Fast Stereo Matching Algorithm Used In Stereo Video Coding Based On Video Plus Depth

Human eyes have the capability of spatial perception on the acquired scene,because they can extract the depth information in the scene. Current stereo visiontechnology can simulate the ability of human eyes system, and calculate the depthinformation via the differences appeared in two images from different cameras. Peoplehave high expectations of stereo vision technology market, such as3D film,stereoscopic video conferencing,3D medical imaging and3D satellite imagery. Stereovision technology is coming to our lives with high speed.Visual stereo matching which finds the corresponding positions in different imagesof a same goal is one of the fundamental and significant technologies of stereo vision.Parallax refers to the geometric difference between the corresponding positions. Inrecent years, stereo matching has become one of the international research hotspots andhas made great progress on algorithm improvements and platform building. The stereomatching algorithm is divided into global stereo matching algorithm and local stereomatching algorithm. And each of them has its own weakness. Local matching algorithmis generally hard to get high precision. Global algorithm can get a better match quality,but its high degree of complexity of the algorithm takes a long time and is difficult toachieve real-time computing. As3D technology continues to evolve, requirements tothe quality and efficiency of stereo matching are constantly improved.Large amount of data and less efficiency of stereo matching algorithm are far frommeeting the needs of real-time transmission. A fast stereo matching algorithm used instereo video coding based on video plus depth is proposed considering stereo matchingefficiency.As stereo video coding based on video plus depth just encodes reference viewpoint(left-viewpoint) and the parallax/depth map of left and right viewpoints, it can greatlyreduce the amount of data transmitted and be considered to a promising method ofstereo video coding.In this paper, stereo vision technology and stereo matching techniques wereintroduced. This paper is based on an algorithm in which color seglnentation and beliefpropagation are combined to solve the problem of each single method. The algorithm inthis paper utilizes motion information generated in the reference view coding ascompensation information in depth coding to perform disparity estimation of Frame I inthe stereo matching process. The information will be reorganized as the disparity estimation of Frame P for the subsequent iteration. Full use of the existing codinginformation reduces the complexity of disparity map generated in Frame P.In this paper, the algorithm is realized by hybrid coding of C and Matlab using themulti-view video sequences “breakdancers” and “book-sale” as experimental videosequences. The experimental results are analyzed from two aspects of the efficiency andquality of the algorithm. Experimental results demonstrate the proposed method candouble the speed of generating the disparity map without reducing the quality ofmatching and greatly improve the efficiency of coding and transmission for thestereoscopic video.