Moving Target Detection And Tracking Algorithm Research Based On Binocular Vision

Moving target detection and tracking is widely used in many fields,such as military guidance,intelligent transportation,video monitoring and so on.It's a major issue on which the researchers will continue to be devoted.But at present,thevastmajorityof the research on detection and tracking algorithm are based on monocular vision,inevitably influenced by light changes,shadows andobscured properties.Binocular stereovision has the advantage over monocular vision of acquiring deep information of the target,and takes a very important part in computer research area.In this paper,binocular vision technology is applied to the detection and tracking on moving targetsin case oflight changing,shadows andobscured properties.The main researches are as follows:Binocular vision camera calibration is researched.Firstly,three commonly used camera calibration methods are analyzed,the transformationalrelation between usual coordinates and each coordinate of binocular stereo vision is established.Secondly,the radial distortion processing is discussed.Thirdly,the homograph matrix and parameters of the Binocular vision system is calculatedby checkerboard calibration method,the internal and external parameters of the cameraare given.The stereo matching algorithm is studied.A region matching algorithmhas been improved on the base of disparity gradient,showing high matching accuracy,fast matching speed and strong noise resistance.The improved similarity measure function based on the sum of squared differences(SSD)is put forward to obtainan accurateparallax value together with disparity gradient constraint to determine the edge or smooth area of the image,and then different window sizes is used to reduce the search space.Moving target detection algorithm based on the binocular vision is researched.Study on symmetrical frame difference and background differenceshows that background difference has better effect,but it still cannot deal with light changes,shadows andobscured properties.Therefore,do “AND”operationbetween the above result and binocular detection results can get good effect.The combination of binocular detection and monocularon moving detection not only solved the problem brought about monocular detection,but also overcomes the disadvantages ofbinocular detection.This paper proposesatarget tracking method combined with binocular vision and target tracking of Camshift.Firstly,moving target is detected byusing the parallax background difference,and afterbackground removingit is tracked throughthe improved Camshift algorithm.In order to overcome the disturbance of light changes and similar color moving targets,kalmanfilteralgorithm isintroduced toforecast the position and velocity of moving targets,then different scale factor is usedto linearly combine the results of Camshift and that of kalman filter algorithm,the trackingreliability can beenhanced.Experimental resultsshows that the proposed detection and tracking method can keep good robustness in the case of complicatedscene,multi-targets or object occlusion.