Study On Motion Trajectory Of Object Based On Binocular Vision

Binocular stereo is an important branch of computer vision, which collects images with two cameras and rebuilds3D information of scene based on the geometric relationship contained. It is used more and more widely in such fields as military, industry and livelihood. This paper is based on the binocular vision technology to expand the study of the trajectory of object, including the camera calibration, the moving target detection and tracking, stereo matching, and moving object three-dimensional trajectory generation and so on. Among which it focuses on the detection and tracking moving object, and stereo matching algorithm. The detailed information is listed as follow:Firstly, on calibration of the camera, this thesis discussed the classification of camera calibration methods and the several commonly used coordinate system, as well as analyzed the linear and nonlinear model of the camera. It detailed elaborated the Zhang Zhengyou checkerboard calibration method for this article which is between the traditional calibration and self-calibration. Matlab toolbox was also used to complete the camera calibration and to obtain the corresponding internal and external parameters of camera, which laid the foundation for calculate3D coordinate points.Secondly, in terms of the detection and tracking of moving objects, this thesis introduced the basic principle of background difference method and the several common modeling methods. By comparing these methods, it selected background modeling method based on statistical average used in the paper. What’s more, it studied the CamShift tracking algorithm in deeply and compared the advantages and disadvantages of the algorithm. For the need of manually choosing tracking target and with the problem of losing target because of similar background colors, this paper combined with CamShift algorithm, the background difference method, which is based on the statistical average modeling proved that the improved algorithm could improve some defects existing in the traditional CamShift algorithm.Next, this paper dealt with the problem of the stereo matching for the target objects when two cameras were used at the same time. Based on the actual research, this thesis used the centroid of the object as feature points which avoided the complex feature extraction. However, the way of taking the center of mass points as match points would produce a lot of mismatching due to noise. In this situation, this paper proposed a method of the contour matching combined with the centroid matching, which used contours for constraints. Experiments proved its effectiveness and it could eliminate many mismatching points.At last, we used the least square method to solve the three-dimensional coordinates of the centroid, by combining the internal and external parameters, as well as the have matched centroid pixel coordinates. This paper successfully emulated the3D trajectory of moving objects. In order to verify the precision of the system, this thesis collected the checkerboard images under different distance and cleared deviations by comparing the experiment results with the known corner distance on the checkerboard. The experiments proved that within a certain distance, this system enjoys great precision and decent practical value.