Research And Simulated Implementation Of People Statistics System Based On Video Image Processing

Along with the development of video image processing technology, the intelligent video surveillance has become a very active new field. And people statistics as an important application direction has been extended to a lot of fields. For instance, the work time and work quota for police can be well arranged by people statistics for the intersections with a busy crowd flow. How to realize people statistics accurately and efficiently in a particular scene is becoming a significant research subject.In order to make the application of people statistics algorithm more general, the paper does not specify a particular scene and just assumes the scene with a monocular camera fixed on the ceiling. Thus, for the scene such as office buildings, squares, and even buses, etc., people statistics can be realized based on the algorithm mentioned in this paper and on the proper processing for the specific scene.According to the three key links--detection, tracking and counting, the algorithm designed in the paper is as follows:The first step is moving body detection. First, motion region information is obtained by frame difference method and edge difference method. After information fusion, morphology processing, region connected, edge connection and filling cavities, more complete information of motion region is gained. Then bi-directional projection is used to determine the position of each motion region in the image. In order to segment each moving body, according to the number of people that can be contained in a video image, two methods are given in the paper:simple segmentation method and circle detection segmentation method based on vote algorithm.The second step is moving body tracking. In order to ensure the integrity of the target information for tracking, two detection lines is set firstly. Then, block color distribution model is adopted to represent each moving body. According to the similarity between two color models represented by Kullback Leibler distance, moving body model for tracking is established. In the use of trust region method to solve the tracking model, the problem is met that iteration cannot converge, due to the fact that points in an image are discrete. Withal, trust region method is improved in the paper, and a new termination condition and a new method for choosing the optimal solution are presented, which accelerate the convergence speed and a more satisfactory tracking result is obtained at the same time.The third step is people statistics. Depending on the detection and tracking result in each frame image, the trajectory of each moving body is associated. And then judgment methods for an effective and complete trajectory are provided and people statistics is realized.At last, the experiment results got from three test video clips prove the effectiveness of the proposed algorithm.