Research On Local Stereo Matching Technologies Of Binocular Stereo Vision

Stereo matching technology,one of the important and difficult fields in computer vision research,is used to extract depth information from stereo image pairs.Local stereo matching algorithm has quick calculating speed,but the disparity map is easily effected by the window size.And it always can't find correct match point in the occlusive area or discontinuous area.In view of these problems,this paper has carried on the following three aspects of research and innovation work:1.This paper proposes a stereo matching algorithm based on steady-state matching probability and edge-aware disparity propagation,using the candidates of disparity to reconstruct matching costs which can remove the influence of the error disparity.It also uses the geodesic distance filter with the information of two directions to generate correct disparities and it need not to specify the matching window size.The simulation results show that it can effectively fill up holes in the disparity map and improve the matching precision of discontinuous area.2.An adaptive edge-aware disparity propagation algorithm is put forward to solve the problems of error disparity propagation.In order to make full use of local information and remove the mutual influence between different disparity planes,this algorithm uses the information of four directions around the pixel to filter and it also has constraints of the maximum window size and geodesic distance.The experimental results show that it can effectively eliminate the error propagation and remain more image details.3.In view of the high resolution and big disparity images,the disparity plane fitting method based on steady-state matching probability is put forward.The algorithm uses Census transformation,super pixel segmentation technology and two layers of MRF energy function to fit disparity plane.The experimental results show that the algorithm can correctly distinguish between two adjacent small objects and has very strong robustness to the images of different brightness and visibility.With using the Middlebury data sets to do experiment and evaluate,the results show that these three kinds of algorithms all can effectively improve the quality of disparity maps.