Research On Crowd Counting And Application Method Of Railway Passenger Station Based On Deep Learning

This thesis combines deep learning intelligent technology to realize crowd counting in passenger stations,which is of great significance for improving the intelligent level of passenger transport management and discovering potential safety hazards in advance.The main research work of this thesis is as follows:(1)Based on the actual situation of railway passenger station,the relevant content of passenger flow density is analyzed,the passenger station is divided into multiple functional areas,the passenger flow characteristics are analyzed in combination with regional characteristics,the current passenger transport management work mode is summarized,and finally,the difficult points of crowd counting in the passenger station are analyzed according to the characteristics of passenger flow,which provides theoretical support for the construction of the algorithm later.(2)Aiming at the problem that multi-scale features of crowd density images in railway passenger stations are difficult to solve,based on the method of density map generation,a multi-context perception crowd counting model is proposed,which realizes the extraction of global feature attention through multi-scale feature extraction module and multi-context information module,guides the network to pay attention to people of different scales,and improves the expression ability of feature map.(3)In order to improve the problem that crowd counting depends on the quality of density map generation,using the idea of P2 PNet point matching framework,a crowd counting model based on multi-level feature fusion generated by point generation is proposed,and the number of people is estimated by generating a set of points and head marker positions for one-to-one matching,and finally,a new network loss function is designed to enhance the positioning accuracy and generalization ability of the model.Based on the video data of Beijing West Railway Station and the crowd pictures of railway passenger stations collected online,and the above two algorithms were trained and tested on Shanghai Tech and UCF open source datasets,and the experimental results verified the counting performance of the model.(4)In order to study the application method of crowd counting in passenger stations,a passenger flow density detection platform is constructed,and according to the actual needs of passenger stations,video-based passenger flow density detection is realized,historical passenger flow information is recorded and stored,and passenger flow-time curves are drawn,and based on the method generated by density map,the crowd heat map is drawn to display the spatial distribution information of passengers in the passenger station,which can basically meet the needs of actual scenarios.