Autonomous Mobile Robot Visual Tracking Measurement And Control Technology Research

Mobile robot is widely used in underwater, space, nuclear industry and military,because it can enter the dangerous environment, where human beings are hard to reach.Under the unmanned control circumstances, the self-vision tracking measurement andcontrol technology of mobile robot can perceive the surrounding environment, determinethe target’s position in space, extract and match target characteristic and realize targettracking by machine vision, so as to improve the mobile robot’s self-organized andself-adapted navigation system.Because of target screened, deformation, light intensity and background interference,the stability and real-time performance of target tracking is very poor. This has become theresearch focus among many scholars.This paper focuses on binocular stereo vision system. Based on parallax and otherprinciples, we gradually achieve: camera calibration, feature points detection andextracting, image’s stereo matching and3D reconstruction. At last, we succeed in trackingthe moving object. The mainly research work of this paper includes:(1) We adopt the Zhengyou Zhang’s Chessboard camera calibration method, which isfamous and widely used, to obtain the intrinsic and external parameters, distortioncoefficient, rotational vector and translational vector through the calibration experiment.(2) According to the situation of there are many pseudo angle points when we usingthe Harris algorithm to extract the angle point of3D object, this paper puts forward animproved H-S corner detection algorithm, which using neighborhood pixels differencemethod and SUSAN algorithm. Those pseudo angle points are eliminated by the selectionof double threshold value. In comparison with the other two algorithms, the experimentalresult shows that this improved algorithm performs well and acquires higher accuracy oncorner detection. In addition, we also research the pyramid L-K optical flow algorithm, andrealize in tracking strong angle points of two consecutive frames.(3) As for the stereo matching of images, we accomplish SIFT feature points matchingexperiment between right and left pictures, and acquire the2D coordinates of eachmatching conjugate points. With these coordinates, we can get the deepness informationafter3D reconstruction based on parallax principle. Finally, we achieve successive trackingof the moving object with the method of SIFT matching and part feature block matching. Furthermore, we also evaluate and analyze the error results.