Research On Subway Passenger Congestion Recognition Based On Deep Learning

At present,the overcrowding problem in subway stations occurs frequently.If these overcrowding problems are not realized and solved in time,stampede and other mass incidents are likely to happen.Therefore,it is necessary to identify the degree of passenger congestion in subway stations quickly and accurately.Deep learning has significant advantages in solving the problem of target recognition.Therefore,this thesis explores subway passenger congestion recognition based on deep learning.The main research work is as follows:(1)Preliminary work of the research.By analyzing the applicability of deep learning theories,the convolutional neural network is used to establish the subway passenger congestion recognition model in this thesis.Public data sets and self-built data sets were collected to provide data support for the experiment of this thesis.(2)Establishment and analysis of subway passenger crowd counting model.Subway passenger crowd counting model is the core and foundation of subway passenger crowding degree recognition model.Firstly,a subway passenger counting model MSF-SPC based on multi-scale feature fusion is proposed in this thesis.Then,in order to solve the problem of feature fusion competition in MSF-SPC model,an improved passenger counting model GAN-SPC based on conditional generation adduction network is proposed.Finally,in order to further improve the counting efficiency of subway passengers,this thesis integrates the MSF-SPC model and GAN-SPC model,and establishes the subway passenger counting model MSFGAN-SPC.At the same time,the public data set and self-built data set are used to evaluate the algorithm performance of the three models respectively.The results show that MSFGAN-SPC model has better experimental effect.(3)Establishment and analysis of subway passenger congestion degree identification model.Based on MSFGAN-SPC model,the subway passenger congestion recognition model MSFGAN-SPCI was established.Then the comparison model is established,and the MSFGAN-SPCI model and the comparison model are used for comparison experiment and the experimental results are analyzed.The experimental results show that the recognition accuracy of MSFGAN-SPCI model on two self-built data sets reaches 97.6% and 98.2% respectively,which is obviously better than the experimental results of the comparison model,and also meets the requirements of accurate recognition in subway application scenarios.In addition,the recognition time of each image of the MSFGAN-SPCI model on the two self-built data sets is 0.11 s and 0.07 s respectively,which also meets the requirement of fast recognition in the application scenario of subway.In summary,MSFGAN-SPCI model has high practical value in the task of subway passenger congestion recognition.