Research On Human Traffic Statistics System Based On Video Surveillance

In the field of computer vision, in recent years, pedestrian flow statistical technique is the focus research. It provides more effective management by using pedestrian flow statistical for public place. In pedestrian statistical technique, human detection and tracking and counting algorithm is the core of the system. This paper obtains video image through camera which is fixed and has a certain angle in both vertical and horizontal directions in order to detect the human movement targets, track the human movement targets and count the human movement targets.In the matter of moving target detection, we study some common moving object detection algorithms: interframe difference method and background reduction method, and analyze the advantages and disadvantages and their applicable places of the two methods.A moving pedestrian detection algorithm based on combining the symmetrical frame difference and background subtraction method is developed in this paper. Mixture gause model is used for updated background. In the matter of pedestrian detection, This paper mainly introduces the advantages of the features of head in the pedestrian target detection based on different scenes. In the matter of head detection, we choose the Haar feature based on Adaboost classifier to detect the head of pedestrian. Firstly, we introduce the Haar feature and integral image and the eigenvalues can be calculated. Then we introduce the methods of training and detection of Adaboost classifier in detail, then we improve the detection method. We propose the idea of contour extraction based on connected regions in the modular of head detection. The cascaded classifier is loaded to make a multi-scale detection in the current frame’s foreground region. By this way, the detection accuracy and speed of the head are highly improved. In the matter of pedestrian tracking, we study some common moving object tracking algorithms and analyze the advantages and disadvantages of the methods. According to the problem of pedestrian shelter, this paper developed the method of combining Meanshift and Kalman for tracking the pedestrians. When the pedestrian flow is high and the pedestrians are severely sheltered, basing on the method of the multi feature matching is propsed for pedestrian flow statistical.In order to verify the effectiveness and real-timeness of the proposed method, a large number of different scenarios and different number of videos are tested, results show that the proposed algorithm has certain anti- interference ability and is accurate, real-time,effective for pedestrian flow statistics.