Research On Passenger Abnormal Behavior Detection In Terminal Video Surveillance System

With the rapid development of China's civil aviation industry and the continuous improvement of airport passenger throughput,terminals are playing an increasingly important role in civil aviation transportation.At the same time,abnormal behavior in the airport terminal occurred from time to time.The traditional monitoring method is still used in airport security at present,which relies on manual real-time monitoring and is time-consuming and laborious,and it is also prone to wrong judgment and missed judgment,so it is difficult to meet the needs of airport security management.In recent years,with the wide application of computer technology,the rapid development of digital image-related technology and the rapid rise of deep learning,the convolutional neural network has become a symbol in the field of deep learning.In order to maintain the normal waiting order and meet the security management requirements of the airport,the deep convolutional neural network is integrated into the airport terminal scene to identify and detect the abnormal behavior of passengers in the terminal so as to improve the traditional monitoring mode in the airport and achieve the effect of intelligent monitoring.By comparing the principles and advantages of common human behavior detection models,it is proposed to detect abnormal behaviors of passengers in airport terminals based on the I3D(Two-Stream Inflated 3D ConvNet)network model.In the case of a small airport data set,the data set exposed by the network is filtered and integrated to expand the data set.Due to the pedestrians are easy to be blocked and the tracking loss rate is high in the pedestrian identification in the application of the traditional I3 D model in the airport security area,an improved I3 D model is designed to solve these problems.Based on the traditional I3 D model,this model introduces the motion model constructed by continuous frames,and based on motion model to realize the recognition of human behavior,which effectively reduces the problems of the traditional I3 D model in the pedestrian recognition application which is easy to be blocked and tracks the high loss rate.Meanwhile,an improvement based on Inception-v3 algorithm is proposed,which speeds up the network operation speed,weakens the occurrence of overfitting phenomenon,and optimizes the structure of Inception Module.Finally,experimental verification was carried out.The accuracy of the improved I3 D network model can reach over 91.5%,and the efficiency can reach 5ms/ PCS.It is proved that the improved I3 D network model can effectively improve the accuracy and efficiency ofvideo identification of airport terminals.In the meanwhile,the average accuracy of different weights on the abnormal behavior detection task was obtained,and the optimal weight of model fusion was selected.