Research On Disparity Extraction Algorithm Based On Binocular Stereo Matching

Stereo vision is one of the focus and hotspots in the field of computer vision because it is closer to the people’s feelings. It is widely used in areas such as military, medical, industrial, and has broad promise application vision and high commercial value. Using stereo matching techniques to estimate disparity is the core issue in stereo vision. Given two images of the same scene from two different viewpoints, the goal of stereo matching is to find the two corresponding pixels and the difference in the positions of their, then, compute the disparity map for the reference image. Its accuracy directly determines the effect of the subsequent three-dimensional reconstruction. Thus, the disparity extracted algorithm based stereo matching technology has important academic value and practical significance.In this paper, disparity extracted algorithm based stereo matching is researched. Firstly, stereo matching algorithm’s research and development status were introduced; the problems stereo matching facing and the aim of this article were illustrated. Subsequently, some of the basic principles and concepts of the stereo matching techniques were also introduced; the commonly used algorithms of stereo matching include region based matching, dynamic programming, graph cuts and belief propagation were discussed. On this basis, we propose the improved method of the region based matching and dynamic programming algorithm.For the selection problem to support region and matching cost of the region based matching algorithm, the article mainly spreads that the WLD (Weber Local Descriptor) texture characteristic can be applied to stereo matching. We use the matching cost which include WLD texture characteristic for constraints, and reliable disparity was also screened by WLD texture characteristic. Support region was determined by the line segmentations for stereo matching.In order to overcome the stripe-like defects in stereo matching algorithm based dynamic programming, we propose optimizing in both rows and columns directions many times for suppress the stripe-like. Firstly, we segment the image by mean-shift algorithm and then use the result to improve the energy function. On this basis, final disparity was figure out by optimizing in both rows and columns directions.Comparing with some traditional algorithms, the effect of the proposed algorithm has significantly improved, which is verified by the experiments.In the final of this paper, the software for disparity extract was introduced. Some commonly used algorithms and the improved algorithm is integrated into the software, together with the visual interface and instructions to facilitate the works of future research.