TCR-LM: Terrain confidence region-guided LiDAR mapping for open-pit mine loading operations

Abstract

Abstract Robust LiDAR mapping in open-pit mine shoveling and loading scenarios is challenging because point-cloud registration is affected by dynamic equipment occlusion, dust-induced returns, platform vibration, and continuously changing terrain. This study proposes TCR-LM, a terrain-confidence-region-guided LiDAR mapping method that improves the reliability of scan-to-map constraints in disturbed shoveling and loading scenarios. TCR-LM uses relatively stable terrain observations as the primary constraints for pose estimation, rather than directly matching raw point clouds contaminated by non-terrain objects. Terrain confidence regions are constructed through continuous terrain-envelope extraction and neighborhood-based confidence refinement to reduce the influence from trucks, buckets, mechanical structures, and dust-contaminated regions. Spatially balanced feature selection and unstable-feature removal near occlusion boundaries are then used to improve the spatial coverage and geometric reliability of the matching constraints. To handle time-varying terrain, local elevation changes are detected from temporal layers in the keyframe map, and outdated terrain points are progressively down-weighted through point-level state updating. Controlled simulation and single-site field experiments show that, under the evaluated conditions, TCR-LM achieves lower trajectory errors in simulation and better relative trajectory consistency in the field than the compared methods. Deployment on a field edge-computing platform indicates its feasibility for quasi-real-time, low-frequency operating-area map updating and candidate terrain-change perception.