Stereo Matching Based On Segment And Research On Stereo Vision

With the unceasing development of the computer vision technology, the stereo vision sensor obtains widespread application. Especially, with the merits of simple organization and easy-operating, the binocular vision sensor is successfully used in the domains of industry examination, object recognition, work piece localization, robot self-leading, ceflight etc. The binocular vision sensor uses two vidicons to imitate and realize the function of human eye, and the mechanism is similar to human eye's visual mechanism.The paper researches the stereo vision technology, including image-segment. And the paper is focused on stereo-matching algorithm, which can be described as follow:(1) On applying stereo vision technology to stereo matching algorithm, the traditional region-based matching algorithm adopts regular window which cannot effectively dispose the joint of foreground and background, and trends to obtain over-smooth results. The method presented in this paper takes the image division region as the matching element. After divided, the foreground and background regions which have biggish gray difference will belong to different regions, therefore the method can solve the problem mentioned before.(2) For large occlusion region and texture flaw region, this paper analyzes the matching rate curve in entire disparity space, presents the concept of believability and neighborhood correlation factor. A region with low believability trends to take the biggest parallax value of neighborhood correlation factors. Because the believabilities of large occlusion region and texture flaw region are generally lower, according to the continuity restriction of Marr vision theory, it's reasonable to make the parallax value equal to the parallax value of its most reliable neighborhood. The experimental result has also proved that it is effective.(3) The paper realizes the watershed image division algorithm, and applies this algorithm to stereo matching.(4) The result of using standard test image to evaluate the stereo matching algorithm of the paper demonstrates that this algorithm can deal with large occlusion region and texture flaw region greatly.(5) Using OpenGL to Experiment the 3D reconstruction basing stereo match results.