| Trinocular stereo vision is the important range measurementtechnique in robots passive range measurements. It indirectly imtatesthe way of human vision,could flexibly and accurately obtain the rangeinformation of scenery.It's so important that it cann't be substituted byother robot vision methods,it's of great value to reasearch on it fromthe point of vision physiology and engineering application. In this paper I first discuss the relative problems in the stereo visionsystem on a base of typical parallel binocular stereo vision model,thenpoint out the shortcomings of binocular stereo vision,so I choose thetrinocular stereo vision as the solution of the problems. Camera calibration is for the purpose of determining the position,the intrinsic and extrinsic parameters of the cameras,in order to establ-ish the correspondence between object point and the point in the images.Camera calibration methods an be classified into two classes:traditionalcalibration and camera self calibration . Constructing an effectivecamera model is not only helpful to reconstructing the scenery but alsohelpful to the matching problems.In chapter 3,a camera model with onerank radial aberration is established,and I use radial alignment constraint method to recalibrate the camera. Stereo matching is the most important and the difficultiest problemin stereo vision.There will be great difference when 3D scenery isprojected into 2D images,and so many factors in the scenery such asillumination conditions,geometry and physical characteristics of thescenery,interferences of noises,aberration and camera parameters,etc,will be represented by the gray values of the image.So it's difficultto perform a matching between images containing so manydisadvantageous factors without mistakes.Any matching algorithm'seffectivity depends on the solution of the following 3 problems:①choosing the correct matching feature;② finding the intrinsiccharacteristics of the feature;③establising the stable algorithm thatcan match the features correctly.All the researches on stereomatching are aiming at the solution of these 3 problems,and so manyalgorithms has been invented,but stereo matching problem hasn't beensovled properly untill now because so many disadvantageous factors areinvolved, especially in complicated scenery, how to improve theablility of the algorithm so that it can be more robust towardsinterferences ,make less mistaking matches ,requires further study .Inchapter 4,i first invent the feature matching algorithm that based on theprojection invariable constraint , and establish the correspondentmathematical model,changes the solution of the matching into thesolution of the minimum value of the function (4.5.3.2-5)in chapter4,theoretically very concise and beautiful.I choose the stereo vision system DIGI-HIBW-60,which is theproduct of Point Grey Research Inc,to construct the experimentalplatform.The system is mainly composed by Digiclops? IEEE-1394digital camera model,4.5meters,6pin, IEEE-1394cable,Digiclops?SDK and Triclops SDK,device drivers and IEEE-1394 OHCI PCI HostAdapter 3 port-400 Mbps card. Triclops Software Development Kitprovides real-time range images using stereo vision technology. Itallows users to accurately measure the distance to every valid pixel inan image.The SDK implements an MMX optimized version of the Sumof Absolute Difference Correlation algorithm providing the user withaccurate and fast depth map generation. Triclops delivers fullfield-of-view range images at speeds up to 30 Hz without sacrificingaccuracy through the use of subpixel stereo matching. When combinedwith the Digiclops Stereo Vision Camera System,the system can beused to the following fields: people tracking,gesture recognition,mobile robotics, mining, aerospace,object modeling and objectrecognition.Finally I wrote the programs about image grabbing,edge detectionand obtaining the disparity and depth map. |
PDF Full Text Request