The Detection And Tracking Of Moving Objects Based On Binocular Stereo Vision

Binocular stereo vision is one of the important contents of research on computer vision.For obtaining images,the binocular stereo vision system directly uses two cameras to simulate human eyes,then extracts useful information,and ultimately gets 3d information of objects in the scene.The binocular stereo vision technology is very convenient and reliable,and has great application value and prospect,which currently is widely used in robot navigation,precision measurement of work piece,target recognition,virtual reality,scene reconstruction,and other fields.To overcome the common issues of the vulnerable to light,shadow and shelter,taking advantage of the binocular stereo vision,the paper try to combine the moving target detection and tracking technology to realize real-time and accurate target detection and tracking.This article focuses on the content as follows: the camera calibration,stereo matching,target detection based on binocular vision and moving target tracking.In camera calibration,four coordinate systems required and the corresponding relationship between each other are introduced,as well as the imaging principle of the pinhole camera model and nonlinear model.Detailed classification of camera calibration method is explained and the commonly used calibration methods: Tsai two-step,plane template calibration method by Zhang Zhengyou which is used in the experiments are detailed studied.After distortion correction,the accuracy camera internal and external parameters are obtained.In image stereo matching,this paper introduces the classification of the stereo matching method.Aiming at solving the existing problem of general methods,a stereo matching algorithm based on control point constraint and regional correlation is proposed,which can effectively improve the matching rate and the matching speed,therefore generating the dense parallax image.In target detection,the Mean Shift algorithm for clustering process is used,and then the threshold segmentation method based on Otsu is adopted to detect the moving object.In view of error detection caused by quasi-circle objects,a simple improved target detection method is designed.In target tracking stage,to solve the shortcomings of the Cam Shift algorithm,a new methodthat combines binocular stereo vision with Cam Shift tracking algorithm is proposed,meanwhile introduce the Kalman filter state prediction.The algorithm makes full use of the space position and velocity of target information,and overcomes the changes of environment light and similar color shade problem,improving the tracking accuracy and robustness evidently.