A Tunnel-Axis-Guided Structure-Aware ICP Algorithm for Mine Roadway Point Cloud Registration

Realistic 3D modeling is an important foundation for mine roadway safety management, engineering design, operation and maintenance, and digital twin construction. A tunnel-axis-guided structure-aware ICP algorithm (TA-ICP) is proposed for mine roadway point clouds with repetitive cross sections and sparse local features. The method constructs PCA-based candidate initial poses, estimates the roadway principal axis, maps the relation between target normals and the axis into a tunnel-axis-guided cross-sectional structural weight, and combines this weight with a Cauchy robust kernel, normal consistency with source normals updated by the current rotation, curvature reliability, and multi-scale point-to-plane optimization. Experiments include known-pose perturbations, end-to-end comparison on eight same-mine sections, fair comparison under a common PCA initialization, ablation, parameter sensitivity, runtime, and robustness tests under noise, outliers, occlusion, and low overlap. TA-ICP maintains a 100% success rate under 10 to 60 degrees of rotational perturbation. On eight measured sections, its 95%-trimmed surface-consistency RMSE is 7.474 +/- 4.721 mm, comparable to FPFH-FGR followed by ICP, whereas unit-initialized ICP and GICP show severe misregistration under large initial offsets. Under a common PCA initialization, differences among local optimization variants are not significant, indicating that the present end-to-end gain mainly comes from reliable initialization rather than a single local weight. The conclusions are limited to same-mine roadway segments with a clear principal axis and sufficient overlap.