The Research On Human Flow Counting Algorithm In Intelligent Video Surveillance

With the growing requirement for safety, more and more cameras are utilized for visual surveillance. Therefore, it is natural to establish a computer vision-based system to make use of such a large amount of visual information.To keep computer aware what happens in the visual field automatically, Intelliengent Video Surveillance System (IVSS) has long been the research focus. Pedestrian counting is a crucial and challenging task in visual surveillance such as security application, pedestrian traffic management and measuring marketing effectiveness. A vision-based people counting system relates to many common research topics including moving object extraction, object tracking, trajectories analysis and so on. The main contributions of this thesis are as follows:1) The background model of mixture of Guassians is adopted to extract the moving object. First, we analyze the impact of key variables in the mixture of Gaussians algorithm. Base on this, an adaptive selection strategy of the number of components of mixture of Guassians model is proposed which decreased the computational complexity and memory requirement with no loss of detecting accuracy.2) A scene model and statistic learning based on method for pedestrian detection is proposed in order to train and update the classifier online. In the method, we use the positive samples and the limited negative samples from each individual scene to train a cascade classifier. Additionally, a boosting architecture is presented to speed up the classifier training.3) In order to overcome the problem that the Camshift algorithm will track failture when the object is occlusioned seriously, we propose the improved method of multiple targets tracking algorithm based on Camshift algorithm combined with Kalman filter.The experiments result shows the proposed algorithm can overcome the occlusion problems effectively and achieve real-time requirmrnts.4) This article constructed a pedestrian flow statistical system based on the above algorithm. The Preliminary experimental results verify the proposed related technologies. At the end of this paper, a summary of the full text and the suggestion of the future work are provided.