Efficient rock joint detection from large-scale 3D point clouds using vectorization and parallel computing approaches

Abstract

The application of three-dimensional (3D) point cloud parametric analyses on exposed rock surfaces, enabled by Light Detection and Ranging (LiDAR) technology, has gained significant popularity due to its efficiency and the high quality of data it provides. However, as research extends to address more regional and complex geological challenges, the demand for algorithms that are both robust and highly efficient in processing large datasets continues to grow. This study proposes an advanced rock joint identification algorithm leveraging artificial neural networks (ANNs), incorporating parallel computing and vectorization of high-performance computing. The algorithm utilizes point cloud attributes—specifically point normal and point curvatures—as input parameters for ANNs, which classify data into rock joints and non-rock joints. Subsequently, individual rock joints are extracted using the density-based spatial clustering of applications with noise (DBSCAN) technique. Principal component analysis (PCA) is subsequently employed to calculate their orientations. By fully utilizing the computational power of parallel computing and vectorization, the algorithm increases the running speed by 3–4 times, enabling the processing of large-scale datasets within seconds. This breakthrough maximizes computational efficiency while maintaining high accuracy (compared with manual measurement, the deviation of the automatic measurement is within 2°), making it an effective solution for large-scale rock joint detection challenges.