Research On Dense Matching Algorithm For Stereo Matching In Binocular Vision

Binocular stereo vision is a method of obtaining three-dimensional geometric information of objects based on the disparity principle. The height of a point is computed from the disparity of the corresponding matched pairs of pixels. Different postprocessing can be applied on objects' three-dimensional information based on different application to obtain desired outcome. Now stereo vision has been widely used in many fields and has been a research hotspot both in and abroad.The thesis focuses on the key and most difficult part of stereo vision:stereo matching. Aiming at the dense matching which is needed, an improved stereo matching algorithm based on graph cut is presented in this paper on basis of in-depth study on all kinds of matching algorithms. Vast computation is a great disadvantage of the existing graph cut based algorithms. In proposed algorithm, the reference image is divided into many connected regions using watershed segmentation algorithm and then the disparity range of each region is computed. When minimizing the energy function,α-expansion operation depends on whether a is in or out the disparity range of the relevant pixel. The large computing cost for traditional graph cut algorithms can be reduced and the matching is speeded up efficiently. The occlusion regions are marked in the matching and will be covered by the closest value, thus the disparity distribution of the entire image is obtained.The experimental results show that the accuracy of the proposed algorithm is high and it will take a shorter time to compute a preferable dense disparity mapping.