Flexible Manipulator Motion Tracking Based On Binocular Vision Odometry

In this paper, we take a flexible manipulator as the research object and manipulator trajectory tracking as the research contents. To face the difficulty of building certain model of flexible structure system, this paper proposes a tracking method based on the binocular stereo camera visual odometry. This method uses binocular camera sampling images to detect SURF feature points and applies the RANSAC algorithm to estimate rigid motion equation parameters after feature matching and feature tracking. Compared with the traditional approaches for flexible manipulator motion trajectory detection, the method proposed in this paper makes use of visual information effectively to improve the precision of the trajectory tracking and to save the computation cost. Experimental results validate the higher precision and usability of the proposed method.This paper is organized as follows:(1) We first conduct the problem of the flexible manipulator trajectory tracking, analyses the significance of the flexible manipulator trajectory tracking, and then introduces the current research situation and difficulty of trajectory tracking. To solve difficulty, we propose visual odometry based on binocular stereo camera, and then discuss the current research status.(2) The paper studies three key process of visual odometry:extracting image feature point, matching feature points, solving rigid motion equation. For the current popular feature point extraction algorithms, the paper mainly studies the SURF feature point extraction algorithm and analyzes advantage of the SURF feature points relative to other feature point. This paper also analyzes the existing problems by using SURF basic matching algorithm and proposes new method of decomposition of the SURF descriptor vector and support-degree based association rules. The experimental simulation results have verified that, compared with the SURF algorithm, the image match correctness of the proposed algorithm is obviously improved. To solve rigid motion equation, this paper first uses least squares method, but it is low precision because of not ruling out the wrong matching point in sample collection. The paper combines RANSAC method to solve motion equation which firstly eliminate error matching point in sample collection and achieve overall minimum through the iteration.(3) Develop visual odometry software and design two groups of comparative tests in a real flexible manipulator experimental platform. Through comparing the experimental data calculated by visual odometry and obtained by standard equipment, the results validate the higher precision and usability of the proposed method.