Research On Stereo Matching Technology Applied In Robot Vision

With the upgrading of electronic products and the continuous development of science and technology level, including the computer technology, since nearly half a century ago, the robot technology keeps on beginning to be mature and its application areas are continuously expanding; at the same time, the rapid developments of industrial automation technology and the growing demands of robot applications make the robot vision technology came into being. In recent years, the robot binocular vision technology has gradually become one of the leaders in the fields of robot vision research for its anthropomorphic vision functions, and this technology has been widely used in automatic navigation, virtual reality, unmanned driving, object perception,3D reconstruction, medical image processing and so on; the stereo matching technology as one of the key technologies of robot binocular stereo vision technology, has an important influence on the robot vision system. In this paper, we have made some researches on the stereo matching technology which applied in the robot vision, and the researches are mainly focus on repeated or mistake matching and other technical problems in the stereo matching process. So far, the main research contents and innovative research works have been completed in this paper are as follows:(1) We have collected some data, queried some literatures, understood the background and significance of the research, and mainly introduced the concepts, working principles, system components and other related contents of the robot binocular vision system from the model and schematic diagram of standard robot binocular vision system; then, reviewed the basic concepts of stereo matching technology and the common stereo matching algorithms, and determined the matching strategy adopted in this paper.(2) Simply introduced the proposed background and features of the Scale Invariant Feature Transform (hereinafter referred to as the SIFT algorithm or SIFT), then described the principle and implementation process of the SIFT algorithm in detail, the process of feature points extraction based on SIFT algorithm; and did some feature extraction experiments based on SIFT algorithm on the simulation platform of MATLAB R2012a. According to the experimental results, the feature extraction experiments based on SIFT algorithm proposed in this paper are successful on the whole and obtain the SIFT features used for the subsequent stereo matching successfully.(3) For there are some duplicate or incorrect matchings and other technical problems existing in the stereo matching in robot vision system, this paper proposes a stereo matching method of based on SIFT algorithm and the cosine similarity matching rule:the method regards the left image of the image pair as a primal matching image, another image is regarded as the matching image, then applies the stereo matching strategy established by the cosine similarity measure between the two SIFT feature vectors to make some stereo matching; and these are the innovative research works of this paper. Subsequently, did the improved SIFT stereo matching experiments based on cosine similarity on the simulation platform of MATLAB R2012a on a personal computer: reviewed the stereo matching strategy proposed in this paper firstly, described the stereo matching strategy based on the cosine similarity measure proposed in this paper; did the stereo matching experiments based on SIFT algorithm and cosine similarity matching rule according to the experimental results of the feature extraction experiments based on SIFT above mentioned, and then made the comparative experiments between the presented SIFT stereo matching algorithm and the SIFT stereo matching algorithm based on euclidean distance. According to the experimental results, the stereo matching experiments based on SIFT are successful overall, and realize the stereo matching based on SIFT successfully, and the experimental results are better than the SIFT stereo matching algorithm based on euclidean distance experimental results. In addition, we give the experimental design of SIFT algorithm realization based on FPGA.The experimental. results show that the proposed improved SIFT stereo matching method based on cosine similarity with a better matching effect, can get more successful matching point, with the strong robustness, effectively reduces the error or repeated matching ratios, and the technology will be more conducive to 3-D reconstruction and localization in the robot vision system.