Research And Application On Stereo Matching Algorithm

Stereo matching obtains a dense disparity map by searching one-to-one correspondence between projection points in the images from different viewpoints of the same actual scene. And stereo matching is the core part of stereo vision. Although people do many years study for stereo matching, considering the influence of brightness, occlusion, textureless regions, depth discontinuities and other factors, it is hard to obtain a dense disparity map with high accuracy. So stereo matching is still a challenging work to study.In this paper, stereo matching relevant theoretical knowledge has been introduced, focused on stereo matching algorithm, and obstacle detection is the applied direction. The traditional stereo matching methods using belief propagation algorithm based on pixels have the deficiencies:high computation and matching errors easily induced by single pixel. To make up for the deficiencies, this paper studies Andreas and relevant algorithm, combines image segmentation and belief propagation, realizes a stereo matching algorithm based image segmentation and belief propagation. This algorithm includes five steps:image segmentation, initial disparity estimate, disparity plane fitting, extracting disparity plane template set and disparity plane template global optimization distribution. In the initial disparity estimate, this paper adopts the matching criterion combining SAD and gradient. At the same time, this step applies crosscheck to checkout disparity in order to detect the occluded region. For improving the accuracy of the disparity plane parameters, only reliable segment region does fit at the step of disparity plane fitting. In disparity plane template assignment, puts forward the concept of supporting point, increasing the weight of reliable pixels for choosing disparity plane template, rustling disparity plane template assigned to the segment regions more accurate. At last, does disparity plane template global optimization assignment by region based belief propagation algorithm. Experiments show that this algorithm has a strong processing capacity for textureless regions and occluded regions.According to the disparity map, this paper structures the U_V disparity map, extracts the straight line information in the U_V disparity map,and through these straight line information detects the potential obstacles.