Research On Stereo Matching Method Based On Dynamic Programming

Stereo matching is a classic problem in binocular stereo vision,which has been the focus and difficulty of research in the field of stereo vision.Domestic and foreign researchers have made a number of algorithms that attempt to solve the problems,which include the mismatching rate of the disparity discontinuities,occlusion, image edge and weak texture regions,as these problems itself has ill-conditioned, so far there is still no perfect solutions for these problems.At the same time how to improve the speed of the stereo matching algorithm in ensuring the accuracy of the premise to the algorithm is the main research stage of the stereo matching.In this paper,the dynamic programming of stereo matching algorithm has been in-depth study,and the mismatch of the streaking problems in weak texture,the image edge region and the disparity discontinuity in same depth region has been analyzed, our algorithm is improved in the traditional dynamic programming algorithm, the paper details are as follows:1. Firstly,the basic principle of binocular stereo vision and theoretical knowledge has been introduced,the current situation at domestic and foreign of research on stereo matching are described, Then the paper also summarizes the present general steps of stereo matching algorithm and general evaluation criteria, through the summary and classification of stereo matching algorithm, finally points out the difficulties and challenges of the current stereo matching.2. Due to the streaking problems in weak texture and the mismatching of the image edge region exist in the traditional dynamic programming algorithm of the stereo matching,this paper proposed a guided filtering based on tree structure dynamic programming stereo matching algorithm.First the matching cost is calculated based on the RGB color image, then the matching cost function to guided filtering, global optimization algorithm based on the parallax direction of the tree line dynamic programming algorithm, the optimal solution corresponding income disparity energy function for the initial disparity map.Finally, after the initial processing to get the final disparity map.Experiments show that the algorithm can effectively improve the streaking problem and the high-mismatch in image edge region problem with traditional dynamic programming.3. Specific to the problem of disparity discontinuous same depth region, we first consider the effect of uneven illumination or other noise in stereo matching,our new algorithm using the normalized moment invariants instead of the central pixel to calculate the stereo matching cost; Then the cost of a three-dimensional matrix of each match consisting of disparity as tree-based dynamic programming algorithm for guided filtering of the initial cost matrix to obtain the initial disparity map.At last is the disparity optimization to the initial map,a technology-based image restoration with a quasi-separation for solving quadratic variable minimum objective function after processing algorithms is introduced in the algorithm to obtain the final disparity map.Experiments show that the algorithm to improve the match rate in weak textures and edges of the image smoother while at the same depth in the area of the disparity, it means the disparity map is more accurate.