Research On Mobile Robot Localization Based On Stereo Visual Odometry

Localization and mapping are important abilities for any robot to have if it wants to navigate intelligently in unknown environment. Localization is the basis for navigation of mobile robots. Recently, with the development of computer vision, vision based localization techniques of mobile robots are developing rapidly. With the large information, rich features and wide adaptive, vision can provide a good source of information to solve the localization problems of mobile robots. But the rich information also bring a lot of problems. So the algorithm should have a good real-time and considerable robustness.This paper focuses on key techniques of localization for mobile robots based on the visual odometer algorithm. The specific measures and steps of the algorithm are analyzed and researched in depth, and the necessary experiments are carried out to verify the correctness and effectiveness of algorithms.Firstly, the techniques of robot localization are introduced. And then this paper gives the summarize of the research significance and the state of the research at home and abroad. In plus, the related issues of binocular stereo vision are discussed, including: the research content; present situation; the basic principles; camera model and the principle of epipolar geometry.Thirdly, some common feature extraction algorithms are analyzed, SIFT feature detecting algorithm is focused on, matching tactics based on BBF nearest search approach are explored. In addition, some experiments are implemented to prove the robustness of the SIFT algorithm. Compared with Harris corner detection, the test result indicates that SIFT feature detection algorithm is more suitable in visual odometry system.Finally, this chapter mainly researches on the algorithm and the implement of robot localization based on visual odometry. An approach based on least median squares (LMedS) to a stereo-based visual odometry system is presented. Through the feature matching and feature tracking, the landmarks 3D information can be obtained, and then combined with LMedS for motion estimation in order to achieve the robot's position and orientation. The experiments demonstrate the algorithm can gain good results for localization.