Research On Stereo Matching Algorithm Based On Binocular Vision

Vision is one of an important means of understanding the world,stereo vision is a research on how to simulate human eyes on the world’s understanding of the subject.Observation of the same scene in two different perspectives,image based on geometric principle of different angles,the parallax,reducing the3D information of scene.Stereo matching is the most difficult and the most important research content in stereo vision,is the core technology of the three-dimensional reconstruction.This paper makes a thorough study of the stereo matching problem in stereo vision.introduces the theoretical basis of stereo vision,and the four elements:stereo matching of feature space,similarity measurement,search strategy and search space are analyzed,at the same time classify the matching algorithm:region matching,feature matching.In region matching algorithms new similarity measurement function is introduced,the different gradient of disparity with different size of window matching,reduce the false match rate.In the feature matching algorithm.due to the false matching and the changes of illumination of the classical SIFT feature matching algorithm,proposes an improved global information into the descriptor,improves accuracy and stability of matching.Finally,analyze the advantages and disadvantages of the two algorithms,based on the complementages strategy.an new matching algorithm is proposed,combining region and feature algorithm,the first step is the edge extraction,based on disparity gradient, the edge of the feature points are extracted by the improved SIFT operator,then the feature points matching,and then using the improved region matching algorithm,this can reduce the search range of region matching.higher speed,and can not only obtain a dense disparity map,but also improve the correct rate.The experimental results show that the correct rate of the new matching algorithm is higher,and the performance is also relatively stable.