Research On Binocular Vision Stereo Matching Algorithm

Stereo matching is one of the research hotspots in current computer vision,and it is also the most important and most difficult step in machine vision.Stereo matching is the fundamental problem in artificial intelligence and pattern recognition which is to match the corresponding points in different scenes;Due to the existence of the factors such as occlusion and noise,it has been thought as a morbid problem.In recent years, many researchers tried to solve this problem by introducing different algorithms,and they got very good results.In this thesis,we studied the existing algorithms on stereo matching,and proposed some improved algorithms,evaluated the results objectively. We list our researches as follows:(1) The common stereo matching algorithms is analyzed,including algorithm based on dynamic programming,algorithm based on belief programming,algorithm based on graph cut and algorithm based on image segmentation,and their virtues and defects is summarized either.(2) A stereo matching algorithm based on dynamic programming and mean shift image segmentation is proposed.Experimental results show that this algorithm improves the precision,especially in the depth discontinuity region.(3) A stereo matching algorithm based on graph cut and image segmentation is proposed,firstly we construct a new energy function,integrate the information of image segmentation,then construct graph based on the new energy function,find the min-cut of the graph usingα-expansion iteratively.Experimental results show that this algorithm improves the precision efficiently.(4) Aimed at the large computational cost and bad real-time performance of current binocular vision stereo matching algorithms,a fast stereo matching algorithm is proposed.Experimental results show that the computational speed has been improved for one order of magnitude while the percentage of bad matching pixels is equivalent to commonly used area-base matching algorithm.Besides,the edge feature is better, thus this method is an effective and feasible stereo matching algorithm with good real-time performance.