| With the rapid development of science and technology, applications of robotic technology are used more and more widely in the current modern society . As an important means of sensing the external environment information, Stereo Vision Technology in Robot are obtained more and more attention. The most abundant external environment information can be obtained through vision sensor, while it also results computational complexity problems in the later processing. By using the stereo vision system, and adopting the appropriate method of stereo matching, three-dimensional information of the outside scenery can be obtained, which has significant meanings to the robot's location, navigation, etc. This paper is focused on the matching problem in the binocular stereo vision technology, and the main work is as following: Firstly, the current domestic and approaches of the domestic and the external in the field of robot stereo vision matching are summarized and analysized, matching accuracy and real-time performance in this field are studied. The mobile robot platform used in this paper—the MT-R hardware structure and Bumblebee2 binocular stereo vision system which are given necessary instructions.Secondly, the types of robot stereo vision technology and relative technologies are illustrated. Focusing on researching the binocular stereo vision feature extraction and matching technology, the feature point extraction of image rotation, scaling, the affine and so on which possesses invariant SIFT and matching algorithms are investigated and the procedure are designed. Through the experiment, the benchmark scale space factor, sampling order and sampling layer in the algorithm which can affect the feature extraction and matching effect are analyzed. According to the experiment results, the appropriate parameters which balanced the number of feature extraction and real-performance are determined under the current experiment environment.Finally, the SURF algorithm which include the same three parameters is researched. Procedures are written according to the SURF algorithm principle and feature points of the binocular image are obtained. Combining with the theory of nearest neighbor method and comparing with the SIFT, the experiment results prove the program get the better real-time performance ,and this method is more appropriate for the location and navigation of the MT-R. |
