Study On Image Matching In The Field Of Stereo Vision Based On Line Feature

Stereo matching is one of the prominent problems in the field of stereo vision. According to the difference of matching primitives, stereo matching approaches are divided into three categories: area-based matching method, phase-based matching method and feature-based matching method. The first one bases on gray information of local area and the second is built upon supposition of partial structure, which is difficult to improve its stability and accuracy. Due to the stability of feature and relative abundance of information, the third one becomes the hotspot of the research.Macro-optimization of the whole stereo vision system is critical to the achievement of 3D information from spatial scene on the purpose of stereo vision. Linear feature can provide more information than dot feature and can be extracted and described conveniently compared with the planar structure .Thus a new matching method of stereo vision based on linear feature is brought forward. This method is easy to optimize calibration and matching unitedly.In this paper, the principle of biocular stereo vision is analyzed, and the key technologies of the stereo matching based on linear feature are studied. At first, an improved method of two-step camera calibration is put forward, which acquires distortion and undistortion parameters respectively and eliminates the control points of large error to iterate the calibration process. By using the method, the iterative speed is accelerated and the precision of calibration parameters is improved. Second, to gain the straight lines made of sequential single pixel width, a line extraction algorithm of phase-based grouping and heuristic searching is used. Then with epipolar and disparity continuity constraints, a stereo matching strategy based on corners and geometric and gray features of linear segments is presented, which realized united optimization of matching results and parameter calibration. Finally by using this approach mentioned in the paper, fine matching results and accuracy calibration parameters can be gotten, and all the results show that this method is feasible.