Lidar-based foreign object detection for railway intrusion under adverse weather conditions

This paper proposes a LiDAR-based foreign object detection method designed for railway intrusion under complex and adverse weather conditions. To improve detection accuracy and environmental adaptability, a real-time point cloud processing framework is developed, supporting high-frequency, long-distance monitoring. The method introduces a spatial modeling strategy and multi-feature filtering to effectively suppress noise interference from heavy rain and other harsh conditions. To further enhance detection confidence, a loose fusion approach with visual data is adopted. Field experiments conducted at the National Railway Test Center and simulated intrusion testing fields demonstrate that the method can achieve zero missed detections and extremely low false alarm rates, even in scenarios with overlapping disturbances such as heavy rain and strong lighting. These results confirm the method’s robustness and feasibility for practical deployment in railway safety monitoring.