Research And Applycation On Target Tracking Based On Kernelized Correlation Filters

Video-based target tracking technology is used to obtain the trajectory of the target in the video and image sequences through computer technology.It is widely used in video surveillance,video compression,human-computer interaction system,intelligent transportation system and other fields.However,due to the complexity background environment and the target's appearance and shape changes during its moving,led to the goal to achieve accurate tracking is a very difficult thing.This paper focuses on some problems of moving target detection and tracking algorithm on target's scaling and occlusion.The main research contents are as follows:1,Research on single-target tracking.In this paper,The Kernelized Correlation Filter(KCF)tracking algorithm was investigated and an occlusion detection mechanism for determining whether the target being occluded by other objects was proposed.Kalman filter was used to predict the target's position once target was under occlusion.Experimental results show that the KCF tracking algorithm with occlusion detection mechanism was quite effective in processing the occlusion problem when the target was in linear motion.Then,the correlation filter and image scale Pyramid was used to estimate the scale of the target,a better solution to target scaling.2,Research on multi-target tracking.First,the depth images was used in the conventional moving target detection algorithm,better solved the interference of moving target's shadow.Then,this paper proposes a multi-target tracking algorithm combined with KCF tracking algorithm and moving target detection.when the target without mutual occlusion,the target area was obtained by the moving target detection,Then,face detected on target area,Kalman filter and data association techniques was used to generate multiple trajectory.Each target was tracked by KCF algorithm independently when multiple targets will be interactive.Experimental results show that multi-target tracking algorithm can better deal with occlusion problems between multiple targets.3.Finally,the application of the system based on KCF target tracking algorithm is designed and implemented,the application contains wechat network alarm video surveillance subsystems,Runner video analysis subsystem and people-counter subsystem.Operating results of the application verify the reliability of the target tracking algorithm presented in real life.