Railway safety is at the core of reform and development, with a particular emphasis on the technological and intelligent construction of intermediate stations. Strengthening the intelligent operation and safety risk control of these intermediate stations aims to improve transportation efficiency, reduce costs, and enhance transportation safety. Currently, there is a lack of effective control measures to prevent unauthorized individuals from entering critical railway areas, especially individuals climbing on trains. Their high concealment makes them difficult to detect, posing a serious threat to train operations and personal safety. Therefore, the use of machine vision technology to monitor individuals climbing on trains at railway intermediate stations is crucial. This paper is based on the YOLOv9 machine vision technology, introducing the Pconv convolution in the backbone network and designing the C2faster module. This approach maintains network accuracy while reducing model complexity. We suggest using the SPP module as a spatial pyramid pooling module to better utilize information between low-level and high-level features, strengthen regional context connections, and construct a warning system for monitoring individuals climbing on trains at railway intermediate stations. Experimental results show that on a self-built dataset, the algorithm achieves an accuracy rate of 85.3%, a recall rate of 66.4%, and a mAP@0.5 of 82.1%. The warning system in this paper can effectively identify individuals climbing on trains, reduce false alarms and missed events in complex environments, and effectively safeguard personal, property, and operational safety in critical railway areas. This promotes the high-quality development of railway intermediate stations.