Study Of Stereo Correspondence Alogrithms Optimization In Binocular Vision

Binocular stereo vision is an important branch of computer vision, and its basic principle is how human feel distance of objects: the projection of scene in our two eyes are different, it has what we call disparity. We get the depth information according to the disparity. As a powerful technology of environmental awareness, it is widely used in autonomous driving, UAV autonomous navigation, unmanned lunar autonomous land vehicle. Stereo correspondence is the most important and difficult technology in stereo vision. It directly affects the precision of 3-D reconstruction, recognition, scene description, image interpretation and so on. In this paper, we do some related research in stereo correspondence, and puts forward a fast adaptive weight correspondence algorithm.Firstly, we conducted a binocular stereo vision system for work, and completed parameter calibration, stereo calibration of the system. Secondly, we studied the stereo correspondence algorithm. Stereo correspondence is to find matching points in two images, and then calculate the disparity. According to the disparity we can get the depth information. Most of the stereo correspondence algorithm proposed is unable to give consideration to both real-time and matching precision. Thus, we put forwards our correspondence algorithm. To evaluate our algorithm, we use three standard image(Tsukuba, Cone and Teddy) on Middleburry website to get their disparity images with our algorithm. Compared with the SAD algorithm, our method perform much better in the discontinued disparity areas.At last, we give out the error percent of our algorithm. Compared with other mainstream algorithm, the accuracy of our algorithm can reach their levels or even better, but the operation time is much less than theirs.