Personnel Abnormal Behavior Detection Of Subway Stations Based On Machine Vision

With the rapid development of our country’s rail transit construction,the subway has gradually become an important means of transportation for people’s daily commuting.While the subway brings convenience to people,it also has many hidden dangers due to the increase in the flow of people and its special environment.At present,the security monitoring in the subway still relies on manual method,which not only consumes a lot of manpower,but also has the problem of misunderstanding,so it is difficult to meet the needs of the actual situation.Therefore,how to accurately and efficiently detect the abnormal behaviors of passengers in the subway environment and timely alarm to reduce the loss caused by abnormal behavior events is of great significance.In this paper,machine vision is used to study the related fields of abnormal behavior detection of pedestrians in subway,and a set of abnormal behavior detection system of subway pedestrians is realized by combining algorithms.Specifically,there are the following aspects:1.An improved YOLOv5 pedestrian detection network is presented.YOLOv5 s is optimized for the characteristics of fast detection speed,small model,and convenient deployment,but low detection accuracy.The evolution process of the IOU loss function and the limitations of each stage are analyzed,and it is improved by introducing Alpha-CIOU,so that the network pays more attention to the goal of high intersection over union.At the same time,the coordinate attention mechanism is embedded to improve the network’s ability to pay attention to important features which improved the detection accuracy of the model.The effectiveness of the improved pedestrian detection network is verified on the Wider Persons dataset and the self-built subway pedestrian dataset.The experimental results show that the improved network can achieve better detection results.2.An abnormal behavior detection network fused with spatial context is presented.Based on the pedestrian detection network,Slow Fast is used to recognize abnormal behaviors such as falling,running,fighting,smoking,etc.Aiming at the problem that the Slow Fast detection network cannot make full use of the spatial context,an abnormal behavior detection network integrating spatial context is constructed.The spatial context network is used in the Slow Fast detection head,so that the network can combine the surrounding environment information to improve the discrimination ability of abnormal behavior.At the same time,non-local operation blocks are embedded in the spatial context network to improve the network’s ability to utilize spatiotemporal features.The experimental results show that the constructed detection network has stronger detection ability on the self-built abnormal behavior dataset.3.A system for detecting abnormal behavior of subway personnel is implemented.The system integrates an abnormal behavior detection algorithm,which can detect and display the access video.When an abnormality occurs,the system will issue an abnormal alarm and record the alarm information at the same time.Through the test,the system can operate normally and has a good detection effect.In this paper,ablation experiments are designed for pedestrian detection and behavior detection in abnormal behavior detection in subway stations to verify the effectiveness of the proposed detection network.Combined with the improved algorithm,a subway personnel abnormal behavior detection system is implemented to help improve the subway’s emergency management capabilities and operation scheduling capabilities.The work of this paper is supported and funded by the construction of the China-ASEAN International Joint Laboratory for Comprehensive Transportation,the Guangxi Science and Technology Base and the Talent Special Project.