An improved multi-rule region growing method for point cloud segmentation of rock structural planes

Abstract. Accurately and efficiently identifying rock mass structural planes and extracting their characteristic information is crucial for rock mass stability assessment. Three-dimensional (3D) laser scanning technology can significantly enhance both the efficiency and accuracy of structural plane surveying; however, current mainstream point cloud segmentation algorithms exhibit notable shortcomings, including blurred recognition of structural plane edges, insufficient segmentation accuracy, and poor integration precision among segmented blocks. To address these problems, this study proposes an improved multi-rule region growing point cloud segmentation method for rock structural planes. Specifically, plane fitting residuals are calculated from the point cloud data, and these residual values are then used to optimize seed point selection, thereby improving the segmentation accuracy of planar point sets. Next, considering the spatial relationship between the location of rock structural plane point clouds and their neighborhoods, a KD-tree data structure is employed to perform voxel downsampling for nearest-neighbor searching, and the RANSAC-based region growing algorithm is further refined. By adjusting the region growing segmentation parameters using multiple feature values and segmenting structural planes based on point cloud normal vector differences and final feature values, the proposed method facilitates the extraction of structural plane orientation, spacing, and extent, improving the overall segmentation quality. Experimental results demonstrate that the error between the segmented rock structural plane area and dimensions obtained by this method and those computed using CAD is only 1.07%, which meets the engineering error tolerance. Consequently, the proposed method provides stable and effective technical support for the identification and segmentation of rock structural planes.