Home

research

1. Algorithm for converting the data frame into matrix format 1 Read one frame of data sent by microcontroller Convert the HEX data for A B and C into binary format 2 Store 96 binary digits 32 for each segment in matrix at co ordinates starting from 0 0 to 7 11 3 Convert HEX data for D into binary format and store remaining 32 binary digits in matrix at co ordinates from 11 0 to 7 0 as shown in fig 3 4 4 Store the entire matrix in file month dd yy hours MATRIX txt Go to step 1 0 0 Start 0 11 7 0 End 7 11 11 0 11 11 Figure 3 4 Figure shows the conversion of data frame into matrix format The data is populated in 128 elements starting from Start up to End The 16 elements shown in lower right corner are unused and left blank 13 For Example the binary data for every segment in the frame SA8 1 000000B00000080C00000080D8 1 OOO000E is Segment A 10000001 00000000 00000000 000000002 Segment B 00000000 00000000 00000000 100000002 Segment C 00000000 00000000 00000000 100000002 Segment D 10000000 00000000 00000000 00000000 A B C cD o o ojo o o lo o ao CD o o o aojo a o afo 0 d 0 o o o aojo a o afo oO 0 o o o ajo a o afo 0 d 0 o o o ajo a O afo oO d 0 o o o aojo a o afo 0 OQ o o o ajo o ojo o oO OQ o o ojo o o ojo o o o o o o O O O OO 0 4 4 4 4 o o Oo O o O GO o Do o o o o o
2. Figure 4 15 shows the scenario for acquiring the data shown in Figure 4 14 The carpet data is displayed on the remote computer without any wired connection to the microcontroller or the Sheevaplug server It receives the data through the socket Internet connection to the server The Sheevaplug server transmits every frame of data and places a time of occurrence at the end of the data frame 44 Data display on remote Computer SS Figure 4 15 Picture shows the scenario in which a person is standing on the carpet sensor making it active Sheevaplug is shown connected to the microcontroller circuit to receive and transmit the data over the Internet Remote computer is shown receiving the data and displaying it We calculated the time elapsed for a frame to go to the remote computer Beginning with the time of occurrence as described above sent to the remote client computer The computer then sends an acknowledgement with the same timestamp Upon receiving it the Sheevaplug server calculates the half of difference between the time of reception of acknowledgement and transmission of the data We used the server s clock time in the scale of nanoseconds for precise calculations For example Transmitted frame SAQOQ000000BO0000000CO0000000D00000000E 384177024645945 Received Acknowledgement ACK 384177024645945 at 384177257609945 Round trip time 384177257609945 384177024645945 ns Time taken to transmi
3. Figure A 7 Figure shows the carpet display after pressing the start button The red color rectangles represent the active sensors with their current activation count ccccccccccccccceceeeeeeeeeeeeeeeeeeeesseseeeseeeeeess 68 Figure A 8 Figure shows the file explorer window Carpet Data for displaying the data files located on the Sheevaplug server User can right click on any CSV file and select Show History menu to retrieve the data stored in the file The terminal is also shown for invoking commands to the Sheevaplug server 69 Figure A 9 Figure shows the Sheevaplug server terminal invoking the required processes 000008 70 vii LIST OF TABLES Table 3 1 Table showing the list of Eclipse frameworks used for the development cccccccccceeeeeeees 11 Table 4 1 Time required ms by a computer program to convert the data received from microcontroller into matrix and CSV formats First two columns describe the time for two segments while the remaining columns describe the time for three and four segments respectively w represents the mean whereas o represents the standard deviation of the observed times over the period of 12 hours 0 0 0 41 Table 4 2 Table indicating the performance of fall detection system for fall experiments performed on the carpet Three types of activities are performed in each direction at 40 times ccccceeeeeeeeeeeeeeeeeeees 51 Tabl
4. in Wearable Computers 2003 Proceedings Seventh IEEE International Symposium on 2003 pp 184 187 60 11 12 13 14 15 16 17 18 19 20 21 U Lindemann A Hock M Stuber W Keck and C Becker Evaluation of a fall detector based on accelerometers A pilot study Medical and Biological Engineering and Computing vol 43 pp 548 551 2005 N Noury P Barralon G Virone P Boissy M Hamel and P Rumeau A smart sensor based on rules and its evaluation in daily routines in Engineering in Medicine and Biology Society 2003 Proceedings of the 25th Annual International Conference of the IEEE 2003 pp 3286 3289 Vol 4 G Wu Distinguishing fall activities from normal activities by velocity characteristics Journal of Biomechanics vol 33 pp 1497 1500 2000 L Chia Wen and L Zhi Hong Automatic Fall Incident Detection in Compressed Video for Intelligent Homecare in Computer Communications and Networks 2007 ICCCN 2007 Proceedings of 16th International Conference on 2007 pp 1172 1177 H Foroughi B S Aski and H Pourreza Intelligent video surveillance for monitoring fall detection of elderly in home environments in Computer and Information Technology 2008 ICCIT 2008 11th International Conference on 2008 pp 219 224 D Anderson J M Keller M Skubic C X1 and H Zhihai Recognizing Falls from Silhouettes in Engineering in Medicine and Biolog
5. 11 and c 11 then go to step 4 3 Ifc lt 11 then increment c by 1 and go to step 2 Else increment r by 1 set c 0 and go to step 2 4 Transform the result list into comma separated value line using openCSV library 36 and store it in month dd yy hours csv We will use the matrix from Figure 3 5 for the demonstration The co ordinates of Is in the matrix are 0 0 0 7 0 11 7 0 11 0 We defined the following CSV format for these co ordinates 1 0 0 1 0 7 1 0 11 1 7 0 1 11 0 Mar 1 2012 4 19 22 PM Figure 3 6 demonstrates the conversion of matrix data in a line of CSV format 15 0 0 A B C 0 11 iyo o aojo o o lo o 08 o o o aojo o oojo o Of O o o o ofo o yy ojlo o A o o o o ofo o A ajo o o o o o o ojo o o of o 0 o o o o o ofo 6 o ojl o o o o o o ofo o o oyo o o To o o Oo ao o d o o a oO D o0 0 0 0 0 0 o owa Of o o 0 O Dio o o MO 0o gpro o 11 0 Oe vs C1 L1 Tisi 1 0 0 1 0 7 1 0 11 1 7 0 1 11 0 Mar 1 2012 4 19 22 PM Figure 3 6 The demonstration of conversion of one frame of matrix data into CSV format 3 5 Sheevaplug computer as a project server The data acquisition and storage system was intended to operate for 24x7 to collect all the movements happening on the carpet In such applications we may not always need the full fledged desktop computer Those are costl
6. that we developed Following is the web address for the website http carpetsmart dyndns server com Figure 4 21 shows the screenshot of the smart carpet website The website worked as a first access point of the system Users can download a required executable from this website and can use it after fulfilling the user authentication requirements O Smart Carpet Server Over amp carpetsmart dyndns server com overview htm wit Smart Carpet Server Center for Eldercare and Rehabilitation Technology Home Overview People Resources Overview Alzheimer s is becoming a most hazardous disease in United States since past several years especially in the age group of above 65 Though Alzheimer s does not have any relation with aging facts sheet from the Alzheimer s association shows that most of the patients are older than the age of 65 The main symptoms of Alzheimer s are loosing control over memory and body movements lacking knowledge of time and space sudden changes in the mood and personality This may result into their unintentional and uncontrolled behavior such as walking and possibly falling on the floor which can lead to serious damages or fractures to the body Such movements by the patients needs to be monitored to prevent these damages There are several systems in place to solve this purpose by monitoring the movements locally However we intended to develop a new system to record the data of patients each and
7. we required data in different formats for its further analysis storage and display This section discusses the details of conversion and storage techniques in two different formats The matrix shows each data scan in the same form as the sensor layout on the floor providing positional information The second format Comma Separated Value CSV takes the advantage of sparse nature of the data and reduces the frame size 3 4 1 Conversion of carpet data in matrix format Each frame of data contains 32 HEX digits representing the floor sensor data for four carpet segments each having 32 sensors For example if a received data frame is SA80000000B00000000C00000000D00000000E then first digit in A 8 10002 informs the activation of first sensor in segment A but all others inactivated The entire frame shows the activation information of all 4 32 sensor segments 1 e 128 sensors in the system This raw data format is very efficient for transmission however it s difficult to correctly locate the person on the carpet Hence we developed the computer program to convert this data in the matrix format We chose the matrix size as 12x12 having 144 elements to accommodate the data for 128 sensors and the 16 blank sensor spaces 12 Figure 3 4 demonstrates the conversion of frame into matrix format The computer uses the data between S and E to fill the matrix as shown It starts from start and continues filling up the elements up to end
8. 0 oO gt o o o o o O D Figure 3 5 The demonstration of conversion of one frame of raw data into matrix format Hence the matrix format clearly shows the position of active sensors in the room and in turn helps identifying the person s location on it However the matrix may contain large number of zeros representing the inactive sensors Such matrices are called as sparse matrices and are very inefficient to store In order to achieve memory efficiency we needed to eliminate all unnecessary Os and decided to store only 1s along with their matrix coordinates Next sections describe the storing technique of sparse matrices in memory efficient and user readable format 14 3 4 2 Conversion of matrix carpet data in Comma Separated Value format A CSV format is an efficient format for storing tabular or matrix data It stores the data in a single line in which each datum separated by commas Because of sparse nature of the data we store only the co ordinates of the Is in the matrix as the records We developed a program to search the co ordinates of Is in horizontal raster scan then create a single line of all these 1s records by separating them using commas The pseudo code is described as follows 1 Let matrix row be r and column be c Let result list be the list to store the co ordinates Set r 0 and c 0 Start reading matrix elements at r c 2 Ifthe value at r c 1 then store the r c co ordinates in the result list If r
9. LAA apse ao Sf m Sheevaplug Server Linux Ubuntu 2 6 30 2 11 PREEMPT Wed Jul 22 19 53 31 MDT 2009 armv5te The programs included with the Ubuntu system are free software the exact distribution terms for each program are described in the individual files in usr share doc copyright Ubuntu comes with ABSOLUTELY NO WARRANTY to the extent permitted by applicable law To access official Ubuntu documentation please wisit http help ubuntu com Last login Mon Jul 16 19 11 30 2012 from 192 168 10 108 root ubuntu cd carpet root ubuntu carpet java jar Collector jar ServerProperties properties Figure A 9 Figure shows the Sheevaplug server terminal invoking the required processes 70
10. believe that the false positive rate increases as the number of people increase In the future detection performance of this system can make use of Receiver Operating Characteristic ROC space to show the tradeoff between true positive and false positive results It helps in evaluating the system s ability to discriminate falls and non falls Also ROC curves obtained by taking weather and light conditions in the room into consideration 4 can help determine the fall false alarms under a range of operating conditions The microcontroller provides floor sensor signal in binary format 1 e the signal is high or low The ongoing research for producing data for the levels between these two extremes will possibly help identifying the intensity of fall weight of a falling person etc This will improve fall detection abilities remarkably Human activity modeling techniques such as Bayesian modeling can also be used for improving fall detection abilities A model can be created using major body parts such as torso hands and legs 58 with other parameters as fatness length and height Every matrix data frame can be summed to prepare distribution that can be used to compare with the model values 1 e relating it to human body shapes for detecting falls and correctly distinguishing falls from other objects This will also help reducing false alarms We used Google Voice messaging service for reporting falls as it is free for calling and te
11. frame of data in the matrix format to determine the size of the smallest connected sub regions to determine the threshold level of connectivity Falls were declared when a connected sub region was found larger than the predefined threshold We also used the rapid change in the number of connected elements between the last frame walking and the first frame of a fall The fall notification is triggered only when the matrix satisfies both the above conditions 1 e detection of largest connected region with rapid change from the previous The tests were carried out in laboratory with a volunteer performing an intentional fall The Sensitivity and False Negative Rate results shown in Table 4 2 Table 4 3 indicates that the proposed 57 algorithm effectively solves the purpose of fall detection The false negatives miss conditions were observed to be very low giving the high accuracy which indicates that the system is highly reliable for a single person We executed false positive fall experiments defined by the optical fall detection system using one volunteer at a time 41 We observed very low amount of false positives Table 4 4 Table 4 5 which indicates the highly specific system These tests were carried out in laboratory by a single person at a time and in a carefully controlled environment Individuals are of variable size and we believe that the threshold setting can help adjust the algorithm We have not tested for two or more people We
12. in which each column corresponds to 2 hexadecimal digits in the raw data Similarly in the lower segment D there are 4 rows in which each row corresponds to 2 hexadecimal digits For example the data for C and D in Figure 4 3contains all zeros hence the matrix equivalent of them is zero The last two digits in A are 38y 1 e 0011 10002 which corresponds to the last column in A Similarly the second two numbers in B are 15y corresponds to 0001 01012 which are mapped in segment B 2 column of the matrix 2012 05 16 12 53 19 628 B C a Hes H a a Hos fe we as z to Po F H Hes He Hes ne H He Hos Pa iA E e e M e e H 3 ko Coe i H ii LA I La Gi Cod Gi Ga Gi Gi GA Gi Oa H a oh i gt Ofe e Olje r i A LA i Li Figure 4 4 Data shown is stored in matrix format The elements with 1 show the active sensors in the carpet and 0 shows the inactive sensors The crossed region represents the floor area with no carpets laid 36 The highlighted frame in Figure 4 5 is achieved from the matrix shown in Figure 4 4 by storing all 1s with their row and column coordinates and neglecting all Os All these formats are stored with the same timestamp which makes easy retrieval 1 5 5 2012 05 16 12 53 19 581 1 3 3 1 5 5 2012 05 16 12 53 19 605 M12 3 ML fe 6 m TI 33 TI 3 4 15 5 15 6 014 39 1 5 5 e le 6 6 6
13. is the rapid rise as shown in the graph The trailing pulse shows the person s rolling state after the fall Observation of data in connected components Fall occurred on 12 49 47 826 PM with largest component size 10 ooo TW doa l a A ee ae hi uan b ah bata aE ma y J Ki ol q he ka S q S q Kaf q Ka AD 3 e a e Pg gh gh ott ot Ov a Sr ee ate eo a s ae nh a at a ca Cj ee a se aon ts A a a Ba id i se g se n g wt k sI Mi k Si MEON S i sie Largest Component Size en on K Fa E GA Ou E ka Time Figure 4 20 Graph for showing the variation in largest component with respect to time The vertical line parallel to ordinate describes the fall instane with maximum component size 10 4 3 2 Performance of the fall detection system We carried out the fall experiments to characterize the performance of fall detection system based on the falls test protocol 41 Table 4 2 shows the results of fall experiments for 4 volunteers each performed at 10 times Each row in table describes the results for a single experiment with all 4 volunteers For example the first row shows that all the falls for first two volunteers were detected while it missed one fall per other two volunteers for fall type standing to falling forward For each experiment we also calculated the False Negative Rate FNR and the Sensitivity and displayed them correspondingly 50 Table 4 2 Table ind
14. microcontroller systems 4 We tested the Sheevaplug server and a remote computer display Importantly we tested the system s ability to detect falls We walked on the carpet and the data was recorded on the server and observed on the remote computer With each footstep the display was expected to change accordingly with a fall on the carpet a mobile phone was expected to receive a notification The experiments were categorized as follows 3 9 1 Observation of carpet data on terminal emulator This experiment was meant to verify if the USB connection produces the same data as observed with RS 232 connection We produced the data by walking on all 4 carpet segments The system was setup as shown in Figure 3 2 and the data was observed on SecureCRT terminal window 3 9 2 Organization of data in files The experiment was carried out to store the data in three different formats raw data matrix format CSV format Sheevaplug server was connected to microcontroller using RS 232 to USB adapter Figure 3 2 We acquired the data during day time by walking and falling on the carpet as well as by leaving the system idle in the night These activities were individually performed by the graduate students in the lab For every hour of data acquisition the program created three different files with the three formats raw 2 data matrix CSV The system was setup to generate the data for 4 weeks and continued running for that time 3 9
15. provided the solution for detecting falls using the floor sensor data We also organized this data in files and developed a display for presenting this data on the remote computer The entire system was built using the Sheevaplug computer after successfully achieving its common ground with the microcontroller system 55 CHAPTER 5 DISCUSSIONS CONCLUSION AND FUTURE WORK We have developed a fall detection and notification system using inexpensive technologies such as aluminum foil sensors plug computer and a Google messaging service We installed the system in our laboratory and tested it for variety of fall situations This chapter discusses our decisions findings and actions Considering the wide range of programming languages and development environments available we used Java v1 5 and Eclipse v3 5 since we have had previous with these resources Eclipse was used to create a software executable and a web application using a single set of source code The executable was targeted only for the Windows systems However other operating systems such as Linux Macintosh and Solaris etc can accomplish the same tasks using Eclipse Delta Pack 42 We programed wall plug computer to acquire and store the carpet data extending previous work 24 Being a novice user of plug computer we chose the cheapest one 1 e Sheevaplug to test its feasibility Initially we observed corrupted data which we corrected after providing a ground connect
16. software executable using Eclipse 2 Detailed directions for using the software executable on remote computer 3 Instructions for initiating the Sheevaplug server This manual is valid only for using on the Windows Operating systems and Eclipse v3 5 with its frameworks described in Table 3 1 A 1 Creating a software executable using Eclipse 1 Download Eclipse v3 5 Galileo from its official website 28 Double click the eclipse exe file and open the attached source code repository as shown in Figure A 1 S Workspace Launcher Select a workspace Eclipse stores your projects in a folder called a workspace Choose a workspace folder to use for this session Workspace C Users Kaustubh Elder Care Use this as the default and do not ask again SZ Gow Figure A 1 Figure shows the launch screen of Eclipse workbench with the source code repository 2 Open the file edu missouri eldercare product from a package explorer on the left side Open the overview tab on the right side Click on Export Eclipse Product Wizard for exporting an executable file as shown in Figure A 2 65 Bjo GPR BP x Package Explorer X Plug ins eS B AppFramework jar batik awt util 1 7 jar batik svggen 1 7 jar batik util 1 7 jar ard build properties calendarcombo 1 0 0 src zip B calendarcombo 1 0 0 jar B comm jar ganttchart 2 0 jar gluegen rtjar grid 1
17. the ground connection provided to it The ground pin was identified after analyzing the circuit diagram and measuring the voltage across the pin and ground Figure 4 8 The Sheevaplug is connected to AC power using the ground connection This connection achieves the common ground between both ends of RS 232 communication To prove the assertion that grounding was required we performed an experiment connecting and then disconnecting the ground connection and observed the difference in the received data Figure 4 9 shows this difference in the data The upper half of Figure 4 9 shows the Sheevaplug terminal showing sensor data corrupted by noise due to lack of common ground while the lower half shows the correct data sent by microcontroller In the latter case the values were zero for all non active sensors except the 2 sensor in last column of B which had been activated 39 Microcontroller data received on HHA saa Bag FR E Sheevaplug without ground LY carpetsmart yn server com adjustment The received data EEL contains large number of invalid 54070 73 gt 3 gt B911 3077C04000020E 2012 05 08 SA7 gt 416207B10 940 lt 1C04000020E 2012 05 08 15 53 53 characters 5407177 gt 00B19770 lt C80000020E 2012 05 08 SA377 7 gt 0083 28 gt 8CO0800020E 2012 05 08 DOUE FEHER Microcontroller data received on 00000000 0000004000000000 2012 05 08 15 53 53 a I Ss 2012 05 08 15 53 53 Sheevaplug after providing
18. the work Secure without losing any references s 3 co ddsnccaasacascdnncdannecadand cenadenncecadensdenhdnadeesaceas eeaheoaseesaderdeeees 55 Figure A 1 Figure shows the launch screen of Eclipse workbench with the source code repository 65 Figure A 2 The Eclipse workbench is shown in the figure with product file opened The highlight areas show the required users click while exporting an executable ccccccccccceeceeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeees 66 Figure A 3 Figure shows the eclipse product configuration screen before exporting Select the parameters AS describe d aDO Ve osc Masanctinndens inate ta uns a E a a a ER 66 Figure A 4 The files and directories that are generated during the software creation are shown in the Figure Click on the SmartCarpet exe file to launch the software cccccccccssesseeessseneeeeeeeeeeeeeseeseeeeens 67 Figure A 5 User authentication window launched after executing the application User needs to satisfy the authentication requirements before using the software ccccccccceeeceeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeees 67 Figure A 6 Figure shows the carpet display in the software 128 rectangles are used to represent each foil sensor in the carpet and the remote file explorer named as Carpet Data is used to browse the file stored on the Sheevaplug server Start button is highlighted on the upper left corner cccceeeeeeeeeeeees 68
19. was used to transmit the data and a client program was used to receive it We performed an experiment by stepping on the carpet and the remote computer was expected to receive and display the data on terminal window We repeated this experiment by connecting multiple computers to the Sheevaplug server 3 9 3 4 Observation of data transmission time A computer program was developed to calculate the expected delay while observing the data from remote computer We used Sheevaplug server on transmitting side and a computer on receiving side The program performs following steps note t as the time when a frame of data was transmitted and t as the time when the acknowledgement of same frame was received The difference t2 t was expected to be a Round Trip Time RTT Hence half of this difference was calculated as expected delay in data The experiment was not performed by varying the geographical distance between the server and the computer which might increase the delay 30 3 9 4 Performance of fall detection and notification system We conducted the experiments to determine the Sensitivity Specificity False Negative Rate FNR and the False Positive Rate FPR of the fall detection system The Sheevaplug server was used to detect and declare the falls and a mobile phone was used as a medium to notify the event We performed following experiments based on the falls test protocol 41 3 9 4 1 Sensitivity analysis of the fall d
20. 0 0 src zip grid 1 0 0jar gwt visualization jar B javacsv jar a javax jar jmathplot jar jogl jar a jxl jar opencsv 2 2 jar org jzy3d 0 9 dependencies zip Window Help rO rQ o y Raga Le B edu missouri eldercare product 2 A Overview A General Information This section describes general information about the product ID edu missouri eldercare product Version 1 0 0 qualifier Name Elder Care Technology _ The product includes native launcher artifacts Product Definition Plug in Development edu missouri eld ication edu missouri eldercare product Eclipse File Edit Source Refactor Navigate Search Project Tomcat Pipeline Run EMML BEO BMGP H F This section describes the launching product extension identifier and application Product edu missouri eldercare application product m Application edu missouri eldercare application application The product configuration is based on plug ins Testing 1 Synchronize this configuration with the product s defining plug in Test the product by launching a runtime instance of it features Exporting 3 Use tk Eclipse Product export wizard o a ee the productt m El Overview Dependencies Configuration Launching Splash Branding Licensing Figure A 2 The Eclipse workbench is shown in the figure with product file opened The highlight areas sho
21. 1 1 Observation of data on terminal emulator using USB interface We replaced the RS 232 serial communication interface which was used 1n earlier implementation 23 with serial to USB adapter This made the smart carpet system 34 compatible with new computers Figure 4 1 shows the data received on computer s terminal emulator Each line in the figure shows one frame of data The sensors activation information for segments A B and C 1s sent between S and E representing starting and ending words respectively The time of occurrence of each frame is appended after comma HWA sa A eis SFR carpetsmart dyndns server com SA02000000B0000004 0C00000000E 2012 05 08 SAOZO00000B0000004 0CO0000000E 2012 05 08 1 SAO2Z020000B0000004 0CO0000000E 2012 05 08 SADOO2Z0000B0000004 0C00000000E 2012 05 08 SA0002 0200B00000040C 00000000E 2012 05 08 1 SADOOOOOO0BOO000004 0COO0000000E 2012 05 08 Print the current sere ssh2 AES 256 CTR 1 22 6 Rows 69 Cols YVT100 CAP NUM na 7 a Figure 4 1 Data received from microcontroller connected on USB interface is displayed on receiving computer s terminal emulator appended with its timestamp We used the USB port for further implementation of this project in particular to observe the data on terminal emulator SecureCRT and the graphical display of the sensors Refer APPENDIX A We also developed memory efficient data stori
22. 17S 2012 S16 12 54 19 6207 MLL sph ML sy TI 4 5 I 46 1 3 3 Y 1 3 5 I 3 6 1 4 3 15 5 1 6 4 1 6 5 1 7 6 2012 05 16 12 53 19 653 Figure 4 5 Data shown is stored in CSV format All 1s from the matrix are stored along with their row and column coordinates The highlighted row shows the CSV conversion of the matrix in Figure 4 4 In case of no movement on the carpet the matrix contains all Os Storing such matrix unnecessarily consumes a lot of memory with no valid information We ignored such instances for CSV formats Hence the data stored in CSV files provides the information of carpet movements and not the idle carpet condition 4 2 Sheevaplug computer as a project server We used Marvell s Sheevaplug with Feroceon 88FR131 1 2 GHz processor 512 MB RAM 512 MB NAND Flash to receive and store the data The plug computers are available in market at price starting from 100 and they occupy space less than 1 2 square foot The implementation of this computer reduces the cost by approx 1000 as well as saving the space per installation Figure 4 6 shows the Sheevaplug connected to microcontroller using the USB interface It receives the data and transmits over the internet using Ethernet connection allowing remote computer to display We used this setup for further experiments in this project 37 Sheevaplug device used for collecting the data from microcontroller by eliminating the ne
23. 2 1 Observation of data conversion time A computer program was developed to calculate the data conversion time It performs following steps note the time of new frame as t the time when matrix and CSV formats became available as tz and tz calculate the differences tz t1 and t t2 and store them in separate file We recorded the times when 2 3 and 4 carpet segments were used In future the system will use more number of carpet segments it is important to calculate the data conversion times 3 9 2 2 Observation of memory required for storing the data Using our hourly generated files in three formats we selected a time length of 12 hours 10 AM 10 PM and obtained the 36 generated data files as discussed in 3 9 2 The files were selected to include walking falling and idle time for the carpet We calculated the average size per hour for raw data files matrix data files and CSV data files Future implementation will require compensation for variable number of sensors in one room since the raw data and the matrix sizes will also vary The expected file size impacts the data sampling and storage rate 28 3 9 3 Performance of Sheevaplug computer 3 9 3 1 Verification of Sheevaplug ground connection This experiment aimed to verify the ground connection which was intended to achieve common ground between both connectors of RS 232 communication cable 1 e the microcontroller and Sheevaplug server We used the same experime
24. EE 2010 pp 2242 2245 U Shriniwar Data control for signal scavenging for a personnel detection system Computer Engineering Computer Engineering University of Missouri Columbia 2010 K Devarakonda Data display for a signal scavenging personnel detection system Computer Engineering Computer Engineering University of Missouri Columbia 2010 Marvell 2010 PLUG COMPUTER BASIC Available http www plugcomputer org downloads plug basic E R Harold Java network programming developing networked applications Beijing u a O Reilly 2004 L L Peterson and B S Davie Computer networks a systems approach Amsterdam Boston Morgan Kaufmann 2012 Eclipse v3 5 Galileo Project Online Available http www eclipse org galileo M Jankowski and J Kuska Connected components labeling algorithms in Mathematica Java C and C 2004 ProlificUSA PL 2303H 2303HX 2303HXD Drivers Online Available http prolificusa com pl 2303hx drivers SWT The Standard Widget Toolkit Online Available Official Web Site www eclipse org swt RAP Rich Ajax Platform Online Available Official Web Site http www eclipse org RAP Target Management RSE Online Available Official Web Site http www eclipse org tm Rich Client Platform Online Available http wiki eclipse org index php Rich Client Platform 62 35 36 37 38 39 40 41 42 SecureCRT T
25. FALL DETECTION SYSTEM USING LOW COST COMPUTING AND ONLINE COMMUNICATION A Thesis presented to the Faculty of the Graduate School University of Missouri Columbia In Partial Fulfillment Of the Requirements for the Degree Master of Science by KAUSTUBH RAGHUNATH GADRE Dr Harry W Tyrer Thesis Supervisor JULY 2012 The undersigned appointed by the dean of the Graduate School have examined the thesis entitled FALL DETECTION SYSTEM USING LOW COST COMPUTING AND ONLINE COMMUNICATION Presented by Kaustubh Raghunath Gadre A candidate for the degree of Master of Science And hereby certify that in their opinion it 1s worthy of acceptance Dr Harry Tyrer Dr Marjorie Skubic Dr John Fresen ACKNOWLEDGEMENTS I would like to take this opportunity to thank all the people who made this thesis and my graduation in University of Missouri possible First of all I would like to express my sincerest gratitude to my thesis adviser Dr Harry W Tyrer for his constant support valuable guidance and encouragement I would like to sincerely thank the members of my thesis committee Dr Marjorie Skubic for providing very insightful comments on my thesis and Dr John Fresen for his valuable suggestions on statistical aspects of this project I would also like to thanks all the earlier graduate students in the lab Uday Shriniwar Rohan Neelgund and Krishna Kishore Devarakonda for sharing their innovative views and hel
26. L ccceeeesseeesseseesseeeens 49 4 3 2 Performance of the fall detection system ccccccccccccceeceeeeeeeeeeeeeeeeeeeeeeeeeeeeeees 50 44 Distribution of software executable and source repository ennnso00000000000000 53 4 4 1 Smart Camper SEL VEL WEDS IE eaa ae E haialiede tal ates ideas aide 53 4 4 2 CNIS Source reposi Ory srecan o aa 54 CHAPTER 5 DISCUSSIONS CONCLUSION AND FUTURE WORK 56 REFERENCE Sron ct tate E E E A O A ihe ates 60 APPENDIC E Scese E ee eee eee 64 APPENDIX A SOFTWARE USER MANDAL QW cccccccccessseeeeecseeeeeeeeenenaes 65 LIST OF FIGURES Figure 3 1 Figure indicates the arrangement of carpet segments Each segment A B C D consists of 32 sensors arranged im 8 columns and 4 10WS aus3cckn cat a SAS A Aa iA 8 Figure 3 2 Figure shows the block diagram of smart carpet system The first two blocks consists of sensors amplifiers and the microcontrollers system which remained same from previous implementation 23 The microcontroller system is connected to computer using serial to USB adapter 10 Figure 3 3 The connection description of Sheevaplug computer Image source 25 eeeeeeeeeees 10 Figure 3 4 Figure shows the conversion of data frame into matrix format The data is populated in 128 elements starting from Start up to End The 16 elements shown in lower right corner are unused and MS TUM VATA eane n E nse ean Geek a a
27. a 53 2 BIS Elder Care Technology oO 0 Oo 0 Oo 0 0 Oo Oo 0 Oo 0 0 0 Oo oO 0 0 Oo 0 0 oO 0 0 0 0 Oo 0 Oo 0 0 Oo 0 oO 0 0 0 Oo 0 Oo 0 0 0 0 Oo 0 oO 0 Oo 0 0 0 Oo 0 0 0 0 0 Oo 0 0 Oo 0 0 0 0 0 0 0 Oo 0 Oo 0 Oo 0 0 0 0 0 0 0 0 0 0 Oo 0 it 0 0 0 Oo 0 0 0 0 0 Oo 0 Oo 0 Oo 0 0 Oo 0 0 0 0 oO 0 Oo 0 Space for Cupboard 0 0 0 0 Oo 0 0 0 0 Oo 0 Oo 0 0 Oo 0 P Figure A 6 Figure shows the carpet display in the software 128 rectangles are used to represent each foil sensor in the carpet and the remote file explorer named as Carpet Data is used to browse the file stored on the Sheevaplug server Start button is highlighted on the upper left corner B Elder Care Technology _ File Window Help Search me wer z 7i Carpet Data 53 a 2 Blg Elder Care Technology Sheevaplug 0 Oo 0 Oo 0 0 0 0 Oo 0 Oo 0 0 oO o 0 0 0 Oo 0 0 o o Mm O 8 o o M o o o o a 3 MMMM gt 0o o o o 2 0 0 0 oO 0 0 o 0 Oo 0 o 0 0 Oo 0 Oo 0 2 o 0 Oo 0 Oo 0 Oo 0 0 0 0 0 o 0 Oo 0 oO 0 0 Oo 0 0 0 0 0 0 0 Oo 0 0 0 Oo 0 0 Space for Cupboard 0 0 0 0 0 Oo 0 0 0 0 0 0 0 0 0 0 Figure A 7 Figure shows the carpet display after pressing the start button The red color rectangles represent the active sensors with their current activation count 4 Right click in the file explorer window and follow these steps a Click on new connec
28. aplug before and after providing the common ground The upper half shows the data received before whereas the lower part shows the data received after Providine the Sround CONMECIOM a 55525 4 eG acs le EEN A N E 40 Figure 4 10 Number of active sensors detected by Sheevaplug before and after establishing the ground CONNEC Oi ursan E E A maanseaeosseneeset 40 Figure 4 11 Graph for comparing the data conversion time v s no of active carpet segments 00 42 Figure 4 12 Graph for the comparison of average memory requirement for storing the information in all formats of data sent by microcontroller Over 12 Dours ccccccccceeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeees 43 Figure 4 13 Secure shell terminal window is shown accessing the Sheevaplug server using DNS name The Sheevaplug can be accessed using this DNS name within the closed network c0eeseseeeeeees 43 Figure 4 14 Sheevaplug acting as a socket connection server and transmitting every received data frame to the Temolely Connected e MICA kertaan ia a a e a a 44 Figure 4 15 Picture shows the scenario in which a person is standing on the carpet sensor making it active Sheevaplug is shown connected to the microcontroller circuit to receive and transmit the data over the Internet Remote computer is shown receiving the data and displaying it nnnnnonnnnnnnnnnnneneneeneenssseeeeeee 45 Figure 4 16 Graph shows the comparison of the o
29. as found to be very high The false negative rate was very low Average Specificity 95 9 Average FPR 4 09 We did not perform the separate set of experiments for evaluating the true negative conditions However we collected carpet data by having few people randomly and individually walking on the carpet during the daytime and leaving the system idle in the night We continued this observation for several weeks We did not receive any fall notification on mobile device Hence no false positive events were observed Nevertheless under real circumstances we will expect to see false positive events These are documented in the discussion chapter 4 4 Distribution of software executable and source repository We created the software executable consisting of a collection of all the computer programs to produce a carpet display using eclipse development environment The executable program worked in such a way that all the programs can be automatically launched with a single user click and the end user does not need to have the detailed knowledge of their execution We describe the details for using the executable SmartCarpet exe in APPENDIX A 4 4 1 Smart Carpet server website We developed our own website and deployed it on Sheevaplug web server The purpose of this website was to provide information about this project such as team 53 members and their roles It also has a download section to easily distribute the executable
30. bserved network latency during one hour The line parallel to abscissa describes the average network latency 335 MS 0 0 0 0 ccceeesseeeeeeeteeeeseseessssssssseeeeeaees 46 Figure 4 17 Figure shows an eclipse web application opened in Internet browser on desktop computer The display contains 128 different blocks each of which shows the state of one carpet SENSOT cee 47 yvi Figure 4 18 The Emergency notification in the form of text message is received on nurses mobile after detecting the fall The bottom part of this picture shows the way data is displayed on remote computer 48 Figure 4 19 Figure shows the matrices at two different instances of fall Largest component is boxed and all other smaller components are encircled Fall notification is sent when the size of boxed component is more than the Threshold value MO asc cs ess acta ctiest estore sa et ieee Seca E A diene ees 49 Figure 4 20 Graph for showing the variation in largest component with respect to time The vertical line parallel to ordinate describes the fall instane with maximum component size 10 ceeeeeeeeeeeees 50 Figure 4 21 Figure shows the website of smart carpet project being hosted on the Sheevaplug server This website acts as the entry point of the entire software system ccccccccccccccceeeeeseeeeeeeeeeeeeeeeeseesesseeeeseeeess 54 Figure 4 22 Picture shows the page of project s Google repository This repository is developed to keep
31. cellphone based messaging services Hence this system should work without any glitch as long as Sheevaplug is connected to Internet 4 3 Analysis of carpet data and fall detection system The purpose of the smart carpet research is fall detection It turns out to have additional value but here we want to test the ability of the carpet system to detect falls We analyzed each frame of data using connected component analysis on each matrix Such connectedness results in detecting falls This section discusses the results of fall detection system 48 4 3 1 Observation of data for largest connected component We observed various fall patterns by having a volunteer perform well described falls on the carpet The fall patterns produce distinctive sub regions in the matrix data A fall pattern consists of regions of matrix elements with value 1 with connectivity size more than 10 Refer Figure 4 19 At every line of data the computer program attempts to find such connected sub regions to detect a fall After detecting the connected region of size 10 or more the program declares that a fall has occurred This information is used to send a message notification as discussed in Section 4 2 6 Figure 4 19 shows the data produced by two different fall patterns observed during experimentation Each matrix has multiple connected components The largest boxed component is used for detecting the fall if it exhibits connected region of 10 or more element
32. ces computer s hardware interface and the network interface libraries The interface libraries provide an access based on the server s Operating System 24 File Carpet Display Explorer Terminal RSE Network Interface Libraries Computer Network Interface Internet Tunnel 0 Remote Server Linux Windows FTP SSH Telnet Figure 3 13 Figure shows the architecture of executable application after its integration with the file explorer and the terminal resources The idea behind adding the file explorer was to provide an ability to select and play the previous carpet data on the same display To perform this user need to open the server connection in the explorer window and then click on Show History after selecting an appropriate file The following is the algorithm for displaying the data stored in a file 1 Read new line of data from the file Truncate the timestamp from it 2 Convert back the line of CSV data into matrix form by reversing the procedure discussed in section 3 4 2 3 Iteratively check the value of every matrix element a If found 1 then change the block color to red b If found 0 then change the block color to green 4 Check if the end of file is reached a If yes go to step 6 b Ifno go to step 1 5 Stop 25 3 7 2 Eclipse web application We already discussed about using Sheevaplug as a ceaseless
33. daily activities and independence of living It is observed that falls account for 40 of injury related deaths 5 and immediate medical attention helps avoid fatalities 5 7 Healthcare cost which includes medication diagnostics hospitalization doctor visits and the cost of care giver can be very expensive 5 Fall detection of the elderly is important reducing the mortality as well as the considerable expenditure in healthcare 8 Many researchers have worked on fall detection systems Accelerometer based wearable devices for detecting falls such as waist worn mercury tilt switches 9 wrist watches 10 mobile phones 2 hearing aids 11 and sensors placed under the armpits 12 have been described The disadvantages of these fall detection systems are that the elderly person is likely to forget carrying or wearing these devices and they may produce false alarms The image processing techniques 13 perform the fall detection by analyzing the horizontal and vertical speeds during fall whereas the video based systems 14 15 monitors the moving objects for predicting human falls Both these methods violate privacy of the personnel University of Missouri researchers developed a video based passive fall detection system which alleviates such privacy concern It distinguishes human shapes from the background by segmentation and then generating silhouette images from the video sequences 16 17 Human activity modeling techniques suc
34. dow is shown accessing the Sheevaplug server using DNS name The Sheevaplug can be accessed using this DNS name within the closed network We successfully accessed the server using DNS name at various locations in the University network including the wired and wireless access points Due to the limitations from University network administration we could not make this server accessible outside 43 the MU network Assignment of static IP or the DNS name within missouri edu can make this happen 4 2 4 Time for transmitting the data to the remote computer Figure 4 14 shows the terminal window on a remote computer It displays data received from Sheevaplug which receives the data from the microcontroller We used one computer having an IP address 10 7 62 28 for receiving this data over the Internet ps www google com voice settings ta p https wwe google com voice OK aiccess Hilly Received rnr_se f 10 7 62 28 connected SAQ0000000B 0000004 0CO0000000E 2012 05 08 15 46 24 889 gt 10 7 62 28 SA00000000B 0000004 0CO0000000E 2012 05 08 15 46 25 019 gt 10 7 62 28 SADDOOO000BO000004 0CO0000000E 2012 05 08 15 46 25 149 gt 10 7 62 28 SAD00004 OOBOOD00004 0CO0000000E 2012 05 08 15 46 25 279 gt 10 7 62 28 ssh2 AES 256 CTR 6 22 68Rows 71 Cols WT100 CAP NUM Figure 4 14 Sheevaplug acting as a socket connection server and transmitting every received data frame to the remotely connected client
35. e 4 3 Table shows the performance parameters of the system for fall experiments The average sensitivity was found to be very high The false negative rate was very LOW cccceeeeeeeeeeeeeeeeeeeeeees 51 Table 4 4 Table indicating the performance of fall detection system for false positive fall experiments performed on the carpet Each experiment is performed twice at 10 times Note that in this the system detects no falls i e in V1 first row of 10 attempts 8 came in with no fall cee ccccceeeeeeeeeeeeeeeeeeees 32 Table 4 5 Table shows the performance parameters of the system for false positive fall experiments The average sensitivity was found to be very high The false negative rate was very lOW cccccceeeeeeees 53 viii FALL DETECTION SYSTEM USING LOW COST COMPUTING AND ONLINE COMMUNICATION Kaustubh Raghunath Gadre Dr Harry W Tyrer Thesis Supervisor ABSTRACT Falls are prevalent among elderly and sometime may result in fatal injuries which impacts their ability of independent living Thus a reliable as well as cost effective fall detection system is required We have developed an inexpensive fall detection system which detects falls and automatically sends notification to the caregiver It uses low cost unobtrusive wall mounted Sheevaplug computer and an Internet based free messaging service The fall detection solution is based on the carpet foil sensors information received from the hardware su
36. e IEEE 2008 pp 4632 4635 R Neelgund Floor sensor development using signal scavenging for personnel detection system Computer Engineering Computer Engineering University of Missouri Columbia 2010 S R Lord C Sherrington and H B Menz 2001 Falls in older people risk factors and strategies for prevention Available http public eblib com EBLPublic PublicView do ptiID 218065 L Liang M Popescu M Skubic M Rantz T Yardibi and P Cuddiby Automatic fall detection based on Doppler radar motion signature in Pervasive Computing Technologies for Healthcare PervasiveHealth 2011 Sth International Conference on 2011 pp 222 225 D Wild U S L Nayak and B Isaacs How Dangerous Are Falls In Old People At Home British Medical Journal Clinical Research Edition vol 282 pp 266 268 1981 N Noury A Fleury P Rumeau A K Bourke G O Laighin V Rialle and J E Lundy Fall detection Principles and Methods in Engineering in Medicine and Biology Society 2007 EMBS 2007 29th Annual International Conference of the IEEE 2007 pp 1663 1666 G Williams K Doughty K Cameron and D A Bradley A smart fall and activity monitor for telecare applications in Engineering in Medicine and Biology Society 1998 Proceedings of the 20th Annual International Conference of the IEEE 1998 pp 1151 1154 vol 3 T Degen H Jaeckel M Rufer and S Wyss SPEEDY a fall detector in a wrist watch
37. ed of full fledged computer Serial to USB adapter replacing the serial communication cable Ethernet connection to Sheevaplug for transmitting the data over the Internet Microcontroller circuit box for sensing the state of active sensors and sending it to the Sheevaplug Figure 4 6 The Sheevaplug is shown connected to the microcontroller system using USB interface for receiving the data and transmitting it over the Internet using Ethernet connection Sheevaplug also performs the analysis of incoming data and notifies the caregiver after detecting the fall 4 2 1 Ground connection provision to Sheevaplug computer The connection from the final microcontroller to the Sheevaplug begins with a wire with an RS 232 connection grounded then a USB connection ungrounded to the Sheevaplug That is the Sheevaplug is not grounded to the microcontroller system and produces invalid characters as a data see Figure 4 9 Figure 4 10 While it was clear this was a grounding problem its resolution was not easily forthcoming Discussions with the manufacturer were helpful but not conclusive We ended up analyzing the Sheevaplug circuit and circuit board to find a suitable ground pin Figure 4 7 shows the Sheevaplug circuit board with its identified ground pin and connected this pin to the AC supply ground as shown in Figure 4 8 38 Ground Pin connection ane Figure 4 7 The Sheevaplug circuit board is shown with
38. eea as aatae cede esa eurtes r S 13 Figure 3 5 The demonstration of conversion of one frame of raw data into matrix format 008 14 Figure 3 6 The demonstration of conversion of one frame of matrix data into CSV format 06 16 Figure 3 7 Figure shows the schematic of Sheevaplug circuit board Source 38 This schematic was used analyzed to analyze the circuit board for finding the ground PIN ccccccccceecceeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeees 17 Figure 3 8 Figure shows the Sheevaplug circuit board with the identified ground pin encircled The wire connected to the ground pin establishes a ground connection with AC supply ground eeee 18 Figure 3 9 Block diagram describes the communication of carpet data from Sheevaplug server to the remote computer The data is communicated over the socket connection between the server and the FEMOLS COMP UUCE lt 5 taustancelabadtewariaarel a nad heins laaaalitbagasin tens a A Sueathatandel a 19 Figure 3 10 Block diagram illustrates the fall situation and the notification system Sheevaplug is responsible for storing the carpet data transmitting it over the Internet and notifying the fall events to the G2 6s 22 hi nee eee ee RE er ee Se oe et a SPN eRe ee eo ee ar ree eRe 21 Figure 3 11 Figure illustrates first two steps of the algorithm The matrix on the left side represents data received from microcontroller The other matrix shows the result of st
39. eeeeeeeeeeeeaaaas 12 3 4 1 Conversion of carpet data in matrix format ccccceessesesessssssssssessessseeseeeeenens 12 oe Conversion of matrix carpet data in Comma Separated Value format 15 3 5 Sheevaplug computer as a project SCTVEL ceea oen a E 16 3 5 1 Ground connection provision to Sheevaplug computer 000000000000000000000000000 16 3 5 2 Domain Name System DNS name assignment for Sheevaplug server 18 3 5 3 Transmission of carpet data to remote computer through socket Internet COMME CUO Mag icitec atch Aina eaccseta awit acetals ta aaenen Sanaa cates ta eoosne aera emia es 19 3 5 4 Emergency notification for fall detection System cccccccesesessssseessesseeseeeens 20 3 6 Analysis of carpet data and fall detection system ccccccccccceeeeeeeeeeeeeeeees 21 3 6 1 Algorithm for fall detection system ccccccccccccecceeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeees 21 3 7 Development of carpet display applications ccs eeeeseeessseeeeeeceeeeeeeeeeeeeaaaaas 23 3 7 1 Felipse desktop applica totic c tec osasetstucsaeusageeesadee oa E EE aes 23 Soke EChHpSe WeDrapplMCALON caure T nascar ities 26 3 8 Distribution of software executable and source repOSItOTy eccceceeeeeeeeees 26 So EPEE nn a sceneries eases ee ata 27 3 9 1 Observation of carpet data on terminal emulator ccceceeeeeessssessssssssssseeeeens 27 3 9 2 Orsaniza
40. en as one of the most dreadful events for elderly seniors especially in their independent living Literature shows that 30 of independently living elderly over 65 years fall at least once every year and 30 of such falls result in disabling injuries Severe fall injuries can also lead to deaths 2 Such fall events may also cause psychological impact that is they always feel fear of falling 3 and experience deteriorating quality of life This research focuses on the development of a system to monitor ambulatory movements of a frail elderly and send this data on the Internet using low cost wall mounted plug computer The online data allows relatives and caregivers to monitor the state of the elderly seniors from the remote locations and the text messages on mobile phones notify them about the fall incidences In this system the elderly person does not need to carry or wear any fall detection sensor and it s called as passive fall detection system It consists of low cost aluminum foil sensors laid under a carpet to detect the presence of a person and the microcontroller system to communicate the presence information to the computer 4 We incorporated a plug computer as a server to receive and store this information in two different matrix and CSV formats The use of Sheevaplug computer in place of desktop computer reduced the cost of this system by around 1000 per installation and also made the system less obtrusive We established a com
41. entation 39 18 3 5 3 Transmission of carpet data to remote computer through socket Internet connection To remotely monitor movements on the carpet it was necessary to communicate the data to the remote computer We established a TCP socket connection between the remote computer and the Sheevaplug server as shown in Figure 3 9 http carpetsmart dyndns server com Micro controller SA00210000B E 384177024645945 System y ACK 384177024645945 v ff Sheevaplug server acquiring and transmitting the carpet Sensors and data to the remote computer Amplifiers through socket Internet connection Figure 3 9 Block diagram describes the communication of carpet data from Sheevaplug server to the remote computer The data is communicated over the socket connection between the server and the remote computer The algorithm for establishing socket connection server on Sheevaplug server 1 Initiate the socket connection on port 99999 2 Check for any incoming client connection request If request found make a connection to the client and store its IP address 3 Acquire one line of carpet data from the data acquisition system Organize the frame of data by appending the time of occurrence Transmit the frame to the connected clients 4 Receive an acknowledgement from each client 5 Go to step 2 The algorithm for socket connection client on remote computer 1 Request a sock
42. ep 2 cccccccccccccccceceeeeeeeeeeeeeeees 22 Figure 3 12 Figure shows the architecture of executable eclipse application The lower block shows the eclipse plug in developed for displaying the carpet data Other blocks represent the eclipse frameworks operatine system libraries and Javacenvironment 2 7 a ie Ee Rs 23 Figure 3 13 Figure shows the architecture of executable application after its integration with the file explorer and the terminal TESOULCES orrainn ea a a a e a e 25 Figure 3 14 Figure shows all the falls we performed First row describes the Standing to falling activities Second row describes the Tripping and falling activities whilst the third row indicates the Sitting to TANNE AGYNESS 32 Figure 3 15 Figure shows the false positive experiments performed on the carpet We performed 11 different experiments to verify the specificity of the system cccccccccccccceeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeees 33 Figure 4 1 Data received from microcontroller connected on USB interface is displayed on receiving computer s terminal emulator appended with its timestaMp cccccccccceeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeees 35 Figure 4 2 Figure shows the person fallen down on the carpet We used this scenario for showing the corresponding data in different formats earnen aina eaaa e Ea a eas 35 Figure 4 3 Each frame of received data is stored along with its timestamp in the separate file created at e
43. et connection to the server at address described in Figure 3 9 on the port number 99999 Wait for server to accept the connection 19 2 Receive a complete frame of data from S to E Send back an acknowledgement ACK to the server 3 Display data locally on the eclipse application display 4 Go to step 2 3 5 4 Emergency notification for fall detection system Google Voice offers a free Internet based telephone service for sending text messages and placing calls by assigning a separate phone number to every Google account holder We created a Google account and obtained a phone number Google Voice also provides a Java library 40 for integrating its telephone service in the user program Using this library we developed a text message notification system for declaring fall events We also defined different text recipient numbers based on the caregivers schedules in the external properties file stored in Sheevaplug The algorithm of the notification system is as follows and operates in the Sheevaplug 1 Initiate a connection to the Google voice server using the phone number 2 Check if there is any fall declared by the fall detection algorithm 3 If fall is detected select recipient s phone number from properties file for the fall time Send the notification text to the selected number wait for 45 seconds 4 Go to step 2 Since the microcontroller sends data at a rate higher than the person s walking speed the fall detect
44. etection system The fall experiments were carried out with 4 volunteers each performing every experiment for 10 times as shown in Figure 3 14 Hence the total numbers of trials were 440 In every experiment when volunteer fell on the carpet the mobile phone was expected to receive a notification message If the notification was received it shows that the trial was detected as fall and we noted it as true positive On the other hand if the notification was not received it shows that the system failed to detect the fall and we noted it as false negative The Sensitivity and the FNR were calculated using the number of detected falls and the number of non detected falls respectively The average FNR and the average sensitivity were calculated to evaluate the overall fall detection performance of the system Following are the formulae used FNR False Negatives Total number of fall experiments Sensitivity True Positives Total number of fall experiments 31 1 Standing to falling 2 Standing to falling 3 Standing to falling toon 4 Standing to falling on forward backward right side left side 5 Tripping and falling 6 Tripping and falling 7 Tripping and falling on 8 Tripping and falling on forward backward right side left side 9 Sitting and falling 10 Sitting and falling on 11 Sitting and falling on forward right side left side Figure 3 14 Figure shows all the falls we performed First row describes
45. every movement and make it available on Internet using cost efficient wallplug computer Availability of the carpet data on Internet allows relatives and caregivers to monitor distantly The real time notifications are sent using SMS after detecting the falls We developed a software installable to make the system reproducible to larger scale at ease An electronic system consisting of aluminium foil sensors is used to detect the presence of a person standing on the carpet All the electronic systems pertaining to all carpets make a wireless systems having separate micro controller for each one A single micro controller is connected to computer using USB interface to transmit the data The data is in the form of continuous stream which needs to be chopped and converted to the required form We developed a computer program to acquire this data and convert it into the store efficient and user readable CSV format Initially we developed the software product to wrap all our computer programs in single easy to use executable We tested this product on several different computers and it could successfully perform all the actions on the data except the connectivity to Internet We gained a confidence after successful testing of this product and that encouraged us to make the system available on Internet We decided to develop a web interface to monitor the carpet system remotely and read the previously stored carpet data We used Eclipse RAP technology which helped u
46. gure 3 15 Figure shows the false positive experiments performed on the carpet We performed 11 different experiments to verify the specificity of the system 33 CHAPTER 4 RESULTS This software system delivers a solution that includes capturing the floor sensor data using USB interface and storing it in various formats The floor sensor data is organized in frames each frame is analyzed at real time and stored The analysis seeks to detect falls on the carpet to provide emergency notifications Furthermore the system provides the internet based communication it establishes the server on a plug computer an inexpensive wall plug mounted computer transmits the carpet data over the internet and allows access for display on the remote computer This chapter discusses the performance and results of all the modules in this software system 4 1 Data acquisition and storage system We developed a computer program to receive a continuous stream of characters sent by the data source 1 e microcontroller We connect the computer using a USB interface and stored the data in user readable format such as Matrix and Comma Separated Value CSV files Every frame in the data stream is first converted into matrix form with the same dimension as the carpet installation segments To form the CSV format the matrix elements having the value 1 are selected and their row and column coordinates are then stored as described in the following sections 4
47. h as graphical models 4 Bayesian networks 18 are used to correctly identify human shapes from these images 19 The system neither processes nor stores the video it only uses silhouette images which maintain user s privacy In 20 a low cost Microsoft Kinect based computer vision system was proposed to monitor elderly activity for 24 7 in low to no light conditions It uses a depth imaging technique for fall detection which is robust to the changes in lighting and shadowing In other work unobtrusive vibration and sound sensor system in which the person does not require wearable devices performs the fall detection using sound that is generated during human falls 3 A system consisting of a circular microphone array detect falls by capturing the sound generated during falls It correctly locates the source of falling sound in attempt to reduce false alarms which also makes the system robust under the noisy situations 21 22 The smart carpet system that we developed monitors the elderly senior without any body worn device or the privacy intruding cameras It automatically detects falls and notifies the same using cellphone text messaging It also consists of a server to store and communicate the data over the Internet to enable remote monitoring Low cost components produce an affordable fall detection system The system was originally developed containing 4 carpet segments each consisting of 4x8 aluminum foil sensors to detec
48. he Usable Flexible SSH Client Online Available Official Web Site http www vandyke com products securecrt index html OpenCSV Project Information Online Available http opencsv sourceforge net project info html J Axelson Serial port complete programming and circuits for RS 232 and RS 485 links and networks lakeview research llc 1998 PlugWiki Enhanced Sheevaplug Installer Application Available http www plugcomputer org plugwiki index php Main_ Page DynamicDNS 2011 Official Ubuntu documentation for DynamicDNS Available https help ubuntu com community DynamicDNS google voice java An Unofficial Java API for Google Voice Online Available http code google com p google voice java M J Rantz M A Aud G Alexander B J Wakefield M Skubic R H Luke D Anderson and J M Keller Falls technology and stunt actors New approaches to fall detection and fall risk assessment Journal of Nursing Care Quality vol 23 p 195 2008 J McAffer and J M Lemieux Eclipse Rich Client Platform Designing Coding and Packaging Java TM Applications Addison Wesley Professional 2005 63 APPENDICES APPENDIX A Software user manual APPENDIX B CD ROM Contents 64 APPENDIX A SOFTWARE USER MANUAL This user manual explains how the software system is used for displaying carpet data on the remote computer It is mainly divided into following three parts 1 Instructions for creating a
49. hese differences are shown in Table 4 1 The mean and standard deviation of the conversion times was over a 12 hours period In the table the first two columns describe the times when two segments were used whereas the remaining columns describe the times when three and four segments were used respectively Table 4 1 Time required ms by a computer program to convert the data received from microcontroller into matrix and CSV formats First two columns describe the time for two segments while the remaining columns describe the time for three and four segments respectively w represents the mean whereas o represents the standard deviation of the observed times over the period of 12 hours Conversion Times ms For 2 installation segments For 3 installation segments For 4 installation segments For Matrix For CSV For Matrix For CSV For Matrix For CSV Format Format Format Format Format Format u 16 842 u 17 648 u 18 091 o 2 398 o 1 959 o 2 83 Each frame of microcontroller data arrived on Sheevaplug every 130 140 ms The conversion times in Table 4 1 are substantially less than this so the system can convert and store every line of received data Figure 4 11 shows a graph of some conversion times It is clear from Figure 4 11 and Table 4 1 the conversion time increases slightly in direct proportion to the number of active segments Hence in order to occupy larger 41 rooms more number of segments can be used The co
50. his system a connection between the microcontroller and the personal computer was established using RS 232 because it takes only 4 pins of microcontroller and it was a most commonly used communication interface 23 However the new cheap wall plug mounted computers like Sheevaplug do not commonly provide RS 232 but do accept USB communication interface We chose Prolific s PL 2303 Serial to USB converter kit 30 to establish connection between the microcontroller and personal computer This replaced the RS 232 connector on computer s end with USB leaving the transmitting end connector as RS 232 Figure 3 2 The instructions and the device drivers for this adapter are available for download for free from prolific website 30 Sensors and Micro controller Prolific Serial to S A m HMH gt Amplifiers System USB adapter N RS 232 lt u USB Connection gt yd Connection Figure 3 2 Figure shows the block diagram of smart carpet system The first two blocks consists of sensors amplifiers and the microcontrollers system which remained same from previous implementation 23 The microcontroller system is connected to computer using serial to USB adapter 3 2 2 Replacement of desktop computer by plug computer We required a low cost low power consuming computer to be used for continuous data collection which will also fit in very small space Wall plug computers fulfill the above requirements We c
51. hm The following describes the major steps in the algorithm 21 1 Read one frame of data from the microcontroller and convert in the matrix format 2 Scan the matrix in a horizontal raster scan and call the current element as pivot element 3 Ifthe pivot element is 1 then check the value of all 8 neighboring elements a If at least one neighbor is found 1 then make the pivot 2 b If all neighbors are 0 then retain the value of pivot element 4 Ifthe size of largest region occupied by contiguous 2s 10 or greater then compare it with the size of previous frame a Ifthe difference is higher than 6 send an alert to the notification system 5 Go to step 1 a a co a a 2 Me oC a Co L ma ia Ooo oe D oe oo oo oraosna D oo ma Cc a a 3 5 a amp oc g 0 g o 1 o 0 1 g g i 0 g g f a a oOo O O O O GO O O O O G amp G f a ia a Figure 3 11 Figure illustrates first two steps of the algorithm The matrix on the left side represents data received from microcontroller The other matrix shows the result of step 2 Figure 3 11 illustrates the data produced during steps 1 and 2 in the above algorithm The first matrix 1s similar to the one shown in Figure 3 5 1 e containing only Os and ls In second step every pivot of 1 which is connected to any other 1 is replaced by 2 as shown in second matrix The matrix sub
52. hose Marvell s Sheevaplug computer which provides a USB 2 0 port for connecting a communication cable from the microcontroller Secure Digital SD card reader for extending the memory and gigabit Ethernet for high speed wired Internet connectivity The plug computer is physically small with dimensions 110mm L x 69 5mm W x 48 5mm L 25 which can be installed at a site near the smart carpet system It is available for 100 that is way cheaper compared to the desktop computers Figure 3 3 shows the description of Sheevaplug connection ports SD Card Reader Mini USB Port 3 a aa a Gigabit a Ss Ethernet l Figure 3 3 The connection description of Sheevaplug computer Image source 25 10 3 3 Development environment and programs used For the software development part of this project we considered our experience and used Eclipse v3 5 Galileo development environment 28 Eclipse environment is a collection of plug ins and frameworks The Plug ins are developed separately by software engineers to add new functionality into the environment whereas the framework is a collection of plug ins that are grouped together by the functionality they provide We developed one such plug in by using following Eclipse Frameworks that are developed by Eclipse Foundation an open source community and made available for use at no cost Table 3 1 Table showing the list of Eclipse frameworks used for the development Eclipse Framework u
53. icating the performance of fall detection system for fall experiments performed on the carpet Three types of activities are performed in each direction at 40 times Performance Fall Experiment Performing Volunteers Parameters Standing to falling forward 10 10 10 10 9 10 9 10 Standing to falling backward 7 10 9 10 10 10 9 10 87 5 0 Standing to falling on right 8 10 10 10 8 10 mee ae Standing to falling on left side 10 10 9 10 10 10 mae ae l 23 ILS Sitting to falling on left side 9 10 10 10 9 10 10 10 Table 4 3 shows the average sensitivity and average FNR values calculated as 92 72 and 7 27 respectively from the results recorded in Table 4 2 Table 4 3 Table shows the performance parameters of the system for fall experiments The average sensitivity was found to be very high The false negative rate was very low Average Sensitivity 92 12 Average FNR 127 We also performed experiments on situations expected to produce false positive data using optical techniques 41 Each row in Table 4 4 shows the results of experiments for 2 volunteers each performed at 10 times For example first row 51 describes the results when the activity standing to sitting legs tucked in was performed The system did not declare it as fall at 8 or 9 out of 10 iterations for two volunteers respectively The total numbers of detected falls for each experiment were used to calculate the Fal
54. ication The lower block shows the eclipse plug in developed for displaying the carpet data Other blocks represent the eclipse frameworks operating system libraries and Java environment 23 3 7 1 1 Development of carpet display Using a socket client program remote computer receives data from the Sheevaplug server To show this data we developed the display containing 128 rectangles Each rectangle corresponds to one floor sensor The color of each rectangle changes with state of a sensor 1 e red if active otherwise green The following operations are performed while displaying the data 1 The remote client activates the socket connection to the Sheevaplug server Once connected the client starts polling for new available data 2 If new data is available the client converts it into matrix format as discussed in section 3 4 1 3 For every matrix element a If found 1 then make the rectangle red b If found 0 then leave the block green 4 Go to step 2 3 7 1 2 Integration of file explorer and the terminal window in the carpet display RSE framework provides the file explorer and the terminal resources to allow users to browse the files located on the server and observe the data coming from the floor sensors Figure 3 13 shows the architecture of the system after integration of these resources It shows that the carpet display the file explorer and the terminal are given an access to the remote server through network interfa
55. ion Section 4 2 1 The use of Sheevaplug computer made the Smart Carpet system less obtrusive and more cost effective The ongoing research on compressing the size of the electronics support system will make the entire system even more cost effective and less obtrusive 56 The large amount of data required complex data handling We decided upon storing the data in matrix text files and comma separated formats The average time spent in storing one frame of data was 17 ms Section 4 2 2 and Table 4 1 This was substantially less than the time between arrivals of two successive frames on Sheevaplug It indicates that the system has significant margin for the time to acquire and store the data In the future the time complexity of this storing system needs to be assessed for larger rooms The data storage server needs will grow requiring upgrade to the Sheevaplug A Database Management System DBMS can be used for storing the data For data organization security and storage we could use the specific schema based DBMS such as Oracle or MySQL With more sensors the database schema needs to be adaptable The carpet data contains the information of the floor sensors which changes rapidly with the activation and deactivation of every sensor Currently the plug computers cannot handle such rapid database queries During the initial steps of fall detection experiments the generated data did not follow any specific patterns Hence we analyzed every
56. ion system could declare multiple falls for a single physical fall To prevent this we decided to suspend the notification system for next 45 seconds Figure 3 10 illustrates the fall detection system 20 1 Amplifier and d Microcontroller N i system Sheeviplug Figure 3 10 Block diagram illustrates the fall situation and the notification system Sheevaplug is responsible for storing the carpet data transmitting it over the Internet and notifying the fall events to the caregiver 3 6 Analysis of carpet data and fall detection system Floor sensors and the microcontroller system used in the carpet are intended to sense the person s presence In addition matrix formatted data provides the location of a person However none of those help knowing the individual s state 1 e standing sitting or lying Hence it is difficult to identify if there is a fall on the carpet We designed an algorithm for detecting falls by manipulating each matrix and comparing it with preceding ones 3 6 1 Algorithm for fall detection system In order to estimate the design parameters of the algorithm we performed intentional falls on the carpet with the help of volunteers After carefully observing the recorded data we found that person in a fallen state occupies a region of approx 10 connected sensors with a change of nearly 6 sensors from the previous frame of data The numbers 6 and 10 were selected as thresholds in the algorit
57. ly running server to acquire the microcontroller data and to notify the emergency situations after detecting falls We went ahead to make it a web server for displaying the carpet data in Internet browser Among various web development techniques Rich Ajax Protocol RAP was found to be most useful especially after having a carpet display developed for showing data on the computer RAP creates a web application by reusing the code developed during carpet display with the addition of RAP Widget Toolkit RWT RWT basically transforms the entire SWT display components for displaying in the browsers This way we developed two different display applications with the single development effort 3 8 Distribution of software executable and source repository For the distribution of executable file we developed a separate project webpage using HTML and hosted on the Sheevaplug server The website contained the project outline information the information of all team members and the software download section for downloading an executable file We also created an online code repository in the Google cloud for backup Future graduate students can use this code repository to continue working on this project 26 3 9 Experiments We carried out experiments to determine the performance of the Sheevaplug computer and the system s fall detection ability The experimental setup contained 4 carpet segments each with 32 foil sensors with amplifier and
58. martCarpet Figure A 4 The files and directories that are generated during the software creation are shown in the Figure Click on the SmartCarpet exe file to launch the software A 2 Directions for using software executable on the remote computer 1 Double click on the SmartCarpet exe file and you should see the user authentication window Figure A 5 The default username and password are admin and admin In the future user credentials can be changed using the configuration listed above ecljgge Product User Name admin Password eeeee Figure A 5 User authentication window launched after executing the application User needs to satisfy the authentication requirements before using the software 2 After entering the username and password press OK button It should launch the window as shown in Figure A 6 128 rectangles on the right side are used for showing the state of carpet foil sensors Initially the rectangles show 0 and then the count starts increasing as person walks on the carpet The remote file explorer is shown on the left side 67 3 Press Start button This initiates the socket client program to receive the carpet data from the Sheevaplug server and display it on the rectangles on right It can be stopped using Stop button ie Elder Care Technology Se File Window Help Search Cster stop we z 6 gt At EO Carpet Dat
59. munication between plug server and the remote computer using Network Socket connection The plug server was used to transmit data to the remote computer We also developed an Eclipse executable application for displaying this data on the remote computer The web application was developed with the same set of source code for displaying this data in Internet browsers For achieving fall detection the Sheevaplug server was programmed to analyze this data for detecting falls The algorithm was based on the Connected Component Labeling algorithm which is commonly used in Image processing techniques We used an Internet based Google Voice messaging service for reporting these falls to the caregiver We successfully developed and tested the fall detection system using low cost Sheevaplug computer and the online communication The observed results were satisfactory in terms of sensitivity and specificity of the fall detection system We recorded part of the fall experiments using a video camera and then created a short video which is uploaded on the video sharing website YouTube com This thesis focuses in part on the development of the inexpensive online carpet monitoring and reporting system and the development of executable carpet display application CHAPTER 2 BACKGROUND STUDY Falls by elderly and the subsequent injuries are a major concern for eldercare communities The injuries mainly include disabling bone fractures that restrict their
60. nection between two processes that are running on two different computers The data acquired on the Sheevaplug was forwarded to the remote computer over this connection 26 27 Eclipse development environment v3 5 was used for exhibiting the data received on the remote computer through the socket connection Eclipse started as an IBM project and then became an open source software development environment Its functionality is divided into plug ins that are mainly written in Java Users can customize the existing plug ins or can add new one for the additional functionality We developed a new plug in containing the carpet display and then created an executable file using the base Eclipse environment 28 For developing the fall detection algorithm we began with graduate student discussions who led us to number of image processing algorithms The Connected Component Labeling algorithm was found suitable as it process the pixelated image which is in the matrix format We had envisioned storing the carpet data in a matrix format and hence developed the fall detection algorithm based on connected component labeling 29 CHAPTER 3 METHODS We developed a smart carpet to detect falls by frail elderly The system has largely been built and this work includes the data reading storage display and communication First we discuss recording the person s movement and storing the floor sensor data Then focus on using Sheevaplug a low cost compu
61. ng techniques as discussed next 4 1 2 Storing the carpet sensor data in different formats Figure 4 2 shows the picture of a person falling on the carpet This fall produced the raw data frames as shown in Figure 4 3 Each frame transformed into a matrix Figure 4 4 and each matrix reduced into a CSV file Figure 4 5 that requires much less storage This section illustrates the various formats of data Figure 4 2 Figure shows the person fallen down on the carpet We used this scenario for showing the corresponding data in different formats 35 SA00000000BOOO40000C0O0000000D00000000EF 2012 05 16 12 53 18 936 SA0OOOOOOOBOOO40000CO0000000DO0000000E 2012 05 16 12 53 19 581 SA0OOOOOLOBOOO40000CO0000000D00000000E 2012 05 16 12 53 13 605 BA00000038810153600C00000000D00000000E 2012 05 16 12 53 19 620 SADOOO00TERO2Z363100C00000000D00000000E 2012 05 16 12 53 19 653 S5A00000004800073018C00000038000000000E 2012 05 16 12 53 13 678 Figure 4 3 Each frame of received data is stored along with its timestamp in the separate file created at every one hour The file name describes the date and time of the occurrence of the data frame The highlighted frame represents the data generated at the fall shown in figure 4 2 Figure 4 4 shows the matrix format of the highlighted frame from Figure 4 3 It is divided in 4 segments of the carpet arrangement discussed in Section 3 1 In the upper 3 segments named A B C there are 4 columns
62. ntal setup previously discussed and shown in Figure 3 2 The test was carried out by connecting and then disconnecting the ground connection the SecureCRT terminal was expected was used to see the difference in carpet data when ungrounded the expected data should be noisy whereas the correct sensor data was expected in ground connected state 3 9 3 2 Verification of Sheevaplug access using DNS name The purpose of this experiment was to determine that we could access the Sheevaplug from a remote computer using the DNS name We followed the instructions provided in the Ubuntu documentation 39 for configuring the DNS name carpetsmart dyndns server com to Sheevaplug server The DNS was configured to access the server using its domain name rather than remembering its IP address ssh command was used to login into the server from remote computer using SecureCRT terminal Command used ssh root carpetsmart dyndns server com The terminal window allows access to the server after satisfying user authentication requirements We repeated this experiment from different wireless and 29 wired Internet access points in the same network We also performed the experiment from the outside access points 3 9 3 3 Observation of data on a remote computer A remote computer was used for receiving the data from Sheevaplug server over the socket Internet connection We used two different programs for establishing this connection A server program
63. nversion time limit will need to be assessed for substantially larger rooms N U1 Comparison of data conversion times For Matrix amp CSV N O m Ui m O f For matrix format U1 N _ e9 om i Z fom Q gt Q O 2222222222333333 33334444444444 Number of active segments Figure 4 11 Graph for comparing the data conversion time v s no of active carpet segments We stored each line of data in three formats and compared storage requirements We obtained data under two different activities 1 e idle carpet condition and moderate or busy walking In the idle condition no active data was generated so we stored it only in the raw data and its matrix equivalent the CSV format was null In a walking activity all three formats were stored the raw and matrix formats required the same space for all activities while the CSV file size grew depending upon the number of active sensors Figure 4 12 shows the average memory requirements calculated over 12 hours of data acquisition clearly CSV storing technique is very memory efficient whereas matrix and raw data formats occupy a lot of space It is convenient to extract a person s movement from the CSV format since it stores only active sensor s data refer Figure 4 5 Matrix format being inefficient in terms of memory simplifies reading and identification of person s location on the carpet Hence matrix and CSV can said t
64. o be a tradeoff between memory usage and ease of locating the person 42 Comparison of memory requirements of data storage techniques No Activity lt Som N e9 Som om Memory req 1417 1417 0 Raw Frame Matrix Format CSV Format Format Format of stored data Figure 4 12 Graph for the comparison of average memory requirement for storing the information in all formats of data sent by microcontroller over 12 hours 4 2 3 Accessing Sheevaplug using dynamic Domain Name System name We assigned a dynamic DNS name to Sheevaplug server for providing easy access to it from any computer in the network Figure 4 13 shows the terminal window used while logging into the DNS enabled Sheevaplug server ANSO Terminal ssh 71x15 mu 170181 users ssh root carpetsmart dyndns server com root carpetsmart dyndns server com s password Linux ubuntu 2 6 30 2 11 PREEMPT Wed Jul 22 19 53 31 MDT 2009 armvStel The programs included with the Ubuntu system are free software the exact distribution terms for each program are described in the individual files in fusr share doc copyright Ubuntu comes with ABSOLUTELY NO WARRANTY to the extent permitted by l applicable law To access official Ubuntu documentation please visit http help ubuntu com i Last login Mon Mar 19 17 08 36 2012 from mwc O627167 dhcp missouri edu root ubuntu j F Figure 4 13 Secure shell terminal win
65. owsers running on the mobile devices and tablets data not shown However being a lightweight and tiny server Sheevaplug took comparatively longer time to respond to each query To make our solution more useful we decided to develop Internet enabled desktop software Refer APPENDIX A 4 2 6 Emergency notification for fall detection system Figure 4 18 shows the emergency notification text received on smartphone which reports the situation using the predefined message format This system is not limited to the usage of smartphone receivers any mobile phone can be used at the receiving end 47 v 900 t0 SO EL Z1L0Z 12 pejoeyep e 30d yews wo wel 9 Aey AepsaupaM g0 91 90 Oo oy lt g 3 4H z i a w Elder Care Technology o o obo a ee gt o o o o 00000N00 Space For Cupboard 0009000009090090 00004A N0000090 orn or a 0000 OC lm nn g ooooNtS o0000 00 0 olk g 000 00 Figure 4 18 The Emergency notification in the form of text message is received on nurses mobile after detecting the fall The bottom part of this picture shows the way data is displayed on remote computer Since we used the Google voice VoIP and did not use any wireless cellphone carrier for sending these notifications the system can be reproduced free of cost on bigger scale Also there were no incidences of poor cellphone network coverage which are usually observed in
66. ping me while learning this project I am also thankful to my project colleagues Karthik Peddi Namratha Sunkara and Suriyakul Chinchao for their valuable contribution in my work I am also thankful to all my iSocial project colleagues and especially Dr Laffey and Dr Schmidt for showing faith all the times My sincere appreciation is to them for providing me a financial assistance through the most part of my graduate studies Last but the most important I owe my deepest gratitude to my parents Without their encouragement faith and financial support this would not have been possible TABLE OF CONTENTS ACKNOWLEDGEMENT Siriei heat e eter ea eed wa oe OET ll Fed ak OF gl Gs BU lhe reer eee a eee nee oe ne Ee RCE ee oe a eee V LISTOR Bs 1D os pee ee na e E RE E E E T E vili ABSTIRAC Pororo a E E A 1X CHAPTER 1 INTRODUC TION aiescccesneiarcccctsngiocecsecacapsvciernca ria cesrecasecsecusans cannes l CHAPTER 2 BACKGROUND STUDY moreris ar E S 4 CHAPTER 3 ME THO D Saccani a A E 8 Dike Maare UDan a AA nseusaautsa oc seueeioneearaneaseas 8 32 Hardware USE encienda a e E Een 9 32a Replacement of RS 232 by Universal Serial Bus USB interface eee 9 32 Replacement of desktop computer by plug computel cccccessesesseseeseeeeens 10 3 3 Development environment and programs used ccccesseeeeeecceeeeeeeeeeeeeaeees 11 3 4 Data acquisition and storage system 0 0 0 ccccccecccssssssssesesseeeeeeecee
67. pport system and uses a connected component algorithm to discover the simultaneous activation of group of contiguous sensors during a fall In addition to the fall detection the floor sensor data was stored on the Sheevaplug server and then communicated to the remote computer using Internet socket connection An executable application was developed using Eclipse environment for providing a visual representation of this data on the remote computer A web application was further developed by reusing the same source code The falls algorithm was tested for 11 different fall scenarios with 4 volunteers Experimental results show that the system is highly reliable and accurate with the average sensitivity as 92 72 and the average specificity as 95 9 for a single person Further research on improving the floor sensor data will also improve the results of fall detection system CHAPTER 1 INTRODUCTION Due to increasing life expectancy the world population above 65 years of age continues to increase it was 10 in the year 2000 and is expected to be 24 by the year 2100 Likewise the average age which was 26 years in 2000 will be 44 years by then 1 Thus the number of elderly people in the community is increasing and it is becoming important to invent inexpensive technologies for helping large number of elderly live independently detect falls quickly reduce the healthcare costs and provides improved caregiver access to elderly Falls are se
68. r which worked as the data collection server for the carpets In the future we suggest developing a separate website for distributing the software 4 4 2 Online source repository We also created an online repository in the Google cloud for backup and future use This repository consists of all the computer programs developed here Google 54 provides the repository services for users to maintain source programs along with its versioning Following is the link for our repository http code google com p alzheimers project O alzheimers project Unive Co C code google com p alzheimers project wi My favorites Sign in PH alzheimers project University of Missouri Search projects Project Home Downloads Wiki Issues Source Summary People Project Information This project is developed as research project developed by team working under Dr Tyrer Project feeds from University of Missouri Columbia in collaboration with the Alzheimer s Association of Columbia Code license Other Open Source See source for details Labels Academic Eclipse RCP Research Alzheimers UniversityofMissouri Tyrer 2 amp Members kaustubh qmail com Links External links Elder Care University of Missouri Figure 4 22 Picture shows the page of project s Google repository This repository is developed to keep the work secure without losing any references Thus we successfully
69. regions with neighboring 2s are identified in next step In short the algorithm declares a fall occurs when the size of 22 largest contiguous sub region is found to be 10 or larger and exceeds the previous frame by 6 such contiguous elements 3 7 Development of carpet display applications For displaying the carpet data on Graphical User Interface GUI we developed two display applications one for displaying it on the computer and the other for Internet browsers This section discusses the steps of development of the applications 3 7 1 Eclipse desktop application We developed an executable application exe to monitor the carpet data received from the Sheevaplug on a remote computer display Figure 3 12 shows the architecture of this application The entire exe file consists of parts from the Frameworks Standard Widget Toolkit SWT Remote System Explorer RSE core Eclipse environment and the other necessary parts These parts are the Operating System libraries and Java Runtime The shaded block represents the carpet display and the file explorer window that we developed as a plug in We used the Rich Client Platform RCP framework to create a single executable file from the plug in and the framework codes Java Runtime Operating Environment System Eclipse Core SWT RSE Eclipse plug in Figure 3 12 Figure shows the architecture of executable eclipse appl
70. root root 192564 Jan 17 23 59 January _17_2012_ 11PM csv 70009 Jan 17 19 59 January_17_2012 7PM csv root root 183107 Jan 17 21 00 January _17_2012 8PM csv _9PM csv Figure A 8 Figure shows the file explorer window Carpet Data for displaying the data files located on the Sheevaplug server User can right click on any CSV file and select Show History menu to retrieve the data stored in the file The terminal is also shown for invoking commands to the Sheevaplug server A 3 Initiation of Sheevaplug server Following are the instructions for initiating the Sheevaplug server for data acquisition communication on the Internet and fall detection 1 Establish the Sheevaplug connections as described in Section 3 5 1 2 Access the Sheevaplug server using a SecureCRT on the remote computer and invoke the ssh command Refer Section 3 9 3 2 and Figure 4 13 3 The Java archive JAR file is stored on the server at location carpet The name of JAR file is Collector jar case sensitive 4 Execute the jar file to initiate the following processes a Data collection and fall detection program 69 b Socket server program c Google voice program 5 Execute the command java jar Collector jar ServerProperties properties 6 The properties file defines the USB port name thresholds for fall detection user credentials for Google voice connection Sheevaplug Server SecureCRT oe a a ad 4 Lo oJ
71. s to reuse maximum part of our existing computer programs for new Internet application The new web application was tested on various Computers as web servers and the results were exactly similar in addition to the remote accessibility to the data We intended to develop the entire system including the micro controller kits and computer in very efficient manner in terms of space power consumption and the most important factor cost After much discussion and readings we decided to use the Sheevaplug wallplug computer as our web server This low power cheap and tiny computer was able to be successfully interfaced to the micro controller We found many glitches in the received data when the plug computer is used We overcame this issue after reading and experimentation with the wallplug computer and by successfully creating the common ground between micro controller and the computer We feel proud to deliver the system to Alzheimer s association to help them save the lives of elderly people suffering from Alzheimer s This dissertation focuses on the development and evolution of the scalable and cheap online carpet monitoring and reporting system Tome 2012 Kaustubh Gadre University of Missouri DMCA and Figure 4 21 Figure shows the website of smart carpet project being hosted on the Sheevaplug server This website acts as the entry point of the entire software system For the current implementation the website is hosted on the Sheevaplug serve
72. s with a rapid change of 6 elements from the previous matrix Note the encircled components these are ignored as fall data They are most likely other appendages of the person falling but can also be due either system noise or the presence of another person etc and the count does not reach the required threshold Kaustubh Documents Elderc Bebe Kaustubh Documents Elderc eE New Frame 2012 05 16 12 49 47 826 a New Frame 2012 05 16 12 56 50 857 00000 0 0 0 0000 0000000 0 0 0 0 O ooo 0 0 oda Do 0000 O O OOO Oo 000 0 oo 0 0 o 0 0000 090 0 0 0 0 00000 o oO 0000 00 00000 00 00 000 o o0 0 000 0 oo 0 0 Of1 0 0 0 OO 0 ooo 0 0 o oo odi D 000000 0 00000 000 00 0 000 0 0 000000 0 00 00 0 0000000 0 0 0 0 O 000000 0 0 0000 000000000 0 0 0 000000 00 0000 000000 00 0 0 0 O 000000 00 0000 0000000 0 0 0 0 O 0000000 00 000 7 0 o0o000000 00 0 OF ror t iili F Figure 4 19 Figure shows the matrices at two different instances of fall Largest component is boxed and all other smaller components are encircled Fall notification is sent when the size of boxed component is more than the threshold value 10 49 Figure 4 20 shows the graph where every point is from a line of data single matrix and represents the largest connected component in the matrix The data in the graph was over a 15 sec period and the fall occurred at time 12 49 47 826 PM as shown in Figure 4 19 and by the line in Figure 4 20 Another indicator of a fall
73. schematic was used analyzed to analyze the circuit board for finding the ground pin The major steps are as follows 1 The layout shows few ground pins We measured voltage between each pin and the ground from the microcontroller system because both of them use a power supply from the same connection 2 When this voltage was found to be zero we identified one pin as ground which was suitable for making a wired connection to AC supply ground Figure 3 8 3 This established common ground between the microcontroller and the Sheevaplug computer 17 Ground Pin mA 3 ae a SES a Figure 3 8 Figure shows the Sheevaplug circuit board with the identified ground pin encircled The wire connected to the ground pin establishes a ground connection with AC supply ground gt Next in order to remotely monitor movements on the carpet we decided to make this data available on the Internet The next two sections discuss the steps taken for communicating the data over the Internet 3 5 2 Domain Name System DNS name assignment for Sheevaplug server To access any server on the Internet it must have a DNS name assigned to it We assigned a DNS name to the Sheevaplug server for making it available in the network with a project related name using dyndns com facility The carpetsmart dyndns server com name will be used in all further sections The detail directions for using the DynDNS are provided at Ubuntu official docum
74. se Positive Rate FPR and the Specificity Table 4 4 Table indicating the performance of fall detection system for false positive fall experiments performed on the carpet Each experiment is performed twice at 10 times Note that in this the system detects no falls i e in V1 first row of 10 attempts 8 came in with no fall False Positive Fall Experiment Performing Performance Volunteers Parameters Specificity From a sit with legs tucked in 8 10 9 10 From screed sit with legs spread 10 om 10 8 10 ced From standing perform sit ups 10 10 10 10 From standing bend down on knees and stoop toa 10 10 10 10 squatting position From standing bend down and kneel on the floor 10 10 10 k 10 Ean From standing bend down and kneel on the floor 8 10 Batad 10 and then lie down on the floor From standing bend down to plug an appliance 10 10 10 10 into electric outlet From standing squat to tie a shoe 9 are 10 10 ook 10 pet From standing walk forward and stop suddenly 10 hia 10 10 10 ka From standing walk towards chair and sit in it 10 10 10 Ka 10 From sitting in chair bend down to pick up the 10 10 foaled 10 book Table 4 5 shows the average specificity and average FPR values calculated as 95 9 and 4 09 respectively from the results recorded in Table 4 4 52 Table 4 5 Table shows the performance parameters of the system for false positive fall experiments The average sensitivity w
75. sed Purpose of use Standard Widget Toolkit SWT 31 Displaying the sensors activation on computer Rich Ajax Protocol RAP 32 Displaying the sensors activation in Internet browser Remote System Explorer RSE 33 Remotely Accessing the carpet data stored on Sheevaplug server Rich Client Protocol RCP 34 Creating a software executable containing all the above plug ins development environment our computer programs In the earlier implementation a Windows utility Hyper Terminal was used for observing the floor sensor data arriving on the computer However this utility was not available in Windows 7 and hence we used another similar utility SecureCRT It is available to download for free on the manufacturer VanDyke website 35 For converting and then storing the carpet data in Comma Separated Value CSV format an openCSV program library was used It is developed by the another open source community and is also available to download for free 36 11 3 4 Data acquisition and storage system The micro controller scans all the sensors in carpet segments A B C and D It compiles a frame of data starting with word S start followed by the data for segments A B C and D and ending with word E end 4 It continuously sends these frames to computer at a speed of 19 2 kbps 23 For acquiring this data on computer we established the connections as shown in Figure 3 2 and displayed the data 24 However
76. t one frame Round trip time 2 116 482 ms 45 Figure 4 16 shows the times we recorded for transmitting one frame of data to the remote computer for one hour It also shows the average time 335 ms which is approximately 2 5 times the rate of data generation i e 140 ms Hence every frame appeared to be delayed by 2 or 3 frames which can be accepted considering the added advantage of remote monitoring This delay is notable in displaying the fall data and the actual fall Network latency observed in one hour duration a DH Vv Pa Q Q SS wf Sam gt Q Z Figure 4 16 Graph shows the comparison of the observed network latency during one hour The line parallel to abscissa describes the average network latency 335 ms 4 2 5 Displaying carpet data in the Internet browser It is important to pass the carpet data over the Internet We installed an Apache Tomcat web server installed on the Sheevaplug server used an eclipse RAP for display when opened in an Internet browser 46 Smart Carpet Display Space for Cupboard Figure 4 17 Figure shows an eclipse web application opened in Internet browser on desktop computer The display contains 128 different blocks each of which shows the state of one carpet sensor Figure 4 17 shows a display of the smart carpet sensors demonstrating that the eclipse web application worked satisfactorily It also worked well on various types of br
77. t the person s presence standing on it The microcontroller selects the amplifier output connected to foil sensors and sends it to the display computer 4 A specific protocol was used for sending the floor sensor data in continuous stream of characters 23 This system was developed to detect the presence of a person and display the information of sensors on a local computer 24 It was not equipped to store and produce it on the Internet or to detect any falls This necessitated a 5 server that detects falls and makes the data available on Internet at all time The server needed to be cost and power efficient Also we needed the software executable for displaying movement on the carpet Global Scale Technologies manufactures the wall mounted plug computer called the Sheevaplug As the name suggests it can be directly hooked onto the AC power on the wall used without any display monitor or I O device It is available in different versions having different technical specification and costs We used the Sheevaplug being one of the first and cheapest among its variants GuruPlug DreamPlug and D2Plug It fulfilled all the requirements to be the Smart Carpet server such as data acquisition storage and communication over the Internet 25 The network sockets were used for transmitting the data from Sheevaplug server to the remote computer and Java programs were developed for both ends of communication The network socket is a direct con
78. ter for data storage and communication We also developed algorithm for fall detection alert signal generation and display application 3 1 Initial Setup The smart carpet system is made of four carpet segments arranged as shown in Figure 3 1 Figure 3 1 Figure indicates the arrangement of carpet segments Each segment A B C D consists of 32 sensors arranged in 8 columns and 4 rows Each segment having 4x8 foil sensors is connected to a circuit of 32 operational amplifiers and a microcontroller Such 4 segments along with their microcontrollers comprise the complete hardware system for covering the entire room Only one 8 microcontroller 1s connected to the computer using a cable having Recommended Standard 232 RS 232 serial connectors on both ends All the microcontrollers communicate with each other using wireless protocol 23 The software system acquires the floor sensors data and displays it on a personal computer 24 It displays the state of sensors and the way person walks on it We identified enhancements for this system as goals for our project which includes storing and communicating the data over the Internet detecting and notifying the fall events and implementing the above features using plug computers Following sections discusses the methods carried out to achieve these goals 3 2 Hardware used 3 2 1 Replacement of RS 232 by Universal Serial Bus USB interface In the earlier implementation of t
79. the ground adjustment All invalid characters have been omitted ssh AES 256 CTR is a 12 Rows 63 Cols Figure 4 9 Comparison of the data eee by Sheevaplug before and after providing the common ground The upper half shows the data received before whereas the lower part shows the data received after providing the ground connection Figure 4 10 shows the graph where each point shows the number of active sensors in each line of raw data in Figure 4 9 These sensors were detected by Sheevaplug before and after connecting the ground pin The ground pin eliminated the large amount of invalid data Observation of data on Sheevaplug ee o Atb a Nal WaN NON TW After tortet title Sumber of sensors detected a active si pi e D AD D a oe am 2 5 AS 8 ets K oe i at ye D aye ay nye ee i Cl ee ed Time Figure 4 10 Number of active sensors detected by Sheevaplug before and after establishing the ground connection 40 4 2 2 Sheevaplug data conversion time and storage requirements It s important to analyze the time and memory requirements of Sheevaplug in the process of real time data acquisition We measured the data conversion times and storage requirements for each storage format using the Sheevaplug on receiving end We calculated the difference between the times at which the raw data frame arrived at computer and the time at which the new format became available obtained mean and standard deviation of t
80. the Standing to falling activities Second row describes the Tripping and falling activities whilst the third row indicates the Sitting to falling activities 3 9 4 2 Specificity analysis of the fall detection system Another set of experiments were performed with the situations that are expected to produce false positive data Figure 3 15 shows the list of experiments we performed with 2 volunteers each performing every experiment at 10 times Following are the formulae used FPR False Positives Total optical false positive experiments Specificity False Positive Rate 32 These experiments were not intended to produce the fall data These were aimed to determine the ability of fall detection system to distinguish other daily activities as non falls In every experiment when volunteer performed the activity the mobile phone was not expected to receive any notification from the server If the notification was received we noted the trial as false positive The specificity and the FPR were calculated using these observations 3 Performing 4 Bend down in squatting position 1 Sitting with legs tucked 2 Sitting with legs in spread sit ups ee 5 Bend down and kneel 6 Bend down kneel and 7 Bend down and plug an 8 Bend down and tie a on floor then lie down on floor appliance shoe 9 Walk straight and 10 Walk towards chair 11 Sit on chair bend down sudden stop and sit on it to pick book Fi
81. tion Select the system type as Linux 68 b Write host name as http carpetsmart dyndns server com and connection name as Sheevaplug Click next Do not change anything in subsequent screens Click Finish Elder Care Technology File Window Help Search Start Stop we mz X gt D Carpet Data 3 a Sheevaplug a h Sftp Files My Home 4 Carpet Data Ea January_13_2012_5PM csv 4 January_13_2012_6PM csv ER January_13_2012_7PM csv E January_13_2012_8PM csv 4 January_13_2012_9PM csv Ea January_17_2012_10PM csv Ea January_17_2012_11PM csv ta January_17_2012_7PM csv E January_17_2012_8PM csv Ea January_17_2012_9PM csv ray January_18_2012_0AM csv 4 January_18_2012_0PM csv ED Januar 18 2012 1OAM csw fL Remote Shell 2 Fi Sheevaplug 32 drwxr xr x 2 root root drwxr xr x 3 root root E rw r r 1 root root E rw r r 1 root root E rw r r i E rw r r 1 root root E rw r r 1 root root E rw r r a E rw r r a EN rw r r 1 root root EN rw r r 2 Command Launch a command shell from the selected directory paa 4096 Mar 19 20 36 4096 Mar 20 17 49 2733 Jan 13 17 57 January_13_2012 5PM csv 90339 Jan 13 18 46 January _13_2012 6PM csv root root 70 Jan 13 19 23 January_13_2012 7PM csv 5167 Jan 13 21 00 January _13_2012 8PM csv 1829 Jan 13 21 00 January _13_2012 root root 192456 Jan 17 23 00 January _17_2012 10PM csv
82. tion OL data TN TLS ea tinh ht iE a IE EEA EAE ARE Ea 27 3 9 3 Performance of Sheevaplis Computer ciao eee 29 3 9 4 Performance of fall detection and notification SYSteM cc ccesssssessssseeeeeeeens 31 CHAPTER 4 R SULTS ereren rennet a tra mE n ERT N Treen ee hema errr eat 34 4 1 Data acquisition and storage system ccccccccsssseesssssseeeececccceeeeeeeeeeeaaeeeseeeeeeeees 34 4 1 1 Observation of data on terminal emulator using USB interface 00000 34 4 1 2 Storing the carpet sensor data in different formats cccccccccccccccceseceeeeeeeeeeeeeees 35 4 2 Sheevaplug computer as a project server s0ceceeeeeeececcceceseceneaasseseeess 37 4 2 1 Ground connection provision to Sheevaplug computer ccceeeeseeeeeeeseeeeeeeees 38 4 2 2 Sheevaplug data conversion time and storage requirements ccccccccccseeeeeeees 41 42 3 Accessing Sheevaplug using dynamic Domain Name System name 43 4 2 4 Time for transmitting the data to the remote Computet ccceseeesesseeeeeeeeeees 44 4 2 5 Displaying carpet data in the Internet DrOWSe cccceeeeeeeseesesesssesseeseeeeeens 46 4 2 6 Emergency notification for fall detection System ccccccceesesessessessssssseeeeeens 47 4 3 Analysis of carpet data and fall detection System cccccccccccsssteeeeeeeeeeees 48 4 3 1 Observation of data for largest connected COMPONEN
83. very one hour The file name describes the date and time of the occurrence of the data frame The highlighted frame represents the data generated at the fall shown in figure 4 2 0 0 ccccceeeeeeeeeeeees 36 Figure 4 4 Data shown is stored in matrix format The elements with 1 show the active sensors in the carpet and 0 shows the inactive sensors The crossed region represents the floor area with no carpets laid Figure 4 5 Data shown is stored in CSV format All 1s from the matrix are stored along with their row and column coordinates The highlighted row shows the CSV conversion of the matrix in Figure 4 4 37 Figure 4 6 The Sheevaplug is shown connected to the microcontroller system using USB interface for receiving the data and transmitting it over the Internet using Ethernet connection Sheevaplug also performs the analysis of incoming data and notifies the caregiver after detecting the fall 38 Figure 4 7 The Sheevaplug circuit board is shown with the ground connection provided to it The ground pin was identified after analyzing the circuit diagram and measuring the voltage across the pin and EOIN trp a ee eit ele oe fa Se it hae wk E EE E a lh eat Be eet cht aed ae 39 Figure 4 8 The Sheevaplug is connected to AC power using the ground connection This connection achieves the common ground between both ends of RS 232 communication ccccccceeeeeeeeeeeeeeees 39 Figure 4 9 Comparison of the data received by Sheev
84. w the required users click while exporting an executable 3 Select the destination directory for Leave exporting the executable Configuration and Root Directory options unchanged Make sure to tick the check box Synchronize before exporting Refer to Figure A 3 Eclipse product Use an existing Eclipse product configuration to export the product in one of the available formats Product Configuration Configuration tion edu missouri eldercare product w Browse Root directory Software Synchronization Synchronization of the product configuration with the product s defining plug in ensures that the plug in does not contain stale data v Synchronize before exporting Destination Directory C Smart Carpet Archive file D Figure A 3 Figure shows the eclipse product configuration screen before exporting Select the parameters as described above 4 Now the software should be created in few minutes at the provided location It should contain following items as shown in Figure A 4 a The executable file SmartCarpet exe b Directories configuration jre plugins workspace 66 c Other configuration files p7 e OGO a gt Computer OS C Smart Carpet gt eclipse gt Organize v Include in library v Share with Burn New folder Wr Favorites Ss _ _ 3 Homegroup configuration jre plugins workspace ElderCare S
85. xting to any telephone number in United States and Canada It also provides an Application Programming Interface API that is developed in Java Several other options were discussed earlier which included Skype MagicJack Vonage and Voxeo 24 Out of which Skype also provides Java API and MagicJack has an unofficial API Voxeo has its own network server Every emergency call needs to be placed through it the server which in turn calls the caregiver All these services are costlier than Google voice which led us to move forward with Google Voice We can implement the notification system on smartphones iPhone Android etc Applications can be developed to push the data to the cellphones Developing the mobile application will help scalability making the system available for larger numbers of end users at very low cost 59 1 2 3 4 5 6 7 8 9 10 REFERENCES U Nations World population to 2300 New York United Nations 2004 D Jiangpeng B Xiaole Y Zhimin S Zhaohui and X Dong PerFallD A pervasive fall detection system using mobile phones in Pervasive Computing and Communications Workshops PERCOM Workshops 2010 amp th IEEE International Conference on 2010 pp 292 297 D Litvak Y Zigel and I Gannot Fall detection of elderly through floor vibrations and sound in Engineering in Medicine and Biology Society 2008 EMBS 2008 30th Annual International Conference of th
86. y Society 2006 EMBS 06 28th Annual International Conference of the IEEE 2006 pp 6388 6391 D Anderson R H Luke J M Keller M Skubic M Rantz and M Aud Linguistic summarization of video for fall detection using voxel person and fuzzy logic Computer Vision and Image Understanding vol 113 pp 80 89 2009 K P Murphy Dynamic bayesian networks representation inference and learning 2002 D Anderson R H Luke J M Keller M Skubic M J Rantz and M A Aud Modeling Human Activity From Voxel Person Using Fuzzy Logic Fuzzy Systems IEEE Transactions on vol 17 pp 39 49 2009 T Gill J M Keller D T Anderson and R H Luke A system for change detection and human recognition in voxel space using the Microsoft Kinect sensor in Applied Imagery Pattern Recognition Workshop AIPR 2011 IEEE 2011 pp 1 8 M Popescu Y Li M Skubic and M Rantz An acoustic fall detector system that uses sound height information to reduce the false alarm rate in Engineering 61 22 23 24 25 26 27 28 29 30 31 32 33 34 in Medicine and Biology Society 2008 EMBS 2008 30th Annual International Conference of the IEEE 2008 pp 4628 4631 L Yun Z Zhiling M Popescu and K C Ho Acoustic fall detection using a circular microphone array in Engineering in Medicine and Biology Society EMBC 2010 Annual International Conference of the IE
87. y as well as they consume lot of power and occupy lot of space Hence we replaced our desktop computer by the plug computer The plug computer is equipped with minimal resources but sufficient for executing our computer programs This section discusses the detail procedures for setting up the project server on Sheevaplug computer 3 5 1 Ground connection provision to Sheevaplug computer We used the Sheevaplug computer to service the data acquisition system and observed large number of invalid characters in the form of noise The similar noise was observed on laptop computer when operating on batteries We excluded noise on the desktop computer and the available documentation 4 23 24 of this project did not mention how to resolve this issue 16 Now RS 232 communications requires both ends at zero volts for communication 37 The Sheevaplug computer does not have a ground connection so that the RS 232 receiving end was left ungrounded This requires grounding of the Sheevaplug signal side To address this issue we analyzed the Sheevaplug circuit board with the help of its layout shown in Figure 3 7 E ALITTI ETT ETT TI ju T j i 4 Jo ak NOMEG earl j P a ae ad Li k Lat l ne ele a ra ee rie eee ELL i ee FH PEPER ETERS FR PAPE es bm Pee PLPC SCLC E IT Geant od TEN wI r Figure 3 7 Figure shows the schematic of Sheevaplug circuit board Source 38 This

research

Contents

Download Pdf Manuals

Related Search

Related Contents