Research on 3D convolutional autoencoder enhanced metro abnormal behavior detection based on multi-level attentional memory

Research on 3D convolutional autoencoder enhanced metro abnormal behavior detection based on multi-level attentional memory

연구 분야: Verification

학회: Multimedia Tools and Applications

Subway is one of the most important rail transit tools in China, which has the advantages of convenience, safety and high efficiency, and subway has also become the main means of transportation for people to travel. However, the subway scene has the characteristics of narrow space and large passenger flow, and abnormal behavior events often occur in the subway scene during passenger rush hours. At present, the abnormal behavior detection methods of manual monitoring in subway scenes have been unable to meet the increasing demand of passenger traffic. In this paper, pedestrian abnormal behavior detection in subway scene is studied, and a self-encoder abnormal behavior detection method based on channel attention mechanism and multi-level memory enhancement is proposed. It solves the problems of excessive generalization ability of traditional convolutional self-encoders and difficulties in extracting pedestrian behavior features under complex subway background, which is verified by experiments. Finally, better performance has been achieved in UCSDPed2 dataset, CUHK Avenue dataset and Chengdu Metro self-built dataset.