Theoretical Study And Algorithm Optimization Of Binocular Stereo Matching

Binocular stereo vision is an important research subject of computer vision, which employs two different location cameras to shoot the same scene. Calculating the parallax of the corresponding points in two images of spatial point,3D coordinates of special point can be obtained by a series of inverse projection transformations. This is similar to the biological stereo vision. Binocular stereo matching is one of the most important problem in stereo vision. It affects the results of the final stereo matching and the objects’3D information in real scene. There are several challenge problems in stereo matching, such as illumination, occlusions and low textures. Stereo matching problems are affected by each other to cause distortion of stereo vision in the complex background of natural environment, especially in real scene.In order to find a solution to solve the problems of stereo matching this paper focuses on three problems in stereo matching, analyzes some kinds of stereo matching algorithms and proposes some reasonable and effective methods.(1) In order to solve the problem of illumination in binocular stereo matching, this paper proposes an algorithm which based on invariant moment to reduce the mismatching of disparity map.(2) The paper proposes an algorithm based on the occlusion. The algorithm is proposed which the tree structures of dynamic programming with self-adaptive parameters and a disparity iterative refinement to improve stereo matching performance. The algorithm model consists of an initialization model and a disparity iterative refinement. To obtain the initial disparity map with "clear" background and whole details, the tree structure of dynamic programming with self-adaptive parameters is presented. A disparity iterative refinement is proposed to estimate the disparity of unreliable pixels in the doubtful regions. The disparity iterative refinement consists of a doubtful regions growing method and a disparity filling based on color similarity.(3) This paper analyzes the reasons of error matching in low textures and proposes a stereo matching algorithm based on the edge maps. Firstly the edges of left and right images based on the sobel operator are extracted, then the cost functions of the corresponding points based on the color information and edge maps are calculated.The results of experiments combine with the Middlebury database and evaluation system show that this algorithms effectively reduce "fatting inflating" by occlusion and the error matching in low textures, calculate a high quality of disparity map.Finally, we build the system of binocular stereo vision. The results show that the binocular stereo vision device, utilizing the proposed methods in this paper, reduces the influence of illumination, has a higher precise experimental results in low textures and ranging precision.