Research On Stereo Matching Algorith Based On Dynamic Programming And Belief Propagation

Binocular vision is the most important distance sensing technology in the field of computer vision research. It can flexibly measure three-dimensional scene information in a variety of conditions to achieve the passive measurement. Other type measurements are no substitute for binocular vision. Stereo matching is the most important and most difficult part in binocular vision. So stereo matching algorithms based on dynamic programming and belief propagation were studied in this article deeply.First, the history and current status of stereo matching algorithms.was introduced, then the deficiencies of traditional dynamic programming algorithm and the belief propagation algorithm were analysed and directions for algorithms improvements were pointed out. Basic theories of stereo matching, including the basic stereo matching principle, the constraints of stereo matching, the classification of stereo matching and the basic steps of stereo matching were also introduced to obtain the necessary theoretical support.Secondly, stereo matching algorithm based on the traditional theory of dynamic programming was discussed in detail. To improve the traditional one-way algorithm optimization strategy, a new stereo matching algorithm based on two-way dynamic programming was proposed. the matching results of the column direction were guided by the results of line direction. our method can effectively restrain the disparity stripes, reduce the false match rate.Again, stereo matching algorithm based on the traditional theory of belief propagation was discussed in detail. The global energy function of the traditional belief propagation algorithm can not effectively balance data items and energy items, two adjustment coefficients were introduced to adjust the energy function, which effectively could inhibit the phenomenon of parallax empty and reduce the false matching rate.Finally, a new matching strategy was proposed.to improve the traditional belief propagation algorithm which had disadvantages in terms of real time matching. Images were segmented into different regions by the mean-shift algorithm, according to the parallax template, partitions were defined, Dense disparity images were generated by minimized the global energy function. The adopted method improved the matching speed and reduced the false match rate.