Research On Abnormal Behavior Detection In The Terminal Under Video Surveillance System

The airport terminal, as one of the important hub of civil aviation traffic and transportation, is prone to abnormal behavior because of high passenger throughput. However, the modern airport terminal seems time-consuming because of thousands of cameras and manual observation, which can hardly satisfy the needs of airport security management. Therefore, with the intelligent video analysis technology, we can actively do real-time monitoring and alarm for passengers’ abnormal behavior, which can effectively assist security guards to deal with the abnormal events and improve the rapid response ability of emergencies. This paper mainly focuses on the key technologies involved in the detection method for three kinds of abnormal behavior about gathering,running and abandoned objects. The main work is as follows.Firstly, the extraction of moving objects under video surveillance is presented. With the foundation of Gaussian Mixture Model(GMM), the updating mechanism of mean and variance is optimized and the HSV shadow removal algorithm is added to the model, which can improve the modeling speed and reduce the interference of shadow, in order to reconstruct the background model of video image and extract the moving object.Secondly, the detection of gathering and running behavior is studied. According to the different performance characteristics of population density and motion characteristics, different index is designed for the detection of abnormal behavior. Crowd Density Index(CDI) is designed with the improved foreground area and the two-dimensional joint entropy. Motion characteristics are described with energy and weighted entropy, which can be calculated by the Pyramid Lucas-Kanade Optical Flow. The effectiveness of algorithm is verified by video sequences of different scenes which can emulate the airport terminal.Finally, the abnormal behavior based on abandoned objects detection is proposed. Two sets of backgrounds for scene are constructed by improved Gaussian Mixture Models with different updated rates,which can remove the interference of moving targets such as pedestrian and extract the short-term static objects. Then the objects are tracked with multi-features to get the time the object exists. If the time is over the threshold, the object is judged as abandoned object. Meanwhile, with the information of the abandoned object and historical images, the passenger’s abnormal behavior is determined when the object is lost and detained. Experimental results based on video sequences of different scenes which emulate the airport terminal show that the proposed algorithm conducts a desirable performance.