Abstract: In order to improve the safety and efficiency of railway station yard shunting operations, this paper designed a highly reliable foreign object intrusion detection system for railway shunting operation, and described the overall architecture, data exchange architecture, and functions of the system to implement accurate identification and real-time warning of personnel, locomotives, and obstacles in complex scenes of the yard, delved into key technologies such as the Real Time Detection Transformer (RTDETR) model and detection process, and conducted field application experiments. The experimental results show that the recognition rate of the system for train sets and personnel reaches 95% and 96% respectively, which has better small object detection performance than YOLO (You Only Look Once) v8 and other algorithms, provides a technical basis for the intelligence of railway shunting. In the future, it is planned to integrate laser radar technology to enhance robustness in harsh environments.