Research On Optical Flow Based Passenger's Abnormal Behavior Recognition Under Airport Terminal

In recent years, with the rapid development of civil aviation industry, the increasing number of passenger flowing led to frequent occurrence of passengers' trouble accident in the airport terminal. Since traditional video surveillance system relies on artificial observations to discover abnormal events, the effectiveness of monitoring does not match the huge number of monitor cameras in terminal buildings. On the contrary, intelligent video surveillance system is able to analyze abnormal behaviors of passengers in real time, warns of the security situation timely, and eases airport security management pressure.By using the video behavior analysis technology, general abnormal behaviors of passengers in the airport terminal such as the fast running, fighting, and the gathering are studied in details on detection and recognition in this paper. The research work is mainly reflected in as follows:Compared with normal walking, running and fighting are defined as motion abnormal behavior. The motion based abnormal behavior recognition is adopted via the direction interval of optical flow. First, the direction of optical flow is assigned to a certain number of angle intervals from 0 degree to 360 degree respectively after optical flows in pedestrian region obtained from the moving object detection. Then, the number and the average amplitude of optical flows are counted out in each angle intervals, which construct feature vectors of motion abnormal behavior. Finally, support vector machine can be the effective classifier for recognizing the normal walking, fast running and fighting of abnormal behavior after trained by optimized feature vectors.According to the similarity of crowd gathering and other behaviors in the low or medium crowd density scenarios, the method based on crowd density and optical flow on the corner is put forward to detect the multi-passenger's gathering abnormal behavior. First, the change of crowd density is estimated by normalized foreground area and two-dimensional joint entropy from binary foreground image. At the same time, kinetic energy and direction entropy are calculated by using motion of optical flow on the foreground corners. With characteristics of gathering in crowd density and movement, these differences of different abnormal behaviors are used to define two novel decision indexes on gathering and running. Thus, gathering, running, fighting as well as normal walking can be identified step by step via experimental thresholds. The proposed method can effectively reduce the number of false judgments caused by the use of single crowd density feature based.Using different abnormal behavior detection proposed carried out the experimental simulation on the multi-standard behavior video database and the actual videos. The experimental results have shown that the proposed methods are helpful to improve the performance on recognition of passenger's abnormal behavior under airport terminals.