User Manual
Contents
1. To use EasyCamCalib the easiest way is to add the Interfaces folder you downloaded to the matlab path This will only input some GUI executables to your path All the functions needed by EasyCamCalib are loaded locally so your matlab path don t get filled with unnecessary files Currently the toolbox needs the Matlab optimization toolbox So to start the software simply call EasyCamCalib from the matlab prompt 2 1 User Interface When you call EasyCamCalib from the matlab prompt you will find a GUI similar to the one of figure 3 The Ul have also a menu bar that will be addressed bellow 2 2 File Menu Start Start the calibration using the files in the calibration list T his menu has the same function as the button 5 of Figures Save Data Save the calibration data in a mat file that can be loaded later Note that during execution as a fail safe EasyCamCalib saves intermediary results in temp CalibData_temp mat Save to txt Save the intrinsic calibration parameters in txt file The parameters are saved in the following order 6 File Edit Tools Refinement BEB le e EE ele Done mi Ki ReProjection Error 2 2 File Menu _Auto Calibration Results Eta 2050 83 O Aspect Ratio 1 00003 Skew 0 000620292 Center 640 2 491 6 Focal Distance 1089 31 Qsi 0 282124 0 1498 Mean RMS Error 0 149868 gt alia Value2. 2 3 3 Aspect Ratio 4 Skew angle 5 Principal point x coordinate 6 Principal point y coordinate 7 Focal distance The remaining values refer to the case of wide angle lenses where there is a circular boundary limiting the meaningful region of the image Load Data Load a previously saved mat file 2 3 Edit Menu Options Open the option UI of Figure4 2 4 Tools Menu Modify Points Manually add or remove points from the calibration pattern When you use this tool all the images in the calibration list will pop up consecutively en abling you to add or remove any points to the calibration process Here are the commands you can use with this tool e Left Click select the nearest corner for insertion and input the co ordinates in the matlab prompt Note that although the coordinates in the calibration plane are referenced in millimetres when inputting coordinates you have to input them in integer units the origin corner is set to 0 0 the next is 1 0 and so on 2 4 Tools Menu Figure 4 Options window 1 Define the calibration chessboard grid size in millimetres The grid is assumed to be square Automatically set the same reference frame in the calibration pattern across multiple images To be able to use this feature the calibration images must fulfill three requirements e The calibration grid must have two diagonally consecutive white squares painted with different c
3. Calibration Images Figure 2 Examples of good calibration images e The angle between the optical axis and the normal to the calibration plane should be higher than 15 i e you must avoid fronto parallel configurations angle 0 in order to have a good decoupling between and the focal distance 1 You also must avoid highly slanted views to avoid bad automatic corner detections Figures 1 and 2 illustrate some good and bad calibration images examples e The number of squares present in the image must be enough to calibrate from a single view The image should contain at least 16 corners Note that the more points you provide to the algorithm the better the projection model will be estimated e The calibration grid must be in the central part of the image An optimal situation is when all the image is filled with the calibration grid If you cannot take calibration images in this conditions try to put the This is not the minimum number of corners required to calibrate an image from a single view 2 USING EASYCAMCALIB calibration grid over a non textured material like a black fabric to avoid bad automatic corner detections One of the few limitations of the software comes from the automatic corner detection used to initialize the calibration If the calibration image does not fulfil the above requirements there is a good chance that the calibration will fail due to bad detected initial corners 2 Using EasyCamCalib
4. _ Bars Figure 3 Main EasyCamCalib window Browser list box All the calibration im ages are chosen from this list box Calibration list All the images you want to use for calibration are listed here Preview window As you click in the browse list box or the calibration list box a small preview off the image is presented here Mean reprojection errors of the calibrated images T he value in bold represents the mean re projection error of the current se lected image Start calibrating the images of the cali bration list Options Button Estimated intrinsic calibration parame ters 10 11 12 13 Estimated extrinsic calibration parame ters relative pose between the camera and the plane After the calibration is over switch to the calibration parameters you want to dis play Main visualization window All the visual results are presented here Angle of the camera relatively to the cali bration plane and distance from the plane origin to the optical center of the camera Visualization buttons Read the tooltips while using the toolbox for more informa tion Return to the calibration image selection stage to perform a new calibration 2 USING EASYCAMCALIB 1 Coupled parameter between the focal distance and distortion coefficient n To This parameter is estimated during the linear calibration in 1 2 Distortion coefficient according to the first order division model
5. rest of the images stays untouched s set the size of the window where we will search for corners when clicking in the image for inputting points e g try to generate more interest points from the ones already existent Use carefully and always save your changes before attempting this Define Origins Manually set the reference frame of the calibration plane for each image You define the origin with the Left Mouse click and the x direction with the Right Mouse button After you are done with one image press space to get to the next image or press q to exit without further changes It is very important that you keep a constant reference frame across all calibration images This reference frame will be used to compute the extrinsic parameters transform from the image plane to the calibration grid Correct Radial Distortion This tools aims at checking the projection model and distortion parameter s estimation When you correct the radial distortion if the model is accurately estimated straight lines in the 3D world will be projected as straight lines in the image plane To use the UI of Figured simply select a calibration file previously saved from the left list box and the image you want to correct from the right list box a small preview will appear on the right Then hit the start button In the case of wide angle lenses where the radial distortion is high select the Arthroscope image source and input a desired u
6. 01 pp 1 125 I 132 vol 1 3 Y Ma S Soatto J Kosecka and S S Sastry An Invitation to 8 D Vision 4 5 From Images to Geometric Models SpringerVerlag 2003 J Y Bouguet Camera Calibration Toolbox for Matlab Online Avail able http www vision caltech edu bouguetj calib_doc index html ref D C Brown Decentering Distortion of Lenses Photometric Engineer ing vol 32 no 3 pp 444 462 1966 17
7. DUCTION e Initial Calibration With the automatic corners detected a first cal ibration is estimated using 1 This calibration will be referred as the Initial Calibration e New Points Generation Using the initial calibration estimation points are generated in the image plane and matched to squares of the calibration grid Note that for lenses with strong radial distortion points generated in the periphery of the image tend to be inaccurate In this case you might need to use the manual selection tool to remove undesired generated points e Final Calibration With the new generated points the calibration parameters are recomputed providing what we will call from now on the Final Calibration e Calibration Refinement The calibration parameters are refined us ing a non linear optimization over the re projection error This is the final result of the calibration and will be referred from now on as the Optimal Calibration 1 3 Capturing Calibration Images For EasyCamCalib to be able to calibrate the camera from a single image the following requirements must be fulfilled Figure 1 Examples of bad calibration images On the left we can see an image in a fronto parallel configuration On the right we can see a highly slanted calibration image Both these images fail to calibrate automatically you can still use the slanted one to calibrate but you will have to manually select some of the input points 1 3 Capturing
8. INSTITUTE OF SYSTEMS AND ROBOTICS UNIVERSITY OF COIMBRA EasyCamCalib User Manual VERSION 1 1 August 24 2011 Contents Contents 1 Introduction 1 1 What is EasyCamCalib 0 0 020000 1 2 Basics of Single Image Calibration 1 3 Capturing Calibration Images 2 Using EasyCamCalib 2 Uoer terae as cfs a8 iw Se eS Se Bed Ee Y 2a SES A he Bio Gi He ok Avot Heide dic l amp ai GR a eee sd 29 AA DA Tool NCH A e al hae oe A ote TR ih as Bok Ue e he oe Be Dio WRCMMCHICING Leds oS ee Ree we Ee A do SS ee 3 Calibration from a Single Image 4 Analysing the Calibration Data 4 1 EasyCamCalib output yy ars ew Eos a ee Ee aS 5 Known Issues Bibliography 13 15 15 16 17 1 Introduction 1 1 What is EasyCamCalib The purpose of EasyCamCalib is to calibrate a camera with radial distortion from a single image of a planar chessboard pattern The application aims to automatically calibrate a camera from an image or a set of images requir ing minimal user intervention The original goal of this application was to calibrate an endoscope high radial distortion using a single image captured by a surgeon on the operating room but the methods are generalized to any camera that presents moderate to high levels of radial distortion Although the algorithm is designed to provide accurate calibration using a single image the accuracy and robustness is increased when using more
9. alibration using all the corners automatic corners new generated points Inside this structure besides the intrinsic extrinsic and distortion parameter you can find the re projection error information OptimCalib Optimal calibration after using the non linear optimizer over FinalCalib Inside this structure besides the intrinsic extrinsic and distortion parameter you can also find the reprojection error information Inside InitCalb FinalCalib and OptimCalib you can find all the relevant calibration information The calibration parameters aspect ratio skew angle focal distance and projection center are identified according to the literature 3 The transformation T gives you the transform between the calibration plane and the camera Besides the parameters defined before you can find a distortion parameter and a parameter 7 Fe as well as the intrinsics matrix K computed as usual in the literature 3 and a matrix K used in other applications 5 Known Issues 16 e The major drawback of this software resides in the automatic corner detection and counting for the first calibration parameters linear esti mation As the application targets a wide range of cameras and lens with different distortions this task is far from being trivial The soft ware must be able to handle illumination variations resolution changes Bibliography different sizes of the squares in the image different amount and effects of dis
10. e again the Levenberg Marquardt algorithm is used as the minimization solution 3 Calibration from a Single Image This is the quick guide to easily calibrate a camera from a single image You can start by choosing the image for the calibration from your dataset using 13 3 CALIBRATION FROM A SINGLE IMAGE the listbox of Figure9a You can also navigate through your file system using the path selector above The next step is to configure the calibration using the Options menu or the button above the S tart button The window of Figure9b will appear and you will be able to define some options File Edit Tools Refinement File Edit Tools Refinement IE Al iE w z 10 E H lArhroscope02_2mm i 6 PointGrey_15mm ti PointGrey_76mm tiff w y 4 CEE ali gt por a Choosing the calibration image from the b Configuring the calibration through the list box options editor Figure 9 Calibrating from a single image Step 1 After the calibration customization you can hit the green Start button and a confirmation dialogue will appear Figurel0a After confirming the calibration setup hit the Proceed button to start the calibration After a while the first step of the calibration is completed At this time you can check the grid points used in the linear estimator and do small adjusts using the Modify Points tool as shown in Figurel0b Select Origins Correct Radial Distortion Homograph
11. he 3D transform between the calibration plane and the camera etc Further instruction about the advanced use of the toolbox can be found in the tutorial video 4 1 EasyCamCalib output All the calibration data is stored in a MATLAB structure that is composed by the following fields e ImageData ImageRGB RGB image 15 5 KNOWN ISSUES ImageGrey Grayscale image Info Some additional information about the calibration image including the grid size resolution etc Hand2Opto if any OptoTracker information is available this 4x4 matrix holds the transformation from the Hand camera to the Optotracker Boundary if the image was acquired with a wide angle lens this field holds all the boundary information conic parameters boundary points etc PosImageAuto Point automatically detected in image coordinates for the Initial Calibration PosPlaneAuto Points automatically detected in calibration plane coordi nates for the Initial Calibration InitCalib Initial calibration using only the automatically detected points Inside this structure besides the intrinsic extrinsic and distortion parameter you can also find the re projection error information PosImage Corners detected in image coordinates after joining automatic corners and new generated corners PosPlane Corners detected in plane coordinates after joining automatic cor ners and new generated corners FinalCalib Final c
12. is homography is computed using the extrinsic parameters of the views and according to 3 Figure7 shows the result of the test in a wide angle lens 4mm arthroscope Note that this test only makes sense if all the images in the dataset gener ated a different calibration see the 1st 2nd Order Refinement 1 by 1 presented bellow If you do this test on a dataset where all images were optimized to gether since the non linear optimization also optimizes the transforms you will be biasing the results Compare with Bouguet This tool launches the Bouguet 4 calibration toolbox over the current data and compares the results with the EasyCamCalib estimations Note that the distortion profile curve has to be converted from the division model 2 used by EasyCamCalib to the Brown s model 5 used in the Bouguet s toolbox which can introduce some error in the distortion parameters Also be aware that to do a fair comparison you have to use more than 10 images and optimize the EasyCamCalib results using all the data see 1st 2nd Order Refinement bellow before running the tool In this comparison only two distortion parameters are used in the Bouguet s calibration estimation and the tangential distortion is ignored Also both aspect ratio and skew angle are fixed to 1 and O respectively 12 2 5 Refinement 2 5 Refinement Ist Order Refinement Perform a non linear optimization over all the input images assuming the first order div
13. ision model for radial distortion Both intrinsics and extrinsic parameters of all images are optimized together over the re projection error of the pixels The Levenberg Marquardt algorithm is used as the minimization solution 2nd Order Refinement Perform a non linear optimization over all the input images assuming the second order division model for radial distortion Both intrinsics and extrinsic parameters of all images are optimized together over the re projection error of the pixels The Levenberg Marquardt algorithm is used as the minimization solution 1st Order Refinement 1 by 1 Perform a non linear optimization over the input images independently as suming the first order division model for radial distortion Both intrinsics and extrinsics parameters of each images are optimized independently over the re projection error of the pixels This means that in the end you will get as many optimized calibrations as the number of images in the dataset The Levenberg Marquardt algorithm is used as the minimization solution 2nd Order Refinement 1 by 1 Perform a non linear optimization over the input images independently as suming the second order division model for radial distortion Both intrinsics and extrinsics parameters of each images are optimized independently over the re projection error of the pixels This means that in the end you will get aS many optimized calibrations as the number of images in the dataset Onc
14. ndistorted image size in pixels 10 _ Calibration File Bl mat 4 CalibBoundary mat CalibData_12 mat CalibData_13 mat CalibData_14 mat CalibData_EasyCam CalibData_EasyCam CalibData EasyCami CalbData_t1l mat CalibData_temp ma CalibDatatodelete mi CalibDatattt mat CaliblmagesScene m_ Image File ES Arhroscope02_2mm Arthroscope01_2mr Options Arthroscopic a Point Grey a Original image 2 4 Tools Menu b Corrected image Figure 6 Result of the radial distortion correction Image01 Image02 Image03 Image04 Image05 Image06 Image07 Image08 Image09 Image 10 Imagell Imagel2 Image13 Imagel4 limane1S 7 E Image01 Image02 Image03 Image04 Image05 Image06 Image07 Image08 Image09 Image10 Imagell Imagel2 Imagels Image14 mane1s i7 Load Calibration Start Figure 7 Homography checker UI 11 2 USING EASYCAMCALIB b Image 2 c Image 2 generated from image 1 trough homography Figure 8 Result of the homography test Homography Checker This tools checks the consistency of the extrinsic parameters through a visual homography test The UI of Figure7 is presented to the user which has to load a calibration file and select two images for the test After the user hits the start button a new image 2 will be generated though homography from image 1 Th
15. ol ors e The colors of the marks should be very distinctive in the HSV color space The colors tested usually present both high Value and Satu ration and a different value of Hue One good example is using light blue and purple as marks colors e The marks should be nearby the center of the image if the image has high distortion 3 Setting this option will enable the cali bration parameters non linear optimiza tion over all the images without needing to run it manually from the menu Defines the image source If Endo scopic FishEye Lens is selected Easy CamCalib assumes that there is a wide angle lens with high radial distortion and will try to fit an ellipse to the boundary contour of the meaningful region of the image If Normal Lens is selected the lens is assumed to have moderate distortion Abort the whole calibration process in case an image fails the calibration Automatically remove the images where the process fails leaving only the good images for the calibration Save the defined options and quit to the main UI 2 USING EASYCAMCALIB e Right Click select the nearest point for removal The point gets surrounded by a yellow square e Middle Click define a box with two middle clicks that select points for removal e p show hide point coordinates e space get to the next image e q finish the manual selection here The modifications you have done so far are kept the
16. than one image 1 2 Basics of Single Image Calibration The EasyCamCalib toolbox is built upon the recent work of Barreto et al 1 where the authors are able to calibrate a camera presenting radial distortion using a single image of a planar chessboard pattern The radial distortion is modeled using the so called division model 2 and the method provides a closed form estimation of the intrinsic parameters and distortion coefficient The fact that the distortion follows a known model provides additional geometric cues for achieving calibration from a single image For further details on the calibration algorithm please refer to 1 The toolbox provides an interface that facilitates the calibration of a cam era from a single image The calibration is performed as follows e Boundary Detection in the case of endoscopic or fish eye lenses The boundary between the meaningful region of the arthroscopic image and the background is defined The boundary information is used to later restrict the automatic corner detection of the chessboard pattern e Automatic Corner Detection The image is searched for plausible corners which are referenced in the chessboard reference frame This detection is based in the entropy of the angles and uses geometric metrics to validate and count the corners Therefore the automatic corner de tection is sensitive to illumination conditions and view angle as referred in section 1 3 1 INTRO
17. tortion background clutter other than the calibration grid different shapes of the grid squares as the perspective distortion changes etc Therefore at this stage the software is not yet completely automatic You will find yourself using the Modify Points tool quite often This issue will be addressed in the next releases e While analysing the calibration data you are able to change the view point in the extrinsic parameters visualizer For some reason the UI has a bug that allows the user to change the viewpoint in all the other figures of the UI When this happens simply change the image you are analysing clicking in a different item of the list box and the axis will go back to normal e It is possible that if some input points are misplaced the New Points Generation step starves your system memory while generating an in finite number of points When this happens you have to restart the application Bibliography El a J Barreto J Roquette P Sturm and F Fonseca Automatic camera calibration applied to medical endoscopy in Proceedings of the 20th British Machine Vision Conference London UK 2009 Online Available http perception inrialpes fr Publications 2009 BRSF09 A Fitzgibbon Simultaneous linear estimation of multiple view geometry and lens distortion in Computer Vision and Pattern Recognition 2001 CVPR 2001 Proceedings of the 2001 IEEE Computer Society Conference on vol 1 20
18. y Checker g aj Arhroscope02_2mm gt Arhroscope02_2mm rthroscope01_2mm Ka rthroscope01_2mm DragonFly_76mm pn DragonFly_76mm pn FishEye_76mm png Ps FishEye_76mm png PointGrey_15mm ti PointGrey_15mm ti PointGrey 76mm tiff PointGrey_76mm ti MA AA gt al al CEA TA a Checking up the calibration setup b Adding Removing points using the man ual tool Figure 10 Calibrating from a single image Step 2 14 After making sure that no wrong input points are being used in the cal ibration you can proceed to the non linear optimizer Use the Refinement menu to choose the appropriate optimizer Figurella In the end you will finish with a complete calibration from a single image Check the results as indicated in Figurellb After that you can return to the first step using the Return button or you can analyse the data as shown later Ay fa Qu ay Jul BR a Running the non linear optimizer b Final result of the calibration Figure 11 Calibrating from a single image Step 3 4 Analysing the Calibration Data After a successful calibration or any time you load a calibration file the Easy CamCalib toolbox allows to visually inspect the data and check the parameters estimation accuracy With the provided tools you can see the re projection errors of each image inspect the input points the parameters change across calibrations t
Download Pdf Manuals
Related Search