Research On Stereo Matching Algorithm In Stereo Vision Ranging

Binocular stereo vision is an important branch of computer vision research, which generates the corresponding disparity map by capturing two images of the same scene with two cameras at different locations and matching the projected point of the same spatial point in the two images and extracts the depth information in the real scene according to the relationship of parallax and depth. Compared with conventional depth measurement, binocular stereo vision has two advantages including non-contact distance and passivity. These advantages make it used widely in the aerospace, industrial production, military confrontation and robot navigation and other fields.The main research of this paper is about binocular stereo vision ranging technology based on the principle of parallax, especially stereo matching algorithm as its key steps. Stereo matching algorithms can generally be divided into two categories: local and global stereo matching algorithm. In recent years, although the binocular stereo vision method made great progress, some issues remain to be better addressed. The theory and implementation techniques of camera calibration, stereo matching and depth extraction are discussed in this paper. The main work and innovations are as follows:1. In order to optimize the problem of insufficient matching accuracy at the edge in the local stereo matching algorithm, we propose a variable window in a local stereo matching algorithm on the basis of fixed window in it. The proposed algorithm uses gray information as a matching cost and the image edge information auxiliary to build dynamic matching window to weaken the adverse effects of the fixed window to get a better quality of initial disparity map.2. Aiming at the problem resulting from the blocked area between left and right images, the proposed algorithm detects the LRC occlusion through two-way matching disparity map obtaining from inputting images from left to right and right to left. Further, a blocked area filling algorithm is proposed to achieve the repairing process of mismatch in blocked area to improve the accuracy of matching occlusion region.Finally, we design experiments by using two identical USB camera binocular to build hardware and software platform, which completes image acquisition, camera calibration, image calibration, stereo matching and the extract of deep information extracted. The experimental and measurement result that the LRC variable occlusion detection window partial matching algorithm improve the accuracy of image matching block area, while improving the overall quality of matching disparity map, and extracting deep information more accurately, so it has some practical value.