A Feature Fast Matching Technique Using Improved Self-Organizing Map for Stereo Vision

Landmark feature match method of robot visual navigation has been studied, An unsupervised learning methods based on binocular vision has been proposed to match Landmark feature fast. Firstly, environmental Information images are obtained by binocular vision, Then SIFT feature vectors are abstracted from binocular vision images, Lastly, Self-Organizing Map is pulled in to match multidimensional feature point fast by using competitive learning methods. Experiments showed that the proposed methods on feature matching with better computation time and effect than the traditional SIFT and SURF of feature matching methods. And it can satisfied the requirement of real-time.