift acceleration versus time for x y and z axis Development of a Wearable Mobility Monitoring System TI Hardware Design and Evaluation 6 2 9 Data Filtering The external board was designed such that each of the accelerometer output signals were passed through an analog low pass filter with a cut off frequency of approxima
44. easi a alto M Ia 16 Figure 2 7 Example of a Wireless Body Area Network of intelligent sensors for patient monitoring reproduced from 84 ccc ceesseceesseceeseceeneecesaaeceeaeeeceeaeeceeeeecseeeeeeeeecseeeeeseeeees 20 Figure 2 8 MaSS Spriing systemi uoo Jo ecru Decet eda cin osos i eios el eed iu dive Pn tes ot eins 24 Pigure 2 9 SenseCam images 129 scho ensem E e ERR Vei de wan awa ee 29 Figure 2 10 Seismic uniaxial accelerometer measuring the component a i of an equivalent acceleration a in the direction i of the sensitive axis of the accelerometer The equivalent acceleration is the sum of the acceleration a of the sensor and the equivalent gravitational acceleration g acting on the seismic mass 9 is the angle between the sensitive axis of the accelerometer and the acceleration a 9 is the angle between the sensitive axis and the gravitational field reproduced from 147 w wwmmmammmammmawemane anamwamini 43 Figure 2 11 Dual or tri axis accelerometer with two axes for measuring tilt reproduced PROTA ID MEER 43 Figure 2 12 Generic classification framework presented by Mathie et al 175 48 Figure 4 1 System Architecture ofa WMMS 4st tuer pe teteMiadve 55 Figure 4 2 Front and side view images of the WMMG sees 56 Figure 4 3 WMMS signal processing and algorithm outline for each data window 58 Development of a Wearable Mobility Monitoring System vi
45. encounters for each of the eight dimensions was measured using a self administrated questionnaire to collect information on activities and trips Subjects were video taped during three trips in the community to record the physical environment associated with community mobility Older adults with mobility issues were characterized by a decrease in the number of trips taken in the community and the number of activities performed during these trips The dimensions that distinguished between an older adult with mobility disability and an older adult without such disabilities were temporal factors physical load terrain and Development of a Wearable Mobility Monitoring System 8 Literature Review postural transition The dimensions that did not distinguish between groups were distance traffic density ambient conditions and attentional demands 2 2 Mobility Measurement The following summarizes existing methods used to measure mobility including functional mobility community ambulation physical activity and human motion analysis The categories presented are observation and clinical tests diaries and questionnaires physiological measurements and biomechanical measurements 2 2 1 Observation and Clinical Tests Observation and clinical tests are performance based measures used to assess an individual s functional mobility These tests are usually easy to perform and are carried out in a clinical environment over a short period Howeve
46. external board and the BlackBerry were pre processed before extracting features from the signals The features were then used as input to an algorithm that determined the state and took a picture if there was a change of state All features extracted for every second of data time stamp and image name were saved to an output file The digital images were stored on an SD card External board BlackBerry Bold raw data data Data Pre Processing E Determination of State Change of State Features generated from BlackBerry and external board data T kea Picture gt the current state and the image name are copied to Output File stored on i BlackBerry SDCard Picture saved on BlackBerry SDCard Yes Figure 4 3 WMMS signal processing and algorithm outline for each data window Development of a Wearable Mobility Monitoring System 58 Methodology 4 4 System Evaluation Outline One of the first steps in developing the WMMS was to select a hub or platform that met our design requirements Therefore a preliminary evaluation was performed to evaluate the BlackBerry smartphone as a hub of a mobility monitoring system Chapter 5 presents the details about the preliminary BlackBerry evaluation The next step presented in Chapter 6 was to design and evaluate hardware for the WMMS Then everything was put together to create the WMMS and the software was developed to capture process and l
47. heavy tree canopy and in dense urban areas 8 GPS accuracy may vary based on atmospheric conditions as well as from signal deflection or obstruction GPS was also found to be unable to detect static activity 127 2 3 4 6 Camera Many cell phones and smartphones include a digital camera Applications that have used cameras in a wearable system are mostly for life log or diary purposes A wearable system to capture audio and visual information corresponding to user experiences was presented in 128 Yamazoe et al s system is worn below chest level and consists of a head detection camera a wide angle camera a microphone and possibly GPS A method to extract meaningful context from life logs and a smartphone was proposed by Lee and Cho 12 The life logs included GPS SMS call charging MP3 photos taken images viewed and weather information SenseCam from Microsoft Research Microsoft Corporation 129 is an example of a wearable digital camera that takes pictures without the user intervention Figure 2 9 The camera contains different sensors such as light intensity and light color sensors a passive Development of a Wearable Mobility Monitoring System 28 Literature Review infrared detector temperature sensor and accelerometers Pictures are taken based on significant changes measured by the sensors and or at specific time interval Microsoft Research has also explored the use of audio level detection audio recording
48. indoor from outdoor Section 2 3 4 7 e Temperature and humidity sensor to give weather information Section 2 3 4 7 e Board shaped in such a way to be fixed on the BlackBerry s holster and without obstructing the camera view of the BlackBerry 6 2 2 Parts Specifications A general system design of the board is presented in Figure 6 2 The complete electrical schematic is shown in Appendix A An image of the board is presented in Figure 6 3 indicating the location of the sensors and other main components The board consists of a microcontroller CY8C27443 Cypress Semiconductor Corporation San Jose CA USA a Bluetooth Module F2M03GLA Free2Move AB Halmstad Sweden a triaxial accelerometer LIS344alh STMicroelectronics Geneva Switzerland a light sensor APDS 9005 Avago Technologies Limited San Jose CA USA and a humidity and temperature sensor SHT71 Sensirion AG Staefa Switzerland The board is powered up with a lithium battery and has a USB rechargeable circuitry This external board could run continuously for approximately 14 hours on one charge Specifications for the main components are presented in Table 6 1 Power and Rechargeable Circuit Accelerometer Lis3444LH Bluetooth Module F2MO3GLA Temperature and Microcontroller Humidity Sensor SHT71 C Y8C27443 24SXl Light Sensor ADPS 9005 Debug Port Figure 6 2 Block diagram of the external board Development of a Wearable Mobility Monitoring System 70
49. measured during the rotation between 1g The output a of one accelerometer can then be expressed as qa 079 2 6 S where u is the un calibrated acceleration However this calibration method requires input from the user and should be performed in a controlled environment Therefore auto calibration procedures have been developed where a specific angular rotation is not required These auto calibration methods are based on the fact that the modulus of the acceleration signal during quasi static movement is equal to g 9 81 m s For a triaxial accelerometer this concept can be expressed as Je a a 1g 2 7 By replacing the three accelerations a ay and a with Equation 2 6 Equation 2 7 can be rewritten as Development of a Wearable Mobility Monitoring System 38 Literature Review 2 8 This concept was used by Lotters et al 167 to create a method for calibrating the sensitivity and the offset of a triaxial accelerometer while in use The method calculated six elements 5 5 5 0 0 0 after detecting quasi static state and only required random movements to be performed Another example is an on the field auto calibration procedure created by Frosio et al 168 Frosio et al s calibration model incorporated the bias offset 0 0 T 0 and scale factor sensitivity for each axis s and the cross axis symmetrical scale factors xx S55 S gt SxS The cross axis scale factors des
50. of a static binary classifier were used as inputs to the HMM classifier Adding that second HMM layer Lester et al improved their classification accuracy by approximately 10 1596 2 4 8 Summary of Data Analysis Accelerometers have been used in many studies to measure mobility identify postures and posture transitions detect falls classify activity and so on Accelerometer specifications for human motion studies may depend on where the sensor is placed on the body and type of activity to be identified Some studies placed sensors at multiple locations but some also proved that it was possible to detect activity with a single accelerometer placed around the center of mass area For an accelerometer placed at the waist for daily activity assessment Bouten et al 80 concluded that an accelerometer should be able to measure acceleration with amplitude ranging from 6 to 6 g and frequency up to 20 Hz Development of a Wearable Mobility Monitoring System 49 Literature Review Methods for calibrating accelerometers vary from simple DC offset removal to more complex automatic signal calibration to correct for drift The DC offset can be removed with a low cut off frequency filter Filtering techniques were also used to remove spikes noise and undesirable frequencies from the raw signals The raw filtered and calibrated acceleration signals are usually divided into small windows from which features can be extracted The different categor
51. subjectively chosen based on common lighting conditions under which the WMMS will operate Having different light condition associated with a real world light intensity value helped determine classification threshold values for indoor and outdoor conditions The board was worn on the right hip of one subject during testing The subject was asked to stay in the same light condition within a circle of approximately 1 5 2 meters of diameter for the whole measurement period but to move and turn around within that circle Five trials for each light condition were completed at different times days and locations Each trial was Development of a Wearable Mobility Monitoring System 74 Hardware Design and Evaluation for one minute The light sensor values were averaged for each light condition From these results thresholds for indoor and outdoor conditions were set to 1000 and 300 respectively More details on the algorithm using these thresholds are provided in Chapter 7 Table 6 2 Average output value of the light sensor mV for different light conditions standard deviation in brackets Light condition Average light sensor Vout mV Outdoor sunny day 1474 0 16 3 Outdoor sunny day in the shade 1214 6 334 4 Outdoor cloudy day 1185 9 451 6 Indoor away from window 74 5 83 9 Indoor cloudy in front of window 252 7 236 1 Indoor sunny day in front of window 531 5 387 5 Outdoor during the night 19 3 7 3 Indoor during t
52. to control the external board were for setting the board s sampling delay and to turn off sampling Command setting sampling delay e Packet 0xC3 0x42 0x02 0x01 Ox delay e delay is the delay between samples in milliseconds For example if the delay byte is set to Ox14 which means 20 milliseconds then the sampling frequency is 50 Hz Development of a Wearable Mobility Monitoring System 73 Hardware Design and Evaluation Command Turning off sampling e Packet OxC3 0x42 0x02 0x01 0x00 e This will turn off sampling 6 2 6 Temperature and Humidity Sensors The temperature and humidity raw data coming from the board was converted using Equation 6 1 and 6 2 to get the temperature in Celsius and the humidity in percent of Relative Humidity 191 The two bytes received from the board rxTemp were processed using Equation 6 1 to give temperature T in Celsius T rxTemp x0 01 39 C 6 1 For humidity the two bytes received rxHum were processed using Equation 6 2 to provide humidity H in of Relative Humidity RH H rxHum x 1 5955e 5 rxHumx0 0367 2 0468 RH 6 2 6 2 7 Light sensor The board provided 3 3 volts and a load resistance of 2 kohms to the light sensor Since the manufacturer did not provide calibration curves for VCC 3 3V a calibration table of different light conditions versus voltage output of the light sensor was created Table 6 2 These different lighting conditions were
53. to obtain valid data e Captures motion data location data and ambient environmental data e Wearable small lightweight does not interfere with range of motion Integrated in one package so that the device is only worn at one location on the body e Power efficiency system lasts one day on one charge e Memory capacity should be at least one day e User friendly for consumer and health care provider e Uses commercially available technology e Follows wireless transmission standard protocols e Inexpensive e Reliable e Safe e Detect a change of state within a 5 tolerance for sensitivity and specificity Development of a Wearable Mobility Monitoring System 54 4 2 Methodology 4 1 2 Software Design Criteria Perform real time processing of incoming data Identify change of state Obtain contextual information automatically when there is a change of state Save processed data and pictures to a file Data security on device and during transmission Application easily upgradeable for future use System Architecture The proposed WMMS system architecture is illustrated in Figure 4 1 A smart phone was used as the platform for the WMMS to perform functions such as capturing processing storing and transmitting motion data and contextual information The system could send community mobility data or emergency events e g fall to a hospital external server Data received at the external server could be further analyzed and feedback
54. used a 20 Hz low pass filter to attenuate frequencies not expected to be caused by body movement Another common filtering technique such as used by Mathie et al 7 is applying a median low pass filter to the signal to remove noise spikes Digital filtering techniques can be used to separate gravitational acceleration from the body movement acceleration Since human movement will never correspond to a DC response it is important to remove the DC offset from the accelerometer output otherwise the measured acceleration could over or underestimate the body movement acceleration 80 Since most daily activity movements appear between 0 3 to 3 5 Hz 165 filters use a cut off frequency between 0 1 to 0 5 Hz 81 The DC component of the acceleration signal can also be represented by the mean of the acceleration over a certain window 156 2 4 5 Data Window In an activity classification system acceleration signals are usually divided into smaller time segments or windows prior to feature extraction The feature set generated from each window can then be used as input to a classification algorithm Preece et al 171 found three windowing techniques that have been used for activity identification sliding windows event defined windows and activity defined windows The sliding windows technique divides the signal into small windows of the same length with no gap in between with the option to overlap windows The sliding window technique is on
55. used to estimate temporal parameters of a gait cycle More examples of event defined window studies have been presented in Preece et al 171 The activity defined window 171 technique detects the time when activity changes and from these times data windows are identified Every window corresponds to a specific activity For example Sekine et al 139 used wavelet analysis to detect the time when changes in walking pattern occurred These times were then used to classify walking pattern such as walking on level ground or ascending and descending stairs 2 4 6 Feature Extraction Many different features can be extracted from an accelerometer signal and then used as inputs to classification algorithms Preece et al 171 presented different feature generation techniques applied to body worn sensor data in the field of activity classification including heuristic features time domain features frequency domain features and time frequency domain The following presents feature extraction techniques that have been applied to accelerometer data The term heuristics features is referred by Preece et al 171 as the features which have been derived from a fundamental and often intuitive understanding of how a specific movement or posture will produce a characteristic body worn sensor signal The first example is extracting the inclination angle from the DC or static component of an accelerometer signal The inclination angle represen
56. values were averaged Table 6 3 shows the results for camera performance evaluation The time the function take a picture was executed and the time the camera was ready again to take a picture was almost 2 seconds This is slow for application where real time processing was one of the criteria These time results will need to be taken into account Development of a Wearable Mobility Monitoring System 78 Hardware Design and Evaluation during WMMS design If the user s state changes between 1 02 seconds windows consecutive pictures cannot be taken Table 6 3 BlackBerry Bold camera performance evaluation results Shutter Lag s Time before camera is ready s Standard Standard Trial 1 0 65 0 07 Trial 2 0 63 0 12 Trial 3 0 70 0 08 Trial 4 0 66 0 08 0 01 Trial 5 0 61 0 03 ESI TOTAL AVERAGE 0 65 0 07 0 86 0 01 6 4 Summary The BlackBerry Bold was chosen as the platform for the WMMS Since access to raw accelerometer data was not available with the 9000 model an external board was added to the design The external board designed to fit on the BlackBerry holster provides motion data and context data such as light intensity humidity and temperature The BlackBerry provides GPS current time and camera functions The light sensor was calibrated with different lighting conditions present in everyday life A threshold value of 1000 was a good estimate for detecting outdoors The low threshold to reset back to the in
57. video and or still image analysis could greatly enhance accuracy and reliability over systems that only rely on inertial sensors 3 1 Application of a Wearable Mobility Monitoring System WMMS A wearable system that can validly monitor mobility in the community and capture the context associated with mobility could benefit people with physical disability by helping the rehabilitation medicine field For instance such a system could help evaluate the progress made during and after rehabilitation help identify mobility issues outside a hospital environment and enhance clinical decision making about the rehabilitation program i e assistive devices exercises etc Measurement of activities avoidance and categorizing activities are other useful information for physical rehabilitation that could be provided by a WMMS A WMMS could also be used as a research tool to evaluate mobility interventions and assessment methods in the community In addition a WMMS could determine the skills required to overcome challenges found in different community environments e g busy city street farm mall etc These results could help improve training or advocate for changes to the environment Additionally exploration of the camera in smartphones to capture context will provide insight on this approach for mobility monitoring applications Development of a Wearable Mobility Monitoring System 52 Rationale 3 2 Objective of the thesis The
58. walked down stairs Skewness increased when walking up stairs but not as much as when going down stairs The same skewness values as upstairs were sometimes observed during normal walking which could result in a false positive change of state detection The high and low thresholds were chosen to detect down stairs and allow the possibility to detect up stairs with minimal false positive stairs detection The high stairs threshold was set to 0 6 and the low stairs threshold was set to 0 2 Because the WMMS was not rigidly attached on the person s waist WMMS movement may add noise to the signal Various smoothing techniques on the skewness signal were tried which seemed to improved the false positive but the time resolution to detect true positive stairs was reduced Therefore since the goal was to take a picture when there was a change of state the skewness signal was not smoothed more than the 1 02 seconds sliding window already applied More advanced data processing could be performed later on the WMMS output to improve stairs detection A double threshold DT algorithm such as the one used for the standard deviation was also applied to the skewness value However since walking up or down stairs is a dynamic state the DT was only applied to the skewness when the standard deviation of the vertical axis was verified to be in a sufficient dynamic level Waiting to be in dynamic state to identify Development of a Wearable Mobility Monito
59. were also calculated The z axis skewness was used by Baek et al 141 to differentiate between walking going up stairs from running Baek et al also used kurtosis to detect upstairs downstairs from walking running From preliminary work in this thesis these two features were combined with y axis skewness to try and to improve upstairs detection and decrease the number of false positive However combining these two extra features was Development of a Wearable Mobility Monitoring System 94 Development of the Prototype WMMS ineffective Therefore they were not added in the algorithm but were kept in the output file for further data processing For further data processing purposes other features kept in the output file are the mean value of the body acceleration of all three axes the temperature and humidity and the GPS location coordinates 7 6 Determination of State and Change of State The algorithm developed determined the user s state every 1 02 seconds and compared the current state with previous states to determine if a change of state occurred The features extracted from the acceleration signals GPS speed and the light intensity were used to set the bits of an 8 bit number representing the user s state STA DYN STAIRS STAND LIE GPS LIGHT SMA PEAK SMA INT If the state was 160 in decimal value which gives 10100000 the person was moving and in a standing position Table 7 1 describes each bit A f
60. 095 Sit to stand 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 0 100 0095 Walking on level ground 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 0 100 0096 Standing waiting for elevator 1 1 1 1 1 1 1 1 1 NA 1 1 1 1 1 14 0 100 00 Walking to get in the elevator 1 1 1 1 1 1 0 0 1 NA 1 1 1 1 1 12 2 85 7196 Walking to get out of elevator and keep Taking elevator to 2 floor 1 1 1 1 1 1 0 0 1 1 1 1 1 1 1 13 2 86 67 walking on level ground 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 0 100 00 Standing waiting for elevator 1 1 1 1 1 ara 1 1 1 1 lege ile 1 1 15 0 100 00 Walking to get in the elevator 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 0 100 00 Taking elevator to 1 floor 1 1 1 1 1 Bm 1 1 1 1 EB 1 1 15 0 100 00 9 xipueddy urojsKg Sun o JA A IqoJA AQLIM Jo 3jueuido oAe TSI walking on level ground 2 13 13 33 Walking to get out of elevator and keep 1 Walking up stairs 1 15 EZ 100 00 meter of level ground Walking on stair intermediate landing 1 5 0 1 1 0 0 0 0 0 0 0 0 1 1 1 1 6 9 40 00 walking on level ground Transition outdoor indoor and keep walking on level ground e ase Walking up stairs 1 1 1 0 0 0 0 0 0 0 0 1 1 1 0 6 9 40 00 Walking on level ground 1 0 1 0 1 rane 0 1 0 0 EXE 0 0 4 11 26 67 Walking down stairs 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 0 100 00 iate GLISSER T sols 9 0 a g 0 1 es 1 10 3 66 67 Walking down stairs 0 0 0
61. 1 found that although SVM method was a powerful classification method few activity classification studies have used that approach Classification system using SVM could also be slow to train Development of a Wearable Mobility Monitoring System 47 Literature Review from parent movement Meme 35 no yes other m sub movement 1 no yes sub movement 1 Figure 2 12 Generic classification framework presented by Mathie et al 175 Research studies have also used artificial neural networks to recognize activity such as Wang et al 177 and Yang et al 178 An artificial neural network is a mathematical model based on the biological neural network It consists of inputs and outputs with a processing layer or hidden layer in between 77 Artificial neural networks are complex and required previous training data Research studies have also used naive bayes classifiers to recognize activity from accelerometer data 96 156 This type of classifier assumes that all attributes of the variables class are independent and learns from training data the probability of each attribute 77 Fuzzy logic is another example of a classification technique that provides a way to arrive at a specific conclusion based upon vague ambiguous imprecise noisy or missing input information 179 Recently Chen et al 180 demonstrated that a classifier based on a fuzzy basic function was able to recognize different human daily
62. 4 2 Frequency and Amplitude Accelerations produced by human movement vary across the body and depend on the activity being performed Acceleration amplitude decreases from ankle to head with the greatest amplitude found in the vertical direction 162 During walking and running Bhattacharya et al 162 found that acceleration amplitude could reach 12g at the ankle 5g at the lower back and 4g at the head g acceleration due to gravity When selecting an accelerometer for human movement studies the choice of the accelerometer amplitude range should be based on the type of activity being studied and the location of the sensor Ermes et al 5 found that an accelerometer of range 2g was insufficient for detecting vigorous exercises therefore they had to use an accelerometer of range 10g instead However a larger range of acceleration results in a decrease in signal resolution but this decrease had a negligible effect on the signal features in Ermes et al s study During different walking speeds 0 99 to 2 35 m s the acceleration frequency spectra measured at the head shoulder and pelvis was between 0 75 to 4 8 Hz 163 These results from Cappozzo also demonstrated that the maximum frequencies measured increased from head to ankle and were the greatest in the vertical direction A study by Antonsson and Mann 164 found that in foot acceleration measurement during walking 9846 of the frequency spectra were less than 10 Hz and 9990
63. 50 159 P Barralon N Vuillerme and N Noury Walk detection with a kinematic sensor frequency and wavelet comparison in Proceedings of the 28th Annual International Conference of the IEEE Engineering in Medicine and Biology Society 2006 pp 1711 1714 160 T Hester D M Sherril M Hamel K Perreault P Boissy and P Bonato Identification of tasks performed by stroke patients using a mobility assistive device in Proceedings of the 28th Annual International Conference of the IEEE Engineering in Medicine and Biology 2006 pp 1501 1504 161 J P rkk M Ermes P Korpipaa J M ntyj rvi J Peltola and I Korhonen Activity classification using realistic data from wearable sensors IEEE Transactions on Information Technology in Biomedicine vol 10 pp 119 128 2006 162 A Bhattacharya E P McCutcheon E Shvartz and J E Greenleaf Body acceleration distribution and O2 uptake in humans during running and jumping Journal of Applied Physiology Respiratory Environmental and Exercise Physiology vol 49 pp 881 887 1980 Development of a Wearable Mobility Monitoring System 142 References 163 A Cappozzo Low frequency self generated vibration during ambulation in normal men Journal of Biomechanics vol 15 pp 599 609 1982 164 E K Antonsson and R W Mann The frequency content of gait Journal of Biomechanics vol 18 pp 39 47 1985 165 M Sun and J O Hill A method for measur
64. 53 F Pitta T Troosters V S Probst M A Spruit M Decramer and R Gosselink Quantifying physical activity in daily life with questionnaires and motion sensors in COPD European Respiratory Journal vol 27 pp 1040 1055 2006 54 A M Jette A R Davies P D Cleary D R Calkins L V Rubenstein A Fink J Kosecoff R T Young R H Brook and T L Delbanco The Functional Status Questionnaire Reliability and validity when used in primary care Journal of General Internal Medicine Official Journal of the Society for Research and Education in Primary Care Internal Medicine vol 1 pp 143 149 1986 55 J S Brach J M VanSwearingen A B Newman and A M Kriska Identifying early decline of physical function in community dwelling older women Performance based and self report measures Physical Therapy vol 82 pp 320 328 2002 56 D B Reuben L A Valle R D Hays and A L Siu Measuring physical function in community dwelling older persons A comparison of self administered interviewer administered and performance based measures Journal of the American Geriatrics Society vol 43 pp 17 23 1995 57 S E Sherman and D Reuben Measures of functional status in community dwelling elders Journal of General Internal Medicine vol 13 pp 817 823 1998 Development of a Wearable Mobility Monitoring System 133 References 58 J F Fries P Spitz R G Kraines and H R Holman Measure
65. 86 7 90 0 4 7 92 9 100 0 96 4 5 1 93 3 100 0 96 7 4 7 100 0 100 0 100 0 0 0 73 3 93 3 83 3 14 1 Standing waiting for elevator 85 7 100 0 92 9 10 1 Walking to get in the elevator 7 7 23 1 15 4 10 9 peer on un keep walking on level ground 15 100 0 100 0 100 0 0 0 Standing waiting for elevator 93 3 100 0 96 7 4 7 Walking to get in the elevator 7 1 35 7 21 4 20 2 Taking elevator to 1 floor 85 7 57 1 71 4 20 2 psr of en en keep walking on level ground 15 86 7 86 7 86 7 0 0 Walking up stairs 100 0 50 0 75 0 35 4 Walking on stair intermediate FE landing level ground for 1 5 m 100 0 100 0 100 0 0 0 Walking up stairs NH 100 096 100 096 100 096 0 096 Walking on level ground 75 096 75 096 75 096 0 096 Walking down stairs 0 096 0 096 0 096 0 096 Development of a Wearable Mobility Monitoring System 119 Technical and Mobility Evaluation of the Prototype WMMS Walking on stair intermediate So landing level ground for 1 5 m 10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 73 3 73 3 73 3 0 0 100 0 100 0 100 0 0 0 100 0 86 7 93 3 9 4 100 0 100 0 100 0 0 0 60 0 46 7 53 3 9 4 Fwakingonramn 33 3 0 0 16 7 23 6 Transition indoor outdoor and keep walking on level ground 7 100 0 71 4 85 7 20 2 Transition outdoor indoor and keep walking on level ground 7 71 4 42 9 57 1 20 2 Transition indoor outdoor and keep walking on level ground 3 100 0 100 0 100 0 0 0
66. A wareness in CHI 2000 Workshop on the what Who Where when and how of Context Awareness 2000 135 M Tentori and J Favela Activity aware computing for healthcare IEEE Pervasive Computing vol 7 pp 51 57 2008 136 B T Korel and S G M Koo Addressing context awareness techniques in body sensor networks in Proceedings of the 21st International Conference on Advanced Information Networking and Applications Workshops Symposia 2007 pp 798 803 137 M J Mor n J R Luque A A Botella E J Cuberos E Casilari and A D az Estrella J2ME and smart phones as platform for a Bluetooth Body Area Network for patient telemonitoring in Proceedings for the 29th Annual International Conference of IEEE Engineering in Medicine and Biology Society 2007 pp 2791 2794 138 M Sekine T Tamura M Akay T Fujimoto T Togawa and Y Fukui Discrimination of walking patterns using wavelet based fractal analysis IEEE Transactions on Neural Systems and Rehabilitation Engineering vol 10 pp 188 196 2002 139 M Sekine T Tamura T Togawa and Y Fukui Classification of waist acceleration signals in a continuous walking record Medical Engineering and Physics vol 22 pp 285 291 2000 140 M J Mathie A C F Coster N H Lovell B G Celler S R Lord and A Tiedemann A pilot study of long term monitoring of human movements in the home using accelerometry Journal of Telemedicine and Telecare vol
67. DEVELOPMENT OF A WEARABLE MOBILITY MONITORING SYSTEM Ga tanne Hach Thesis submitted to the Faculty of Graduate and Postdoctoral Studies in partial fulfillment of the requirements for the degree of MASTER OF APPLIED SCIENCE in Biomedical Engineering Ottawa Carleton Institute for Biomedical Engineering University of Ottawa Ga tanne Hach Ottawa Canada 2010 Abstract Monitoring mobility at home and in the community and understanding the environment and context in which mobility occurred is essential for rehabilitation medicine This thesis introduces a Wearable Mobility Monitoring System WMMS for objective measurement of community mobility This prototype WMMS was created using a smartphone based approach that allowed for an all in one WMMS The wearable system is worn freely on a person s belt like a normal phone The WMMS was designed to monitor a user s mobility state and to take a photograph when a change of state was detected These photographs are used to identify the context of mobility events i e using an elevator walking up down stairs type of walking surface Mobility evaluation using the proposed WMMS was performed on five able bodied subjects System performance for detecting changes of state and the ability to identify context from the photographs was analyzed The WMMS demonstrated good potential for community mobility monitoring Development of a Wearable Mobility Monitoring System ii
68. Geriatrics Society vol 52 pp 625 634 2004 Development of a Wearable Mobility Monitoring System 132 References 47 T Giantomaso L Makowsky N L Ashworth and R Sankaran The validity of patient and physician estimates of walking distance Clinical Rehabilitation vol 17 pp 394 401 2003 48 P J Rathouz J D Kasper S L Zeger L Ferrucci K Bandeen Roche D L Miglioretti and L P Fried Short term consistency in self reported physical functioning among elderly women The women s health and aging study American Journal of Epidemiology vol 147 pp 764 773 1998 49 S Mudge and N S Stott Outcome measures to assess walking ability following stroke a systematic review of the literature Physiotherapy vol 93 pp 189 200 2007 50 M J Follick D K Ahern and N Laser Wolston Evaluation of a daily activity diary for chronic pain patients Pain vol 19 pp 373 382 1984 51 R M P Moore D Berlowitz L Denehy B Jackson and C F B S McDonald Comparison of pedometer and activity diary for measurement of physical activity in chronic obstructive pulmonary disease Journal of Cardiopulmonary Rehabilitation amp Prevention vol 29 pp 57 61 January February 2009 52 O R Pearson M E Busse R W M Van Deursen and C M Wiles Quantification of walking mobility in neurological disorders QJM Monthly Journal of the Association of Physicians vol 97 pp 463 475 2004
69. MEE D OTIAWA oras b n oed Ottawa Hospital Research Ethics Boards Conseils d thique en recherches 761 Parkdale Avenue Suite 106 Ottawa Ontario K1Y 1J7 613 798 6555 ext 14902 Fax 613 761 4311 http Awav ohri ca ohreb Friday December 04 2009 Dr Edward Lemaire The Ottawa Hospital Rehabilitation Centre Institute for Rehabilitation Research and Development Room 1402 505 Smyth Road Ottawa ON K1H 8M2 Dear Dr Lemaire Re Protocol 2009846 01H Evaluation of a Wearable Mobility Monitoring System Protocol approval valid until Thursday February 04 2010 Thank you for the letter from Gaetanne Hache dated December 4 2009 am pleased to inform you that this protocol underwent expedited review by the Ottawa Hospital Research Ethics Board OHREB and is approved for two months to begin recruiting English speaking participants Approval is conditional upon the receipt of the University of Ottawa Health Sciences and Sciences administrative approval No changes amendments or addenda may be made to the protocol or the consent form without the OHREB s review and approval Approval is for the following COREB application English Recruitment Script received December 3 2009 French Recruitment Script received December 4 2009 English Information Sheet and Consent Form dated December 4 2009 Upon receipt and review of the French consent form the study expiry date may be extended to December 3 2010 one year and the r
70. Person standing inside Transition open door to walk a Initiation Person standing inside b Termination Start of forward walking progression Walk 20 meters a Initiation Start of forward walking progression b Termination End of forward walking progression Turn around a Initiation End of forward walking progression b Termination Facing opposite direction Transition turn around to walk a Initiation Facing opposite direction b Termination Start of forward walking progression Walk 20 meters towards the front door a Initiation Start of forward walking progression b Termination Inside stepping outside Transition inside to outside automatic door a Initiation Inside stepping outside b Termination Start of forward walking progression Walk 30 outside towards the car a Initiation Start of forward walking progression b Termination End of forward walking progression when arrive at the car Transition walk to open car door a Initiation End of forward walking progression when arrive at the car b Termination Start opening car door Opening car door a Initiation Start opening car door b Termination Initiation of hip flexion at the start of stand to sit transition Stand to sit transition get in the car a Initiation Initiation of hip flexion at the start of stand to sit transition b Termination Seated position in the car Sitting in the car a Initiation Seated position in the car b Termination Seated position start to o
71. Roetenberg H J Luinge C T M Baten and P H Veltink Compensation of magnetic disturbances improves inertial and magnetic sensing of human body segment orientation IEEE Transactions on Neural Systems and Rehabilitation Engineering vol 13 pp 395 405 2005 118 H S Zhu J J Wertsch G F Harris J D Loftsgaarden and M B Price Foot pressure distribution during walking and shuffling Archives of Physical Medicine and Rehabilitation vol 72 pp 390 397 May 1991 119 P Cavanagh and J Ulbrecht Clinical plantar pressure measurement in diabetes rationale and methodology The Foot vol 4 pp 123 135 1994 Development of a Wearable Mobility Monitoring System 138 References 120 A D Townshend C J Worringham and I B Stewart Assessment of speed and position during human locomotion using nondifferential GPS Medicine and Science in Sports and Exercise vol 40 pp 124 132 2008 121 A Le Faucheur P Abraham V Jaquinandi P Bouy J L Saumet and B Noury Desvaux Measurement of walking distance and speed in patients with peripheral arterial disease A novel method using a global positioning system Circulation vol 117 pp 897 904 2008 122 P J Troped M S Oliveira C E Matthews E K Cromley S J Melly and B A Craig Prediction of activity mode with global positioning system and accelerometer data Medicine and Science in Sports and Exercise vol 40 pp 972 978 2008
72. System iv Chapterd Me ethodoloBEys ass ocod dures Rawat wheels ae SE UE 54 Al Design Griteria iiie reete tete p ie b eite odd ee Lo end Ebo ge 54 4 1 1 System Design Criteria iuueni eene deret Er tne end eoe LEUR ee ria 54 4 1 2 Software Desipti Criteria ioter d rere dette rete Ree Dd ete 55 42 System Architecture eee e rere Brent dee en ehe get apodo pene SE de e ebbe by ede 55 4 3 Determination of Change of State eee esessecsecssecssecssecssecsseceeeeseeeeaeeeaeeeneeeneeenaes 57 4 3 1 Mobility Tasks and Context Classification eeseeseeeeeeeeeeee 57 4 3 2 Algorithm Outline xt Eee per elidel hed a ieee 58 4A System Evaluation Outlines cc c cre2tscsscaleesiitgrsestseectabecdacdetsctadoediscaueestucesubaacauteeasouesen sears 59 Chapter 5 Preliminary Evaluation of the BlackBerry for WMMS 60 5 1 Biomechanical Parameters Calculations eese eene 61 2 2 XDuS Kit dao e e ei e IRE Lae eet eae Land 63 25 3 Java Programming ute eee et tre tee be aee ep esie oe ce ere iubere eaa 64 25 4 Test Procedure lioe reete ete darted Sonne pe edo dos RE EET NES 64 5 5 Preliminary Evaluation Results eese ene en rennen rennen 65 5 6 Preliminary Evaluation Discussion eese ener rennen enne 66 5 7 Sufntlaty on e ehe OR EP D e ere bee deni ein 67 Chapter 6 Hardware Design and Evaluation eeeeeeeeeeeee
73. Table of Contents ilg d ii Ei MILI TT EET m vii List Of PI SUTES P Vili PICTON Y RN X AeknoWwIede mik 25 AAA xii Chapter I Introduction uade bier Meroe ve diae dea EUER NUR RU de ve pe ERE 1 EI iContributions uie eee eene eee ee a a ee ae UR elena aes 2 1 2 Scope Of the THESIS e Ra e iae diet e emo 3 1 3 Ovyerview oLbthe tliesls interne fet pe epe De et dulce iedeete repete ode reete eere ie ed 4 Chapter 2 Literature Review iios ni kaa 2 2 Community Mobility eene ege Re tide de idee tpe De ete die ride Led 5 2 1 1 International Classification of Functioning Disability and Health 6 2 1 2 Dimensions of Mobility Framework eese eene nennen 7 2 2 Mobility Measurement Ed e Pep e Edge 9 2 2 1 Observation and Clinical Tests AE nennen nennen 9 2 2 1 1 Dynamic Gait Index DGD pena u ianen ER a A EENET EE 9 2 2 1 2 Functional Gait Assessment FGA sees eene iiki irer ennt enne 9 2 2 1 3 Community Balance and Mobility Scale CB amp M sees 10 2 2 1 4 Bete Balance Scale edet erede etn Hee eoe 10 22 05 Timed Up and Go Testo ie e tee ett e EG ee PL eee oe e UR Linee 10 2 2 1 6 6 Minutes Walk Test iR pde wee Ede dee 11 22 17 Tinetti Assessment DI ool ite eee e Petri i e 11 2 2 1 8 Functional Independence Measure ws wmmmsmmmamenzanzanzanzznmnimamanm am
74. acts the ramp Walk up the ramp a Initiation Lead leg contacts the ramp b Termination End of forward walking progression on the ramp Turn around a Initiation End of forward walking progression b Termination Facing opposite direction Transition turn around to walk a Initiation Facing opposite direction b Termination Lead leg contacts the ramp Walk down the ramp a Initiation Lead leg contacts the ramp b Termination Lead leg contacts level ground Development of a Wearable Mobility Monitoring System 108 49 50 51 52 53 54 DD 56 57 58 59 60 61 62 63 64 Technical and Mobility Evaluation of the Prototype WMMS Walk 15 meters towards the exit door a Initiation Lead leg contacts level ground b Termination End of forward walking progression and start pushing on the door to go outside Open the door to go outside and transition inside to outside a Initiation End of forward walking progression and start pushing on the door to go outside b Termination Person standing outside Transition open door to walk a Initiation Person standing outside b Termination Start of forward walking progression Walk 60 meters on paved path way towards the front door a Initiation Start of forward walking progression b Termination Pulling on the door to go inside Open the door to go inside and transition outside to inside a Initiation Pulling on the door to go inside b Termination
75. ailable http www who int classifications icf en Accessed 18 Mar 2009 20 E Stanko P Goldie and M Nayler Development of a new mobility scale for people living in the community after stroke Content validity Australian Journal of Physiotherapy vol 47 pp 201 208 2001 21 A Shumway Cook A E Patla A Stewart L Ferrucci M A Ciol and J M Guralnik Environmental demands associated with community mobility in older adults with and without mobility disabilities Physical Therapy vol 82 pp 670 681 2002 22 A M Myers P J Holliday K A Harvey and K S Hutchinson Functional performance measures Are they superior to self assessments Journals of Gerontology vol 48 1993 Development of a Wearable Mobility Monitoring System 130 References 23 A Patla Mobility in complex environments implications for clinical assessment and rehabilitation Journal of Neurologic Physical Therapy vol 25 pp 82 90 2001 24 A Shumway Cook M Baldwin N L Polissar and W Gruber Predicting the probability for falls in community dwelling older adults Physical Therapy vol 77 pp 812 819 1997 25 J McConvey and S E Bennett Reliability of the dynamic gait index in individuals with multiple sclerosis Archives of Physical Medicine and Rehabilitation vol 86 pp 130 133 2005 26 L E Dibble and M Lange Predicting falls in individuals with Parkinson disease A reconsideration of cli
76. airs nN Walking on stair intermediate landing level ground for 1 5 meters Walking up stairs oo Walking on level ground No Walking down stairs N Walking on stair intermediate landing level ground for 1 5 meters N Walking down stairs N N Walking on level ground N W Stand to lie transition N F Lying N CA Lie to Stand transition N oN Walking on level ground N l Walking on ramp N oo Walking on level ground N No Transition indoor outdoor and keep walking on level ground 99 j Transition outdoor indoor and keep walking on level ground io Transition indoor outdoor and keep walking on level ground o2 N Stand to sit transition to get in the car W o2 Sitting in the car Development of a Wearable Mobility Monitoring System 111 Technical and Mobility Evaluation of the Prototype WMMS 34 Starts of car ride 35 Stop of car ride 36 Sit to stand transition 37 Walking on level ground 38 Transition outdoor indoor and keep walking on level ground 39 Standing Changes of state timing from digital video was compared with the WMMS change of state timestamps WMMS data output was analyzed window by window All data windows were analysed to determine if the state for that window was a true or false negative True positives occurred when a change of state occurred the algorithm ide
77. and GPS location The camera can take up to 3000 images per day A recent study proposed an automatic event segmentation method for the Figure 2 9 SenseCam images 129 SenseCam using content and contextual information 130 SenseCam has been particularly explored for its memory aid application 131 133 A list of publications related to SenseCam is presented on the Microsoft research website 129 In research applications video cameras are often used to validate other mobility assessment methods For instance participants have been videotaped during community excursions to validate self report mobility tools 61 2 3 4 7 Ambient Sensors Ambient sensors are sensors that can measure different properties related to the surrounding conditions and environments Light humidity temperature acoustic and barometric pressure sensors are example of ambient sensors These sensors are used in context awareness systems to add more information about the context that can help to better identify location and recognize activity 14 15 Light sensors such as photodiodes color sensors IR and UV sensors can help differentiate between indoors and outdoors Temperature and humidity sensors can help detect weather characteristics such as raining or cold and differentiate between indoor and outdoor activities 2 3 5 Context Awareness A context aware system was defined by Dey and Abowd 134 as a system that uses context to provide rel
78. ard walking progression Walk 60 meters until the elevator a Initiation Start of forward walking progression b Termination End of forward walking progression and moving to press elevator button Transition walk to wait for elevator a Initiation End of forward walking progression and moving to press elevator button b Termination Standing Standing waiting for elevator a Initiation Standing b Termination Start of forward walking progression to get inside the elevator Get in the elevator a Initiation Start of forward walking progression to get inside the elevator b Termination Standing inside the elevator Take the elevator to the second floor a Initiation Standing inside the elevator b Termination Start of forward walking progression to get outside the elevator Get out of the elevator and walk 15 meters a Initiation Start of forward walking progression to get outside the elevator b Termination End of forward walking progression Turn around a Initiation End of forward walking progression b Termination Facing elevator Transition turn around to walk a Initiation Facing elevator b Termination Start of forward walking progression Walk 15 meters towards the elevator a Initiation Start of forward walking progression b Termination End of forward walking progression and moving to press elevator button Transition walk to wait for elevator a Initiation End of forward walking progression and moving to press elevator button b
79. assessment period The mobility envelope was found to be smaller for a frail individual compare to a healthy individual GPS receivers can also be used to complement accelerometer data by providing the locations where physical activity occurs 8 and also to help better recognize activities 122 Development of a Wearable Mobility Monitoring System 27 Literature Review Many smartphones and mobile phones are now integrated with GPS receivers which offer a feasible way to collect location information for contextual health research Using the GPS enabled BlackBerry 7520 Wiehe et al 123 tracked adolescent travel patterns and gathered daily diary GPS data MacLellan et al 124 used a smartphone GPS receiver and the activPal 79 in order to help people to examine their activity pattern and potentially provide indications where environmental barriers could occur GPS was found to be a promising tool to characterize exposure to social and physical environments in studies of older adults living in diverse communities 125 Other GPS applications are in wearable activity recognition systems to help detecting more types of activity such as cycling outdoors 5 Also GPS can be used in life log applications 12 and to annotate text notes and photos to location in mobile phones 126 Despite all the advantages and uses of GPS some limitations exist when recording positions for indoor and for some outdoor environments such as under
80. ata to andimagenames as raw output file to output file i 3 GPS i SavefiletoSD Flag 2 data Parse to integer I I nteger I Value i l Ch ange i of State i i EE i i card I I Create one output I i i i l i i 1 Figure 7 11 Overview of programming flow Development of a Wearable Mobility Monitoring System 101 Development of the Prototype WMMS 7 8 Summary The prototype WMMS was designed to determine a user s state detect changes of state and take a picture when a change of state occurred The data used in the algorithm were coming from the external board and the BlackBerry The raw acceleration signal was divided into its dynamic and static components using a digital low pass filter Signal features were extracted from these two components and then input to the algorithm The features selected for this prototype WMMS were standard deviation of the y axis acceleration inclination angle skewness of the y axis acceleration signal magnitude area SMA light intensity and GPS speed The standard deviation was selected to detect changes of state caused by start stop actions the inclination angle detected postural changes skewness detected changes of state caused by walking on stairs SMA detected a change in movement intensity and postural transition light intensity differentiated between indoor and outdoor states and GPS speed detected when
81. ation principle a user s location on earth can be determined GPS works anywhere on earth any time and no subscription fee or setup charge is required to use GPS services However the performance of GPS receivers is reduced during situation where their view of the sky is obstructed e g indoors close to tall building cloudy Determining the speed of displacement from a GPS receiver is usually based on the Doppler Effect which is the measurement of the rate of change in the satellite s signal frequency caused by the movement of the GPS receiver The speed of displacement can also be calculated by the change of distance divided by the change of time but it is usually less accurate than using the Doppler Effect 120 Many mobility monitoring studies have used GPS systems For human locomotion non differential GPS receivers can provide accurate speed displacement and position information 120 GPS was recently found to potentially provide valid information on walking capacity in patients with peripheral arterial disease 121 In human tracking GPS technology offers a great opportunity to help understanding how environmental factors can influence a person s mobility Frank and Patla 17 proposed a mobility envelope measured from excursions in the community over a week as a potential outcome measure for mobility Frank and Patla s mobility envelope is the length of the outer perimeter of spatial excursions made by the individual during the
82. ational Conference on Multimedia and Ubiquitous Engineering 2009 pp 386 91 103 A F Dalton C N Scanaill S Carew D Lyons and G laighin A clinical evaluation of a remote mobility monitoring system based on SMS messaging in Proceedings of the 29th Annual International Conference of IEEE Engineering in Medicine and Biology Society 2007 pp 2327 2330 104 Network dictionary WPAN Wireless Personal Area Network Communication Technologies Network dictionary 2004 Online Available http www networkdictionary com wireless WPAN php Accessed 28 Apr 2009 105 K Hung Y T Zhang and B Tai Wearable medical devices for tele home healthcare in Proceedings of the 26th Annual International Conference of the IEEE Engineering in Medicine and Biology Society 2004 pp 5384 5387 106 K Hung and Y T Zhang Usage of Bluetooth in wireless sensors for tele healthcare in Proceedings of the 24th Annual International Conference of the IEEE Engineering in Medicine and Biology 2002 pp 1881 1882 Development of a Wearable Mobility Monitoring System 137 References 107 W Y Wong M S Wong and K H Lo Clinical applications of sensors for human posture and movement analysis A review Prosthetics and Orthotics International vol 31 pp 62 75 2007 108 W Zijlstra and K Aminian Mobility assessment in older people New possibilities and challenges European Journal of Ageing vol 4 pp 3 12 2007
83. been used by Baek et al 141 to discriminate between walking running and walking up down stairs Percentiles of the acceleration signals have also been used by Maurer et al 15 for similar applications Development of a Wearable Mobility Monitoring System 44 Literature Review Differentiation among activities that involve translation in just one dimension could be done by calculating the correlation of the accelerometer signal for each pair of axes such as presented by Ravi et al 96 For example walking and running can be distinguished from stair climbing using correlation Walking and running usually involve translation in one dimension whereas climbing involves translation in more than one dimension The correlation of the accelerometer signal corresponds to the ratio of the covariance and the product of the standard deviations Equation 2 15 cowl y 2 15 O O y Corr x y Despite the processing time efficiency of using time domain features they do not give information on the cyclic behaviour of the acceleration signal caused by dynamic activities e g walking running Therefore recent studies have used frequency domain features To generate these features the signal must first be converted into the frequency domain A common technique used for this conversion is the Fast Fourier Transform FFT The FFT compares a family of sine functions at harmonically related frequencies by multiplying the waveform with s
84. bject and the overall values are given in Table 8 3 An overall sensitivity of Development of a Wearable Mobility Monitoring System 114 Technical and Mobility Evaluation of the Prototype WMMS 77 7 X 2 596 and a specificity of 96 4 2 2 were obtained Sensitivity and specificity results for each trial are given in Appendix B The sensitivity and the specificity were also calculated for each of the mobility tasks and are given in Table 8 4 Results per mobility task for each trial are given in Appendix C The lowest performances were obtained for going up stairs 13 396 walking on a ramp 40 0 and transitioning from indoor to outdoor 46 7 for the first time going outside and 20 096 for the second time going outside and outdoor to indoor 46 746 for first time going inside and 26 7 for second time going inside For the first outdoor activity the subjects walked through an unobstructed courtyard In the second outdoor scenario the subjects walked under a building overpass to the car Lighting was different between the two scenarios The subjects were walking indoors before and after these four activities Therefore if a change of state was not detected the following walking indoor change of state was also not identified since the system believed that the subject was still walking indoors This resulted in lower performance values If these low results were to be removed from the overall performance a sensitivity of 93
85. body could be an issue for the camera and the light sensor Since the WMMS was worn on the waist the user s clothes could cover the camera view and the light sensor unintentionally especially during winter Limitations during image evaluation were also present All the images were always in order and the same scenes were evaluated for each trial The evaluators could have become better at identifying the context from the pictures after evaluating results from several subjects Development of a Wearable Mobility Monitoring System 126 Conclusion Chapter 9 Conclusion Maintaining independent mobility at home and in the community plays an important role in an individual s independence quality of life and health and in the lives of their family and the people around them Measuring mobility and the environment in which mobility events takes place can help with these roles Our WMMS approach to respond to the need for community mobility assessment tools shows great potential The BlackBerry handheld device proved to be a viable platform for this WMMS application In addition to industry standard tools for development secure communications and image capture the multitasking device demonstrated good capability for data capture real time processing and data storage Adding the camera to the WMMS suggested that images could help identify mobility tasks such as walking up stairs and taking an elevator The images also helped to identify
86. cessed data the GPS coordinates and the GPS acquisition time were saved to a file on the smart phone s SD card After completing data collection the file was downloaded to a personal computer via USB to visualize the results Motion Capture Hub Xbus Master 5 MTx BlackBerry 8800 PC E Ve x Bluetooth USB O OOO GPS Data OutputFile Orientation Data Time Figure 5 1 System architecture for the preliminary testing Development of a Wearable Mobility Monitoring System 60 Preliminary Evaluation of the BlackBerry for WMMS 5 1 Biomechanical Parameters Calculations The proof of concept WMMS system calculated biomechanical parameters such as joint angles of both knees and hips The sensor placement for this application is shown on Figure 5 2 and Figure 5 3 The Cardan Euler technique was used which is one of the most widely used methods in biomechanics to calculate 3D joint angles 183 For each joint the relative orientation between the distal sensor coordinate system and the proximal sensor coordinate system was determined by computing the rotation transformation matrix RTM of that particular joint For the knee joints the distal sensor was on the lower leg and the proximal sensor was on the upper leg For the hip joints the distal sensor was placed on the upper leg and the proximal sensor was on sacrum Motion Trackers Lower Back Z axis Y axis is pointing in Y axis is ointing in P 3 X axis Y ax
87. cognition behaviour communication and community functioning 44 More details of FIM can be found in 45 Development of a Wearable Mobility Monitoring System 11 Literature Review 2 2 2 Diaries and Questionnaires Diaries and questionnaires are used to assess mobility disability or disability in activities of daily living ADL by having the participants report on whether they have difficulties or need help in performing ADL or mobility related tasks 46 These two approaches provide complementary information to performance based mobility tests because these methods can capture a person s perception of their ability to perform daily activities and capture details on the environmental impact on mobility However self reports and questionnaires on ADL disability are known for their compromised reliability due to under or over reporting 47 and their limited reliability in a frail older population 48 Despite these disadvantages questionnaires remain one of the few ways to understand mobility performance in the community 49 The following will describe some of these methods 2 2 2 1 Diaries Diaries have been used to assess mobility in the community Follick et al 50 asked patients to record three times a day in half hour blocks over 24 hours the time spent lying sitting standing walking and sleeping In a recent study by Moore et al 51 the activity diary appeared to have greater promise than pedometers step counters fo
88. could be given back to the user if required Community WMMS Wireless E Capturing qd P Mobility and N d Context Data Server Sending Data Figure 4 1 System Architecture of a WMMS In this research the WMMS consisted of a central node or hub that captured processed and logged the motion and contextual data An external sensor board was added to the design since the current central node Blackberry Bold did not provide access to raw accelerometer Development of a Wearable Mobility Monitoring System 55 Methodology data The external board was designed to fit on the BlackBerry Bold holster to simulate an all in one WMMS Figure 4 2 The board captured motion data accelerometer ambient data light intensity temperature and humidity The central node provided GPS location data and speed time and digital photo images contextual information The WMMS was designed to be worn on the waist which is a common location to wear a mobile or smartphone and a validated site for accelerometer data collection for mobility measurement Section 2 4 1 The WMMS determined the user s state and took a digital picture whenever a change of state occurred The mobility state was determined within a one second window and then copied to a file along with contextual information for that second Front View Side View Figure 4 2 Front and side view images of the WMMS Development of a Wearable Mobility Monitoring Syste
89. cribe two axis misalignment and crosstalk between channels caused by the sensor electronics 169 This method of using nine elements resulted in higher accuracy than both the factory calibration and the six elements model 168 The choice of calibration method depends of the type of application When an application needs to estimate the distance traveled from double integration of the acceleration signal the error from offset drift may cause the position measurement to diverge in just a few seconds 170 A drift correction technique was studied by Yun et al 170 where the drift was corrected by detecting periods where velocity is zero i e stance phase during walking Finally in other cases the application may only require an offset removal at the start of a data measurement session 148 2 4 4 Filtering Techniques The output signal of an accelerometer worn on the body is composed of the acceleration due to body movement gravitational acceleration and noise Undesirable accelerations could come from external vibration such as vehicle s acceleration bouncing of the sensor against objects jolting of the sensor caused by loose attachment etc 80 If the frequency range of the noise does not interfere with human body acceleration filtering techniques could attenuate the noise in the accelerometer s output signal 80 For example Development of a Wearable Mobility Monitoring System 39 Literature Review Bouten et al 80
90. ction to the gravity vector With the board in that position the acceleration was measured The board was then rotated 180 degrees such that its x axis was in the same direction as the gravity vector the acceleration was again measured The offset value of the X axis was obtained by adding the maximum acceleration measured value Umax and the minimum acceleration measured value Umin divided by two Equation 2 5 Then the x axis sensitivity was obtained by subtracting the minimum acceleration measured value from the maximum acceleration measured value and dividing by two Equation 2 4 The calculated offset and the sensitivity values were used to calculate acceleration in g prior to data processing Equation 2 6 Accelerometer calibration and re calibration is often needed to correct for signal drift Section 2 4 1 4 Drift of the acceleration DC component was tested during five trials of 2 hours each During each trial acceleration data was collected where the WMMS was run without moving the external sensor board The drift was calculated by subtracting the minimum value from the maximum value of the mean DC acceleration The average drift value and the standard deviation for the three axes were 0 0023 0 0010 g hour for x axis 0 0029 0 0008 g hour for y axis and 0 0040 0 0016 g hour for z axis From these drifting rates the inclination angle calculation might vary by no more than 5 degrees after 12 hours From these results it was
91. ctions such as squared root and absolute value Additionally the BlackBerry data encryption built in option was selected to ensure privacy and security of the data Encryption was set so that reading and downloading the output file from the BlackBerry to the computer required a password as well as the same handheld device used to store the data The use of the BlackBerry Bold was also password protected An overview of the programming flow chart for the WMMS Java application is presented in Figure 7 11 The BluetoothListener interface from the Bluetooth API had a built in method Development of a Wearable Mobility Monitoring System 99 Development of the Prototype WMMS called datareceived that was automatically run when data was detected on the Bluetooth port When data were received our data processing method was run Every received byte was processed before reading more data from the Bluetooth port The received bytes were first parsed to verify CRC Cyclic Redundancy Check If the CRC test passed the data were parsed into six integer numbers AccX AccY AccZ Light Temperature Humidity and Battery At this point depending on the selected option the raw data could be copied in a circular queue which could then be emptied by a separate thread to copy the data to a raw data output file stored on the BlackBerry SD card The other option was to proceed with data processing With the processing option selected the acceleration
92. d etii 150 ADpendix Orie Lado a abit ede a ac ade eh ded c duis 151 ZJADDeudbc Eb do cot ocn n dicat dou fox toot oit otis ed bo Led 154 Appendix EF 4 cule D P 162 Development of a Wearable Mobility Monitoring System vi List of Tables Table 2 1 Comparison of different features of common wireless technologies 85 104 22 Table 2 2 Example of laboratory and clinical studies using accelerometers for movement and mobility analysis List modified from Godfrey et al 77 sees 34 Table 5 1 Preliminary BlackBerry evaluation results eee 66 Table 6 1 Summary of specifications for main component of the external sensors board 71 Table 6 2 Average output value of the light sensor mV for different light conditions standard deviation an brackets ni v ei t n averte reise 75 Table 6 3 BlackBerry Bold camera performance evaluation results 79 Table 7 1 Description of the state DIIS iiio IE ere iwa 96 Table 7 2 Section of a WMMS output file to demonstrate timing of the picture taken 98 Table 8 1 Results for the BlackBerry Bold battery evaluation esses 104 Table 8 2 Changes of state and context to be identified from WMM pictures 113 Table 8 3 Summary performance results for the each subject sess 116 Table 8 4 Performance resu
93. d with the digital video camera was used to determine the time value of when a change of state occurred The timing for all tasks were determined based on the initiation and termination details given in the list presented above For this thesis the possible changes of state caused by opening a door and turning around were not evaluated These possible changes of state were not in the scope of this WMMS prototype In addition to be able to compare one trial to another changes of state created by extra mobility tasks were removed from the evaluation i e subject movements not related to the protocol The following list is the mobility tasks that were included in the evaluation of the WMMS going from one task to another should trigger a change of state providing 38 changes of state per trial Development of a Wearable Mobility Monitoring System 110 Technical and Mobility Evaluation of the Prototype WMMS Standing Walking on level ground Stand to sit transition Sitting Sit to stand Walking on level ground Standing waiting for elevator Walking to get in the elevator SOO TM OON UA um WAN Taking elevator to second floor Walking to get out of elevator and keep walking on level ground Standing waiting for elevator N Walking to get in the elevator W Taking elevator to first floor AR Walking to get out of elevator and keep walking on level ground jak n Walking up st
94. differentiate between static and dynamic states The y axis standard deviation was passed through a double threshold DT algorithm Figure 7 3 Figure 7 4 shows an example of the y axis acceleration standard deviation during dynamic walking and static state With the DT algorithm if the state starts with static state it will stay static until the signal cross the dynamic threshold Then the state will be set to dynamic and will stay dynamic until the signal goes below the static threshold The dynamic threshold was set to 0 120g and the static threshold was set to 0 075g These threshold values were estimated based on preliminary testing of the WMMS Development of a Wearable Mobility Monitoring System 84 Development of the Prototype WMMS Standard deviation of y axis acceleration STDY STDY gt Dynamic Threshold No STDY lt Static Threshold State Static State Previous state State Dynamic Figure 7 3 Flowchart of the double threshold DT algorithm applied to the standard deviation of the y axis acceleration Development of a Wearable Mobility Monitoring System 85 Development of the Prototype WMMS Standard deviation of y axis acceleration versus time Dynamic Dynamic 3 0 3 Static J 0 25 0 2 0 15 Dynamic Threshol Standard deviation g atic Threshold Time seconds Figure 7 4 Standard deviation of y axis acceleration dur
95. door state was 300 Accelerometer calibration was only required once prior to use Testing for drift demonstrated that there was no need to recalibrate during use This was expected since a low drift accelerometer was placed on the board The BlackBerry camera test indicated that a picture could not be taken for every window of 1 02 seconds This limited the real time processing aspect of the WMMS Development of a Wearable Mobility Monitoring System 79 Development of the Prototype WMMS Chapter 7 Development of the Prototype WMMS This chapter describes the development of the prototype WMMS including the methods to generate the different signal features and how each feature is used to determine the user s state For this prototype WMMS the selected features were mostly time domain features and some heuristic features Section 2 4 6 such as inclination angle standard deviation of y axis skewness of y axis signal magnitude area SMA light intensity and GPS speed Farther in this chapter the algorithm to determine the state and the change of state of the user is given 7 1 Data Pre processing The raw acceleration data received on the BlackBerry were calibrated as explained in Section 6 2 8 The calibrated acceleration data were then passed through a median filter n 3 to remove spikes 7 Since the external board uses a variable capacitance accelerometer Section 2 3 4 1 the acceleration signal was composed of accelerati
96. dwidth cable replacement Low bandwidth sensors and automation medical monitoring home security High bandwidth applications sending data over wireless internet Another wireless protocol is IEEE 802 15 3 or UWB This standard operates in the 3 1 10 6 GHz frequency band Because of UWB s large bandwidth and since unlicensed and licensed frequencies are covered UWB systems are constrained in their output power which in turn limits their range 85 For applications such as WBSN this standard was found to be too complex in hardware and protocol Having a wide bandwidth was also not required for WBSN applications 85 Development of a Wearable Mobility Monitoring System 22 Literature Review Bluetooth also known as IEEE 802 15 1 standard is designed for short distance and small devices to replace cables between electronic lightweight devices e g mouse keyboard and headset Bluetooth can operate at a range of 10m and up to 100m depending of its class This standard provides small low cost and low power radio modules and is attractive for its technique of frequency hopping which increases security and privacy in radio transmissions 105 106 The maximum Bluetooth data rate is approximately 3Mbps 104 Despite the advantages of ZigBee Bluetooth is still a commonly used standard in WBAN design due to its present penetration in the market and its related commercial support 98 Smartphones exclusively use Blue
97. e 2 t 0 10 20 30 40 50 t t 60 70 80 90 100 Time seconds j 1 110 120 130 140 150 Figure 7 6 Example of a skewness curve for y axis acceleration The top graph is the skewness only The bottom graph is the skewness curve but with some dynamic static and stairs states identified The dotted line shows when the dynamic level was identified 1 e when the skewness values was analyzed for stairs or not stairs state Development of a Wearable Mobility Monitoring System 89 Development of the Prototype WMMS 7 2 4 Signal Magnitude Area SMA The SMA of the three acceleration signals x y z was used by Mathie et al 7 and Karantonis et al 9 to measure mobility SMA was shown to detect both amplitude and duration variation in the acceleration signal which could help detect the type of activity 7 SMA normalized to the length t can be calculated using Equation 7 8 1 SMA f 1a dt Va re la 2 7 8 t t 0 t 0 t 0 T where ft is the time in seconds and ax ay and a are the acceleration of x y and z axis respectively The integration technique used to calculate SMA in Equation 7 8 was based on Simpson s rule f ydx where n is the number of equal steps and y the acceleration ay ay or a With a sampling SE y 45 Xi uos 25 Ying 7 9 m 1 m 2 frequency of 50 Hz a 1 second window gives 50 samples and 49 steps Since Simpson s rule requires an even number of steps
98. e digital camera was synchronized with the WMMS by having the subject to block the light sensor with their hand for 5 seconds when starting data collection Digital video was necessary to validate change of state detection to determine the change of state timing and to provide context information The following is the list of tasks that the subjects were asked to perform The list is divided to facilitate video time segmenting of the different tasks Development of a Wearable Mobility Monitoring System 105 10 11 12 13 14 15 16 Technical and Mobility Evaluation of the Prototype WMMS From standing position walk for 25 meters a Initiation Start of forward walking progression b Termination End of forward walking progression Transition walk to stand to sit transition a Initiation End of forward walking progression b Termination Initiation of hip flexion at the start of stand to sit transition Stand to sit transition a Initiation Initiation of hip flexion at the start of stand to sit transition b Termination Seated position on chair Sitting for 30 seconds a Initiation Seated position on chair b Termination Initiation of trunk flexion and buttock lifting from chair Sit to stand transition a Initiation Initiation of trunk flexion and buttock lifting from chair b Termination Standing position Transition Sit to stand transition to walk a Initiation Standing position b Termination Start of forw
99. e of the most used approaches in activity classification studies because of its simplicity 171 Additionally pre processing of the sensor signal is not required with the sliding window technique making this approach effective for real time applications 171 A non overlapping window of approximately one second has often been used to detect static and dynamic states identify postures and postural transitions identify activities and detect falls 5 9 148 149 Furthermore Mathie et al 7 found that the optimal size was between 0 8 to 1 4 seconds for such classification systems However windows of different sizes and degree of overlap have been successful such as non overlapping 2 seconds window by Baek et al 141 a 5096 overlapping window of 5 12 seconds by Ravi et al 96 and a 6 12 seconds window by Bao and Intille 156 An advantage of having a larger window is that Development of a Wearable Mobility Monitoring System 40 Literature Review cyclic information could be captured for activities such as walking running and climbing stairs The event defined windows method needs pre processing to detect specific events for instance heel strike or toe off 171 The windows are defined from the timing of these events therefore window length may vary depending on the location of the events in the signal An example of a study detecting heel strike and toe off events is the one by Aminian et al 145 where event timing was
100. e password protected and keyboard lock AES or Triple DES encryption when integrated with Blackberry Enterprise Server Battery Life 4 5 hours of talk time and 13 5 hours of standby time Memory GB of onboard memory 128 MB of Flash memory and expandable memory support for microSD card Processor speed 624 MHz Operating System 4 6 0 244 External Board While the cutting edge and future smartphones have integrated accelerometers and the potential to test ambient light via the integrated camera an external board with mobility analysis sensors was used in this thesis The external sensors were required because a BlackBerry smartphone with all the required capabilities was not on the market during the development phase i e accelerometer GPS Wi Fi Bluetooth camera The external board design integrated into the phone s holster provided a flexible approach to add other measurement sensors or tools in the future 6 2 1 Design Criteria The custom made external board design criteria were Bluetooth serial port profile communication to allow communication with the BlackBerry smartphone Rechargeable battery that can last at least a day Triaxial accelerometer with a range of 6g and able to detect frequency up to 20 Hz as discussed in Section 2 4 1 2 This is to detect motion of the user Development of a Wearable Mobility Monitoring System 69 Hardware Design and Evaluation e Light sensor to help in differentiating
101. e the sensors can sample process log and communicate wirelessly to send one or more physiological or environmental parameters to a personal server 84 Figure 2 7 shows an example of a typical WBAN system architecture for patient monitoring as presented by Jovanov et al 84 The first level consists of physiological sensors second level is the personal server and the third level is the health care servers and related services Another example is the WiMoCA from Farella et al 89 that is a custom made WBSN where the sensing node consists of a triaxial integrated MEMS micro electro mechanical system accelerometer The WiMoCa system s ability to handle diverse application requirements such as posture detection system bio feedback application and gait analysis was recently demonstrated by Farella et al 94 Development of a Wearable Mobility Monitoring System 19 Literature Review ECG amp Tilt sensor SpO2 amp Motion sensor Body Area 5 Network Bluetooth Motion sensors f TTA KC OF EAN L 3 AAt OW se Network coordinator amp temperature humidity sensor Physician Figure 2 7 Example of a Wireless Body Area Network of intelligent sensors for patient monitoring reproduced from 84 2 3 2 Personal Server The use of a PDA personal digital assistant mobile phone and smartphone as the central node or personal sever in WBSN or WBAN is becoming very popular PDAs have
102. earable Mobility Monitoring System 16 Literature Review 2 2 4 2 Accelerometer Based Activity Monitor Many commercially available systems for research and individual health care monitoring incorporating accelerometers are presented by Godfrey et al 77 Examples include a waist mounted RT3 tri axial device Stayhealthy Inc Monrovia CA USA 78 for calorie monitoring and the activPAL Pal Technologies Ltd Glasgow United Kingdom 79 used to detect time spent sitting lying standing and stepping Inertial sensors applications in wearable system will be discussed in Section 2 3 4 2 2 4 3 Physiological Measurements Metabolic energy expenditure is a standard physical activity measure 80 81 Measurement of heart rates muscle activity EMG and pulmonary ventilation volume are examples of physiological measures used for this purpose 82 However these objective measures usually have a high cost per measurement 6 In addition these methods might need sensors attached directly to the skin at precise locations on the body such as for EMG This might not be suitable for a wearable long term monitoring mobility system 2 2 5 Summary of Mobility Measurement Observation and clinical mobility assessment tools are performance based measures that evaluate functional mobility and predict how a person will perform in the community However good outcomes from standardized clinical measures do not always result in independent community ambu
103. echnology Lab toward the bed a Initiation Start forward walking progression outside the stairwell b Termination End of forward walking progression Transition walk to stand to lie transition a Initiation End of forward walking progression b Termination Initiation of hip flexion at the start of stand to lie transition Stand to lie transition a Initiation Initiation of hip flexion at the start of stand to lie transition b Termination Lying position on bed Lying on back for 30 seconds a Initiation Lying position on bed b Termination Initiation of upper body movement off the bed at the start of lie to stand transition Lie to stand transition a Initiation Initiation of upper body movement off the bed at the start of lie to stand transition b Termination Standing position Transition lie to stand transition to walk a Initiation Standing position b Termination Start of forward walking progression Walk 30 meters towards the hall way and keep walking in left direction a Initiation Start of forward walking progression b Termination End of forward walking progression Turn around a Initiation End of forward walking progression b Termination Facing opposite direction Transition turn around to walk a Initiation Facing opposite direction b Termination Start of forward walking progression Walk 25 meters inside the Rehab Technology towards the ramp a Initiation Start of forward walking progression b Termination Lead leg cont
104. ecruitment of French speaking participants may begin When submitting French documetation to the OHREB confirm it has been translated or approved by Eric Lepine email all documentation except validated questionnaires to Eric at elepine ohri ca The validation date should be indicated on the bottom of all consent forms and information sheets see copy attached The Ottawa Hospital Research Ethics Board is constituted in accordance with and operates in compliance with the requirements of the Tri Council Policy Statement Ethical Conduct for Research Involving Humans Health Canada Good Clinical Practice Consolidated Guideline Part C Division 5 of the Food and Drug Regulations of Health Canada and the provisions of the Ontario Health Information Protection Act 2004 and its applicable Regulations SB Raphael Saginur M D Chairman Ottawa Hospital Research Ethics Board Encl M Development of a Wearable Mobility Monitoring System 165
105. ed Development of a Wearable Mobility Monitoring System 42 Literature Review Figure 2 10 Seismic uniaxial accelerometer measuring the component a of an equivalent acceleration a in the direction 4 of the sensitive axis of the accelerometer The equivalent acceleration is the sum of the acceleration a of the sensor and the equivalent gravitational acceleration g acting on the seismic mass Q is the angle between the sensitive axis of the accelerometer and the acceleration is the angle between the sensitive axis and the gravitational field reproduced from 147 1g Figure 2 11 Dual or tri axis accelerometer with two axes for measuring tilt reproduced from 172 Another example of heuristic features is the signal magnitude area SMA of the acceleration signal This feature is extracted from the AC or dynamic component of the acceleration signal SMA has been used to estimate the energy expenditure EE of physical activity and to quantify the acceleration amplitude The relationship between SMA of a triaxial accelerometer signal and EE has been demonstrated by Bouten et al 80 SMA was further used to discriminate between rest and activity periods in similar studies such as Mathie et al Development of a Wearable Mobility Monitoring System 43 Literature Review 7 and Karantonis et al 9 Equation 2 13 represents the normalized SMA used by Mathie et al 7 and Karantonis et al 9 sma la Wr la
106. ed one thread to read incoming data from the Bluetooth port and then parse the data The checksum was calculated for every sample to verify that there were no errors If the checksum was correct data bytes were converted to float numbers and then the biomechanical parameters calculations were completed The resulting joint angles were then put in a writing queue waiting to be copied to a file A second thread took data from the writing queue and then copied the data to a file along with the most recent GPS data Creating and writing files on the BlackBerry SD card were performed using the FileConnection interface from the javax microedition io file package The GPS data was obtained using the LocationListener interface from the javax microedition location package 5 4 Test Procedure Static and dynamic trials were performed In the static trials the Xbus kit and the BlackBerry were placed on a desk for the full duration An adapter connected to the wall AC outlet powered the Xbus Master In dynamic trials the sensors were attached on a subject s lower limbs and hip Figure 5 3 to simulate real world orientation angle measurements The Xbus kit was battery powered for the dynamic trials For the static trials the Xbus Master was set to sample data at 50 Hz and at 25 Hz 5 trials per frequency The Java application received the data from the Xbus for as long as there was no error sent by the Xbus Master A timer overflow error error code 28
107. ed sensors in elderly mobility monitoring were accelerometers gyroscopes magnetometers and pressure sensors or foot switches 108 Development of a Wearable Mobility Monitoring System 23 Literature Review The following gives an overview of four types of sensors that are the most relevant for mobility monitoring applications They are accelerometers gyroscopes magnetometers and pressure sensors Other wearable sensors that are described below are those that could detect contextual information such as GPS camera and ambient sensors 2 3 4 1 Accelerometers Accelerometers are low cost flexible small devices that offer great potential in human motion detection and x t a other clinical applications These sensors are the most commonly used wearable sensor in the field of activity recognition 81 109 Accelerometers applications include movement classification physical activity level assessment metabolic energy expenditure estimation and N assessment of balance gait and sit to stand transfers 81 Figure 2 8 Mass spring system Many of these applications use a single accelerometer attached to the waist Accelerometers were suggested to be a suitable tool for long term monitoring of free living subjects 81 Other applications in the rehabilitation field are gait analysis balance evaluation fall risk assessment and mobility monitoring 77 110 111 An accelerometer detects acceleration or deceleration a
108. ed to correct for the drifts Development of a Wearable Mobility Monitoring System 25 Literature Review 2 3 4 2 Gyroscope Gyroscopes sensors can measure angular rotation of body segments when attached to the segment with their axis parallel to the segment axis Gyroscopes that use a vibrating mechanical element to sense angular velocity have been used in mobility assessment applications 108 These sensors can measure transitions between postures by measuring the Coriolis acceleration from rotational angular velocity Unlike the accelerometer gravitational acceleration has no effect on gyroscopes Gyroscopes are often combined with accelerometers in human motion studies Some recent examples of their applications are in recording of human body segment orientation 113 identification of gait event for drop foot 114 calculation of 3D knee joint angles 115 and also in the detection of pre falls 116 The drawbacks of vibrating element gyroscopes are power consumption price drift and sensitivity to shock 109 2 3 4 3 Magnetometer Magnetometers can be used to measure a change in rotation of the body segment with respect to the earth s magnetic field The basic principle of these sensors corresponds to the magneto resistive effect which is the property to change the resistance with a change in magnetic induction Magnetometer sensors are sometimes combined with inertial sensors gyroscope and accelerometer to correct gyroscopes dri
109. edical Systems pp 1 9 2009 89 E Farella A Pieracci D Brunelli L Benini B Ricc and A Acquaviva Design and implementation of WiMoCA node for a body area wireless sensor network in Proceedings of the 2005 Systems Communications 2005 pp 342 347 90 S Farshchi P H Nuyujukian A Pesterev I Mody and J W Judy A TinyOS enabled MICA2 based wireless neural interface JEEE Transactions on Biomedical Engineering vol 53 pp 1416 1424 2006 91 A Milenkovi C Otto and E Jovanov Wireless sensor networks for personal health monitoring Issues and an implementation Computer Communications vol 29 pp 2521 2533 2006 92 E Mont n J F Hernandez J M Blasco T Herv J Micallef I Grech A Brincat and V Traver Body area network for wireless patient monitoring JET Communications vol 2 pp 215 222 2008 93 M R Yuce P C Ng and J Y Khan Monitoring of physiological parameters from multiple patients using wireless sensor network Journal of Medical Systems vol 32 pp 433 441 2008 94 E Farella A Pieracci L Benini L Rocchi and A Acquaviva Interfacing human and computer with wireless body area sensor networks The WiMoCA solution Multimedia Tools and Applications vol 38 pp 337 363 2008 95 S W Lee and K Mase Activity and location recognition using wearable sensors IEEE Pervasive Computing vol 1 pp 24 32 2002 Development of a Wearable Mobili
110. ee nennen eene ren rennen 90 37 E T a eeu eben ise utu deo ited edad 92 D XE c E NYA 93 7 5 Unused Features ote he tee i e ARE Meet e OE Ree 94 7 6 Determination of State and Change of State sese 95 7 Software development eer ede ete dte uence pe e T doe en ede edge kasia 99 T7 8 SUMMALY 22 1 ee tue adn iae denn 102 Chapter 8 Technical and Mobility Evaluation of the Prototype WMMS 103 8 1 EBechnical Evaluation 5 ne ect eere 103 8 2 Mobility Evaluation 3 enne ee e eit deserit ete nd 105 8 2 1 Sub Jectsi iet eae ned eee am e tee A E A AR TE R 105 8 2 2 Data Collection iit e te pte en reb ete ecce eee dett ie bela 105 8 2 5 Data Analysis 5t eu eH oae eee oe ete 110 8 2 4 Change of State Detection Results eese rennen rennen 114 8 2 5 BlackBerry Image Evaluation Results eee 118 8 3 Mobility Task Discussion eene eee eee ide eet aa 121 8 3 1 Use of Images in WMMS 0 eee iii nennen tenerent enne 121 8 3 2 WMMS Change of State Detection ener rennen 123 8 313 Exmitationg hen ee itt i Pee itt recent iere AUWA 126 Chapter9 WonclusiOn auo etes Eterna eta Edu aedi ades dedu goes 127 ON E t re Work 1o o5eend ditte rete reete debi eo ita akasha bassi 127 orga M EE 129 Appendix PMc scit ont enn duni cte uoc M ci LAU E MALLA EU DAI LU CU cedd 146 ADDendiz B tee ect ea il e et ea mu toa
111. ell with start forward walking progression Walk 15 meters a Initiation Exit stairwell with start forward walking progression b Termination End of forward walking progression Turn around a Initiation End of forward walking progression b Termination Facing opposite direction Transition turn around to walk a Initiation Facing opposite direction b Termination Start of forward walking progression Walk 15 meters towards the stairwell a Initiation Start of forward walking progression b Termination Start pushing on the door of the stairwell Open door and enter stairwell a Initiation Start pushing on the door of the stairwell b Termination Lead leg contacts a stair Walk down stairs 16 steps a Initiation Lead leg contacts a stair b Termination Trail leg off of last stair Walk on stair intermediate landing level ground for approx 1 5 meter a Initiation Trail leg off of last stair b Termination Lead leg contacts a stair Development of a Wearable Mobility Monitoring System 107 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 Technical and Mobility Evaluation of the Prototype WMMS Walk down stairs 13 steps a Initiation Lead leg contacts a stair b Termination Trail leg off of last stair Open door and turn right a Initiation Trail leg off of last stair b Termination Start forward walking progression outside the stairwell Walk 20 meters inside the Rehab T
112. ennneren 68 Gel Platform eiie hne e ttp edens 68 6 1 1 BlackBerry Bold Specifications and Features esee 68 6 2 External Board WI AA E o ei ere e e dieere 69 6 2 1 D sigen Critertaa te idee dete mte hades ree aula Defect ttg ede hades ree gabe REE rona 69 6 2 2 Parts Specifications iesise ectetur t eroe detiene iere pen angie 70 6 2 3 Board Functionality iis eee Reese 73 6 2 4 Packet FU uec EE Hee ee ente cos 73 6 2 5 Commands ee mL ee He CD d eh bee 73 6 2 6 Temperature and Humidity Sensors esee 74 62 T Light se SOEt eee Ip EE EE E ERE ERO du 74 6 2 8 Accelerometer CNDA t een tlt egre eden tege 75 6 2 0 Data Filtering ier Etpe eere irre tege counsel cane 78 6 3 Hardware Evaluations 4 3 once ebat A eer 78 6 3 Camera iei end Ue i E RUE Haee dee dente cos 78 GA SUIBIBAEy cbe m trei e Re e UE e ettet eh bete en 79 Development of a Wearable Mobility Monitoring System V Chapter 7 Development of the Prototype WMMS esses 80 TL Data Pre processimng 2 metit ri ge ect be tee bb P Ese loe eet Hoe Hl depo ena 80 7 2 Accelerometer Feature Generation eeeeeessesseseeeee eene enne 81 T 2 15 Inclination Angle eee b tee teg ete tete Put bee heben 81 1 23 22 Standard Deviation eed edet tte sonnei pee Sod dee ence de ee kae dnas 84 Hed SKEWNESS erae nea e Een 86 7 24 Signal Magnitude Area SMA sssssessssseesseeeeeeeeee
113. ent of a Wearable Mobility Monitoring System 50 Rationale Chapter3 Rationale As noted in Chapter 1 mobility deficits are a large and increasing problem in our aging society A decrease in mobility can reduce independence for activities of daily living produce deterioration in health status and diminish quality of life One of the main rehabilitation program goals is to achieve independent community mobility To understand how people move we must be able to measure mobility at home outside the home and in the community A better understanding of the challenges encountered in these three environments and the skills required to overcome these challenges can help healthcare providers make informed decisions that enable individuals to attain independent community mobility Unfortunately the current tools for measuring mobility outside of a laboratory or clinic are insufficient Therefore there is a need to develop assessment tools that can monitor mobility at home and in the community and provide insight on the context environment in which the activity takes place Current mobility assessment methods include observational and clinical tests diaries and questionnaires biomechanical and physiological measurement and activity monitoring Mobility assessment limitations are presented in Section 2 2 A wearable system approach for mobility assessment presents many advantages and allows a person s mobility to be measured anywhere Challen
114. entified from the pictures Depending on the mobility task context detection from the pictures was required to consider the context successfully identified Table 8 2 gives the list of the context to identify for each mobility task Context Indoor Outdoor Pave ment In a Car Door Ceiling Door Elevator Grass Unknown Figure 8 2 Example of the spreadsheet used by the pictures evaluators Table 8 2 Changes of state and context to be identified from WMM pictures Taking elevator to 1 floor Indoor elevator Walking to get out of elevator and keep Indoor floor walking on level ground Development of a Wearable Mobility Monitoring System 113 Technical and Mobility Evaluation of the Prototype WMMS diede re e landing level Walking on level ground Bia Mosca Md landing level Walking on level ground Stand to lie transition Lie to Stand transition Walking on level ground Walking on level ground Transition indoor outdoor and keep walking AA AA pavement on level ground a Transition outdoor indoor and keep walking indoarsfigar on level ground Transition indoor outdoor and keep walking Outdoor pavement on level ground p Walking on level ground Outdoor pavement Transition outdoor indoor and keep walking Indoor floor on level ground i 8 2 4 Change of State Detection Results For every trial WMMS sensitivity and specificity were calculated The average values for each su
115. ese false positives could be removed later with more offline processing Signal Magnitude Area SMA of acceleration signals versus Time 1 4 4 Lyingon the floor 1 2 4 Lyingon Getting up abed Getting up SMA g Sittin Getting up 0 4 down PeakThreshold Walking Walking High Threshold Low Threshold 0 10 20 30 40 50 60 70 80 90 Time seconds Figure 7 7 SMA of a person walking then sitting standing up walking lying down on a bed getting up from the bed lying on the floor and getting up again A DT algorithm was used to determine increases in intensity and peak detection Figure 7 8 illustrates the DT algorithm flowchart applied to the SMA feature When a peak was detected the next data window was not classified as a peak again until the signal went below the low threshold This avoided inappropriately switching from state peak to state no peak with increased in intensity and then to no peak with normal intensity since each windows is independently analysed However if the transition was slow and a change happens across windows it was possible to detect the state no peak with increase in intensity just before detecting the state peak These false positives could be removed later with more offline processing Development of a Wearable Mobility Monitoring System 9 Development of the Prototype WMMS During preliminary testing of the SMA algorithm it was also observed that the state
116. et al 143 for detection of various posture falls and gait disabilities used triaxial acceleration data taken at the abdominal level Wearing a single sensor at other locations rather than the center of mass region has been explored as well For example one sensor on the thigh has been used to study leg movement during walking 144 145 a triaxial accelerometer placed on the dorsum of the hand has recently been studied for the evaluation of Parkinson disease 146 Accelerometers placed at multiple locations on the body have also been used in many studies Table 2 2 One common configuration is having one accelerometer placed on the chest or trunk and one on the thigh This configuration has demonstrated capability in detecting sitting standing and lying and in detecting walking and postural transitions 147 150 Development of a Wearable Mobility Monitoring System 33 urojs g FUNUN AITIQOP AqL A Jo jueurdo oA oq vt Table 2 2 Example of laboratory and clinical studies using accelerometers for movement and mobility analysis List modified from Godfrey et al 77 Year 1997 1998 1999 Author Vetlink et al 147 Bouten et al 80 Bussmann et al 151 Foerster et al 152 Yoshida et al 153 Najafi et al 154 fiSensor placement sternum 1 thigh 1 waist lower back 2 upper legs 2 sternum HR 1 sternum 1 wrist upper thigh llower leg HR 1 centre of abdomen
117. evant information and or services to the user where relevancy depends on the user s task and any information that can be used to characterize the situation of an Development of a Wearable Mobility Monitoring System 29 Literature Review entity An entity is a person place or object that is considered relevant to the interaction between a user and an application including the user and application themselves In other Words context aware systems could monitor a user s activity location and physiological parameters and ambient conditions Then the system could adapt its behaviour based on the information Context awareness wearable systems have been used in activity and location recognition 12 14 15 in health pervasive environments 135 and in recognizing emergency situations by distinguish user motion states 13 Many context awareness approaches related to activity recognition use multiple sensors to recognize a wide range of activities However they also need more complex classification approaches such as artificial neural networks Bayesian networks and hidden Markov models 136 2 3 6 Summary of Wearable Systems In mobility monitoring a wearable system worn on the body can be used to continuously monitor biomechanical parameters regardless of the user s location Many social and technical challenges exist with wearable systems such as privacy and security power requirements portability acceptance and adhe
118. f the Prototype WMMS the camera program running When the full program for the WMMS was running the shortest time interval to take a picture was every 3 seconds Table 7 2 Results from images taken when walking on a ramp did not match the criterion level of accuracy Similar to the points made above possible reasons were the low light condition at the ramp s location and possibly the angle of view In addition when walking on a ramp it might not be possible to see an inclination especially if the image only shows a small section of the ramp As suggested for stairs descent short video and multiple pictures might contribute to better identification of the context The car was well identified from the images taken During mobility monitoring this could provide contextual information on the type of vehicle the person was using 1 e bus train car etc As an example if a person with mobility deficits takes the bus to go to the store or see friends instead of staying at home this could suggest some level of community mobility independence Our results from the image evaluation demonstrated that the walking surface i e floor pavement could be identified from the images From the study by Shummay Cook et al 21 terrain was one of the factors that differentiated an older adult with mobility disability and an older adult without such disabilities The type of terrain is also an important factor in accidental falls 1 e icy pat
119. fferent types of data that were collected further algorithms could be developed to expand on the types of activities and improve context recognition Development of a Wearable Mobility Monitoring System 3 Introduction 1 3 Overview of the thesis After the introduction Chapter 2 provides a literature review related to mobility assessment From that review Chapter 3 gives the rationale for this research Chapters 4 to 8 cover the methodology Chapter 4 starts with the design criteria for the development of a wearable mobility monitoring system and gives an overview of the development and evaluation process Chapter 5 covers a preliminary study that evaluated the BlackBerry smartphone as a hub for a WMMS Chapter 6 presents the hardware design and evaluation Chapter 7 describes the development of the WMMS including the algorithms and methods used to detect a change of state Chapter 8 presents the technical and mobility evaluation of the WMMS Finally Chapter 9 gives an overall conclusion of the thesis Development of a Wearable Mobility Monitoring System 4 Literature Review Chapter2 Literature Review This chapter reviews the literature on methods and technologies for monitoring and assessing a person s mobility This chapter is divided into four main sections community mobility and the importance of the environment in which mobility takes place 2 1 current mobility measurement methods and technologies 2 2 wearable technologie
120. for real time processing application since signal pre processing is not required to detect events or activity periods Section 2 4 5 The 1 02 seconds window size was chosen based on the work from Mathie et al 7 who found that the optimal window size for activity classification was between 0 8 to 1 4 seconds For temperature and humidity sensor data pre processing corresponded to the conversion of these two data into a temperature value in Celsius and a humidity value in percentage of Relative Humidity These conversions are explained in Section 6 2 6 Filtering was not required for the temperature and humidity data since these values were only updated every 4 seconds 0 25 Hz As for the light sensor the non overlapping sliding window of 1 02 seconds was applied to the light sensor which acted as a moving average filter 7 2 Accelerometer Feature Generation 7 2 1 Inclination Angle The inclination angle was added to the algorithm to help classify posture 9 147 149 and identify postural transition 155 For this prototype WMMS the posture was either standing lying on the back or somewhere in between e g sitting Development of a Wearable Mobility Monitoring System 81 Development of the Prototype WMMS The static components of the acceleration signals which were obtained from the RC low pass filter were averaged over the 1 02 seconds window The inclination angle was calculated for every window period The angle calculati
121. ft about the vertical axis 117 However a drawback of magnetometers is their sensitivity to nearby iron and local magnetic fields Magnetometers also need to be calibrated for any change of location 109 2 3 4 4 Foot Pressure Pressure sensors or foot switches can be used to measure gait temporal parameters when attached to the sole 118 The pressure is measured from the force deformation properties of a specific material For instance the deformation caused by pressure can be measured from capacitance and resistance changes where both decrease with compression Another example is piezoelectric polymers that generate more charge with compression 119 Their applicability in pathological gait is limited by many problems including the inability to measure shear forces calibration issues sensors change calibration when bent or Development of a Wearable Mobility Monitoring System 26 Literature Review due to temperature effects difficulty with sensor positioning and to connect attachments mechanical failure and subject acceptance 109 2 3 4 5 GPS The global positioning system or GPS consists of a constellation of 24 satellites plus 6 spare ones orbiting the earth and continuously sending signals to ground stations A GPS receiver will detect several GPS satellite signals and will calculate how far they are by comparing the time the signal was sent from the satellite and the time the signal was received Using the triangul
122. g the Java command Devicelnfo getBatteryLevel This Java command was called every minute inside the WMMS application to verify the battery level of the BlackBerry Five trials were run and the results are presented in Table 8 1 The starting and ending battery levels were the first and last battery level value captured during a trial respectively Total battery usage was calculated by subtracting the ending level from the starting level and dividing by the starting level Then the battery usage per hour was calculated by dividing the total battery usage with the total time of the trial The battery usage averaged 29 per hour Figure 8 1 give an example of one of the battery voltage curve obtained during this evaluation Table 8 1 presents the trials results During battery tests data loss was also evaluated No data loss were observed in any of the trials Development of a Wearable Mobility Monitoring System 103 Technical and Mobility Evaluation of the Prototype WMMS Table 8 1 Results for the BlackBerry Bold battery evaluation Starting Ending Total Time Total Battery Usage Battery Level Battery Level hours Battery per hour 96 96 Usage Yo hour 100 58 1 54 42 27 100 2 3 23 98 30 99 6 3 20 94 29 100 6 3 12 94 30 100 6 3 12 94 30 Battery usage curve of the BlackBerry Bold versus time Battery Level 96 Time hours Figure 8 1 BlackBerry battery with full WMMS app
123. ges encountered with wearable systems include their portability power consumption privacy and security acceptance and adherence Section 2 3 6 Recent technological advances in sensor miniaturization wireless communication power consumption smartphones and handheld devices have helped overcome many of these challenges These advances lead to the development of wearable systems that detect and recognize a person s activity and provide contextual information However many of the reviewed studies involving both activities and context measurement were not intended for mobility monitoring of a person with physical disabilities Section 2 3 6 Development of a Wearable Mobility Monitoring System 51 Rationale Smartphones are considered a viable wearable system platform to monitor mobility in the community Such phones are small lightweight and have good battery life sufficient processing power large memory capacity and multiple networking capabilities These phones can also include technologies appropriate for mobility monitoring such as a camera GPS and accelerometer The advantages of using accelerometers in mobility monitoring have been well documented Section 2 3 4 1 Light humidity and temperature sensors can also be included in the wearable system to add more details on weather and ambient condition However the use of the camera video for wearable context sensitive mobility assessment has not been previously reported Wearable
124. h unlevel ground Injurious falls are related to many health problems and are a leading cause of hospitalization in the elderly 194 Adding instability detection and capturing information on the type of terrain could be a valuable feature for a WMMS to help understanding the underlying causes of falls and help with fall prevention The use of images to capture context and environment in mobility monitoring could also help to monitor activity avoidance Mobility disability has been characterized by a reduction in the number and type of environment challenges 60 Activity avoidance could lead to a reduction of movement which could lead to further deterioration in physical status and social interactions Development of a Wearable Mobility Monitoring System 122 Technical and Mobility Evaluation of the Prototype WMMS 8 3 2 WMMS Change of State Detection Some of the methods used in this thesis to identify a user s state replicate results from previous studies For instance Lyons et al 149 obtained an accuracy of 97 to detect static or dynamic states using the standard deviation of the vertical axis of a thigh accelerometer For our WMMS we used the standard deviation of the vertical acceleration at the waist and were able to detect if the subject started stopped moving with a sensitivity of 97 4 5 396 This is a good result considering that the device holster was worn on a belt and not fixed still to the person s body This finding
125. h mobility issues has increased from 10 5 to 11 since 2001 most likely due to Canada s aging population 2 Mobility disabilities can affect an individual s quality of life health productivity independence and also affect the lives of their family and the people around them Preserving mobility is paramount in order to stay independent and active at home and in the community Accurate mobility assessment is required for decision making in rehabilitation medicine Such assessments can be used to determine mobility issues outside a hospital environment evaluate the progress made during and after rehabilitation and enhance clinical decision making about a rehabilitation program i e assistive devices exercises treatment etc Currently many different types of mobility assessments are performed in clinical setting and are supervised by the rehabilitation physician These assessments include clinical tests quantitative measures and subjective feedback from the client Although clinical mobility tests have their value these easy to apply assessment tools may not be appropriate for determining the contributing factors for independent community walking and the impact of the environment on the individual s mobility 3 4 Monitoring the mobility outside a clinical setting is important because mobility in the real world is typically different from the mobility measured in the clinic 5 Wearable technology can be developed to evaluate mob
126. he night light off 17 3 0 5 Indoor during the night light on 28 3 13 5 Pitch dark in black box 17 3 0 5 6 2 8 Accelerometer Calibration A variable capacitance accelerometer which has the property to measure both DC and AC acceleration was used for the WMMS Section 2 3 4 1 An advantage of measuring DC acceleration is the ability to calculate inclination angle However having a DC component creates a signal offset which as mentioned by Bouten et al 80 should be corrected to avoid over or under estimation of the measured acceleration The other calibration parameter necessary for the acceleration calculation is sensor sensitivity Sensitivity describes the accelerometer gain Despite the factory calibration for offset and sensitivity re calibration was recommended after mounting the sensor onto the board because this process could have modified the factory values Re calibration also defined the orientation of the accelerometer axes with respect to the external board axes Development of a Wearable Mobility Monitoring System 75 Hardware Design and Evaluation Accelerometer sensitivity and offset values for each axis x y z were calculated prior to the WMMS evaluation The calibration method was described on the manufacturer datasheet 166 The method is described here using the x axis as an example the same procedure applies to y and z axis The board was oriented such that its x axis was pointing in the opposite dire
127. he walking around obstacles task from the original DGI was removed since this task was considered to be of insufficient difficulty The FGA demonstrated similar reliability to the DGI and was considered to have acceptable reliability and validity as a clinical gait measure for patients with vestibular disorders 29 2 2 1 3 Community Balance and Mobility Scale CB amp M The Community Balance and Mobility Scale CB amp M was designed to evaluate balance and mobility in high functioning ambulatory patients who have persistent balance problems 30 CB amp M is a multiple components test that measures performance on thirteen physical tasks unilateral stance tandem walking 180 degree tandem pivot lateral foot scooting hopping forward crouch and walk lateral dodging walking and looking running with controlled stop forward to backward walking walk look and carry descending stairs step ups x 1 step This measure was a reliable and a valid scale for the traumatic brain injury population 31 but could also be appropriated for clients with other diagnoses 32 2 2 1 4 Berg Balance Scale The Berg Balance Scale BBS is a 14 item clinical tool developed to measure functional balance in an older population 33 The items include a sitting task transfer tasks sitting to standing standing to sitting and other standing tasks unsupported with eyes closed with feet together tandem on one leg and other mobility tasks turning trunk
128. hics file no H 09 09 15 Evaluation of a Wearable Mobility Monitoring System Dear Ms Hach Dr Lemaire and Ms Baddour Thank you for the protocol documents and the Certificate of Approval from the Ottawa Hospital REB This is to confirm that in accordance with the agreement between the University of Ottawa and The Ottawa Hospital the University of Ottawa has authorized the Ottawa Hospital REB to act as Board of Record for the review and oversight of research involving human subjects conducted at or through the hospital Copies of annual reports and renewals of Ottawa Hospital REB approvals must be provided to our office We remind you of your obligation to Follow all procedures of the Ottawa Hospital REB including reporting and renewal procedures Submit to the authority of the Ottawa Hospital REB and that you are subject to Ottawa Hospital REB requirements including without limitation the requirement to modify or stop the research on demand of the Ottawa Hospital REB If you have any questions please contact our ethics office at 562 5841 Development of a Wearable Mobility Monitoring System 163 Appendix E Universit d Ottawa University of Ottawa Sincerely yours Catherine Paquet Assistant director Ethics Development of a Wearable Mobility Monitoring System 164 Appendix E m Fh a CU a s nivieiity Of SHAWA HEART INSTITUTE Cad uOttawa INSTITUT DI CARDIOLOGIE bt CONWEA
129. hysical disability walking speed or gait pattern showed minimal change from walking on Development of a Wearable Mobility Monitoring System 124 Technical and Mobility Evaluation of the Prototype WMMS level ground to walking on the ramp observed from video data The ramp inclination angle was also moderate approx 7 degree angle In older populations or individuals with mobility disabilities a slow almost stopping movement could be present before attempting walking up a ramp or even stairs As mentioned earlier our WMMS was accurate in detecting static and dynamic movement therefore a picture could be taken to help identifying the mobility task Change in posture angle could be explored since pelvic tilt may be present as the person leans forward and backward during ramp ascent or descent A change in height such as proposed for stairs ascent could be appropriate for larger inclines i e hill Adding other sensors could be explored as well Sensors on the thigh or even the calf might give more biomechanical information when walking on a ramp The light sensor was added to the WMMS to detect outdoor and indoor conditions Our approach of selecting outdoor indoor thresholds did not perform as well as anticipated A change in light intensity level could have been a better measure instead of using fixed outdoor indoor thresholds since changes could be detected on overcast cloudy days The smartphone approach worn at the waist might also have
130. i et al 102 the diversity of mobile devices decreases the portability of Java ME applications Some of the causative factors are the different device features memory size limitations function additions and deletions and device specific bugs 102 Custom made hubs have also been developed for wearable mobility monitoring Dalton et al 103 developed a mobility monitoring portable system that included a Global System for Mobile communications GSM modem and used short message service SMS to send accelerometer data to a remote server for further analysis and data storage For other monitoring systems that do not use GSM networks data loss could occur when the system devices are out of range of their receiver station However with a GSM modem Dalton s system did not suffer from this type of data loss In the development of their WBAN Mont n et al 92 designed a personal data processing unit PDPU for their hub Advantages of PDPU are a better control of the device ability to use the best wireless standards and elimination of the other applications that a cell phone provides but are not required for the monitoring application The disadvantages are the resources time and money it takes to design such a system 2 3 3 Wireless Standards Three popular wireless standards are typically used in WBAN design Bluetooth ZigBee and Wi Fi These three standards operate in the unlicensed 2 4 GHz spectrum called ISM band industrial sc
131. ication and Regression Trees Community Balance and Mobility Scale Custom Decision Tree Chronic Obstructive Pulmonary Disease Cyclic Redundancy Check Continuous Wavelet Transform Direct Current Dynamic Gait Index Double Threshold Discrete Wavelet Transform The Environmental Analysis of Mobility Questionnaire Electrocardiogram Energy Expenditure Energy Expenditure due to physical activity Functional Assessment Measure Fast Fourier Transform Functional Gait Assessment Functional Independence Measure Functional Status Questionnaire Global Positioning System Global System for Mobile communications Health Assessment Questionnaire Hidden Markov Model Instrumental Activity of Daily Living Development of a Wearable Mobility Monitoring System IBL ICF IEEE ISM J2ME MEMS MMS NN PA PDA PDPU RSS RTM SMA SMS SMV STDY SVM UWB WBAN WBSN WLAN WMMS Instance Based Learning The International Classification of Functioning Disability and Health Institute of Electrical and Electronics Engineers Time integrals from separate measurement direction Industrial Scientific and Medical Band Java 2 Micro Edition Micro Electro Mechanical System Multimedia Messaging Service Neural Network Physical Activity Personal Digital Assistant Personal Data Processing Unit Root Sum of Square Rotation Transformation Matrix Signal Magnitude Area Short Message Service Signal Magnitude Vector Standard Deviation of Y axis
132. ie stand Activity retest standing sitting lying back on walking running upstairs downstairs Mobility monitoring of elderly in clinical environment stroke patient sit stand lying postures Various length of median filter window widths and thresholds mean energy expenditure integral area Wavelet transform DWT thresholds visual observation Mean energy frequency domain entropy correlation of acceleration data classifiers C4 5 decision tree decision table naive Bayers classifier instance based learning IBL Kalman filtering optical reference system Vicon Best estimate mid point thresholds mean standard deviation observed comparison 1 minute resolution Mean standard deviation skewness kurtosis eccentricity histograms neural networks Means and standard deviations thresholding best estimate and mid point comparison with manual recordings of patient activity MMY ANPI urojs amp s SuuojmuoJA i IqoJA 9 qe1e9 AA L JO juoeuido oAo q 9g Barralon et al 158 Postural states walking postural transitions Angles inclinations frequency analysis FFT thresholds video 2006 Barralon et al 159 2006 NiScanaill et al 150 2006 Hester et al 160 2006 Parkka et al 161 chest under arm Walk 76 pit postures 8096 1 under left arm pit DWT 78 5 sensitivity 67 7 specificity 1 trunk 1 thigh 1 wrist 1 ankle
133. ientific and medical band Another common wireless standard is ultra wideband UWB but it is less popular in the design of WBSN Table 2 1 summarises the different standards ZigBee was designed specifically for control and sensor networks This standard is intended for short range communication and is characterized by very low power consumption A ZigBee node can run on batteries for several months or years Data rate is limited to Development of a Wearable Mobility Monitoring System 21 Literature Review 250Kbps in the global 2 4 GHz spectrum ZigBee also operates at the 915 MHz America and 868 MHz Europe spectrum ZigBee appears to be a promising wireless standard for WBAN 84 92 Compared to Bluetooth ZigBee is less complex and consumes less power ZigBee is also less prone to interference with other devices in the same frequency range 85 Table 2 1 Comparison of different features of common wireless technologies 85 104 Para meters Bluetooth IEEE UWB WiMedia or ZigBee Wi Fi IEEE 802 15 1 IEEE 802 15 3 8 802 11 Battery Life Days Days Years Hours Cost per 6 6 3 9 Module Complexity of Complex Simple Simple Very Complex Mac and physical layer Radio spectrum 2 4 GHz 3 1 10 6 GHz 868 MHz 915 MHz 2 4 GHz 2 4 GHz Maximum data 3 Mbps 1 Gbps 250 Kbps 54 Mbps rate 7 nodes Unknown 64000 nodes 32 nodes 64 128 bits 128 bits AES 128 bits AES WEP keys Application Low bandwidth cable replacement High ban
134. ies of features are heuristic features time domain features frequency domain features and time frequency domain Usually the time domain features do not required as much processing power as the frequency analysis methods which is important when designing real time portable application using low power and memory devices However frequency domain features have the advantage of detecting cyclic motion such as in walking and running Features showing both time and frequency characteristics can also be obtained from wavelet analysis methods However wavelet analysis may be inferior to frequency domain features to detect dynamic activity Data transfer to a personal computer is often required to perform more advanced signal processing techniques and to better analyze the signal 9 155 160 After a set of features have been generated and selected they can be used as inputs for a classification algorithm Simple algorithms based on threshold and hierarchical tree configurations have been successfully used to detect different activities postures falls etc These methods are often implemented in applications using low memory and processing power devices Other advanced methods have been used such as decision tree k nearest neighbor support vector machine neural network naive bayes fuzzy logic and Markov chains Many of these methods have demonstrated good classification accuracy but may require more processing power or training data Developm
135. ii Figure 5 1 System architecture for the preliminary testing eene 60 Figure 5 2 Sensor placement for the calculation of biomechanical parameters 61 Figure 5 3 Sensor Pl ACCUSING sa al Ss Soca tol ones oic nt boc eee T a E Siete 63 Figure 6 1 Front side and back view of BlackBerry Bold 181 sse 68 Figure 6 2 Block diagram of the external board eoe eed tes 70 Figure 6 3 Image of the board with all the sensors identified ssessss 71 Figure 6 4 Examples of the drift acceleration versus time for x y and z axis 77 Figure 7 1 Inclination angle measurement method In standing position inclination angle is TS0 HOT AI vmm msc pubes E tera duos ANA KA 82 Figure 7 2 Position classification method ose ee Ba eae 83 Figure 7 3 Flowchart of the double threshold DT algorithm applied to the standard deviation OF the y axis aceeleratiQfi u i cues esee et ue evaneagsacaaaes LOPR eU engages wwesadeatuanecuantemateaee 85 Figure 7 4 Standard deviation of y axis acceleration during level ground walking dynamic followed by a short period of standing static and then back to walking 86 Figure 7 5 Algorithm flow chart for skewness of y axis acceleration 88 Figure 7 6 Example of a skewness curve for y axis acceleration The top graph is the skewnes
136. ility in any location or environment Wearable mobility monitoring systems are designed to be worn on the body and allow mobility monitoring in the person s home and the community 6 Development of a Wearable Mobility Monitoring System 1 Introduction Many wearable mobility monitoring studies measure biomechanical and or location parameters 5 7 10 but most lack environmental or contextual information In community mobility monitoring contextual information is important since it could provide insight on where how and on what a person is moving A camera could provide contextual information from a person s surrounding environment Example of wearable systems that use contextual information are context aware systems 11 and life logs 12 but they are not meant for community mobility monitoring for people with physical disabilities Some context aware wearable systems use context information to better recognize activities 13 15 but the environmental characteristics in which activities take place are not analyzed for their impact on mobility There is a need for an assessment tool that could monitor mobility within the home envi ronment and the community for a long period and provide information on the context in which mobility occurred This tool could help clinical professionals and rehabilitation researchers to determine appropriate training to enhance mobility in the community and could help identify mobility challenges The
137. ility telemonitoring of the elderly in their living environment Annals of Biomedical Engineering vol 34 pp 547 563 2006 7 M J Mathie A C F Coster N H Lovell and B G Celler Detection of daily physical activities using a triaxial accelerometer Medical and Biological Engineering and Computing vol 41 pp 296 301 2003 8 D A Rodr guez A L Brown and P J Troped Portable global positioning units to complement accelerometry based physical activity monitors Medicine and Science in Sports and Exercise vol 37 pp S572 S581 2005 9 D M Karantonis M R Narayanan M Mathie N H Lovell and B G Celler Implementation of a real time human movement classifier using a triaxial accelerometer for ambulatory monitoring IEEE Transactions on Information Technology in Biomedicine vol 10 pp 156 167 2006 10 E Farella A Pieracci L Benini and A Acquaviva A wireless body area sensor network for posture detection in 71th IEEE Symposium on Computers and Communications ISCC 2006 2006 pp 454 459 Development of a Wearable Mobility Monitoring System 129 References 11 C Randell and H Muller Context awareness by analysing accelerometer data in The Fourth International Symposium on Wearable Computers 2000 pp 175 176 12 Y Lee and S B Cho Extracting meaningful contexts from mobile life log in Intelligent Data Engineering and Automated Learning IDEAL 2007 2007 pp 750 759
138. ing data to generate the decision tree Bao and Intille 156 have compared different classifiers such as decision tables instance based learning C4 5 and naive bayes C4 5 had the best overall recognition accuracy of 84 for the detection of 20 daily activities The custom decision tree automatic generated tree CART and neural network were explored by Parkka et al 161 The custom decision tree had the best classification results in recognizing most activities except walking and biking but overall the automatic decision tree had a better result total of 8696 compared to 82 for custom tree and 82 for neural network The K nearest neighbour approach for classifying activity was first used by Foester et al 152 With the k nearest neighbour a feature space is created from training data points Each data point corresponds to a particular activity An unknown window of sensor data can be classified by finding which training data point is the closest in the feature space Although this method could detect a wide range of different activities the execution time is slower than the decision tree 171 In addition in the study by Bao and Intille the k nearest neighbour obtained lower recognition accuracy than the decision tree approach Lau et al 176 demonstrated the high performance and consistency of the support vector machine SVM to classify different walking conditions using accelerometer and gyroscope sensors Preece et al 17
139. ing in a car the vehicle context was identified at 86 796 The pictures taken during the start of the car ride obtained a result of 100 096 For the end of the car ride after the car stopped pictures was not always taken while sitting in the car due to the GPS sampling rate i e the 9 second GPS analysis interval created a delay where the picture would be taken after the person left the car and was already starting to walk Therefore the evaluators had to identify the type of ground for those particular images The success rate was 84 4 Identifying the context from the images taken during the transition from outdoor indoor were low at 57 1 for the first time going inside and 37 5 for the second time The success rate for the indoor outdoor transitions were better with 85 7 for the first time going outside and 100 096 the second time The low results for the transition outdoor indoor could be caused from the decreased light intensity when approaching the door of the building from outside Development of a Wearable Mobility Monitoring System 118 Technical and Mobility Evaluation of the Prototype WMMS Outdoor indoor transitions sometimes happened before the person actually stepped inside which made identifying indoor or outdoor very difficult Table 8 5 Summary results for the picture evaluation Total Successfully identifying context Change of State ane Evaluator Evaluator A Standard Pictures 1 2 verage deviation 93 3
140. ing level ground walking dynamic followed by a short period of standing static and then back to walking 7 2 3 Skewness One of the changes of state that the WMMS was aiming to detect was going up or down stairs The skewness value of the vertical acceleration is a time domain feature that was used by Baek et al 141 to differentiate walking running from going up down stairs The skewness of the y axis was calculated as follows LA 7 6 n NAA skewness n Da 2 A where n is the number of point x the y axis acceleration at point i and o and x are the standard deviation and the mean of the y axis acceleration signal respectively Equation 7 6 can be rearranged as Equation 7 7 for programming purposes Development of a Wearable Mobility Monitoring System 86 Development of the Prototype WMMS 7 7 Y 3xY x 2nx n i l i l n 1 n 2 o skewness Figure 7 6 gives an example of the signal when walking and when walking up and down stairs The top curve shows the skewness curve only The bottom curve shows the same skewness curve but with the dynamic level identified by the dashed line A dashed curve value of 2 means the desired dynamic level was reached and the stairs detection algorithm determined if the state was stairs or no stairs If the dashed curve value was 0 the state was determined as no stairs Based on preliminary work a skewness value larger than 1 was observed when a person
141. ing mechanical work and work efficiency during human activities Journal of Biomechanics vol 26 pp 229 241 1993 166 STMicroelectronics MEMS Inertial Sensor High Performance 3 Axis 2 6g Ultracompact Linear Accelerometer LIS344ALH Datasheet Rev 3 Geneva Switzerland S TMicroelectronics 2008 167 J C L tters J Schipper P H Veltink W Olthuis and P Bergveld Procedure for in use calibration of triaxial accelerometers in medical applications Sensors and Actuators A Physical vol 68 pp 221 228 1998 168 I Frosio F Pedersini and N A Borghese Autocalibration of MEMS accelerometers IEEE Transactions on Instrumentation and Measurement vol 58 pp 2034 2041 2008 169 T Mineta S Kobayashi Y Watanabe S Kanauchi I Nakagawa E Suganuma and M Esashi Three axis capacitive accelerometer with uniform axial sensitivities Journal of Micromechanics and Microengineering vol 6 pp 431 435 1996 170 X Yun E R Bachmann H Moore IV and J Calusdian Self contained position tracking of human movement using small inertial magnetic sensor modules in Proceedings of the IEEE International Conference on Robotics and Automation 2007 pp 2526 2533 171 S J Preece J Y Goulermas L P J Kenney D Howard K Meijer and R Crompton Activity identification using body mounted sensors A review of classification techniques Physiological Measurement vol 30 pp R1 R33 2009 172 F
142. ing studied 81 Accelerometers have been attached to different parts of the body and in various numbers depending of the application In studies using a single location to study whole body movement the sensor is usually placed as close as possible to the center of mass e g trunk under arm waist One reason for this placement is that the body parts in that region move during most daily activities 80 Bouten et al 80 studied accelerometer placement at the trunk for physical activity assessment Studies by Sekine et al 138 139 demonstrated that walking on level ground and walking on stairways could be distinguished with a single waist mounted accelerometer Work from Mathie et al 7 140 and Karantonis et al 9 showed that with only a waist mounted triaxial accelerometer it is possible to detect between periods of rest and activity and also to identify postural orientation falls and estimate energy expenditure Using a two axis accelerometer worn at the waist Baek et al 141 was able to obtain an overall Development of a Wearable Mobility Monitoring System 32 Literature Review classification rate of 97 596 for activities such as standing sitting lying walking running upstairs and downstairs The discrimination of falls from activity of daily living using a single triaxial accelerometer worn at the trunk was successfully 100 demonstrated by Bourke et al 142 A wearable surveillance system developed by Yoshida
143. integer values were calibrated median filtered and divided into the static and dynamic component using a low pass filter Calculation of different variables necessary to compute features as well as integration of acceleration signal were performed as well When all the received bytes were processed then more bytes were received on the Bluetooth port and the same process started again until the number of samples reached the selected window size When one window of data was processed other types of processing were performed From the variables computed the features were calculated Then these features were passed through the algorithm to determine the state and change of state of the user From the change of state result another Java function determined if a picture should be taken Finally an output sample object was created which contained all the features computed image name user state GPS data and time frame This sample object was put in a circular queue which was emptied by a separate thread that copied the data to an output file stored on the BlackBerry SD card Development of a Wearable Mobility Monitoring System 100 Development of the Prototype WMMS Main Program C stan D Output Raw Data Flag true Processing Flag true Thread 1 Create Create Thread2 k Data received on F ii Bluetooth port i Copy data A processingresults including state E Copy raw d
144. inusoidal functions and then averaging From the FFT output Bao and Intille 156 extracted the energy sum of the squared FFT coefficient and the frequency domain entropy normalized information entropy of the FFT components The dominant frequencies in the signal have also been observed by Barralon et al 158 and Hester et al 160 Frequency domain features give information about the frequency components contained in a signal however they do not provide the time at which those components occurred Information on signal time and frequency content is important in signal analyses where frequency changes over time e g human movement Using wavelet analysis time frequency features can be used to investigate both time and frequency characteristics Similar to the Fourier transform the use of wavelets also requires signal decomposition into simple elements but it is more efficient than the Fourier transform for signals dominated by transient behaviour or discontinuities such as human movement 155 Wavelet transforms also use simple basis functions instead of a sinusoidal signal A variety of time frequency Development of a Wearable Mobility Monitoring System 45 Literature Review features using wavelet transform is presented in Preece et al 171 174 Preece et al 174 found that wavelet analysis was not as accurate as the frequency domain features for classifying dynamic activities although wavelet analysis can be used to cha
145. ion 4 1 However the results from this preliminary study showed that the BlackBerry s battery might last for less than seven hours This issue could be resolved by upgrading the battery to a larger capacity Development of a Wearable Mobility Monitoring System 66 Preliminary Evaluation of the BlackBerry for WMMS Java programming problems with conversion of float numbers to a string resulted in excessively long execution times causing the Xbus Master to stop sending data String conversion was required for data display purposes To solve this problem integer numbers were used instead of float numbers The conversion of integer to string was less time consuming for the Java application 5 7 Summary A proof of concept system that calculated biomechanical parameters of the human body was created The objective was to evaluate the BlackBerry as a Wearable Mobility Monitoring System platform The BlackBerry device demonstrated capability and good potential as a WMMS hub Many of the problems encountered during data collection were due to the motion capture system Thus the choice of external sensors for long term monitoring should be made with care Based on this analysis proceeding with BlackBerry as a development and WMM application platform was supported Development of a Wearable Mobility Monitoring System 67 Hardware Design and Evaluation Chapter 6 Hardware Design and Evaluation 6 1 Platform The BlackBerry 9000 Bold
146. ion Tracking System Non visual motion tracking systems do not use cameras to detect human motion Inertial sensor based systems are a commonly used non visual system These systems are based on inertial sensors such as accelerometers and gyroscopes biomechanical models Figure 2 4 Motion track MTx from Xsens Technologies Kit XSens Motion Technologies Netherlands which reproduced from 65 and sensor fusion algorithms An example is the XBus consists of a portable unit XBus Master collecting data from multiple or single motion tracker devices MTx 65 MTx Figure 2 4 are attached to different body segments and can measure 3D rate of turn acceleration and earth magnetic field These data are combined using a Kalman Filter technique to calculate 3D orientation of the MTx unit A literature survey by Zhou and Hu 63 provides more details on these Development of a Wearable Mobility Monitoring System 14 Literature Review systems as well as other sensing techniques used for non visual motion tracking systems including magnetic acoustic ultrasonic EMG and data gloves 2 2 3 3 Force Plates Force plates also called force platforms are the most common force transducers in gait analysis This instrument consists of a plate flush with the ground instrumented with strain gauges or piezoelectric transducers and measures 3D ground reaction forces and moment as the subject makes contact with the plate Force plates a
147. ionality The board power is turned on by flipping a switch installed on the board To start and stop sampling of the sensor data commands that set the sampling delay are sent to the microcontroller Communication with the external board is done via Bluetooth or the debug serial port Data from the accelerometer and the light sensor are first sampled by the microcontroller at a rate of 130 Hz The temperature and humidity sensors are sampled by the microcontroller at 0 25 Hz These data are stored in a buffer on the microcontroller Then at every sampling delay the last data stored in the buffer are sent to the host BlackBerry via Bluetooth In this thesis the sampling delay was set to 20 ms 50 Hz 6 2 4 Packet Format The external board sends a 21 bytes data packet to the host BlackBerry or personal computer using Bluetooth Serial Port Profile SPP protocol or RS232 serial protocol Header Packet Type Packet Length Sample X axis Y axis 2 bytes 1 byte 1 byte Number Acceleration Acceleration 1 byte 2 bytes 2 bytes Z axis Light Intensity Temperature Humidity Battery CRC Acceleration 2 bytes 2 byte 2 byte Voltage 2 bytes 2 bytes 2 bytes The header bytes are 0xC3 and 0x42 The packet type can be either 0x01 for data packet or 0x02 for control packet All the sensor data are sent to the host as integer values 2 bytes 6 2 5 Commands Commands available
148. is Z axis is pointing out X axis Y axis Z axis is pointing out X axis X axis Right Side Left Side Figure 5 2 Sensor placement for the calculation of biomechanical parameters Development of a Wearable Mobility Monitoring System 61 Preliminary Evaluation of the BlackBerry for WMMS Both sensors have a rotation matrix relative to the global coordinate system G R 245 24 1 24 4 2q4 4 24 43 2409 SR 2q 9 2409 295 2q3 1 24 4 24qq 5 1 24 93 2909 2439 2409 245 2q 1 where 4 4 4 4 are the quaternion numbers of one MTx sensor The subscript S represents the sensor coordinate system and G the global coordinate system The RTM for one joint i e knee or hip is then calculated with matrix manipulation RTM R R proximal S distal 5 2 KR Ry Rg RTM i R R Ry R4 5 3 Ra Ry Ry where S proxima and Sisar represent the coordinate systems of both the proximal and distal TO sensors respectively ind is the rotation matrix of the distal coordinate system relative to the proximal coordinate system From the resulting RTM the Euler angles can be calculated al R Prei tan 1 192 5 4 distal Ree S proxima S 1 en 9 EE R4 5 5 ab R 5 6 menia yy tan amp distal R 7 Development of a Wearable Mobility Monitoring System 62 Preliminary Evaluation of the BlackBerry for WMMS The Euler angles 0 v are also called roll pitch and yaw respectively Roll is
149. ive learning algorithm for constructing neural classifiers Pattern Recognition Letters vol 29 pp 2213 2220 2008 179 Steven D Kaehler Fuzzy Logic An Introduction Part 1 Encoder The Newsletter of Seattle Robotics Society Available http www seattlerobotics org Encoder mar98 fuz fl partl htmI INTRODUCTION Accessed 12 Apr 2009 180 Y P Chen J Y Yang S N Liou G Y Lee and J S Wang Online classifier construction algorithm for human activity detection using a tri axial accelerometer Applied Mathematics and Computation vol 205 pp 849 860 2008 181 Wikipedia Markov Chain Wikipedia The Free Encyclopedia Online Available http en wikipedia org wiki Markov chain Accessed 12 Oct 2009 182 J He H Li and J Tan Real time daily activity classification with wireless sensor networks using Hidden Markov Model in Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society 2007 pp 3192 3195 183 J Hamill and W S Selie Joint angles in Research Methods in Biomechanics D E Robertson G E Caldwell J Hamill G Kamen and S N Whittlesey Eds Champaign Illinois Human Kinetics 2004 pp 45 51 184 Xsens Technologies B V MT Low Level Communication Documentation Document MTO010IP Revision H The Netherlands Xsens Technologies B V 2008 185 Xsens Technologies B V MTi and MTx User Manual Document MTO100P Revision K
150. l to detect if the person is in a vehicle such as car bus train and so on This feature was passed through a DT algorithm such as the one used for standard deviation The low threshold value was set to 1 m s and the high threshold value was set to 7 m s With this algorithm a change of state could be triggered when the car stops at a stop sign or slows down sufficiently However since the GPS data is refreshed every 9 seconds the algorithm might miss some stopping instances This could help in decreasing false positive changes of state while riding in a car 7 5 Unused Features Other features have been generated from the accelerometer data but were not used in the algorithm to detect changes of state The correlation between x and y y and z and x and z were generated The correlation values have been used by Ravi et al 96 since these features could detect activities that involve translations in one dimension i e differentiation walking from going up down stairs However in our research with a window of 1 02 seconds correlation values did not help to detect stairs In the work from Ravi et al the correlation values were calculated over a window of 5 12 seconds This window size was not adequate for our research since we wanted real time processing Further data processing using correlation values could be done offline in the future The skewness value of the forward axis z axis and the kurtosis of the vertical axis y axis
151. lation 16 The complexity of the person s environment found within and outside of the home cannot be fully represented by these tools Laboratory based instruments to measure biomechanical parameters are usually very accurate but are limited by space requirements setup time setup capabilities i e may not accommodate stairs inclines uneven ground etc and cost Therefore motion laboratory systems are seldom used for community mobility analysis applications Activity motoring instruments have the advantage of being wearable and can monitor mobility for a long time in the person s own environment However they usually measure one aspect of physical activity and they do not have information on where the activity took place i e context Development of a Wearable Mobility Monitoring System 17 Literature Review 2 3 Wearable Mobility Monitoring Systems A wearable system is designed to be worn on the body and allow continuous monitoring of biomechanical and physiological data regardless of the user s location while he or she goes about their normal daily activities 6 72 83 Some advantages of using wearable systems to measure mobility are direct access to biomechanical parameters data logging and processing can be done anywhere and technological advances are leading to a reduced size weight and cost 6 Compared to laboratory based systems wearable technologies take less setup time since multiple sensors and equipme
152. lication running Trial 2 Development of a Wearable Mobility Monitoring System 104 Technical and Mobility Evaluation of the Prototype WMMS 8 2 Mobility Evaluation 8 2 1 Subjects A sample of five subjects 3 males 2 females age 36 6 6 4 years height 173 8 13 2 cm weight 69 3 16 1 kg was recruited from the staff at The Ottawa Hospital Rehabilitation Center Ottawa Canada and the community Consent forms were obtained from all the participants prior to the trial People with injuries or a gait deficit were excluded at this stage of the testing All the participants were able bodied without abnormal gait patterns 8 2 2 Data Collection Data collection took placed inside The Ottawa Hospital Rehabilitation Center hallways elevator stairs and Rehabilitation Technology Lab and outside The Ottawa Hospital Rehabilitation Center on the paved pathway The last part of the data collection involved taking a car ride as a passenger or driver around the Ottawa Hospital campus on the Ring road The subjects were asked to wear the WMMS on their waist attached on a belt on their right hip with the device pointing forward No additional instructions were given for positioning the instrumented holster The subjects were asked to follow a pre determined path with a series of mobility tasks Each subject followed verbal instructions indicating the next mobility task For every trial the subjects were filmed with a digital camera Th
153. long each of its axes A system can detect posture by measuring acceleration due to gravity or can detect motion by measuring dynamic acceleration Different classes of accelerometers exist but the common sensors for human motion detection are strain gauge piezoresistive capacitive and piezoelectric 111 Although each class has their own techniques to measure acceleration the mass spring system model is often used to describe the mechanism of accelerometers Figure 2 8 Accelerometers operate under the principle of Hooke s law Equation 2 1 and Newton s Dd law of motion Equation 2 2 When the mass spring is subjected to a compression or stretching force due to movement the spring generates a restoring force proportional to the amount of compression or stretch With known values for mass m and spring stiffness k the resultant acceleration of the mass element can be determined from the displacement x characteristics Equation 2 3 Development of a Wearable Mobility Monitoring System 24 Literature Review F k 2 1 F ma 2 2 ja 2 3 m Accelerometer performance may vary between the different classes Piezoelectric accelerometers use the piezoelectric effect to measure acceleration The piezoelectric effect generates voltage from mechanically stressing crystals such as quartz Accelerometers using this technique typically have higher frequency response than strain gauge accelerometers but poor static response Therefore
154. lowchart including all selected features and their methods to determine the user s state is presented in Figure 7 10 A change of state was determined by subtracting the three previous states from the current state If the answer was different from zero for one of the subtractions a change of state had occurred As a result of a change of state the algorithm determined if a picture should be taken Development of a Wearable Mobility Monitoring System 95 Development of the Prototype WMMS Table 7 1 Description of the state bits STA DYN Standard deviation of y axis to determine if static or dynamic Skewness of y axis to STAIRS determine if going up down stairs STAND Inclination angle indicating standing position GPS speed LIGHT Light intensity value SMA PEAK SMA peak detection SMA INT SMA intensity Inclination angle indicating LIE gs lying position Development of a Wearable Mobility Monitoring System Description If 0 person in static mode not moving if 1 person in dynamic mode moving If 0 person is walking if 1 person is walking up down stairs If 0 person is not in standing position if 1 person is in standing position If 0 person is not in standing position if 1 person is in standing position If 0 person is walking if 1 person could be in vehicle If 0 person is inside if 1 person could be outside If 0 no peak in SMA if 1 peak occurred and person might be sitting or get
155. lts for each of the mobility taskS swwwmwwmmswmmewa 116 Table 8 5 Summary results for the picture evaluation see 119 Table B 1 Compiled results for each trial of the five subjects sess 150 Table C 1 Sensitivity values for each of the mobility tasks for each of the trials 151 Table D 1 Picture evaluation results from evaluator 1 eee 154 Table D 2 Picture evaluation results from evaluator 2 eee 158 Development of a Wearable Mobility Monitoring System vii List of Figures Figure 2 1 Interaction between ICF components reproduced from 18 7 Figure 2 2 Dimensions of Mobility framework reproduced from 1 8 Figure 2 3 Vicon Motion System 62 else set DR enu Suc LORI ENSE eR iEUE Pere as tug 14 Figure 2 4 Motion track MTx from Xsens Technologies reproduced from 65 14 Figure 2 5 Examples of Force Plates On the left is model BP400600 from AMTI 66 with dimensions 8 26 x 60 x 40 cm On the right is a smaller force plate from Bertec Corporation ora aun 15 Figure 2 6 On the left example of pressure mat and software analysis using the emed at m model from Novel 69 On the right example of foot pressure insole from the F Scan Lite Xetsalek System 05
156. ly a result of Development of a Wearable Mobility Monitoring System 5 Literature Review the individual alone but is a combination of relationships between the individual and external factors The ICF model encourages clinicians to acknowledge elements in the physical environment that can facilitate or impede a client s ability to ambulate in their community The eight environmental mobility dimensions provide a framework for assessing the impact of the environment in specific areas The two models are sometimes used such as by Corrigan and McBurney 4 to evaluate the effectiveness of mobility assessment tools to determine community ambulation status The following summarizes these two models 2 1 1 International Classification of Functioning Disability and Health The International Classification of Functioning Disability and Health ICF is a classification system that provides a unified and standard language and framework to describe health and health related states 18 The ICF belongs to the World Health Organization WHO family of international classifications 19 The ICF has two parts each divided into two components 1 functioning and disability which comprises body functions and structures activities and participation 2 contextual factors which comprises environmental factors and personal factors The ICF is used to describe and evaluate disability using the complex relationships between an individual s health condition
157. m 56 Methodology 4 3 Determination of Change of State In this research a change of state was defined as the user s change of movement intensity of movement and or position The WMMS was designed to detect the following changes of state e Start Stop moving e g walking running cleaning e Going up or down stairs ramp hill e Posture change e g standing sitting lying e Speed increase e g bus car e Light intensity change e g indoor outdoor e Posture transitions e g stand to sit sit to stand stand to lie lie to stand e Increase in movement intensity e g stairs 4 3 1 Mobility Tasks and Context Classification To detect mobility tasks and identify the context associated with the mobility tasks the WMMS should detect a change of state when transitioning between mobility tasks which signal the smartphone to take a picture to capture the context and help identify the mobility task The WMMS was evaluated for its capability to detect the following list of mobility tasks and contexts e Walking on a level ground e Walking on a ramp e Walking up and down stairs Inside a building e Outside the building on paved pathway e Taking the elevator e Riding in car e Sitting e Lying e Standing Development of a Wearable Mobility Monitoring System 57 Methodology 4 3 2 Algorithm Outline Figure 4 3 presents the outline of the WMMS signal processing algorithm and data flow Data coming from the
158. ment of patient outcome in arthritis Arthritis and Rheumatism vol 23 pp 137 145 1980 59 F Wolfe S M Kleinheksel M A Cathey D J Hawley P W Spitz and J F Fries The clinical value of the Stanford Health Assessment Questionnaire Functional Disability Index in patients with rheumatoid arthritis The Journal of Rheumatology vol 15 pp 1480 1488 Oct 1988 60 A Shumway Cook A Patla A Stewart L Ferrucci M A Ciol and J M Guralnik Environmental components of mobility disability in community living older persons Journal of the American Geriatrics Society vol 51 pp 393 398 2003 61 A Shumway Cook A Patla A L Stewart L Ferrucci M A Ciol and J M Guralnik Assessing environmentally determined mobility disability Self report versus observed community mobility Journal of the American Geriatrics Society vol 53 pp 700 704 2005 62 Vicon Motion Systems Available http www vicon com Accessed 18 Mar 2009 63 H Zhou and H Hu Human motion tracking for rehabilitation A survey Biomedical Signal Processing and Control vol 3 pp 1 18 2008 64 B Rosenhahn T Brox U Kersting A Smith J Gurney and R Klette A system for marker less motion capture K nstliche Intelligenz vol 20 pp 45 51 2006 65 Xsens Technologies B V Xsens motion technologies Xsens Online Available http www xsens com en home php Accessed 19 Mar 2009 66 Advanced Mecha
159. n instead of atand2 However the method using atan required more steps and more processing time to get the inclination angle with a range of 0 to 360 degrees The atan method had to identify in which quadrant the point z y was in and then apply a certain offset based on the quadrant 172 High Standing Threshold Low Standing Threshold kia 180 NU N Low Lying Threshold i Z axis 270 t 0 90 i degrees 02270 egg degree degrees WA 1 degrees High LyingThreshold ETTI TEk 0 0 degree Standing Position Lying Position Figure 7 2 Position classification method For every 1 02 seconds window the averaged inclination angle was compared with a high and low standing threshold to verify if the person was in a standing position If the person was not standing the angle was compared with a high and low lying threshold to verify if the person was lying on their back If not then the position was determined to be somewhere in between Figure 7 2 demonstrates the two states and the range of angles The threshold values to detect these two postures were based on the study by Culhane et al 148 that found that their best estimate approach to determine thresholds demonstrated higher detection accuracy compared to using mid point tolerances values Therefore with the assumption that the sensor is perfectly mounted on the person the angular range for standing position was set to 200 to 160 deg
160. n quickly walk for a period of 6 minutes The 6 minute walk test was recently recommended as a clinical measure for community ambulation 38 2 2 1 7 Tinetti Assessment Tool The Tinetti Assessment Tool 39 is a widely used tool to assess balance and gait in elderly patients and identify patients at risk of falling The tool is divided into two parts balance assessment and gait assessment The balance part consists of evaluating the patient performing different static positions and position changes such as sitting balance arising from a chair immediate and prolonged standing balance withstanding a nudge on the sternum balance with eyes closed turning balance and sitting down The gait part consists of observing different components of gait and scoring them as normal or abnormal 40 41 2 2 1 8 Functional Independence Measure The Functional Independence Measure FIM is a tool used to quantify physical and cognitive disability in terms of level of care required FIM is a widely adopted tool in rehabilitation facilities 42 The FIM consists of 18 items covering independence in self care sphincter control mobility locomotion communication and cognition 43 Each item can be rated from observations patient interview or medical records The rating is based on performance rather than the capacity Alternative forms of the FIM include the Functional Assessment Measure FAM which consists of the FIM plus 12 new items in the areas of
161. na 11 2 2 2 Diaries and O estionnaires oci edite neci ceste det keel ed coo aves Candid de 12 2 22 DIAEIeS 0i TE tn iet ete eee sce Perte tele epe ee eie to dee MUA tee 12 2 2 2 2 Functional Status Questionnaire e eseeeosseessssereesseeessseeessseeessseeeessreessseeessseeee 12 2 2 2 3 Health Assessment Questionnaire enne ener 13 2 2 2 4 Environmental Analysis of Mobility Questionnaire eee 13 2 2 3 Technologies for Biomechanical Measurements eese 13 2 2 3 1 Visual Motion Tracking System esee 14 Development of a Wearable Mobility Monitoring System iii 2 2 3 2 Non Visual Motion Tracking System eseeeeeeeeeeeeeneeenne 14 2 2 3 3 Force Plates eer tee eo e DR in e eet ere detto 15 22 34 Foot Pressure Analysis ee re eet i eene etr eee ep eene 15 224 Activity Momtonn eere edt eet 16 2 2 4 1 Pedometers 5 teu eee d io Re HERREN 16 2 2 4 2 Accelerometer Based Activity Monitor eeeeeeeeeeeeeeeee 17 2 2 4 3 Physiological Measurement eese nennen 17 2 2 5 Summary of Mobility Measurement wsswemmenmenamanennmanzanesnmnimamanin mamia 17 2 3 Wearable Mobility Monitoring Systems esses 18 2 3 1 Wireless Body Sensor Network WBSN cccscceesseceeeeeeseceeaeeceeeeeesaeeeeaeeeeneeees 19 2 3 2 Personal Server een eee Gin uentis 20 2 3 3 Wireless Standards c eee he e tette e ee
162. ng discrete mobility tasks in a controlled laboratory setting 9 Furthermore to better validate our smartphone approach only one accelerometer was used and our protocol did not control the fixation and location of the WMMS Wearing the WMMS on the right hip attached to the belt was the only requirement given to the subject Development of a Wearable Mobility Monitoring System 123 Technical and Mobility Evaluation of the Prototype WMMS The change of state caused by walking on level ground to walking down stairs was detected at 10090 However the stair intermediate landing was not detected all the time therefore the following walking down stairs was detected at a lower rate since it was considered the same stair descent event as the top stair section If a subject was walking on stairs at a faster speed the WMMS may not have enough time to detect a change within a one second window While the detection of stairs landing could be of interest our currents methods did detect the entire stair descent For walking up stairs the WMMS performed poorly at detecting the change of state 13 396 As with stair descent skewness was used to detect stair ascent The choice of the skewness feature was based on the work by Baek et al 141 which obtained a classification rate of 93 for upstairs and 87 for down stairs The evaluation by Baek et al was performed on a single subject and involved the subject performing discrete tasks as opposed to
163. nical Technology Inc Model BP400600 AMTI Online Available http amti biz Accessed 18 Mar 2009 67 Bertec Corporation Gait amp Biomechanics A Movement in Force Online Available http www bertec com gait biomechanics htm Accessed 18 Mar 2009 68 Tekscan Inc F Scan Lite VersaTek System Clinical and Research Solutions Available http www tekscan com medical system fscan litel html Accessed 18 Mar 2009 69 Novel Product Information System I emed Novel Online Available http www novel de productinfo systems emed htm Accessed 18 Mar 2009 70 R Casabur Activity monitoring in assessing activities of daily living Journal of Chronic Obstructive Pulmonary Disease vol 4 pp 251 255 2007 71 Orthocare Innovations StepWatch Orthocare Innovations 2007 Online Available http www orthocareinnovations com pages stepwatch trade Accessed 11 Nov 2009 Development of a Wearable Mobility Monitoring System 134 References 72 E D de Bruin A Hartmann D Uebelhart K Murer and W Zijlstra Wearable systems for monitoring mobility related activities in older people A systematic review Clinical Rehabilitation vol 22 pp 878 895 2008 73 A P Marsh R M Vance T L Frederick S A Hesselmann and W J Rejeski Objective assessment of activity in older adults at risk for mobility disability Medicine and Science in Sports and Exercise vol 39 pp 1020 1026 2007
164. nical balance measures Journal of Neurologic Physical Therapy JNPT vol 30 pp 60 67 2006 27 J Jonsdottir and D Cattaneo Reliability and validity of the Dynamic Gait Index in persons with chronic stroke Archives of Physical Medicine and Rehabilitation vol 88 pp 1410 1415 2007 28 T Herman N Inbar Borovsky M Brozgol N Giladi and J M Hausdorff The Dynamic Gait Index in healthy older adults The role of stair climbing fear of falling and gender Gait and Posture vol 29 pp 237 241 2009 29 D M Wrisley G F Marchetti D K Kuharsky and S L Whitney Reliability internal consistency and validity of data obtained with the functional gait assessment Physical Therapy vol 84 pp 906 918 2004 30 J Howe E Inness M Verrier and J Williams Development of the Community Balance and Mobility Scale CB amp M for the Traumatic Brain Injury TBI in American Congress of Rehabilitation Medicine 1999 31 J A Howe E L Inness A Venturini J I Williams and M C Verrier The Community Balance and Mobility Scale A balance measure for individuals with traumatic brain injury Clinical Rehabilitation vol 20 pp 885 895 2006 32 E L Inness J A Howe E Niechwiej Szwedo S Jaglal W E McIlroy and M C Verrier Measuring balance and mobility after traumatic brain injury further validation of the Community Balance amp Mobility Scale CB amp M Archives of Physical Medicine
165. nt do not have to be attached to the subject and software applications do not need to be started for every session 84 However technical and social challenges exist for wearable mobility monitoring These challenges include Privacy and security Some of the big issues with wearable monitoring system are those of privacy and security such as eavesdropping identity spoofing and redirection of private data to unauthorized persons 85 Appropriate methods of data encryption can help improve security and privacy However developing security and privacy solutions for wireless sensor networks applied to biomedical applications are faced with many obstacles such as limited resources fault tolerance interference and attacks confidentiality and physical security 86 Power requirements For long term monitoring a wearable system must last long enough to capture all of the data However adding larger batteries creates a trade off between more power and a small lightweight wearable system Another issue is with wireless communication that usually increases the system s power requirements Sending processed data instead of raw data could help decrease power consumption creating a trade off between communication and data computation 84 Portability For continuous and long term monitoring wearable systems need to be small lightweight and should not interfere with movement The type of sensors location of sensors and transmission charac
166. ntified a change of state and WMMS took a picture False positives occurred when the algorithm identified a change of state but there was no real change of state True negatives occurred when there was no change of state and the algorithm did not detect a change of state Finally false negatives occurred when there was a change of state but the algorithm did not detect the change The number of true and false positives and true and false negatives were used to calculate WMMS sensitivity and the specificity Equations 8 1 and 8 2 TruePositives 8 1 Sensitivity x100 TruePosives FalseNegatives TrueNegatives 8 2 Specificity x100 TrueNegatives FalsePositives Two research assistants independently evaluated the BlackBerry Bold images The evaluators were asked to identify the context i e stairs elevator ramp floor outdoor etc from the digital images Only the images taken due to a real change of state true positives were evaluated The evaluators were given a list of context options to choose from Figure 8 2 show an example of the spreadsheet that the evaluators filled out for every trial The evaluators were not informed of the mobility tasks represented by the images prior to Development of a Wearable Mobility Monitoring System 112 Technical and Mobility Evaluation of the Prototype WMMS evaluation The results from the two evaluators were then analyzed to determine if context was successfully id
167. og data Chapter 7 presents the details about data processing and algorithm Chapter 8 presents the technical evaluation and the mobility evaluation from five healthy subjects of the WMMS Development of a Wearable Mobility Monitoring System 59 Preliminary Evaluation of the BlackBerry for WMMS Chapter 5 Preliminary Evaluation of the BlackBerry for WMMS A proof of concept WMMS system was assembled consisting of a Blackberry 8800 handheld Research In Motion Ontario Canada serving as a hub or central node and a commercial motion capture system Xbus Kit Xsens Technologies Netherland The purpose was to evaluate the BlackBerry smartphone as a platform for a WMMS The choice for the BlackBerry model 8800 was based on the currently available Java development environment and application programming interface API Figure 5 1 illustrates the proof of concept system architecture Five motion trackers MTx were connected to the Xbus Master in a daisy chain configuration The BlackBerry 8800 used Bluetooth to communicate with the Xbus Master during motion capture to configure and initialize the Xbus Master and the five MTx sensors Motion data was in orientation mode expressed in quaternion units Another command was sent to the Xbus Master from the BlackBerry to start data capture Processing the incoming motion data was performed by the BlackBerry to calculate Euler angles for both knees and hips four sets of Euler angles in total The pro
168. on Complex Medical Engineering 2009 147 P H Veltink H B J Bussmann W De Vries W L J Martens and R C Van Lummel Detection of static and dynamic activities using uniaxial accelerometers IEEE Transactions on Rehabilitation Engineering vol 4 pp 375 385 1996 148 K M Culhane G M Lyons D Hilton P A Grace and D Lyons Long term mobility monitoring of older adults using accelerometers in a clinical environment Clinical Rehabilitation vol 18 pp 335 343 2004 149 G M Lyons K M Culhane D Hilton P A Grace and D Lyons A description of an accelerometer based mobility monitoring technique Medical Engineering and Physics vol 27 pp 497 504 2005 150 C Ni Scanaill B Ahearne and G M Lyons Long term telemonitoring of mobility trends of elderly people using SMS messaging IEEE Transactions on Information Technology in Biomedicine vol 10 pp 412 413 2006 151 J B J Bussmann J H M Tulen E C G Van Herel and H J Stam Quantification of physical activities by means of ambulatory accelerometry A validation study Psychophysiology vol 35 pp 488 496 1998 Development of a Wearable Mobility Monitoring System 141 References 152 F Foerster M Smeja and J Fahrenberg Detection of posture and motion by accelerometry a validation study in ambulatory monitoring Computers in Human Behavior vol 15 pp 571 583 1999 153 Y Yoshida Y Yonezawa K Sata I
169. on was based on the two axes method presented in application note AN3461 from Freescale Semiconductor 172 Using two axes instead of one to calculate inclination angle improved resolution and provided a 360 degree range of inclination angle The vertical y axis and horizontal forward z axis axes were used Figure 7 1 illustrates the method pz 180 degrees Y axis i D 180 degrees XV i b inclination angle pz 0 degree Quadrant 3 Quadrant 2 Y Z Y Z Y axis O 270 degrees D 90 degrees Z axis Quadrant 4 Quadrant 1 Y Z Y Z te TITI LA oincination angle M 0 degree Figure 7 1 Inclination angle measurement method In standing position inclination angle is 180 degrees The Java function atand2 was used to calculate the inclination angle Equation 7 3 atand2 GAz GAy 7 3 where GAz and GAy are the averaged static accelerations of z axis and y axis respectively The atand2 function returns the arctangent of GAz GAy with the resulting angle ranging Development of a Wearable Mobility Monitoring System 82 Development of the Prototype WMMS between 180 to 180 degrees However for convenience an offset of 180 degrees was added to the inclination angle to give a range of 0 to 360 degrees and to measure 0 degrees when the y axis was pointing down Figure 7 1 Another possible option was to use the Java function ata
170. one with all necessary features was unavailable at the start of this thesis an external board was added to the design The external board could make the device slightly heavier and less comfortable for the user There is also the possibility of losing the Bluetooth connection and missing important data However new smartphones have emerged that could solve this problem by providing raw acceleration data GPS signals were not always present during data collection A waiting period of more than 30 minutes to get signal was not always practical Using cell site methods to improve GPS detection should be explored New smartphones could potentially perform better as well BlackBerry Bold 9000 battery usage was 2996 per hour Table 8 1 This is not sufficient for long term monitoring because at this rate only 3 hours of monitoring can be expected A larger capacity battery would be required for longer monitoring Not using Bluetooth could potentially slow down the battery usage however accessing raw accelerometer data from the BlackBerry would be expected to draw additional power from the battery BlackBerry camera performance showed that a picture could only be taken after 3 seconds i e during the third one second window This delay may cause some images to miss details related to the mobility task Additionally the camera did not perform well under low light conditions causing images to be blurry and dark The location of WMMS on the
171. ons caused by gravity static and accelerations caused by movement dynamic Mathie et al 81 mentioned that these two acceleration components can be separated by filtering the signal with a cut off frequency between 0 1 to 0 5 Hz In this thesis a RC low pass digital filter with a cut off frequency of 0 25 Hz was applied to the median filtered acceleration signal to extract the static component The dynamic component was then obtained by subtracting the static component from the median filtered signal The pseudo code used to simulate the RC low pass filter effect is 192 for i from 1 ton yli y i 1 a xfi y i 1 return y where x is the median filtered signal y the static component and a the smoothing factor The smoothing factor can be expressed as Development of a Wearable Mobility Monitoring System 80 Development of the Prototype WMMS dt 7 1 a dt RC where dt is the sampling delay and RC the time constant The cutoff frequency is expressed as oll 7 2 2aRC For a cutoff frequency of 0 25 Hz and a sampling delay of 0 020 second the time constant RC was 0 64 second and the smoothing factor a was 0 0304 To determine the state features were extracted from the static and dynamic components over a non overlapping sliding window of 1 02 seconds With a sampling frequency of 50 Hz 1 02 seconds corresponds to 51 samples As mentioned by Preece et al 171 a sliding window is well suited
172. ot switches with accelerometers being the most commonly used These sensors have been explored by many in applications such as movement classification activity recognition assessment of balance gait and transitions and fall detection However many of these studies are missing environmental or contextual information related to the user s activities 5 7 9 10 110 Other studies have used GPS to monitor mobility or travelling patterns in the community 17 123 but details on the type of activities performed were not considered GPS is also used to complement motion data and improve activity recognition 5 8 GPS can provide contextual information such as location but its accuracy depends of the number of satellites it can detect GPS typically does not work indoors Other context information such as light temperature and sounds provides context awareness for wearable systems Context aware wearable systems used context information to better recognize activities 12 15 but the environmental characteristics in which activities took place were not analyzed for their impact on mobility A camera is an interesting sensor to include in a wearable system since a picture or video can give information on the user s surroundings Studies that used camera GPS and other context data are mostly oriented to life log applications 12 130 To the best of our knowledge the use of a camera in a wearable system to capture the context in which
173. pen the door Development of a Wearable Mobility Monitoring System 109 Technical and Mobility Evaluation of the Prototype WMMS 65 Car ride one loop around the Ottawa Hospital campus Ring Road a Initiation Car starts moving b Termination Car is in park mode 66 Opening car door to get out a Initiation Seated position start to open the door b Termination From seated position initiation of trunk flexion and buttock lifting from car seat 67 Sit to stand transition get out of the car a Initiation From seated position initiation of trunk flexion and buttock lifting from chair b Termination Standing position outside the car 68 Transition get out of the car to walk a Initiation Standing position outside the car b Termination Start of forward walking progression 69 Walk 30 meters towards the Ottawa Hospital Rehabilitation Center entrance a Initiation Start of forward walking progression b Termination End of forward walking progression 70 Transition outside to inside automatic door a Initiation Outside stepping inside b Termination Start of forward walking progression 71 Walk 5 meters a Initiation Start of forward walking progression b Termination End of forward walking progression 72 Turn around a Initiation End of forward walking progression b Termination Facing opposite direction 73 Standing a Initiation Facing opposite direction b Termination Standing 8 2 3 Data Analysis Data collecte
174. purpose of the thesis is to develop and validate a wearable system that will monitor mobility in the community The wearable system must be light and portable easy to use and contained at one body location The WMMS was developed to meet the following objectives e Detect in real time a user s change of state related to mobility and context e Take a picture for every valid change of state to identify the mobility context and environment e Validate the system with a normal population From the WMMS developed in this research it was hypothesised that a change of state can be identified with 95 specificity and 95 sensitivity and that images can be correctly categorized 95 of the time Development of a Wearable Mobility Monitoring System 53 Methodology Chapter 4 Methodology The following section contains the design criteria for a Wearable Mobility Monitoring System WMMS This chapter includes an overview of the system architecture materials data processing methods and system evaluation methods 4 1 Design Criteria A high compliance WMMS must be lightweight wearable easy to place on the person easy to use and located at one location on the body The objectives of the system were also to identify changes of state and take pictures to capture the context The following list of criteria was used in the design of a Wearable Mobility Monitoring System WMMS 4 1 1 System Design Criteria e Minimum number of sensors
175. r they are highly dependent on the test administrator s subjectivity and reaction time Furthermore as suggested by Myers et al 22 the individual s performance at the time of assessment may not be representative of their usual performance As pointed out by Patla 23 the environment in which the assessment takes place is usually a flat well lit area which is an exception in community mobility The following describes some common observational and clinical tests that measure mobility 2 2 1 1 Dynamic Gait Index DGI The Dynamic Gait Index DGI evaluates postural stability in older adults over eight different tasks including walking at different speeds walking while turning the head ambulating over and around obstacles ascending and descending stairs and making quick turns Each task is scored on a scale of 0 to 3 with a maximum possible score of 24 A score less than 19 indicates a high risk of falling during gait 24 28 2 2 1 2 Functional Gait Assessment FGA The Functional Gait Assessment FGA is a 10 item gait assessment based on the DGI Wrisley et al 29 created and validated the FGA This test includes seven out of eight DGI tasks and three new tasks gait with narrow base of support ambulating backwards and Development of a Wearable Mobility Monitoring System 9 Literature Review gait with eyes closed These new tasks were added since they were observed to be difficult for people with vestibular disorders T
176. r measuring free living daily activities in a chronic obstructive pulmonary disease COPD population However diaries require a high level of adherence from the patients and are retrospective and subjective 6 52 Diaries are known for their potential recall bias and misreporting of activity level which affect their accuracy 53 2 2 2 2 Functional Status Questionnaire The Functional Status Questionnaire FSQ is a comprehensive self report functional assessment of patients receiving ambulatory care 54 The FSQ is divided into five main sections physical function of the activities of daily living psychological function role function social function and a variety of performance measures In mobility studies researchers sometimes used only FSQ subscales that relate to physical activities such as ADL Instrumental Activity of Daily Living ADL and social activity 55 57 The ADL subscale consists of questions about activities such as dressing bathing transfers and Development of a Wearable Mobility Monitoring System 12 Literature Review mobility The IADL subscale covers activities such as shopping using public transportation and maintaining a household The social activity subscale is related to social interaction such as the person s ability to visit with family and friends 2 2 2 3 Health Assessment Questionnaire The Health Assessment Questionnaire HAO was first developed to assess IADL in arthritis patients 58
177. racterize non stationary signals 2 4 7 Activity Classification After features have been extracted from the accelerometer signals they can be used as input for activity classification algorithms The following presents classification algorithms that have been used in activity identification Thresholds are one of the simplest methods to extract activity information from the accelerometer signals Signal properties or features e g mean standard deviation vertical velocity are compared with thresholds to determine if a particular activity is present in the data window For example static and dynamic movement can be distinguished by comparing the signal s standard deviation with a threshold value as demonstrated by Veltink et al 147 and Mathie et al 7 Threshold methods applied to inclination angle can also detect different postures as shown in the studies by Cuhrane et al 148 and Najafi et al 155 Fall detection has also been studied by Bourke et al 142 where heuristic features were used with thresholds Threshold methods are often chosen for real time processing applications to be performed by low memory and low processing capability devices such as microcontroller embedded portable units 9 Classification systems using a hierarchical approach are very popular A hierarchical decision tree starts with a top level broad classification e g rest and active followed by more detailed sub classifications at lower levels
178. re often found in gait analysis laboratories and come in different sizes and prices Figure 2 5 Figure 2 5 Examples of Force Plates On the left is model BP400600 from AMTI 66 with dimensions 8 26 x 60 x 40 cm On the right is a smaller force plate from Bertec Corporation 67 2 2 3 4 Foot Pressure Analysis Foot pressure analysis systems measure load distribution under the plantar surface of the foot Two types of systems exist pressure mat or pressure insole Figure 2 6 A pressure mat is similar to a force plate since the mat is placed on the ground and the subject walks onto the mat Pressure insoles are placed directly in the footwear which provides portable pressure measurement between the foot and the shoe i e forces are not dampened by the footwear Examples of commercial manufacturers are TekScan Inc Massachusetts USA 68 and Novel Munich Germany 69 Both companies provide a variety of foot pressure systems including pressure mats and pressure insoles Development of a Wearable Mobility Monitoring System 15 Literature Review Figure 2 6 On the left example of pressure mat and software analysis using the emed at m model from Novel 69 On the right example of foot pressure insole from the F Scan Lite VersaTek System 68 2 2 4 Activity Monitoring A good level of physical activity is usually associated with positive health benefits Therefore the assessment of the physical activity is sometimes used a
179. rees Similarly range for lying position was set to 300 to 240 degrees However during preliminary testing it was observed that certain sitting and lying positions had an angle value very close to the thresholds causing false positive Development of a Wearable Mobility Monitoring System 83 Development of the Prototype WMMS changes of state to occur The sitting posture was sometimes identified as lying and the lying position was outside the range This was due to the way the WMMS is worn on the waist During sitting the leg may touch the WMMS which may caused extra inclination angle of the system During lying if the person had their legs bent this may also caused extra inclination Therefore the lying thresholds were adjusted to 320 and 250 7 2 2 Standard Deviation Another feature that was chosen to determine the user s state is the standard deviation 80 141 147 149 The standard deviations for the three axes were calculated i g 7 4 o meri x where n is the number of point x the acceleration at point i and x the mean of the using the following equation acceleration signal The Equation 7 4 can then be rearranged to the following equation for programming purposes 7 5 In this thesis since most daily activities such as walking sitting lying down and going up down stairs can be observed by a change of acceleration on the vertical axis only the vertical acceleration y axis was required to
180. reescale Semiconductor Application Note 3461 Rev 2 Tilt Sensing using Accelerometers Sensors pp 2 4 173 A K Bourke K J O Donovan and G Laighin The identification of vertical velocity profiles using an inertial sensor to investigate pre impact detection of falls Medical Engineering and Physics vol 30 pp 937 946 2008 174 S J Preece J Y Goulermas L P J Kenney and D Howard A comparison of feature extraction methods for the classification of dynamic activities from accelerometer data IEEE Transactions on Biomedical Engineering vol 56 pp 871 879 2009 Development of a Wearable Mobility Monitoring System 143 References 175 M J Mathie B G Celler N H Lovell and A C F Coster Classification of basic daily movements using a triaxial accelerometer Medical and Biological Engineering and Computing vol 42 pp 679 687 2004 176 H Y Lau K Y Tong and H Zhu Support vector machine for classification of walking conditions using miniature kinematic sensors Medical and Biological Engineering and Computing vol 46 pp 563 573 2008 177 S Wang J Yang N Chen X Chen and Q Zhang Human activity recognition with user free accelerometers in the sensor networks in Proceedings of the 2005 International Conference on Neural Networks and Brain 2005 pp 1212 1217 178 J Y Yang J S Wang and Y P Chen Using acceleration measurements for activity recognition An effect
181. rence Recent technological advances in wireless communications sensor miniaturization and smartphone processing power could help overcome some of these challenges and offer great potential in the development of wearable systems for mobility monitoring Research in the field of wireless body sensor networks WBSN and wireless body area networks WBAN could allow healthcare to be delivered outside the hospital 1 e at the patient s home and in the community 84 94 The hub or personal server of a WBSN or WBAN could be a PDA mobile and smartphone or custom made hub Smartphones are particularly attractive in the development of wearable systems due to their increasing processing power effective display and user interface and features such as GPS accelerometers and camera Wearable system using smartphones may also improve the user acceptance Development of a Wearable Mobility Monitoring System 30 Literature Review Advances in wireless technology could allow wearable systems to eliminate the use of cables Wireless wearable systems could be more comfortable to wear less obtrusive and less encumbering with the user s movement 85 Many smartphones are equipped with wireless technologies such as Bluetooth to communicate between sensors and phones and Wi Fi to communicate with an external server via the internet 84 137 Commonly used wearable sensors for mobility monitoring are accelerometers gyroscope magnetometer and fo
182. ring System 87 Development of the Prototype WMMS stairs was also to decrease the false positives detection of stairs caused by peak in the skewness signal observed during the stop and start of dynamic motions The high and low thresholds of the standard deviation used to determine the sufficient dynamic level for stairs detection were 0 3g and 0 2g respectively Figure 7 5 illustrates the DT algorithm applied to the y axis acceleration skewness Skewness SKEWY and Standard deviation STDY of y axis acceleration STDY gt Dynamic Threshold 2 No STDY Static Threshold 2 SKEWY High Stairs Threshold SKEWY gt Low Stairs Threshold State Stairs State Previous state State No Stairs Figure 7 5 Algorithm flow chart for skewness of y axis acceleration Development of a Wearable Mobility Monitoring System 88 Development of the Prototype WMMS Skewness of y axis acceleration versus time 2 5 4 0 5 Skewness 0 5 4 1 5 d 22 4 t t t Skewness t 1 10 20 30 40 50 Stairs ri 25 Dynamic II s 1 5 IIET 0 5 ad ASA rr Skewness Dynamic m 60 70 Time seconds Fi D 80 90 100 110 120 130 140 150 Stairs Static i DII Stairs High Threshold Stairs Low Threshold Skewness sss Dynamic Stat
183. roblems in elderly patients Journal of the American Geriatrics Society vol 34 pp 119 126 1986 40 L D Abbruzzese The Tinetti performance oriented mobility assessment tool American Journal of Nursing vol 98 pp 16J 16L 1998 41 A Yelnik and I Bonan Clinical tools for assessing balance disorders Neurophysiologie Clinique vol 38 pp 439 445 2008 42 B B Hamilton C V Granger F S Sherwin M Zielezny and J S Tashman A uniform national data system for medical rehabilitation in Rehabilitation Outcomes Analysis and Measurement 1987 pp 137 147 43 C V Granger and B B Hamilton The uniform data systems for medical rehabilitation report of first admissions for 1991 American Journal of Physical Medicine amp Rehabilitation vol 72 pp 33 1993 44 K M Hall N Mann W M High Jr J Wright J S Kreutzer and D Wood Functional measures after traumatic brain injury Ceiling effects of FIM FIM FAM DRS and CIQ The Journal of Head Trauma Rehabilitation vol 11 pp 27 1996 45 I McDowell Measuring Health A Guide to Rating Scales and Questionnaires Third ed New York New York Oxford University Press 2006 46 L Ferrucci J M Guralnik S Studenski L P Fried G B Cutler Jr and J D Walston Designing randomized controlled trials aimed at preventing or delaying functional decline and disability in frail older persons A consensus report Journal of the American
184. s 2 3 and data analysis techniques related to mobility monitoring studies 2 4 2 1 Community Mobility Independent ambulation within the home and the community is an important rehabilitation goal for a person with physical impairments 16 Lord et al defined community ambulation 16 as independent mobility outside the home which includes the ability to confidently negotiate uneven terrain private venues shopping centres and other public venues This definition was based on the environments that participants considered the most important Patla and Shumway Cook 1 defined community mobility as the locomotion in environments outside the home or residence The achievement of independent community mobility is dependent on various factors Frank and Patla 17 mentioned that community mobility depends on 1 The skills and abilities of the performer 2 Requirement of the task activity 3 Challenges of the environment The importance to account for the environmental factors when assessing mobility has been previously emphasized by two well known models The International Classification of Functioning Disability and Health ICF from the World Health Organisation 18 covers aspects of a person s health including mobility The Dimensions of Mobility framework from Patla and Shumway Cook 1 focuses on the person s mobility The main idea behind theses two models or frameworks is that a person s health condition is not on
185. s 21 2 3 4 Wearable Sensors eiie iie us 23 DSA Aecelerot etets iie tete totes eb tei bei te petet bones 24 2 3 4 2 GytOSCODE cei oe RE tU Ga ONE RU ee es 26 2 3 4 3 Magnetorneter eec eene tete lee ei teeth te Dee AN 26 2 3 4 4 Foot Press te ee dete e tue dana ee Er e e ebbe Y eee 26 DBA US AA tae ie ime He eR tet red ass 27 2 3 4 0 AUA AAA 28 2 3 4 7 Ambient Sensors e ete Sah eid kina eta eh 29 2 3 9 Context AWAarelless i oo eet NAI geeen betae ee Papeete ebbe tede o cien thease 29 2 3 0 Summary of Wearable Systems essent 30 2 4 Data Analysis Algorithms e iier eee eere cest eere de eto Ho even 31 2 41 Accelerometer Placement ko seeds matewa eraik aiiai ie ii i Ea 32 2 42 Freg ency and Amplitude ai eee ueteri t 37 DAS s Calibration art ette de d Net edet aet 37 2 44 Filtering Techniques Sot eee Eden at 39 2 445 Data Window inert teilte ra betae Pepe inge boe bee REEE lee iege 40 24 6 Feature Extraction eed enn tede In Ree tete tee ec teta 41 24 7 Activity Classification ss eerie e e cer AR AAEE 46 2 4 8 Summary of Data Analysis eee ceseceseceseceseceseeeseeeeneesneeeseeeaeecaaecnaecsaeeaeenaeens 49 Chapter 3 Rationale sia iiie oes sie eq E ESL Pent AE ES 51 3 1 Application of a Wearable Mobility Monitoring System WMMS esses 52 3 2 Objective of the thesis ie eed ete eme et 53 Development of a Wearable Mobility Monitoring
186. s an indication of health status The following presents monitoring devices used in research clinical and commercial settings 2 2 4 1 Pedometers Pedometers are a well known type of activity monitor 70 These devices are usually worn at the waist with some models worn on the ankle or the calf and they estimate activity by sensing steps during walking More advanced pedometer models may include synchronization of step count measurements to a 24 hour clock such as the Step Watch 3 Activity Monitor Orthocare Innovation Oklahoma City OK USA 71 However pedometers may be poor at identifying other activities e g bicycle riding In addition pedometers cannot provide information on static activities Additionally studies have found that pedometers are not a good choice when assessing physical activity in older adults at risk of mobility disability because pedometers underestimate the number of steps during slow walking 72 73 Despite these limitations pedometers were still found to be a valid simple and inexpensive method for assessing physical activity in research and practice 74 and for detecting differences in ambulatory activity according to age and functional limitations 75 Recently a more precise step counter 0 546 error was developed by Giansanti et al 76 This step counter uses calf muscle expansion measured with a force resistive sensor to define a step for people with Parkinson s disease Development of a W
187. s compared to the first set of static trials 5 5 Preliminary Evaluation Results Table 5 1 shows the average time and the standard deviation for the static and dynamic trials as well as the number of trials that stopped due to error The timer overflow error caused the Xbus Master to stop sending motion data For both the static and dynamic trials the application was able to run longer without error at 25 Hz than at 50 Hz Only one trial at 50 Hz ran without error The other 50 Hz trials stopped due to the same timer overflow error At 25 Hz the dynamic trials had only one stop due to this error compared to two stops during the static trials In addition the averaged time was smaller during the dynamic rather than the static trials The Xbus Master s batteries were not able to last more than 1 5 hours causing this smaller average time For the static minimal trials the average time was slightly better than the normal static trials at 50 Hz However the application still stopped due to the Xbus timer overflow error No data were lost for all trials The BlackBerry battery trials indicated an average usage of 12 1 2 6 per hour At this rate the BlackBerry would run out of battery power after approximately 6 8 hours Development of a Wearable Mobility Monitoring System 65 Preliminary Evaluation of the BlackBerry for WMMS Table 5 1 Preliminary BlackBerry evaluation results Standard Number of stops Deviation due to Xsens min
188. s only The bottom graph is the skewness curve but with some dynamic static and stairs states identified The dotted line shows when the dynamic level was identified i e when the skewness values was analyzed for stairs or not stairs State 89 Figure 7 7 SMA of a person walking then sitting standing up walking lying down on a bed getting up from the bed lying on the floor and getting up again essse 91 Figure 7 8 Flowchart of the SMA algorithm ie eet oed e ibo e dece ee 92 Figure 7 9 Example of the light intensity feature signal while performing mobility tasks indoors abd DI Ka iso see EA pep UR DIVI PHA UNI NONI Mte qM EIS 93 Figure 7 10 State determination algorithm DT stands for double threshold 97 Figure 7 11 Overview of programming flow eese emen 101 Figure 8 1 BlackBerry battery with full WMMS application running Trial 2 104 Figure 8 2 Example of the spreadsheet used by the pictures evaluators 113 Development of a Wearable Mobility Monitoring System ix 3D AC ADL ADT API BBS CART CB amp M CDT COPD CRC CWT DC DGI DT DWT EAMQ ECG EE EE act FAM FFT FGA FIM FSQ GPS GSM HAQ HMM IADL Acronyms Three dimensional Alternating Current Activities of Daily Living Automatic Decision Tree Application Programming Interface Berg Balance Scale Classif
189. stand 15 0 Walking on level ground 15 0 Standing waiting for elevator 14 0 Walking to get in the elevator 12 2 Taking elevator to 2 floor 13 2 MEE RE EUM L o Standing waiting for elevator 15 0 Walking to get in the elevator 15 0 Taking elevator to 1 floor 15 0 Walking to get out of elevator and keep walking on level ground 19 0 100 0 Walking up stairs 2 13 13 3 Development of a Wearable Mobility Monitoring System 116 Walking on stair intermediate landing level ground for 1 5 meters Walking up stairs Walking on level ground Walking down stairs Walking on stair intermediate landing level ground for 1 5 meters Walking down stairs Walking on level ground Stand to lie transition Lying Lie to Stand transition Walking on level ground Walking on ramp Walking on level ground Transition indoor outdoor and keep walking on level ground Transition outdoor indoor and keep walking on level ground Transition indoor outdoor and keep walking on level ground Stand to sit transition to get in the car Sitting in the car Starts of car ride Stop of car ride Sit to stand transition Walking on level ground Transition outdoor indoor and keep walking on level ground Standing Development of a Wearable Mobility Monitoring System Technical and Mobility Evaluation of the Prototype WMMS 40 096 40 0 26 7 100 0 66 7 66 7 100 0 100 0 100 0 100 0 100 0 40 0 46 7 46 7
190. suggests that a phone integrated with an accelerometer could detect changes from static to dynamic movement i e start to walking standing still slowing down We also used a double threshold algorithm instead of only one threshold which provided a degree of variability in the signal and helped to decrease the number of false positive results Changes of state due to postural change i e stand to sit sitting lie to stand etc were detected with a sensitivity of 97 8 4 7 These results compared favourably with previous studies such as Karantonis et al 9 where a 94 2 accuracy was found for detecting tasks related to postural orientation Using threshold methods Culhane et al 148 detected sitting at 92 standing at 95 and lying at 98 However their results were obtained from two accelerometers one on the trunk and one on the thigh Even though our algorithm detected changes of state due to postural change our approach was not evaluated for its accuracy to classify the posture From our observations our methods might not be precise enough to classify all posture The way the WMMS was worn on the hip may have caused false positives during sitting and lying due to the device holster s free movement the leg pushing on the device the person s belt location and sitting angle However our evaluation protocol provided a real time situation where the mobility tasks were performed consecutively and freely instead of performi
191. tal 3 Total 4 Total 5 Total OVERALL Appendix B True False True False Positive Positive Negative Negative 808 11 807 8 768 9 Average 807 7 843 4 Average 914 Average Standard Deviation Average ij Sami 2 30 a 25 Bs 29 2 30 3 31 ph ag 7a ie 3 28 HD 1 26 2 391 3 34 ip 8 2 30 3 29 ES INE ESSEN Standard Deviation Development of a Wearable Mobility Monitoring System Sensitivity Specificity 71 05 78 95 76 3296 75 44 4 02 76 3296 81 0876 81 5876 79 66 2 91 76 3296 76 3296 73 68 75 44 1 52 72 22 81 08 89 47 80 93 8 63 76 3296 78 95 76 3296 77 19 1 52 77 73 2 49 Appendix B 96 88 96 19 96 24 96 44 0 39 93 56 92 52 93 83 93 30 0 69 97 19 97 39 95 38 96 65 1 11 94 74 97 82 95 69 96 08 1 58 99 67 99 89 98 92 99 49 0 51 96 39 2 20 150 uiojs amp s Suriojuo A Ki rqoJA QLI AA Jo juouido oAo T ISI Appendix C Table C 1 Sensitivity values for each of the mobility tasks for each of the trials Subject 1 Subject 2 Subject 3 Subject 4 Subject5 Sensitivity Change of State True False per Positive Negative mobility 1 S 2 3 1 2 3 1 2 3 1 2 3 task Walking on level ground 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 0 100 00 Stand to sit transition 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 14 1 93 33 Sitting 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 15 0 100 0
192. tely 145 Hz before being sent to the BlackBerry Each low pass filter was located on the output of each axis and was composed of an internal output resistor of 110 kohms typical value and an external load capacitor of 10 nF The light temperature and humidity sensors signals were not analog filtered Some digital filtering of the external board sensors data were performed by the Java application developed for the WMMS and will be described in the next Chapter 6 3 Hardware Evaluation 6 3 1 Camera The BlackBerry Bold camera was evaluated for shutter lag which is the time between calling the take a picture function and the time the picture was taken The time before the camera is ready to take another picture was also evaluated A Java application was developed to take a picture continuously until manually stopped The picture encoding was set to jpeg with size 640x480 pixels and quality set to normal The memory size of a picture with this encoding was 10 to 70 Kbytes The time before and after the picture was taken was measured using the Java function System currentTimeMillis Five trials were performed with 20 pictures taken per trial During the trials the BlackBerry was held by a user The shutter lag values were averaged From the same trials the time before the camera was ready to take another picture was calculated by subtracting the time after the previous picture was taken from the time before taking the next picture These time
193. teristics are important factors to consider when designing wearable systems as it could affect the portability 84 Development of a Wearable Mobility Monitoring System 18 Literature Review Acceptance adherence User acceptance is an important determinant of operational feasibility 72 A potential solution is integrating sensors into devices that people already use such as mobile phones As suggested by Lester et al 87 the mobile phone approach is more likely to have better acceptance and adherence as these consumer devices do not make them look different A wearable system that is easy to setup and start will improve the acceptance and adherence of the system Recent technological advances in wireless communications sensor miniaturization and smartphone processing power offer great potential in the development of wearable systems for mobility monitoring and also to overcome some of the challenges related to wearable systems The following give an overview of technologies that are relevant for this research 2 3 1 Wireless Body Sensor Network WBSN Wireless body sensor networks WBSN and wireless body area networks WBAN can monitor human behaviour to allow the shift of health assessment from hospitals to the community 85 Wearable health monitoring systems using technologies of WBSN and WBAN have been introduced in 84 88 93 WBSN and WBAN typically consist of one or multiple sensors worn on the body wher
194. the activities evaluated in this thesis However other external sensors could be integrated into the WMMS using the new WMMS software and Bluetooth communications such as for pressure or electromyography analyses Development of a Wearable Mobility Monitoring System 128 References References 1 A E Patla and A Shumway Cook Dimensions of mobility Defining the complexity and difficulty associated with community mobility Journal of Aging and Physical Activity vol 7 pp 7 19 1999 2 Statistics Canada 2007 Participation and Activity Limitation Survey 2006 Analytical Report Minister of Industry Ottawa Online Available http dsp psd pwegsc gc ca collection_2007 statcan 89 628 X 89 628 XIE2007002 pdf Accessed 25 Nov 2009 3 T Lam V K Noonan and J J Eng A systematic review of functional ambulation outcome measures in spinal cord injury Spinal Cord vol 46 pp 246 254 2008 4 R Corrigan and H McBurney Community ambulation Environmental impacts and assessment inadequacies Disability and Rehabilitation vol 30 pp 1411 1419 2008 5 M Ermes J P rkk J M ntyj rvi and I Korhonen Detection of daily activities and sports with wearable sensors in controlled and uncontrolled conditions IEEE Transactions on Information Technology in Biomedicine vol 12 pp 20 26 2008 6 C N Scanaill S Carew P Barralon N Noury D Lyons and G M Lyons A review of approaches to mob
195. the real world evaluation employed in this thesis Therefore the results from Baek et al may have been overstated Other differences with our methods are that the location of their accelerometer was worn on the lateral side of the pelvis instead of the front side Baek et al also used a 2 second window more features and more complex algorithms such as a neural network To improve the stairs ascent detection other time domain features have been explored such as skewness of the forward axis and kurtosis based on Baek et al 141 but they did not provide better results A study by Ravi et al 96 also suggested calculating correlation values between two axes to detect activity that involved 2D translations Again these values did not show improvement for detecting stairs ascent Another method that could be explored is double integration of the vertical acceleration to evaluate changes in height triggering a change of state due to stairs or inclines More complex algorithms could be added to the design since the newer generation of smartphones have greatly enhanced processing power Currently literature is lacking on ramp detection using accelerometer signals Therefore the stair detection methods were explored for the ramp detection application i e skewness However the skewness approach was poor for detecting a change of state from level ground walking to ramp ascent or descent Since the evaluation was performed on subject with no p
196. the rotation around the x axis pitch the rotation around the y axis and yaw the rotation around the z axis Figure 5 3 Sensor placement 5 2 Xbus Kit The Xbus kit consists of an Xbus Master XM B XB3 and five MTx motion trackers MTx 49A53G25 184 186 The five MTx and the Xbus Master are interconnected in a daisy chained configuration The Xbus Master delivers power to the five motion trackers and retrieves the sampled data With the output mode set to orientation mode with quaternion units each MTx data record contains four float numbers Each float number is 4 bytes long and corresponds with the single precision floating point value as defined in the IEEE 754 standard For every data sample the packet sent is a total of 87 bytes 4 bytes 4 float number 5 sensors 7 bytes for header The message structure contains the following fields PREAMBLE BID MID LEN DATA CHECKSUM Development of a Wearable Mobility Monitoring System 63 Preliminary Evaluation of the BlackBerry for WMMS 5 3 Java Programming A Java application was developed using the BlackBerry Java Development Environment version 4 5 0 7 The Java application was then uploaded to the BlackBerry platform through the BlackBerry Desktop Manager The BlackBerry API application programming interface net rim device api bluetooth was used to initiate a Bluetooth serial port connection and to write and read data from the port The Java application us
197. the type of ground or terrain which is important for mobility monitoring The algorithms developed to detect change of state were satisfactory however with increased processing power in the next generation of smartphones more complex signal processing methods could be employed to improve results Overall our WMMS has good potential for community mobility monitoring The smartphone approach provides an accessible and cost effective option that can easily be implemented in society However the limitations should be addressed to improve performance Interesting future work exists for the WMMS 9 1 Future Work Improvement to the change of state algorithm is necessary to detect going up stairs the ramp and the indoor outdoor Additional signal processing could be added offline to Development of a Wearable Mobility Monitoring System 127 Conclusion improve classification of the raw data An automated process to identify context from the images can also be considered in future research Developing a better software interface would be important as well as post processing software for data and images so that rehabilitation specialists could easily interpret the community mobility data Implementing the change of state algorithm to the new generation of BlackBerry smartphones should be considered since new versions provide raw accelerometer data and improved camera performance This will remove the need for the external board for
198. ther individuals These dimensions capture the external demands for independent community mobility Therefore with this model disability level is expressed as the range of environmental contexts where the tasks required to perform daily activities can be carried out as opposed to expressing disability level by the number of tasks a person can or cannot do 1 Development of a Wearable Mobility Monitoring System 7 Literature Review Minimum Walking Distance Traffic Level Time Constraints Postural Transitions Ambient Condition Attentional Demands Terrain Characteristics External Physical Load Figure 2 2 Dimensions of Mobility framework reproduced from 1 Interestingly around the same timeframe a study by Stanko et al 20 used an open ended questionnaire to ask 15 physiotherapists which tasks and destinations are important to include in a new outcome measure The paper mentioned that the study was completed before the dimension of mobility model was published and therefore the respondents were not influenced by that research The responses obtained identified items in each of the eight dimensions which clearly emphasized the role of the environment in defining mobility The Dimension of Mobility framework was explored further by Shummay Cook et al 21 who examined environmental challenges that older adult with and without mobility impairments would encounter while walking in the community The frequency of
199. ting up If 1 increased in acceleration intensity If 0 back to normal intensity 96 Development of the Prototype WMMS External board BlackBerry Bold Light Sensor Data Acceleration Signals GPS Data Data Pre processing Features Generation Standard deviation Y axis DT Algorithm Signal Magnitude Area SMA DT Algorithm Inclination Angle Skewness Y axis DT Algorithm In Standing Range No SMA PEAK 0 SMA STA STA STAND 1 STAND 0 PEAK 1 DYN 1 DYN 0 STAIRS 1 STAIRS 0 DT Algorithm Speed GPS Light intensity DT Algorithm No No No Yes SMA SMA Increased intensity In Lying Range Figure 7 10 State determination algorithm DT stands for double threshold Development of a Wearable Mobility Monitoring System 97 Development of the Prototype WMMS From the camera performance test in Chapter 6 approximately 0 7 second was required to take a picture and the BlackBerry Bold camera needed another 0 9 second before it was ready to take another picture During that time the BlackBerry Bold was busy and no data was received and processed causing the data to accumulate in a buffer The affected timing could be demonstrated by observing the time frame of every window of data from the WMMS output file as presented in Table 7 2 The section of the WMMS output file in Table 7 2 was recorded with a sampling rate of
200. tion 10 digital and two 8bit analog I O Enhanced Data Rate EDR compliant for both 2Mbps and 3Mbps modulation modes Serial interface up to 4Mbps No additional Bluetooth qualification needed Physical size LxWxH mm 28 5x15 2x2 0 Weight 1 2grams Supply voltage regulated 3 1 3 6 VDC 2 4 V to 3 6 V single supply operation 2 g 6 g user selectable full scale Maximum bandwidth of 1 8kHz Low power consumption Output voltage offset and sensitivity are ratiometric to the supply voltage Sensitivity at Full scale 6g Typical Vdd 15 V g Sensitivity change Vs Temperature 0 01 C Zero g Level Voffset at Full scale 2g Vdd 2V Zero g Level change Vs Temperature 0 4mg C Weight 0 040 grams Physical size LxWxH mm 4x4x1 5 VCC supply 1 8 to 5 5V Low sensitivity variation across various light sources Peak sensitivity wavelength typical 500nm Physical size LxWxH mm 1 50x1 60x0 55 Photo current with Lux 100 from incandescent lamp typical 44 uA Dark current Lux 0 typical 300 nA Relative Humidity Resolution typical 12bit 0 05 RH Accuracy typical 3 0 RH Repeatability 0 1 RH Response time typical 8s Operating Rage 0 100 RH Temperature Resolution 14 bit 0 01 C Accuracy typical 0 4 C Repeatability 0 1 C Operating range 40 to 123 8 C Response time 5 30s Development of a Wearable Mobility Monitoring System 72 Hardware Design and Evaluation 6 2 3 Board Funct
201. tool could also help monitor progress or deterioration thereby providing an indication of treatment effectiveness 1 1 Contributions This thesis presents a Wearable Mobility Monitoring System WMMS to monitor a person s mobility at home outside the home and in the community Our proposed WMMS provides solutions to the limitations of current assessment tools by providing unsupervised objective mobility measurements in a cost effective way The WMMS also provides information on the context and environment in which mobility event takes place which could identify mobility challenges in a person s own environment The WMMS was developed using a smartphone based approach which takes advantage of the smartphone s available features such as GPS camera Bluetooth and Wi Fi to create an all in one WMMS The WMMS is worn comfortably and freely on a person s belt just like a normal phone A Smart Holster was developed to hold the phone at the hip and provide Development of a Wearable Mobility Monitoring System 2 Introduction additional sensor data such as accelerometer light sensor and temperature humidity Sensor To the best of our knowledge an all in one wearable system using a smartphone to monitor a person s mobility in his or her everyday environment as well as using a camera to provide insight on the environment and context has not been explored 1 2 Scope ofthe Thesis The WMMS was designed to monitor a user s mobili
202. tooth to communicate with external sensors and a Bluetooth Java API Application Programming Interface already exists Wi Fi is based on the IEEE 802 11 family of standards The Wi Fi standard allows a personal server to connect to a WLAN Wireless Local Area Network In medical applications Wi Fi could be used to send data from a WBAN via the internet to a remote heath care server Many recent smartphones have this wireless technology Wi Fi is usually not a good candidate for communication between sensors and a central node due to the power requirements 85 A WBAN or WBSN usually requires sensor nodes to be ultra low power which implies that signals from stronger sources may interfere with the sensor signal and could result in sensor data loss 85 2 3 4 Wearable Sensors Wearable sensors or body fixed sensors are attached on the body to monitor the person s kinematics and physiologic parameters as well as contextual information Recent technological advances have produced low cost and miniature sensors which have created great opportunities in designing a wearable system for health monitoring Various wearable sensors have been used for tracking human posture and movement Wong et al 107 presented five sensor classes in their review 1 accelerometers 2 gyroscopes 3 flexible angular sensor 4 electromagnetic tracking systems and 5 sensing fabrics with accelerometers being the most commonly used The main types of body fix
203. top of car ride IC 1 IC IC NA 1 1 0 1 IC IC 1 1 1 1 10 8 80 096 Sit to stand NOP NOP transition 0 0 1 L 9 9 1 1 1 1 IC IC 1 is 9 69 296 Walking outside NOP o on level ground IC 1 1 0 0 1 1 1 1 1 1 1 1 0 1 15 11 73 396 Transition outdoor indoor pa d y os s 0 yi ps Pp 0 1 1 bud ird ne 4 2 50 0 and keep walking on level ground Standing 1 1 1 1 1 1 1 1 1 1 NA 1 1 1 1 14 14 100 0 Total Number of Pictures 27 30 29 29 30 31 29 29 28 26 30 34 29 30 29 440 Tael Numae ci 16 23 18 18 18 19 24 23 23 15 22 28 24 26 24 321 Success Total 96 of o 59 3 76 7 62 1 62 1 60 0 61 3 82 8 79 3 82 1 57 7 73 3 82 4 82 8 86 7 82 8 73 0 Successfully Identifying Context Yo Yo 96 Yo Yo Yo Yo Yo Yo Yo 96 Yo Yo 96 q xipueddy Appendix E Appendix E This appendix contains the ethics approval letters from University of Ottawa Research Ethics Board and the Ottawa Hospital Research Ethics Board Development of a Wearable Mobility Monitoring System 162 Appendix E Universit d Ottawa University of Ottawa December 10 2009 Ga tanne Hach Nathalie Baddour Department of Medical Engineering Department of Medical Engineering University of Ottawa University of Ottawa ghache toh on ca nbaddour eng uottawa ca Edward Lemaire The Ottawa Hospital Rehabilitation Centre elemaire toh on ca re U of O Et
204. toring System 98 Development of the Prototype WMMS 7 7 Software development The software part of the WMMS was developed using Java Eclipse and the BlackBerry Java Development Environment component package version 4 6 1 The Java application was then uploaded to the BlackBerry platform through the BlackBerry Desktop Manager The BlackBerry APIs application programming interface and the Java packages that were used for this Java application are 193 e net rim device api bluetooth to initiate a Bluetooth serial port connection and to write and read data from the port e net rim device api math Fixed32 to execute specific math functions such as arctan2 e net rim device api ui to provide functionality to construct the user interface e net rim device api util to provide utility methods and interfaces such as arrays and data buffer e net rim device api system to provide system level functionality such as the control of the BlackBerry backlight and information on the battery level status e javax microedition io Connector and javax microedition io FileConnection to copy data and images to output files stored on SDcard or device memory e javax microedition media to take picture with the BlackBerry Bold integrated camera e javax microedition location with the LocationListener interface to obtain GPS location coordinates and speed e java io to provide system input and output to data stream e java lang math for other math fun
205. ts the orientation of the segment with respect to the gravitational field as illustrated in Figure 2 10 The inclination angle 9 can be calculated using Equation 2 9 Development of a Wearable Mobility Monitoring System 41 Literature Review p cos ED where a is the measured acceleration and g the gravitational acceleration 9 81 m s This feature has been used to detect postures 9 147 149 and also to identify postural transition 155 However the technique presented in Figure 2 10 and Equation 2 9 only uses one axis for the angle calculation and is subject to resolution problems when the measured acceleration is near 1g or 1 g 172 The one axis technique only allows for a 180 degree range To fix the resolution and range problem Freescale Semiconductor 172 described a method of calculating inclination angle using two axes Figure 2 11 Using basic trigonometry the acceleration in the x axis can be expressed with the following equation Ay sin 2 10 Similarly the acceleration in the y axis can be expressed with the following equation A cos 0 2 11 then by combining Eguation 2 10 and 2 11 the following eguation is obtained fx ano GA A Y With the two axes technique a 360 degree range can be measured using the sign of the acceleration of both x and y axis From the sign of the accelerations the quadrant in which the tilt occurred can be identified and the proper tilt angle can be determin
206. ty Monitoring System 136 References 96 N Ravi N Dandekar P Mysore and M L Littman Activity recognition from accelerometer data in Proceedings of the National Conference on Artificial Intelligence 2005 pp 1541 97 Wikipedia Smartphone Wikipedia The Free Encyclopedia Online Available http en wikipedia org wiki Smartphone Accessed 16 Sep 2009 98 M J Moron J R Luque A A Botella E J Cuberos E Casilari and A Diaz Estrella A smart phone based personal area network for remote monitoring of biosignals in Proceedings for the International Federation for Medical and Biological Engineering 2007 pp 116 99 P Van De Ven J Nelson A Bourke and G O Laighin A wearable wireless platform for fall and mobility monitoring in Zst International Conference on Pervasive Technologies Related to Assistive Environments 2008 100 P Roncagliolo L Arredondo and A Gonz lez Biomedical signal acquisition processing and transmission using smartphone Journal of Physics Conference Series vol 90 2007 101 M D Bloice F Wotawa and A Holzinger Java s alternatives and the limitations of java when writing cross platform applications for mobile devices in the medical domain in 31st International Conference on Information Technology Interfaces 2009 pp 47 54 102 X Zhang D Cao and H Mei Improve the portability of J2ME applications an architecture driven approach in Third Intern
207. ty state and to take a photograph when a user s change of state related to mobility was detected The taken photographs assist in defining the context of the mobility event 1 e using an elevator walking up a ramp type of walking surface etc The changes of state that were evaluated in this thesis were starting or stopping an activity e g walking running cleaning sitting down lying down getting up i e from chair bed going up and down stairs using transportation e g bus car biking and moving between indoors and outdoors The WMMS was intended for people with physical mobility disabilities or at risk to develop mobility disabilities but who are still mobile in the community People with age related pathologies such as stroke osteoarthritis and other physical illness which are often associated with a reduction in mobility could also benefit from this wearable system The validation process was performed on five able bodied subjects The subjects were asked to do a series of predefined mobility tasks such as walking going up down stairs walking up down a ramp sitting lying walking outside taking the elevator and riding in a car The system was evaluated for sensitivity and specificity for detecting changes of state The pictures were evaluated for their usefulness in defining the context of the mobility event For this pilot study the system was not intended to recognize all activities However from the di
208. typically occurred when the Xbus Master did not receive the Motion Tracker response within the measurement period 186 This was an internal error with the XBus system Following this error the Xbus Master stopped sending data and the BlackBerry application had to be re started When no error occurred after 2 5 hours and the application was still running data Development of a Wearable Mobility Monitoring System 64 Preliminary Evaluation of the BlackBerry for WMMS collection was stopped manually For each trial the time the system ran without error the BlackBerry battery level before and after each trial the amount of data loss and the error that made the Xbus Master stop were evaluated Following the 50 Hz and 25 Hz static data collection trials another five static trials were run at 50 Hz but with minimal processing e g static minimal trials For these trials the Java application was modified to only receive motion data no biomechanical parameters were calculated no GPS data were received and no data file was created This was to verify that the Java application was not causing the Xbus Master to stop early during data collection Finally dynamic trials were performed to simulate real orientation angle measurements The sensors were attached on a subject s lower limbs and hip Figure 5 3 The Xbus Master was powered by battery Five trials were run at 50 Hz and 25 Hz for as long as possible This set of dynamic trials wa
209. utes Error 0x1C Average Time Description of Trial minutes Static 50 Hz Static Minimal 50 Hz Dynamic 50 Hz Dynamic 25 Hz 5 6 Preliminary Evaluation Discussion The error sent by the Xbus Master was always error code 28 implying that a timer overflow occurred during measurement i e the Motion Tracker response was not received by the Xbus Master within the measurement period 186 Ignoring this error instead of having the application stopped would have been ideal A few missing data points would have not been as critical as missing a large amount of data due to the application stopping However the Xbus kit was a commercial system that provided minimal control of error handling between the XBus and the MTx sensors Since lowering the sampling frequency showed a decreased in error occurrence a value lower than 25 Hz could have potentially avoided the error However in human motion measurement using accelerometers a sampling frequency lower than 25 Hz might not be sufficient Section 2 4 2 Results from the static minimal trials showed that removing processing logging sensor data and including GPS data did not improve the total sampling time The error code was always the same i e timer overflow The results suggest that the problems encountered during measurement were a result of external sensor errors One of the design criteria for the WMMS is that battery should last at least one day on one charge Sect
210. were less than 15 Hz The major energy band caused by daily activities was found by Sun and Hill to be between 0 3 to 3 5 Hz 165 Many studies related to the measurement of frequency and amplitude spectra of human body accelerations including the ones in the above were reviewed by Bouten et al 80 to determine the appropriate specifications to use for their accelerometer For daily activity assessment Bouten et al concluded that body fixed accelerometers placed at the waist must be able to measure acceleration with amplitude ranging from 6 to 46 g and frequencies up to 20 Hz 2 4 3 Calibration Accelerometer calibration is usually required to correct for DC offset and signal drift Having a DC component in the signal allows for easier calibration of the sensor One simple Development of a Wearable Mobility Monitoring System 37 Literature Review calibration method is based on rotation of the sensor to known angles For example under static conditions if the axis of interest is pointed towards the center of the Earth the output should equal 1g If the axis is then rotated by 180 degrees its output should equal 1g This 1g rotation method is often suggested by manufacturers to calculate the sensitivity s of a particular axis of the sensor 166 U max z U min 2 4 2 Ss The offset o can also be corrected using a similar equation EE 2 5 2 Oo Where Umar and Umin are the maximum and minimum acceleration
211. with feet fixed retrieving object from floor turning 360 degrees stool stepping and reaching forward while standing Each item is scored on a scale from zero to four with a maximum possible score of 56 A score of 0 to 20 represents balance impairment 21 to 40 represents acceptable balance and 41 to 56 represents good balance Although originally designed for older adults a recent systematic review by Blum and Korner Bitensky 34 about the BBS psychometric properties for stroke rehabilitation suggested that the BBS is a valuable tool for assessing clinical change in balance after stroke 2 2 1 5 Timed Up and Go Test The Timed Up and Go Test was originally called Get Up and Go Test 35 but the name changed after the test was validated with a timed score 36 This simple test Development of a Wearable Mobility Monitoring System 10 Literature Review consisted of asking the patient who is sitting in an armchair to stand up walk three meters turn around walk back to the chair and sit down The time taken to execute this task can predict the person s ability to go outside alone safely 2 2 1 6 6 Minutes Walk Test The 6 minutes walk test was developed to measure functional capacity of people with respiratory and cardiac conditions The test is usually performed indoors on a long flat and straight path but could also be done outdoors if the weather is comfortable 37 This test measures the distance that a patient ca
212. wog yao v 5 id 79snvixa td Per gt gt umogiemod 390v 94 COIX3 Id amp Id Hz DW 1193738 7399 KC Kviva He aeo 1487 sueiquis r 01535383350 v za 5 re amp x05 s Xd uiooienig E gt gt ise doom SAfd gt gt zzy XL upooiig gt cu zezSH 1 SLO woo enig K aBeuon 10 3390v seus 399 Genoa iar iuerquiy SLM uonane gt Kanon S9HO01dL T noA tal smia amp uopsauuog yioozenjg Injssasong gt uonsauuog qicoje e uang eser 148 Development of a Wearable Mobility Monitoring System Appendix A empon qioo1enig auo 101dS i I I E I gt I po i wg 8002 Te J8qum oN Kepu LII a lt 00 gt s na a qumw iueumcog azr OSIW lds ISOW Ids 889 Ids X10 WOd 1n0 Wod sn GEBIESNMA ON ON 2 uoromuuo9 weoientg iwang ason gt adt QNO OSIW X99 Kway uocum gt gt Xi uoan z Wd uonje z K sio wooieng 2 gt gt Siu uoowng z gt vonauuog poong sses z KC uopsauuog ipone wan aser z z XL woo z S19 wooente gt gt S4 ISOW siu noone z z ieseu upoojenie gt gt 1383H 0HI lt Kumoowng z 90 OND Shed gt vopsauuog yoong ysseoong z aenog o Teacyzdo 149 Development of a Wearable Mobility Monitoring System Table B 1 Compiled results for each trial of the five subjects Subject Trial 1 Total 2 To
213. zepines Measured vs observed readings L1 distances standard deviations resolutions of 20s and 40s Rectified and integrate output of accelerometer Activity obtained from addition of integrated outputs for 1 minute Discrete wavelet transform DWT optical reference system Vicon MOIADY ANPI urojs amp s SuuojruoJA AITIQOJ 9 qe1e9 AA L JO juoeuido oAo q SE 2003 2004 Mathie et al 7 Najafi et al 155 Bao et al 156 Luinge et al 157 Lyons et al 149 Baek et al 141 Culhane et al 148 1 front of waist 1 chest 1 wrist 1 waist 1 upper arm 1 thigh 1 leg upper back 1 pelvis sternum 1 upper thigh 1 waist 1 chest 1 thigh Sensitivity 0 98 specificity 0 88 0 94 Postural transition 99 average sensitivity and specificity 94 and 9596 Ranging from 41 4296 to 97 49 Sit 93 stand 95 lying 84 97 5 Activity of daily living ADL 11 discrete dynamic activities sit to stand stand to sit walk 12 distinct rest periods stand sit Sitting standing lying walking postural transitions gyroscope Walking sit and relax stand watch television run stretch scrubbing fold laundry brush teeth ride elevator walk carry read cycle climb stairs vacuuming lie down strength training etc Posture inclination of trunk and pelvis Posture and movement detection static and dynamic activities postures sit l

development of a wearable mobility monitoring system

Contents

Download Pdf Manuals

Related Search

Related Contents