Design And Implementation Of Crowd Anomaly Detection System

Most crowd anomaly monitoring system need the corresponding motion tenplate through learning and training and the abnormal decision is made by the movemem pattern matching.The design and implementation of crowd anomaly monitoring system which is based on the ststistical analysis of motion leature points is presented,withoiu the trained motion tempalte to make correct decision.Firstly,the Harris operator is used to extract the preliminary feature points,which falls into twO eategories:one belongs to the region of baekground and the other belongs to region offoreground.The latter is used for crowd anomaly detwection method,and the former eeds to be filtered out.Therefore,followed by the moving foreground extraction algorithm-Mixed Gaussian background modeling method,the next step is the moving foregound region extraction.And with the help of the moving foregrund region has been extracted from feature points set,the feature points set is optimized to filter the background featurle points,while the moving foreground feature points are left.Then,Lucas Kanade point matching alghrithm is used to track the fetature point,while the position chaniges between mo successive frame is calculated,which is assigned to the matched feature points,At the stage of crowd abnormal detection,the quantity of motion feature points of different motion intensity is normalized to the ratio of the whole points set.To add the ration of abnormal motion intensity,and compare with the giving threshold,the crowd abnormal behavior decision is made.Finally,the crowd abnormal monintoring system is implemented by programming in the Visual C++ 2010 development platform.The UMN video data sets is used to test thm practical effect of the system.The result shows that the crowd anomaly monitoring system works correctly.