Detection of Obstacle Intrusion in Front of Train Based on Vehicle-borne LiDAR

Obstacle intrusion threatens the running safety of automatically-operated train seriously. This paper proposes a method to detect obstacle intrusion in front of the trains based on vehicle-borne LiDAR. Firstly, a composite rail track model is established from the track points extracted from the laser-point cloud, and the rail track parameters in XOZ and XOY planes are estimated based on the RANSAC algorithm. Next, the point cloud segmentation is conducted on the ground ,with the consideration of the analysis results on the radial slope and the rail track data generated by fitting, resulting in the obstacle-related point cloud that is clustered using DBSCAN with a dynamic neighborhood radius. Finally, a convex hull of obstacles is extracted for interpolation sampling, yielding a dense edge point set to be used to identify obstacle intrusion. This method was proven in the field onboard tests,with a detection accuracy of above 99.4% for track, and higher than 99.7% for trains, poles and towers, and pedestrians,