关键词: 机器人视觉, 立体匹配, 尺度不变特征变换（SIFT）, 余弦相似度

Abstract: Aiming at the problem of repeated or error stereo matching in the robot vision system, this paper presents a stereo matching method based on Scale Invariant Feature Transform（SIFT algorithm for short） and cosine similarity matching rules. The method regards the image which has more feature vector in the left and right images as a reference image, the other image as the matching image; then makes stereo matching with the matching rules which are designed by the cosine similarity between the feature vectors of the two matching images. Experimental results show that the improved stereo matching method can effectively reduce the matching error or repeat ratio, and has strong robustness, better matching effect, more conducive to three-dimensional reconstruction and localization of the robot vision system.

Key words: robot vision, stereo matching, Scale Invariant Feature Transform（SIFT）, cosine similarity