Pedestrian Detection And Tracking System Based On Intelligent Video Analysis

With the national systematic projects of smart community and safe city vigorously constructed,demands for high-quality and huge-quantity of the security monitoring are growing day by day,thus more and more cameras and monitoring nodes are put into use,which undoubtedly increased the burden on the monitoring staff,and thus cameras with more intelligent analysis and processing function become an important future development direction.The further open construction of community will bring the reform and upgrading of the security monitoring around smart community,among which pedestrian tracking and behavioral warning analysis highlight the important position.Therefore,the thesis proposed a system about pedestrian detection and tracking,moving target detection based on intelligent video analysis.The system totally consists of three functional parts,the first module is about pedestrian detection in video,the second module is about specified single target Pedestrian Detection and Tracking System Based on Intelligent Video Analysispedestrian tracking,and the last one is about moving target detection of ROIs.In the module of pedestrian detection,traditional method——HOG+SVM is used to achieve functional design and implementation.By training difficulties and using NMS rectangular frame fusion mechanism to improve the performance of the classifier to achieve pedestrian detection based on video.The average detection rate in the actual scenes is 41.66 ms/frame,the average accuracy rate of the detection achieves 99.49%,the average recall rate is 72.21% and the average combined effect of the detection achieves 83.68%.A tracking scheme is designed to track pedestrians based on particle filter with color feature histogram,which can achieve specified single target pedestrian tracking in video sequences.System tests showed that the average tracking speed both in the actual and non-actual scenes remains below 8.5 ms/frame,is able to meet the real-time tracking demands.The module of ROI intrusion detection adopts the method of inter-frame difference to achieve the monitoring and analysis of the designated area in the video sequences.The system can automatically avoid the normal situation,but when the intrusion occurs,the system sends out the alarm information and detects the motive foreground.This function can be widely used to detect a variety of irregularities like intrusion,trip line crossing,windows climbing in community video surveillance.Firstly,designing and achieving the above three functional modules.Then,integrating them to form the interface of the system client.The research of the thesis is of certain practical value to behavioral warning analysis in community video surveillance.