Improved DASP-guided normal vector refinement for automatic discontinuity identification in 3D rock mass point clouds

Abstract

The spatial orientation of rock discontinuities is a fundamental parameter for evaluating rock mass stability and designing engineering support systems. This study proposes an automatic discontinuity identification method for three-dimensional rock mass point clouds based on improved density analysis of stereographic projections (DASP)-guided normal vector refinement. An adaptive voxel partitioning strategy and an improved DASP algorithm are employed to automatically determine the number of clusters and the initial orientation centers for K-means clustering. Furthermore, a prior orientation-guided normal vector diagnosis and selective re-estimation strategy is developed to mitigate the influence of mixed normal vectors on discontinuity identification. Validation using both a publicly available dataset and an in situ tunnel-face point cloud demonstrates that the proposed method achieves mean absolute errors as low as 0.89° and 1.05° for dip direction and dip angle, respectively. The proposed approach enables efficient discontinuity identification and automatic extraction of orientation parameters from complex rock mass point clouds.