The research on recognition of passenger abnormal behavior in urban rail transit stations based on conditional generative adversarial networks

With the exponential growth of urban rail transit passenger flow, manual monitoring for passenger abnormal behavior recognition is prone to missed detections and misjudgments due to visual fatigue. Developing rapid and accurate intelligent recognition method for passenger abnormal behavior has therefore become crucial for enhancing operational management and ensuring passenger safety. To address these challenges. First, a U-shaped generator network based on Swin Transformer architecture was constructed to overcome the spatial limitations of conventional convolutional kernels, thereby substantially enhancing the model's capability to extract global information from station surveillance video sequences. Second, the framework incorporates a Patch GAN local discriminator network combined with Lite FlowNet to rectify local and motion distortions in generated frame, thereby enhancing passenger behavioral modeling accuracy. Finally, passenger abnormal behaviors are recognized by computing the Peak Signal-to-Noise Ratio (PSNR) between original surveillance and generated frame. To validate the proposed method's performance, comparative experiments were conducted on both public datasets (UCSD Ped2, CUHK Avenue, Subway Exit) and our self-built PAB dataset. The proposed method achieved AUC values of 96.2 %, 86.6 %, 99.4 %, and 84.3 % respectively, demonstrating superior abnormal behavior recognition performance compared to baseline methods. Experimental results demonstrate that the proposed method achieves superior recognition accuracy and robust applicability for passenger abnormal behaviors in complex urban rail transit environments with long-narrow platform configurations and high passenger density conditions.