Research On People Counting System Key Technology Based On Hot Spot

Intelligent monitoring is the development trend of the future monitoring industry.The statistics of People Counting is of great significance in the field of intelligent monitoring.However,due to the complex background environment and the occlusion phenomenon in pedestrian movement,the accuracy of the current method is not high.The scope of crossing-line counting method application has some restrictions at present.Therefore,the paper focuses on the pedestrian detection and pedestrian multi-target tracking problem in People Counting statistics,and puts forward the corresponding technical method according to the number statistics in the region.In this paper,the main work and achievements are as follows:First,we propose a new auto learning rate foreground extraction algorithm to reduce the scope of head-shoulder targets in detecting.The algorithm is used to extract the head-shoulders initial region to avoid the influence of background.The algorithm combines the advantages of the frame difference algorithm and the fixed learning rate algorithm,which can automatically select the higher learning rate when the background is unstable,accelerate the background convergence,and choose the lower learning rate when the pedestrian movement is on the video screen for the integrity of the pedestrian foreground profile,and analyze the pedestrian foreground to predict the area where the head-shoulders may appear.Experiments show that the detection algorithm using the head and shoulders is about 100 milliseconds faster than the usual single target detection time.Second,head-shoulder targets can be detected based on predicted area which computed in first step.Through utilizing 2000 head-shoulder positive samples and 5000 negative samples and extract HOG features to train the linear classifier,and then use multi-scale sliding window scanning head-shoulder hot spot to detect head and shoulder targets,and finally the results of the test try to get the size and location of the final head-shoulders rectangle.Third,the proposed algorithm is used to update the target with distance of the head-shoulder targets between two next to door frames.At the same time,KCF tracking algorithm is used as the complement of the detection algorithm to improve the robustness of the tracking algorithm.When targets are occurred by each other,for purpose of avoiding tracking lost with multi-targets,the information of foreground and spatial position are used to track.By tracking the head-shoulders target to obtain its historical trajectory,and then according to the number of hot spot statistics statistics,so that the hot spot within the crowd statistics.At last,a people count analysis system was developed to calibrate area of interest in video surveillance,and save the statistics data in the local file.Users can not only real-time check the result in the statistical process,also show historical data in a way of Visualization to achieve intuitive understanding and analysis of people counting.