Long-range outdoor localization of a mobile robot using a binocular camera

In this paper, an efficient stereo vision based visual odometry algorithm is proposed to solve the long-range outdoor localization problem of a mobile robot using a binocular camera. An improved method of feature matching and tracking based on SIFT algorithm is presented. The color information is used to effectively eliminate wrong feature matching, and the BBF tree is adopted to speed up the search process in the feature matching. The consistency of space position of feature points in previous and current frame is checked to filtering the wrong-matched points in the feature tracking algorithm. Hence the real-time performance and accuracy of the matching and tracking algorithm are improved. A hierarchical motion estimation method is also presented. Firstly the least squares principle combined with RANSAC filtering is employed to obtain the initial pose estimation. Secondly the two-stage bundle adjustment is used to optimize the motion estimation results. Furthermore Kalman filter is used to fuse the visual information with inertial navigation to improve the robustness and stability of overall position systems. Experimental results show the reliability and effectiveness of the proposed algorithm.