Risk Prediction Of Crowd In Subway Station Based On Video Image Processing And Crowd Evacuation

With the rapid development of urban rail transit,more and more people choose to travel by subway.Once the danger occurs,it will pose a great threat to persons and property.At present,much of the research in this area falls into single prediction(such as large activities in gym)or cause analysis after accident,which lacks effective real-time monitoring and warning method for congestion risk.In this paper,a new method of evaluating congestion risk parameters in subway station based on video image is proposed.This method solves the problem about the scene with complex population,large flow and long duration and realizes real-time monitoring and warning of population state,and based on this,a set of evacuation routes is designed.This paper takes the actual crowd of Xizhimen subway station as the research object,carrying on the video collection and image processing.The main research content is as follows:1.Determining the method of real-time congestion monitoring and prediction of risk.This paper starts from the three commonly used risk prediction methods,analyzes the advantages and disadvantages of them,and finally proposes a real-time monitoring method combining video image processing.2.Studying the algorithm of detecting population density.Based on the existing problems of traditional algorithms,an algorithm combining texture analysis and neural network is proposed to predict the population density in a certain region.The algorithm uses texture analysis to transform multiple pictures into parameters to describe pictures,so as to analyze the different crowd conditions of subway stations,so as to weaken the mutual occlusion between people and the influence of cameras.The transformation results are input into the neural network for training and forming the sample library,which means the algorithm flow of combining the texture analysis with the neural network is achieved.In the actual detection,the predicted images are automatically entered into the neural network library in the way of texture analysis parameters to determine the number of people and results are obtained.The algorithm is compared with the traditional density estimation algorithms based on pixel statistics and Hough transformation.The results show that the accuracy of the algorithm can reach eighty percent,which is much higher than that of the existing traditional algorithms.3.Studying the algorithm of detecting crowd velocity.Through tracking the crowd in the video,the paper can determine the specific flow rate of it.Two algorithms of target tracking are implemented:mean-shift algorithm and Kalman filter tracking algorithm.Based on the background of the subway station,this paper has classified experiments:single tracking under fixed background,single tracking under changing background and multi persons tracking under fixed background.4.Studying the evacuation route of the crowd.Firstly the people of Xizhimen subway stations were classified;Then the whole structure of XiZhiMen subway station were investigated;Changes of human flow at different time were investigated;at last,this paper developed an evacuation route for Xizhimen railway station.