Research On Crowd Abnormal Behavior Dection Method Based On Optical Flow Feature

With the development of the social economy,there are more and more crowded public places,how to effective monitoring of the population has become a prominent issue of public safety.Intelligent video surveillance uses computer vision,image processing and pattern recognition techniques to monitor crowd status,which can effectively detect abnormal behavior in the crowd.In this paper,a group abnormal behavior detection method based on optical flow characteristics is designed based on the research and analysis of abnormal behavior detection.This paper firstly analyzes the existing methods of motion target detection,the method of mixed gauss model used in moving target detection,and it is used to smooth the noise processing of the target using the median filter,connectivity and accuracy can be obtained more ideal target area of motion.In the extraction of target characteristics,based on the analysis of the Harris corner extraction algorithm,designed an improved multi-scale Harris corner extraction algorithm,feature extraction using the method in the number of more adequate and more stable quality.Second,in the group anomaly detection,this paper use the pyramid Lucas-Kanade optical flow method to extract the feature points matching and tracking,optical flow field can then be obtained by optical flow vector group,optical flow field in motio n feature extraction group,the extraction of motion characteristics including the average kinetic energy and direction of movement of the entropy,the characteristics of the movement is calculated value compared with preset threshold to determine whether there is abnormal behavior,when comparing the judgment based on the movement characteristics of five consecutive frames value compared with the threshold,if the value is greater than the threshold for five consecutive frames,the judge group exists abnormal behavior.Finally,in order to verify the design group anomaly detection method of performance,this article through to the UMN tested experimental data set,the experimental results show that the design group anomaly detection methods in terms of accuracy and real-time are ideal.