Automatic detection of traces in 3D point clouds of rock tunnel faces using a novel roughness: CANUPO method

Trace detection in tunnel faces is critical for preventing failures such as spalling in tunnel excavations. This paper proposes a novel method, referred to as the Roughness-CANUPO (R–C) Method, for the automatic detection of these traces in 3D point clouds obtained from tunnel faces, in which its combines roughness analysis with the CANUPO machine learning algorithm to improve the accuracy and efficiency of trace detection. In this study, images were collected from six sites and point cloud data of 6 sites were obtained by 3D reconstruction. Points representing rock masses in the collected 3D point cloud data can be filtered by roughness analysis in the R–C methods. Roughness analysis can not only significantly reduce the analysis time of CANUPO machine learning, but also increase the accuracy of CANUPO analysis. In this paper, the local neighbourhood radius (LNR) and filtering ratio optimized in the roughness analysis and the roughness for filtering discontinuities were evaluated. The filtered point cloud data automatically extract only the trace point cloud by the CANUPO classifier. The extracted trace was projected onto the original data to accurately designate the location of the trace. Conducting the R–C method yields a significant result regarding discontinuity traces classification by reaching the level of accuracy ranging from 95.9% to 98.9%.