An unsupervised method for rock discontinuities rapid characterization from 3D point clouds under noise

Abstract

The presence of rock discontinuities significantly affects the strength, permeability, and stability of rock masses, so it is crucial to characterize them accurately and efficiently. However, existing discontinuity characterization methods mainly focus on improving automation level and computational efficiency and pay little attention to the effect of noise, which may lead to unsatisfactory results. To address this, we present an unsupervised method to characterize rock discontinuities. The method contains five denoising operations which greatly mitigate the effects of noise while mapping discontinuities rapidly. The proposed approach comprises four steps: (1) remote sensing technology used to collect point cloud data; (2) a Modified Mean Shift algorithm (MMS) including five denoising operations developed to identify group discontinuity; (3) a density-based cluster algorithm employed to detect individual discontinuity; and (4) plane fitting algorithm applied to calculate the orientation of each discontinuity. The working procedure was displayed using a regular cube, and the proposed approach was validated on two real road-cut slopes. The average error degrees in dip angle and dip direction are within 3° for both cases, indicating the excellent accuracy of the presented method. Then, a sensitivity analysis of the parameters in the MMS was conducted, and some recommendations for the selection of these parameters were given (ηmax was suggested to be 0.02–0.2, while R was recommended to be 0.2–0.3). Finally, the computational efficiency of the presented method was compared with the other two algorithms, and the results show that the computational efficiency was improved by 13 and 30 times respectively, which proves that it is computationally efficient. This approach gives geological and geotechnical engineers a new solution to rapidly map rock structures under noise, which is significant for improving the accuracy of rock characterization.