Real - Time Multi - Target Tracking System In Complex Environment

In the recent years, the number of surveillance cameras is increasing fast, but the cameras can’t really play a role, because the security guards can’t timely process all the surveillance video, so in most cases they have to look up the videos after the security incidents. So using the video analysis algorithm to detect and analyze the real-time videos, to handle the events timely become a pressing demand. Because of the above reasons, video analysis technique has become a research focus in recent years. The actual monitoring environment is very complex, such as illumination suddenly changing, branches swinging or in some dark environment the target is not so easy to discover and the shape, moving path and speed varies so much between the targets, so it’s very hard to predict. So to solve the surveillance problem with the video analysis algorithm is a very challenging task. At first, the alarm that has been given by the video analysis system must be very reliable. If the system always sends false alarm caused by the environment noise, no one wants to use the video analysis system. Second, the video analysis must can detect most of the security incidents, otherwise a lot of security incidents can’t be detected by the system will make the system fail. Third, the system must work very efficiently with few computing resources, otherwise, A lot of computing resource will cause the high cost, and affect video analysis system to be widely used.This paper presents a multi-target real-time tracking system working under complex environment, compared to the current published research in the field, to be able to more accurately track invasion target in the video surveillance scene, and the efficiency of the algorithm has been greatly improved. First, the dynamic background modeling in reference to the current the mainstream background modeling method, the codebook is selected as a base background modeling algorithm. This paper did a lot of improvement for the background updating method, that made the computation efficiency and foreground of the codebook detection accuracy relative to the original algorithm has been greatly improved, and the main performance is better than other mainstream background modeling algorithm. At the same time, this paper uses particle filter algorithm as the multi-target tracking method and made large improvement for resampling method that can maintain the effectiveness and diversity of the particles better than the original method.