关键词: 移动机器人, 双目视觉, 特征点, 定位

Abstract: Aimed at the problem of mobile robot localization in indoor environment, a binocular vision method of SLAM is researched, which can be applied to the incomprehensive and complicated environment with dense feature points. Based on EKF-SLAM method, a sparse control mechanism of feature points, which is characterized by restricting the distributed-density of the feature points on the global map in terms of the position space and the feature space, is proposed and introduced, thereby enormous computation cost caused by the fast increase in the number of feature points is avoided, and the data relation accuracy get improved. Moreover, the dimensions of SIFT feature vector is decreased to reduce computational complexity, and multiple conditional constraints is added into the process of matching such feature points in the left and right images to enhance the precision of the stereo matching. The experimental results indicate that, utilizing the methods proposed, the distributed-uniformity of the feature points can be realized and the cost of localization time is condensed on the premise of satisfying the requirements of localization accuracy.

Key words: mobile robot, binocular vision, feature points, localization