Research On Abnormal Behavior Detection And Recognition For Intelligent Surveillance

In recent years, with the rapid development of information technology and the escalation of endangering public security incident, intelligent video surveillance technology has attracted more and more attention. Compared with the pressing technology needs, traditional intelligent control system has a lot of shortcomings. One of the most important points is that the level of intelligence system is too low, rely mainly on human duty, and witch not only costs a lot of manpower and financial resources, but also is difficult to ensure the sensitive, real-time and stability of monitoring. So the problems to be solved are how to increase the monitoring level and the real-time of monitoring system.The abnormal behavior directly related to people was deteced and recognized from two aspects in this thesis which is at the base of image processing and pattern recognition; in the first aspect, the abnormal behavior of single pedestrian in scene was detected and recognized; the core task of the second aspect is to detect and recognize the abnormal behavior of multi-person. The main contents of this thesis as follows:(1) Moving object detection algorithm for outdoor. In this thesis, noise interference being in conventional background subtraction was eliminated partly through the methods such as adaptive background modeling, HSV color space conversion and shadow removing, thus, the exact location and region of moving object (mainly the moving staff) in the background was got.(2) The detection and recognition of multi-behavior such as walk, running, jumping, running and jumping, bending, waving of individual behavior was achieved by the learning for a large of video samples, With Hu invariant moments, aspect ratio, velocity as basic features, using two classifications including support vector machines and nearest neighbor method based on Hausdorff distance.(3) Multi-person abnormal events were recognized. First, motion vector distribution feature in the scene was got by light flow method, and on the basis of motion vector distribution feature the basic feature characterizing multi-person event was got through the calculation of potential energy and kinetic energy in moving area; through the analysis of time distribution curve of potential energy and kinetic energy in different individual event, the detection and recognition of multi-person abnormal events such as meet, fights and meeting were achieved by the threshold value method.The simulation results show that the proposed method can detect and recognize the six personal behaviors and three kinds of groups abnormal behavior appeared in video accurately, and has high stability and reliability.