Research On Image Matching Of Binocular Stereo Vision

With the development of the computer vision, the stereo vision especially binocular stereo Vvision is applied widely for many fields such as objection recognition, virtulity reality, industry examination, navigator of robotics, aviation and spaceflight. Stereo image matching is the one of key techniques in stereo vision. The development of stereo vision depends on the efficient solution of image matching.This thesis is mainly about the area based stereo matching. The theory and methodology of stereo matching are studied. The main factors of increasing veracity of stereo matching and decreasing complexity of stereo matching are then analzied. Rectification is a very important procedure to reduce the complexity of stereo matching. The rectification algorithm for binocular stereo vision is firstly studied. Finally, the software workbench of matching is developed in VC++6.0.During matching, area based stereo matching needs to get information from the neighbor region when the depth of a point is computed. The selection of supported window size and matching criteria are researched also. The combation of disparity consistency and matching cost are used to detect occlusion area. Fast and effective image segmentation is applied for extracting textureless and texture regions. The matching problem of textureless regions is then solved with constraint of disparity consistency and credibility concept. At the meantime, high credible seeds are obtained. Based on the smooth constraint of matching and disparity of seeds, dense disparity map is obtained via the adaptive-window based region growing. Complexity of matching is reduced by using seeds and transform of integral image.The method proposed in the paper is used to process the synthesized and the real stereo pairs. Experimental results have shown that the algorithm is efficiency and accuracy.