Analysis Of Civil Aviation Passenger Group Behavior Based On Video Data Mining

Civil aviation transport is subject to many factors and is prone to irregular flights.The contradiction between irregular flights and airlines or airport disposal mechanisms and passenger satisfaction has led to frequent incidents of crowd conflicts in terminal buildings in recent years,Therefore,it is urgent to research and analyze the behavior characteristics of passengers in terminal buildings to realize the early warning of mass incidents.This article studies the passenger crowd behavior in the terminal from the perspective of video surveillance.Based on the commonly used mixed Gaussian background model foreground detection method,this paper proposes a shadow removal algorithm based on YCbCr color space and topology cut,to achieve the goal of filling gaps and edge optimization.Secondly,due to the difference of the surveillance video images of the terminal building,we obtain the location information of the passengers based on the human head characteristics for the surveillance images with low density.For the surveillance images with high density,the image texture eigenvalues are calculated based on the gray level co-occurrence matrix algorithm,and the target number of the image is estimated,the location information of the passengers is obtained by combining the statistical distribution of foreground pixels.And,based on the location information,clustering clusters are clustered based on the clustering algorithm.Then,the four key groups that are significantly affected by the group conflict events are calculated,that is,clustering density,clustering contour,aggregation rate,aggregate moving speed,etc.Lastly,for the four kinds of characteristic data of the video surveillance picture in a certain period in three sensitive areas of mass incidents in the monitoring area of an airport terminal,BP neural network classifier is constructed and the positive and negative image feature data samples of the three types of area are respectively studied,to find the deep association between feature data and crowd conflict events.Based on the terminal surveillance video images,foreground detection,passenger location information acquisition,crowd feature extraction,and early warning model building,the model test results show that early detection of automatic crowd conflict events can be achieved in three sensitive areas of the terminal.And all three models have a high rate of registration,and an acceptable rate of omission and false alarm rate of crowd conflict events.