Matching Algorithm Based On Binocular Stereo Vision Digital Image Area

Along with the rapid development of computer hardware technology and thehigh performance digital processing system being popular, stereo vision ushered in anew wave of research. Stereo vision is an important branch of the machine visionwith very high research value and broad application prospect. The most commonfields of stereo vision are horizontal binocular stereo vision and binocular stereomatching technology. Stereo matching algorithm is usually divided into twocategories. One is a region based stereo matching method and the other is calledfeature points oriented matching method. Region based stereo matching algorithm isto obtain dense disparity map, and feature based stereo matching algorithm forsparse disparity map. This paper only studies the binocular stereo matchingalgorithm based on region area methods.Generally, region based stereo matching algorithm can be divided into foursteps：the design and calculation of the cost function, matching area cost aggregation,disparity space for the cost function the optimal solution (local optimal or globaloptimal) and correction of the disparity map. This4step as the research order andclues, are studied as well as the ultimate form of the general understanding of theregional matching algorithm.Firstly, a comprehensive understanding of the camera imaging model, principleof in-depth understanding of digital imaging of bionic eye system are obtained. Onthis basis, research and elaborate the direction and current status of the research ofstereo vision system, by using mathematical tools on binocular stereo matchingprinciple and its important role in stereo vision. Secondly, in-depth understandingand introduction of binocular stereo matching principle are done and a series ofpathological classical matching algorithms are faced, how to quickly and accuratelysolve these problems become the main purpose of stereo matching algorithm. Then,in-depth study of the matching algorithm based on local region matching method.Research on the general steps of matching algorithm in local area has beencompleted. And matching similarity measure function (also known as the matching cost function) as one of the core factors has been made as a summary and analysis.The light gray level is not sensitive to the change of the matching cost function usingthe NCC filter box technology (Normalized Cross-Correlation) algorithm tosignificantly reduce the classical matching algorithm time complexity. And putforward a kind of derivative function improved attenuation of the matching costfunction to achieve the algorithm of robust noise performance improvement, and itsapplication in a local region matching algorithm based on adaptive window. Then,in-depth study of matching algorithm based on global region is done. The generalmodel is summarized and described the global matching algorithm and using thebasic idea of the global optimum. A method is proposed to obtain the contourauxiliary based on the wavelet transform into double times dynamic programmingstereo matching algorithm. Finally, summarize the general steps of correctionalgorithm for region matching methods as the last step. A kind of disparity meanvector drift correction algorithm is proposed to realize the optimization techniquebased on the results of matching.To sum up, in this paper, the general steps of region matching algorithm are thelead of the study. Several new methods and algorithms in the following matchingfields are presented: matching algorithms based on local area, matching algorithmsbased on global area and matching correction algorithms.