Research On Crowd Density Detection On Subway Platform Based On Convolutional Neural Network And Transfer Learning

With the rapid development of information technology,deep learning methods have been widely used in the field of public transportation in recent years.Especially in the field of intelligent rail transit video surveillance,deep learning has become an important direction.Due to its powerful feature extraction capabilities,deep learning is particularly suitable for the process of crowd density detection on subway(urban rail transit)stations,and has important application value.this is also an important entry point for the combination of artificial intelligence and intelligent rail transit.Based on the actual needs of crowd density detection on subway platforms,and deep learning techniques such as convolutional neural networks and transfer learning,the following work has been done in the research on crowd density detection on subway platforms: A training and verifying crowd counting data set is established,which is collected from multiple monitoring videos of subway platforms at different times,including 700 typical platform scenes.The data set has the advantages of large changes in the range of crowds,varying degrees of crowding,and large-scale changes.Compared with the public crowd counting data set,it is more suitable for evaluating the practical application performance of the algorithm proposed in this paper in the subway field.On this basis,aiming at the problem of insufficient data in the target domain,a crowd counting method is proposed based on small pre-training samples,and a spatial full convolutional neural network model is constructed.Experiments show that pre-training on the large-scale synthetic data set GCC and then fine-tuning with a small amount of labeled target domain data can significantly improve the counting performance of the crowd counting model and reduce the counting error.Finally,in response to the problem of reacquiring label data in new scenarios,a domain-adapted transfer learning crowd counting method is proposed based on structural consistency and adversarial principles.In a new scene,this method can train the model without any new target domain labeled data.The structural consistency Cycle GAN designed in this paper can significantly reduce the difference between the source domain and the target domain,thereby realizing domain adaptation.Multiple sets of simulation experiments are carried out in the public data set and actual subway scenes,verifying that the above methods can guarantee good counting performance.In summary,taking the research of convolutional neural network in crowd density detection and transfer learning in image style transfer as the theoretical basis,combining with the needs of subway station crowd density detection,a subway station crowd density detection model is established.On this basis,it solves the needs of crowd density detection in different subway scenarios.The research in this paper is the application of crowd density detection technology in the field of rail transit.It uses monitoring distributed on the platform to realize crowd counting and completes passenger flow statistics,crowd density estimation,and crowd crowding risk warning functions.It is useful to improve the intelligent level of urban rail transit.It is of great significance to improve the quality of subway services and build an intelligent rail transit system.