A Handheld LiDAR-Based Semantic Automatic Segmentation Method for Complex Railroad Line Model Reconstruction

To ensure efficient railroad operation and maintenance management, the accurate reconstruction of BIM models is a crucial step. This paper proposes workflow for automated segmentation structures using point cloud data, without relying on intensity or trajectory information. The consists four main components: adaptive denoising, scene segmentation, structure combined with deep learning, model reconstruction. proposed was validated two datasets significant differences in line data. results demonstrated improvements both efficiency accuracy compared to existing methods. techniques enable direct processing from raw data results, providing support parameterized modeling greatly reducing manual time. algorithms achieved an intersection over union (IoU) 0.9 various 450-m-long line. Furthermore, single-track railroads, time within 1 min per kilometer, average mean (MIoU) 0.9518 1.0000, respectively.