Research On Detection And Identification Algorithm Of Human Abnormal Behavior

Recently, to meet the growing social demands for safety monitoring, researching on intelligent video surveillance system appears increasingly important to deal with so much data information. Moving object detection is the basis of intelligent surveillance, and abnormal behavior identification is the purpose of that. Meanwhile, recognition of pedestrian congestion is one of the most important and difficult issue of abnormal behavior recognition. In this paper, we pay attention on the algorithm of moving object detection and abnormal crowd identification. Then we apply the improved algorithm on the real video surveillance system.The main points in this paper are as follows:(1) The basic flows of abnormal behavior recognition are introduced here, including imaging preprocessing, moving object detection, moving object tracing, behavior recognition, etc. The elaboration on the main approaches and classfication of each component lay a solid theoretical foundation for subsequent work.(2) Combining the advantages and disadvantages of background subtraction method and frame difference method, this paper has proposed an improved method, which could adapt to the changes of illuminations automatically. As background subtraction method is sensitive to illumination and the test results of the frame difference method is prone to empty. It is approved that the experimental result of this improved method is great whether the light condition is change or not.(3) From the aspect of energy and image entropy, a crowd identification algorithm is proposed. The energy can represent the number of moving object and entropy can indicate the degree of aggregation. The algorithm idea is simple, low algorithm complexity, and the experiment result is good, strong real-time performance, which can be very useful for real-time intelligent monitoring system.(4) Based on algorithm in this paper, an experimental intelligent surveillance system is designed in matlab environment using PC and USB camera hardware, which can detect the group behavior and give an alarm.