Distributed Multi-camera Coordinated Moving Target Detection

With the popularity of video monitoring, how to realize the intelligent monitoring has become a hot topic. Intelligent video surveillance system includes target detection, target identification, target tracking and so on. Target detection is the foundation, and its detection results will affect the subsequent processing of other functions.In this paper, common target detection algorithm were studied and analyzed,for example, frame difference method, optical flow, single Gaussian background model, Gaussian mixture background model and kernel density algorithm. On the basis of single Gaussian background model algorithm, an innovative method to solve the poor adaptive problems has been proposed. The detection method combines single Gaussian background model, three frame difference method and motion picture history. Firstly, N frames are used to establish the initial model. Then frame difference method is adapted to get the background area. Secondly, background is divided into a large area of the stationary area, historic changing area and historic area of the changing one. Thirdly, the still large area and rapidly changing area are given fixed update rate and fast changing area is given the update rate of linear attenuation which is in accordance with the time distribution. Then, background model parameters are updated. Lastly, the moving target is obtained by using the background subtraction. Experiments show mat the improved algorithm has improved the adaptability of background model, and the detecting effect is better.In addition, we propose a cooperative distributed multi-camera video surveillance system. The system makes multiple camera images into a unified panorama by using the technology which includes the image matching and image matching. By the processing of target's disappearance and confirmation of the same target, the system is used for target detection and tracking via cooperation between multiple cameras. Finally, experiments show that the improved algorithm proposed is effective.