Research On People Detection And Counting Based On Video Analysis

Video processing is a studying hot spot by many researchers, which is widely used in vehicle detection and counting and achieves a good system performance. However, it is still in an immature stage for counting of pedestrian flow based on video processing. The difficulties lie in the fact that not only the pedestrians are non-rigid object, but also the clothing, the color of the hair and the pose of the pedestrians are quite different. What’s more, the surroundings change all the time. These problems make the detection of pedestrians more challenging.The article mainly study the algorithm of counting of pedestrian counting in crowded situation based on video detection and tracking with a single camera. The head is to be detected from the top the pedestrian after theoretical and practical research, and the cascade Ada Boost method is used to train a head classifier based on LBP feature. In order not to count the same target repeatedly, the position information of the same target needs to be predicted in different frame, so this article proposes a robust pedestrian tracking method based on Haar feature matching. The tracking method regard the motion of pedestrian as uniform linear motion in a short interval according the practical situation, and build the state equation, then the kalman is used the predict the position of the target and the appearance model is selected to search and match the target. After that the whole flow the pedestrian counting algorithm is introduced and the importance of the ROI setting in reducing the computation cost and improve the detection rate is specified. At last the basic prototype of the hardware system of video system and software operating system are realized. The graphics information and the counting result of every camera can be monitored from the software operating system. A number of experiments are done to evaluate the performance of the algorithm of detecting, tracking and counting of pedestrian, numerical results show that the pedestrian counting system reaches the requirement of practical application in detecting performant, counting rate and algorithm efficiency.