Risk Identification And Early Warning System For Abnormal Behavior Of Station Passengers Based On Video Data

The fundamental mission of urban rail transit is to ensure that passengers arrive at their destination comfortably on time.In this process,passengers as the object of their services,also have a great impact on the operational safety of urban rail transit.The behavior of a single passenger will affect the passenger flow of the entire area.It affects the safety of the entire area,so it is of great significance to study the abnormal behavior of passengers.Therefore,this paper aims to define the abnormal behavior of passengers,build a risk classification method for abnormal behavior of passengers in stations,and form a passenger flow status based on video data.This paper researches and develops a prototype system for risk identification and early warning of abnormal passenger behavior in stations,and provides help for the safe operation of urban rail transit stations.The main work of this paper is as follows:(1)A risk classification method for abnormal behavior of station passengers is proposedOn the basis of defining the abnormal behavior of passengers,the object of attention in the abnormal behavior of passengers in the urban rail station is clarified,the area of the urban rail station is divided,and the passenger flow status is classified,and the area facing the urban rail station is formed,considering the passenger flow status.Behavioral risk grading methodology.(2)Formed a passenger flow state identification method based on optimized YOLOv5 and Deep SortBased on the optimization of the YOLOv5 algorithm,the identification method of passenger flow density and passenger flow duration is formed.The m AP50-95 index of the optimized algorithm on the test set has been improved from 0.5816 to 0.5962,and the average detection accuracy of passengers in the actual scene in each area of the station It is 86.82%;the Deep Sort multi-target tracking algorithm is combined with the homography matrix to form a passenger flow speed identification method.After verification,the accuracy of passenger speed identification can reach 93.2%.(3)A method for identifying abnormal passenger behavior based on Open Pose and target tracking is proposedCombining the Open Pose key point extraction algorithm with the random forest classification algorithm,the abnormal behavior and the normal behavior are distinguished and identified by constructing the key features of the abnormal behavior of passengers,and the abnormal behavior recognition of passengers based on actions such as fighting,falling,and throwing is formed.The average recognition rate reached93.38%;the passenger movement trajectory was analyzed based on YOLOv5 and Deep Sort,and a retrograde and stranded passenger abnormal behavior identification method based on trajectory identification was formed.After verification,the accuracy of retrograde and stranded behavior identification was 94.2 % and 99.99%.(4)Designed and developed a prototype system for risk identification and early warning of abnormal behavior of passengers in the stationStarting from the urban rail station area-oriented and considering the passenger flow status,the station passenger abnormal behavior risk classification method and the passenger flow status and passenger abnormal behavior identification method proposed in this paper are based on the demand analysis and architecture design of the station passenger abnormal behavior risk identification and early warning system.A prototype system was developed.