Railway Sleeper Segmentation Method Based on Train-Borne LiDAR Data

This is the authors' abstract. We don't add key points for this paper.

Addressing the issue of sleeper interference with ballast calculation, a sleeper segmentation method based on point cloud profile data and point cloud differential images is first proposed. The rapid segmentation method for point cloud profiles analyzes the geometric characteristics and distribution patterns of sleepers, shielding the sleeper area in the train-borne point cloud in real time to reduce errors in the calculation of surplus and deficient ballast. This rapid segmentation method maintains efficient and accurate segmentation performance even at a speed of 160 km/h. For higher segmentation accuracy requirements in offline analysis, this paper proposes a deep learning sleeper segmentation method based on point cloud differential images. Compared with traditional 3D segmentation models based on point cloud data annotation, this method maps 3D point clouds to differential images, nearly doubling the segmentation speed while maintaining essentially the same segmentation accuracy.