Research On Identifying And Alerting Technologies For Crowd Security Of Airport Terminal

At present, with the rapid growth of civil aviation industry, the increasing number of passengers leads to a lot of conflict events which bring a lot of pressure to the airport security management. Although most of the modern airports have equipped the video surveillance system(VSS), the abnormal warning is still a manual work in the traditional VSS. For some airports with thousands of cameras, this work will be very time consuming. Meanwhile, with the development of the computer digital image processing, the new generation intelligent video surveillance system(IVSS) is becoming a hotspot. The IVSS can achieve the abnormal warning work independently with the algorithms of image processing and pattern recognition, which can reduce the time consuming obviously. After analyzing the domestic and overseas research actuality of IVSS and the requirements of civil airports security, some key technologies of security identifying and alerting in airport terminal are invested in this dissertation.As the base of intelligent video processing, the foreground extraction technology is studied firstly. Based on the classic Gaussian Mixture Model(GMM), a modified GMM is proposed. The standard deviation has been replaced by the mean deviation in the new model to reduce the amount of calculation. Meanwhile, the HSV shadow removal algorithm is added to the model to reduce the interference of shadow. Furthermore, a normalized foreground area calculation method is designed to correct the bias of people counting when the foreground area is inputted as a feature for the perspective effect.Secondly, three types of people counting algorithms have been proposed to solve the problem of air travllers counting from the angle of crowd density estimation. The first algorithm is designed for people counting in some low or middle denstity scenes. In this algorithm, a size adaptive sliding window is built to search the binary foreground, and the pedestrian target in this window can be detected fastly by a set of conditions. The second algorithm is designed for people counting in some high density or serious occlusion scenes. In this algorithm, an occlusion coefficient is calculated by the normalized foreground area and the optimized corner information, which can reduce the occlusion influence obviously in people counting. The third algorithm is designed for people counting in the multiple cameras VSS. In this algorithm, based on the homography, the object matching strategy which considering the factor of occlusion, is designed for improving the people counting results of single camera and achieving the overall people counting of multiple cameras.Thirdly, for the requirements of the airport terminal crowd security, two abnormal behavior detection algorithms have been proposed to sovle the problem of passenger gathering and fighting behavior detection. The first algorithm is designed to detect crowd gathering behavior. In this approach, based on the normalized foreground area and the two-dimensional joint entropy of binary foreground, a crowd gathering detection model is designed to estimate the crowd abnormality effectively. The second algorithm is designed to detect the pedestrian fighting behavior. In this approach, the foreground optical flow fields are caculated to extract the entropy of the optical flow direction. With this feature, the pedestrian fighting behavior can be detected reliably.Finally, a prototype system for crowd security identifying and alerting is built for airport terminal crowd security management. Through analyzing the video data from the VSS, this system can achieve the function of people counting and abnormal behavior detection based on the above algorithms. And benifited by the implement of concurrent synchronization technology, this system can deal with multiple video analyzing effectively at the same time.