LiDAR-Based Obstacle Detection Using Trajectory Maps

Addressing the relatively singular operational scenarios of heavy-haul trains and the pressing demand for high-precision environmental perception, this paper proposes a LiDAR-based obstacle detection method utilizing a trajectory map. The method employs a two-stage processing framework: firstly, a high-precision trajectory map is constructed offline to provide accurate prior knowledge for online real-time perception; subsequently, during the real-time operation phase, the system deeply integrates 3D point cloud data acquired by LiDAR with centimeter-level positioning information provided by RTK. Through a series of algorithmic modules, including data preprocessing, railway centerline extraction and vehicle gauge generation constrained by the trajectory map, ground segmentation, object clustering, and feature-based object classification, comprehensive and accurate perception of the railway environment is achieved. Test results from actual railway scenarios demonstrate that the system can effectively and stably detect a dummy at 80 m and a train carriage at 150 m, and can effectively distinguish different types of targets such as pedestrians, trains, and objects outside the gauge. This provides robust support for assisting safe train operations.