Abstract: The rapid development of digital photogrammetry and LiDAR technologies has made underground three-dimen sional data acquisition more convenient and efficient.The foundation for obtaining complete 3D real-world maps of mine tunnels lies in the high-precision registration of sequential point clouds,with the core being the estimation of spatial transformation pa rameters between point clouds to achieve unified spatial coordinates across multiple stations.However,the dusty and noisy envi ronment in underground excavation areas,along with deteriorating roadway structures,makes it difficult for existing point cloud registration methods to establish stable corresponding features,resulting in low registration accuracy and poor quality of 3D tun nel modeling.To improve the robustness and precision of point cloud registration in excavated roadways,a node association based algorithm (Robust-Fast Global Registration,R-FGR) is proposed.This algorithm first extracts point cloud features using Fast Point Feature Histograms (FPFH) descriptors,then employs a k-nearest neighbors matching strategy to obtain initial cor respondences among multi-station point clouds.It subsequently calculates the association degrees between matched points and extracts reliable candidate inliers based on these degrees.Finally,a robust loss function is constructed using progressive non convex optimization,and the global optimal registration result is determined iteratively by minimizing the objective function.A case study was conducted at the Dongshan Coal Mine in Heilongjiang,where various registration algorithms were compared.Re sults show that the proposed method achieves the highest robustness and registration accuracy,enabling automatic alignment of multi-temporal point clouds and effectively supporting 3D roadway mapping and dynamic updating tasks.

Key words: LiDAR,point cloudsregistration,optimizationmethod,point cloudsprocessing