Research On Analysis Technique For Binocular Stereo Vision

Binocular stereo vision is one of the most active branches of computer vision research, it is also the important symbol of the aptitude robots development. It can be defined as the process which based on the different locations'vision system composed of dual CCD cameral or personal computer's moving or rotating the same scene shot, through image acquisition,camera calibration,feature extraction and matching,space points and then by calculating the disparity in the two images,to obtain the scene depth information:This paper is on the basis of a lot of reading related literature and analysis on related technologies of computer vision, focusing on the debate of feature point's extraction and matching. And it also conducts the following two parts as the key research:Firstly, a CCD calibrating approach in the midst of traditional and self-calibrating methods is advanced to separate the inner and the outer parameters. Based on perspective model, it was proposed that drawing camera inner parameters with physics method, The outer parameter model of two CCD can be constructed by selecting appropriate reference frame and analyzing experiment data. The method is simple, speediness, understandability, and practicality. Secondly, compare and introduce a number of the corners detection algorithm, and proposes threshold selection method base on the SUSAN corner detection which based on the average gray scale image, in order to make the threshold's selection not aimless any more,and also guarantees the practicality and applicability when using SUSAN algorithm to corner detection. Finally, the matching algorithm in this paper on the basics of feature points as the basic primitive, integrate with regional normalized gray correlation matching method,and use the strategy of "two-directions match consensus" and "maximum priority "to get the best match effect.A large number of experiments'results indicate that it can reduce the time of extraction of feature points and matching, improve the matching accuracy, obtain a good result.