Research And Implementation Of Multi-Camera Tracking Algorithm

With the rapid development of computer vision,people's requirement of the intelligent video surveillance becomes higher.Because of the limitations of monitoring scope and the flexibility of special target,the traditional fixed single camera has gradually cannot meet the needs of the many occasions of monitoring.The multi-camera monitoring system can Make up the lack of ingle camera,such as the limitations of monitoring scope,the blocking of the objects,continuity tracking of objects and 3-d calibration.The purpose of this thesis is to build a multi-camera monitoring system which not only can monitor a larger scene but also can complete the active tracking of special target,so two goals can be achieved at the same time,monitoring more wide and monitoring more clear.Monitoring more wide is completed by cooperative tracking based on several cameras which have Overlap vision,monitoring clearer is completed by the intelligent control system based on a PTZ camera.The main research content of this thesis is as follows:(1)An improved Vibe algorithm is proposed to solve ghost problem and shadow problem.To remove the ghost of Vibe algorithm,a method which combines method of frame difference and Vibe algorithm is presented.To solve the shadow problem of Vibe,the gray component is replaced by S component which belong to HSV space to be used to build model.When the moving target detection is completed,a method which combines Kalman filter and blobs detected by improved Vibe algorithm is presented to complete Moving target tracking.(2)An algorithm for target handover is proposed for the cooperative tracking based on several cameras which have Overlap vision,it uses both of Coordinates features and Color features into consideration.Firstly,a more accurate homography matrix is got by a method which combined SIFT algorithm and ROI(area-of-interest),just keep the point pairs which belongs to ROI,and then the homography matrix is calculated by RANSAC algorithm and the remaining point pairs.When there is a goal in the Overlap vision of a camera,the goal's Coordinate range in another camera coordinates is calculated,if there are multiple goals in the Coordinate range,the further confirmation is got by color histogram matching to avoid mistake.(3)A PTZ is integrated into the monitoring static network for Close-up.Firstly,the monitoring scene of the monitoring network should be split into multiple grids.Secondly,PTZ can be matched with the monitoring network by its preset,and then PTZ can find the special target in the monitoring network anytime.Lastly,PTZ can complete the active tracking of special target by Camshift algorithm,the intelligent control algorithm on PTZ can ensure that the special target is in the middle of the view from beginning to end.