Detection And Recognition Of Passenger Flow Congestion In Subway Station

With the continuous improvement of China ’s economic level and the acceleration of urbanization,urban rail transit has become one of the main ways for urban residents to travel with its advantages of large capacity,high efficiency and green environmental protection.The construction of urban rail transit is conducive to the development of urbanized metropolises,solving urban congestion problems,and improving the carrying capacity of urban vehicles.It plays an important role in urban modern development infrastructure.At present,China ’s rail transit industry has ushered in a period of rapid development.Many cities have opened subways and light rails.The coverage area of the subway network is becoming more and more extensive,which greatly facilitates public travel.The detection and identification of passenger congestion in subway stations is an important part of the safety management of subway stations.It is the basis for ensuring the safe travel of passengers and improving the operational efficiency and service level of stations.This paper studies the problem of passenger flow congestion detection and recognition based on video of subway stations.Firstly,the problem of passenger detection in subway station,the passenger detection algorithm is studied based on target detection technology and the passenger detection model is constructed.Then,the extraction scheme of passenger flow characteristic parameters is designed,and the passenger flow feature extraction is realized.Then,the passenger flow congestion state of the subway station is divided,and the passenger flow congestion state identification model is constructed to complete the identification of the congestion state.The main work of this paper is as follows :(1)Aiming at the problem of passenger detection in subway stations,this paper proposes a real-time detection method based on surveillance video.Considering the characteristics of high passenger flow density and serious occlusion in subway stations,a real-time detection model of passenger flow congestion in subway stations is constructed based on improved YOLOv3.CIo U is used instead of Io U to calculate the regression loss function of the target bounding box,which improves the positioning accuracy of the target prediction box.The Focal Loss loss function is added to guide the model to focus on the training of difficult-to-classify samples with high occlusion.The K-means clustering algorithm is introduced to re-cluster on the self-made data set to obtain a priori box suitable for subway station passenger flow.The proposed model is trained and verified by using the self-made subway passenger dataset PFS.The experimental results show that the average accuracy of the proposed method is improved from 78.81 % to 80.45 % compared with the YOLOv3 algorithm,and the average detection speed is improved from 26.133 fps to 26.368 fps.The real-time and accuracy of passenger detection are improved.(2)Aiming at the problem of passenger flow characteristic parameter extraction,a subway station passenger detection and tracking model based on YOLOv5-Deepsort is proposed to extract three passenger flow characteristic parameters : passenger flow,density and speed.The passenger collision line counting method based on virtual coil is adopted.The gray change of virtual coil is used as the standard of passenger counting,and the passenger flow is counted by counter.The speed of the passenger is calculated based on the center particle movement method,and the movement of the center particle is used to replace the walking process of the entire passenger.The proportional relationship between the pixel coordinate displacement of the passenger and the actual coordinate displacement is obtained,and then the real-time speed of each passenger is obtained according to the expression.Based on the experimental marking method,the scale of the image monitoring area and the real area is obtained,and then the passenger flow density of the monitoring area is calculated according to the passenger flow data.Based on Pytorch,this paper builds a passenger flow characteristic parameter extraction model,and obtains the passenger flow characteristic parameter data,which provides a data basis for further identification of passenger flow congestion.(3)Based on the extracted passenger flow characteristic parameters,the FCM clustering method is used to complete the division of the passenger flow congestion state of the subway station.The passenger flow congestion state is divided into smooth state,general state,congestion state and severe congestion state.Subway station passenger congestion identification model : A strategy based on fuzzy C-means(FCM)clustering of passenger flow characteristic parameters and neural network-based congestion state learning and identification is proposed.FCM is used to cluster the passenger flow,speed and density passenger flow characteristic parameters,grasp the law and distribution characteristics of passenger flow data,and divide the passenger flow data in advance.Based on the neural network,the correlation between the divided congestion state and the data is studied,and the congestion state is identified according to the real-time passenger flow data.A neural network-based passenger flow congestion state recognition model for different areas in subway stations is established.The model considers the difference of passenger flow characteristic parameters selected for passenger flow congestion state identification in different regions.Different passenger flow characteristic parameters are adopted in the waiting area and the running area of the platform,and the real-time identification of passenger flow congestion state in each region is realized.Combined with the actual extracted passenger flow characteristic parameters,the model is trained and verified by cases.The experimental results show that the model can accurately identify the passenger flow congestion state in different areas of the subway station,and the recognition accuracy is high.In this paper,the detection and recognition of passenger flow congestion state in subway stations based on video images are completed.A passenger detection model based on video images is proposed.The extraction scheme of passenger flow characteristic parameters is given.The passenger flow congestion state recognition model in different areas of subway stations is established,and the subway station passenger flow data set is made to complete the training and verification of the model.The method proposed in this paper can obtain the congestion state of each area in the subway station in real time.The research work can assist the subway station operation management department to grasp the distribution and change of passenger flow in the station in real time,and provide support for the management of passenger flow control,safety warning,monitoring and management of the subway station.It has certain practical significance for ensuring the safe operation of urban rail.