Optical Flow Field Analysis And Abnormal Behavior Detection Based On Crowd Movement

Public Security has always been one of the important factors that affect social stability. In crowded areas, people are more likely to occur accidents such as congestion and stampede, etc. And evacuation issues in hazardous environments are directly related to the stability of the community. At the same time, they threat the safety of people’s life and property. Therefore, the study of intelligent video surveillance system has been continuously developed, because of its important role for urban transport and security work. Moving target detection and tracking, behavior analysis and recognition and detection of abnormal behavior are key technologies of intelligent monitoring.In this paper, we develop a research on abnormal behavior detection including cross the street and retrograde. On the basis of optical flow algorithm, we improve the target detection algorithm with a Line Integral Convolution. After Line Integral Convolution, texture images are obtained, so image segmentation algorithms are explored. Finally, the force models are studied to detect the abnormal behavior, and to achieve the purpose of distinguishing pedestrians in different directions.The main contents of this paper include:(1) We use Line Integral Convolution to detect moving targets based on visual ideological of flow field. And we set a research on improved algorithm of optical flow method, which is vector accumulation based on optical flow to detect moving objects in the crowd scene.(2) Image segmentation algorithm is used after we obtain texture image by Line Integral Convolution. After that, binary processing is used to obtain moving target area which is better and more completely.(3) To achieve the goal of distinguishing pedestrians in different directions, we map the optical flow field.(4) We use the repulsive force model to analysis abnormal crowd behavior of moving targets detected in video sequences. Then we verify the effectiveness of the algorithm through the steps above.Through the experimental results we can know that the proposed algorithm can completely detect moving target area, distinguish pedestrians of different directions and complete the detection of abnormal behavior.