| While building China's strength in transportation,railway first.At present,China's railway development has entered the fast lane,railway modernization has jumped to a new level,and a number of railway technologies have realized the perfect leap from following and running to leading.By the end of 2021,China's railway mileage in operation has reached 150,000 km,including 40,000 km of high-speed railway.In terms of passenger transport,high-speed railway accounts for about a quarter of the total length of national railways,taking up 70% of passenger transport capacity.Even under the severe impact of COVID-19 on the transport industry,In 2021,the number of railway passenger trips reached 2.53 billion,up to 72.9 percent of the pre-epidemic figure of 2019.In the face of such a large scale of passenger distribution,transit transport task,the number of railway station construction is also increasing.At present,the whole road has more than 3,200 passenger stations in operation,including more than2,000 high-speed passenger stations.The passenger operation management of high-speed railway passenger stations plays a key role in the overall passenger transport operation.However,compared with ordinary passenger stations,they also face the operation and management requirements of large-scale passenger gathering and distribution organization,high-frequency train receiving and dispatching operation,high-reliability passenger safety guarantee,high-standard passenger service experience and efficient emergency handling.At present,the integrated video surveillance system of the station plays a certain role in the management of passenger services in the station.Through the video surveillance deployed in key areas such as entrance and exit of the station,security inspection area,waiting room,passenger landing area and platform,the staff of the station can directly observe the current passenger flow in each monitoring area.Check the safety behavior in the passenger station by video patrol.However,due to the large number of video surveillance in the station,the management personnel only monitor and patrol through the video surveillance screen,and it is easy to cause visual fatigue in the manual monitoring work for a long time,and it is difficult to master more information about the situation of passengers in the station in detail.A large number of video monitoring data for the data stored in unstructured way,limited to playback view simple functions,such as video data by the problem of inadequate mining,so urgently need based on the intelligent video image analysis method to improve the utilization rate of the station video monitoring data,to improve passenger service in the process of the station operation experience can assign.With the rapid development of artificial intelligence,deep learning algorithm models based on video images show good performance in the field of vision.Combined with the current algorithm theoretical model in the field of deep learning,this paper analyzes and processes the massive video surveillance data of high-speed railway stations,takes passenger flow as the main research object and identifies and analyzes multiple application scenarios in the station,which is of great significance to improve the supervision efficiency and intelligence level of station video surveillance.The main research work of this paper includes the following aspects:(1)Demand analysis and overall architecture design of intelligent passenger flow identification in high-speed railway stations.According to current status of high-speed rail stations in the intelligent building,intelligent construction for the current station in the process of problems,on the basis of extracting based on passenger traffic intelligent identification of the specific application scenarios,and combined with the station control level and passenger service experience in the process of operation for demand matrix modeling intelligent application scenario.At the same time from the data,computing power,platform,application and other levels of intelligent passenger flow identification process of the overall architecture design.(2)Interpretability of key scenes of high-speed railway station video images based on semantic segmentation.In view of the characteristics of large coverage of monitoring areas inside and outside the high-speed railway station house and multiple monitoring scenes,the research and construction of image semantic segmentation technology for key scenes inside and outside the station is based on the classic semantic segmentation model(FCN),which integrates the characteristics of capsule network classification and enhances the information expression ability in the process of scene feature extraction.The semantic segmentation of video surveillance images in key areas such as station square,ticket hall,waiting room,elevator landing area and platform is realized to improve the ability of video semantic resolution and analysis,and the semantic segmentation is better than the classical semantic segmentation model in pixel accuracy.(3)Passenger population density estimation under complex station background based on deep convolutional neural network.For high-speed passenger traffic in accumulation and distribution characteristics of station,adaptive research constructs the high-speed rail station scene traffic density estimation model(ASCCNet),solves the passenger traffic in the station under different density distribution of passenger flow density estimation,the experiment proved that under the sparse and dense scene model can show the high accuracy;For station under the super vision a single video camera in the process of passenger flow density estimation,passengers between shade,dimension change on the influence of the density estimation,the fusion feature map projection built many perspectives of passenger flow density estimation model(MVPFC),through multi-channel video image multi-scale feature extraction,projection,feature fusion and other steps,The passenger flow density estimation is realized by combining the passenger feature maps from multiple perspectives into a larger feature map,and the feasibility of passenger density estimation under the fusion of multiple perspectives is proved through experiments.(4)Research on accurate identification of railway platform line and passengers at both ends of platform based on meta-learning.In order to solve the "false alarm" problem caused by platform line crossing and terminal intrusion detection,a two-stage algorithm model based on meta-learning was proposed for personnel identification classification and detection.Firstly,multi-scale feature extraction is carried out on the support set based on the classical target detection framework(YOLOv3),and the location and size feature information of pedestrians in the station platform line and platform end area is learned.Secondly,by introducing the learning and training process of meta-learning model and constructing multiple sub-tasks,after learning the feature extraction of a small number of labeled samples and comparing with the similarity of the feature extraction of the new samples,the target classification can be achieved,and good results can be achieved in the actual classification and recognition process of passengers and staff.(5)Research on travelers' abnormal behavior patterns based on deep learning.For passengers in the station by drop and waiting in the process of key areas prone to the risk of personal safety hidden danger,through the station passengers had abnormal behavior in the process of safety analysis,put forward an improved model based on slowfast,realize and passengers in an escalator running in standing in groups fight two abnormal behavior pattern recognition.On the basis of the original slowfast,aiming at the time-consuming problem of passenger target recognition,the target recognition extraction modules of YOLOv3 and YOLOV5 are used to improve respectively.The results show that the timeliness of abnormal behavior detection in both scenarios is improved. |
PDF Full Text Request