Abstract: Passenger density analysis is an important foundation for subway operation management, ensuring passenger safety, and building a passenger flow big data platform. This paper designed a passenger flow density analysis system based on the improved YOLOv5s model in response to the operator's demand for station and train carriage passenger flow density management. The system improved the YOLOv5s object detection model, introduced attention mechanism, and improved the backbone network structure to maintain model lightweight while improving crowd detection accuracy and inference speed. The test application of the passenger information system project based on the Beijing New Airport Line shows that the system has fast recognition speed and high analysis accuracy, which is conducive to the comprehensive supervision of passenger flow by subway operations.