Research On Target Detection And Tracking Algorithm Based On Video Surveillance

In modern society,intelligent monitoring systems and video image processing technologies have been widely used in many scientific and social fields.Among them,moving target detection and tracking is the process of predicting and analyzing the target behavior in the video.However,due to the diversity of monitoring scenarios,different monitoring equipment,and the problems of rotation,occlusion,proportion change,illumination change,and dynamic and static backgrounds,the problem of target detection and target tracking in video has always been a difficult and hot topic in the industry.The target detection and tracking algorithm with strong robustness and accuracy still has a strong challenge.In view of the above problems,this paper starts with the classical algorithm,combines the existing technology and practical application scenarios,and studies the difficult issues such as the occlusion situation in the target detection and tracking algorithm.The main results are as follows:1.In the target detection module,this paper studies the classical dynamic target detection algorithm: frame difference method,background difference method,optical flow method,and based on this,studies a target detection algorithm suitable for both dynamic and static backgrounds..Experiments show that the proposed algorithm is suitable for both target extraction under dynamic and static backgrounds,and the target area obtained is more complete and clearer in contour.2.In the generation tracking,this paper studies the Kalman and Meanshift algorithms and optimization algorithms.In the case of short occlusion,Kalman's good prediction mechanism was added to the Meanshift iterative mechanism to achieve real-time prediction and tracking of the target position.Experiments show that the algorithm has good tracking effect in short occlusion scenes.3.For discriminant tracking,based on the Kernel Correlation Filtering(KCF)algorithm,this paper proposes a KCF optimization algorithm with a confidence judgment mechanism.By setting the confidence value R and the reference sample,it is determined whether there is occlusion in the current frame.In the occlusion case,the reference sample is used instead of the candidate sample to improve the target tracking problem under occlusion conditions.This paper selects a large number of video sets with occlusion properties to test the performance of the algorithm.Through OPE evaluation,compared with the KCF algorithm,the accuracy of the algorithm is improved by 7.1%,and the success rate is increased by 5.5%.