This paper presents a real-time identification method for abnormal behaviors of passengers in metro ve-hicles based on deep learning,aiming at realizing real-time monitoring in edge computing environment.Under the constraint of edge computing,an optimized YOLO model is adopted,which combines the Transformer module to en-hance the ability to capture behavior characteristics and achieve static information detection.Furthermore,Openpose and LSTM models are used to process key point distribution information of multi-frame images,so as to realize dy-namic behavior detection.The system directly obtains real-time video from monitoring cameras for analysis,and gives alarm for any passenger's abnormal behavior,providing effective support for safety monitoring.The system a-chieves a 94.6％detection accuracy and a detection speed of 9.23 frames per second in the test dataset,which proves its practicability and efficiency in the resource-limited edge computing environment and provides theoretical and practical reference for the construction of intelligent detection system for metro vehicles.
关键词
深度学习/行为识别/边缘计算/目标检测
Key words
deep learning/behavior identification/edge computing/target detection
维普
万方数据