Research On Target Tracking Algorithm Based On Multi-camera Collaboration

The multi-camera tracking and monitoring system can monitor the instruction targets all day long in the unattended outdoor natural environments,and perform computer vision processing on the video sequences of the targets it monitors to detect and track moving targets.Under the premise of limited environment,by properly arranging the camera positions,it can not only ensure the advantages of a wide field of view and comprehensive monitoring of the monitoring network,but also improve the tracking accuracy of the system and reduce the frequency of tracking failures in military,environment,family life and other The commercial field has a wide range of applications.First of all,for the video target tracking under the camera,the frame difference method is used to detect moving targets in the video sequence.The Kalman Filter(KF)algorithm and Extended Kalman Filter(EKF)are proposed respectively The algorithm tracks the target and compares and analyzes the two algorithms through MATLAB software simulation.It is concluded that the EKF algorithm has higher accuracy than the KF algorithm.At the same time,an improvement is proposed to reduce the tracking accuracy caused by occlusion during target tracking.Combining EKF with Mean Shift algorithm,it is concluded that the method is highly robust to target occlusion.Secondly,in order to solve the problem of coordinated scheduling among multiple cameras,according to its limited field of view and unstable target motion,this thesis uses a multi-camera monitoring system with overlapping views.In order to achieve the target handover between cameras,based on the association between the camera coordinate systems,a method of target handover based on homography transformation is proposed.For this method,a scale invariant feature transform(SIFT)algorithm and a speeded up robust feature(SURF)algorithm are proposed to perform feature point matching.Simulation analysis shows that SURF algorithm has higher matching accuracy than SIFT algorithm.Then this thesis further optimizes based on the SURFalgorithm,and combines the Random Sampling Consistency Algorithm(RANSAC)to automatically delete the mismatched point pairs,which greatly increases the accuracy of the matching and laid a good foundation for realization of the target handover.This thesis uses this improved method to simulate the process of image fusion and target handover,which has better stability and robustness than single camera.