Jetson Nano-Based Subway Station Area Crossing Detection

This paper aims to address the dangerous behavior of passengers crossing the platform cordon in subway stations. It presents a subway station area crossing detection algorithm based on the YOLOv5s deep learning algorithm. In this paper, the network structure of the YOLOv5s algorithm, the detection principle, and the training process are studied in detail, and 4000 pedestrian pictures are collected, labeled using the labimg tool, and made into a dataset that conforms to the format of YOLOv5s algorithm, and in order to facilitate the installation, the algorithm is ultimately deployed to the embedded end of the jetson nano, which is collected and processed in real-time by the CSI camera, to meet the needs of detection speed and portable installation of equipment in practical applications, and use TensorRT for inference acceleration, to improve the FPS value from 7 to 15, while using the Mask mask function in openCV, to achieve efficient and accurate detection of regional transgression behavior.