UAV close-range photogrammetry combined with adaptive octree-based point cloud filtering for rock joint extraction of open-pit mine slopes

Abstract Manual joint surveys on open-pit mine slopes are high-risk and inefficient, whereas conventional remote sensing and laser scanning are costly and limited in fine-crack detection. To enable low-cost, high-precision, noncontact structural plane interpretation, this study targets the southern slope of the Shitoumei No. 1 open-pit coal mine and presents a joint recognition method that integrates unmanned aerial vehicle (UAV) close-range photogrammetry with an octree-based adaptive statistical outlier removal (SOR) filter using two-layer locality-sensitive hashing (2 L-LSH). High-overlap close-range images from real-time kinematic (RTK)-equipped UAVs are reconstructed into centimeter-scale dense point clouds. An octree-based multiscale density-driven adaptive filter, combined with 2 L-LSH for accelerated nearest-neighbor search, resolves the noise‒crack preservation trade-off caused by uneven point–cloud density and enhances computational efficiency. Four clustering algorithms, namely, discontinuity set extractor (DSE), K-means, region growing, and fuzzy C-means (FCM), were compared; FCM was optimized to semiautomatically extract five dominant joint sets with orientations, trace lengths, and spacings. Compared with the conventional SOR filter, the proposed method achieves a 188-fold increase in computational speed and yields a 126% improvement in comprehensive filtering performance. The optimized FCM achieves a mean orientation deviation of 3.2° and a misclassification rate of 3.3%. This workflow enables safe and efficient acquisition of rock-mass joint geometric parameters, as demonstrated for the investigated Shitoumei No. 1 open-pit mine site.