The Design And Implementation Of Human Abnormal Behavior Detection System Based On OpenCV

Public security has always been a social problem. We can see the cameras elsewhere in our life, which shows video surveillance system is very popular. They provide security for our lives and they are useful for investigation. Meanwhile, image processing, information technology and the video surveillance system are gradually combined to provide intelligent detection and warning mechanism for people. If the video monitoring system automatically detected abnormal behavior and alarm people, efficiency will be greatly raised. To detect the abnormal behavior early will also minimize the hazards which abnormal behavior brought to public safety. The design of the study is mainly about certain abnormal behaviors detection through the analysis of the movement characteristics of people in the video, and issue a warning timely. This paper focuses on the use of optical flow method to do motion characteristics analysis and abnormal behavior detection.The abnormal behaviors can have different definitions for different scenes. In this paper, the following abnormal behaviors are mainly focused on: pedestrian fights violence, pedestrian crossing the forbidden-in area, pedestrian retrograde in certain situations. For each kind of abnormal behavior, abnormal judgment model is established and applied in anomaly detection system. Although there are many moving target detection, identification and analysis methods, in this paper the optical flow method is focused on. The optical flow information generated by the movement of human behavior is analyzed to determine whether the abnormal behavior or not. The amount of global optical flow computation is large. In order to make calculate easier, pyramid Lucas-Kanade algorithm can be used to detect the optical flow of motion. The tracking process of Pyramid Lucas-Kanade algorithm is an iterative calculation process, and the convergence speed is relatively fast. Feature points are necessary for tracking. The feature points are needed to detect in the image before using the optical flow. The number of feature points can be controlled and the sum of them is lesser, so the calculation of the algorithm is not very large. Harris corner detection method is used. The detected Harris corners are used as feature points of a frame. These features are used to carry out subsequent analysis, so as to improve the efficiency of detection. In this paper, some properties of the feature points are controlled, and some of the methods are optimized to make the tracking effect better. Through the Pyramid Lucas-Kanade method we can track the optical flow of these features points. We get the optical flow characteristics of feature points in the image to estimate motion. Finally, the optical flow histogram is introduced to describe the motion characteristics of the moving object more clearly. This paper uses the amplitude-based direction histogram of optical flow, which not only reflects the distribution, but also can be combined with the amplitude analysis of motion information. Meanwhile the method for calculating the kinetic energy is defined. Energy values of feature points are calculated. For each kind of abnormal motion model, according to the optical flow histogram feature and the feature of energy, different abnormal behavior detection model is established to separately detect the fights, the abnormal behavior of pedestrian crossing and pedestrians retrograde. When the system detects the possible abnormal behavior, an alarm is automatically sent out. The design is implemented using Open CV, and the corresponding improvement is made according to the specific situation to effectively realize the detection of pedestrian abnormal behavior.