The Research Of Stereo Matching And Moving Object Tracking Under Road Scene

As an important part of computer vision, binocular stereo vision is applied in moreand more fields in film, medical, game, surveillance and is receiving more and moreattention. In binocular stereo vision, stereo image matching is used for searching thecorresponding feature points. With the increase of image pixel and image acquisitionspeed, stereo matching rate drops greatly. Stereo matching technology is the key point tolimit the large-scale application of stereo vision. Aiming at the problem of stereomatching’s poor speed, this paper focus on this issue. The main research work in thispaper is:1. This paper discusses the structure of binocular stereo camera model, includingvarious physical quantities of the stereo cameras based on the model. Using Bouguet’sMatlab calibration toolbox to implement calibration process, and discusses therelationship between the depth and disparity. Based on the stereo camera model, we getparameters of stereo cameras, we rectify the stereo image to prepare for reducing thestereo match time.2. Comparing the advantages and disadvantages of global constraints stereomatching and local constraints stereo matching algorithm, we adopt local constraints ofstereo matching algorithm to match the stereo images’ corresponding feature points, thenusing sub pixel estimation and dynamic programming optimization to improve thematching result. According to the characteristics of the stereo matching, this paperadopts a tracking algorithm based on point matching to reduce the operation time.3. We made an animation using3DS MAX, and design an object tracking algorithm based on feature points matching tracking. We test the algorithm on the animation, theresult prove the validity and reliability of the algorithm.