Multi-Camera Tracking Based On Multi-Agent System

Object tracking among multiple cameras,that is,using multi-camera cooperation to achieve the goal of monitoring and tracking process,is an important research direction of large-scale intelligent surveillance system.The multi-camera monitoring system faces many problems.First,because of the complex data to be processed,it is difficult to be real-time processing;second,the success rate of object matching is extremely related to as to the lighting conditions or the size of object and many other reasons.Therefore,how to use a distributed technology to cooperate the multi-camera,achieving the goal of intelligent tracking,is a main problem of multi-camera system.According to these problems,this paper has done the study about multi-camera system based on distributed multi-agent technology,combined with the computer vision,such as object recognition and so on.The main work is as follows:(1)The detection and tracking algorithm for single camera end is the base of the underlying vision algorithm for multi camera system.A pedestrian detection and tracking process for practical applications is proposed in this paper.The specific process of single camera detection and tracking is designed,including moving object extraction,shadow removal,pedestrian recognition and pedestrian tracking.A specific algorithm is combined for each link through comparative analysis.(2)A robust and fast pedestrian correlation algorithm among multi-camera is the key to ensure the continuous tracking of pedestrians.Therefore,this paper proposes a method of combining the roughly feature matching and accurately feature matching.The HSV feature and pHash feature are combined for roughly matching.The similar targets are matched accurately by ORB algorithm.It is proved that the algorithm can improve the efficiency of the system while improving the accuracy of pedestrian correlation.(3)The object tracking in the multi-camera system requires a real-time and efficient multi-camera collaborative algorithm.Based on multi-agent technology,this paper proposes an efficient interactive communication method for multi cameras,and designs a BDI model for tracking tasks in the single Agent level.Each Agent combines its own knowledge and external information to match the target state by quickly adjusting its own state in the current system.It is proved that the scheme can not only make the utilization of the system up to the highest,but also improve the target tracking rate.(4)This paper implements the multi-agent framework by Java based on JADE open source framework.The computer vision algorithms are realized by C language combined with OpenCV,which generates.dll files for Java terminal JNI calls.In this paper,a multi-camera object tracking system based on multi-agent system is implemented.The system realizes target tracking under three cameras in the experimental environment.