Evaluation of Wavelet and Non-Local Mean Denoising of Terrestrial Laser Scanning Data for Small-Scale Joint Roughness Estimation

Abstract

ABSTRACT:Terrestrial Laser Scanning (TLS) is a well-known remote sensing tool that enables precise 3D acquisition of surface morphology from distances of a few meters to a few kilometres. The morphological representations obtained are important in engineering geology and rock mechanics, where surface morphology details are of particular interest in rock stability problems and engineering construction. The actual size of the discernible surface detail depends on the instrument range error (noise effect) and effective data resolution (smoothing effect). Range error can be (partly) removed by applying a denoising method. Based on the positive results from previous studies, two denoising methods, namely 2D wavelet transform (WT) and non-local mean (NLM), are tested here, with the goal of obtaining roughness estimations that are suitable in the context of rock engineering practice. Both methods are applied in two variants: conventional Discrete WT (DWT) and Stationary WT (SWT), classic NLM (NLM) and probabilistic NLM (PNLM). The noise effect and denoising performance are studied in relation to the TLS effective data resolution. Analyses are performed on the reference data acquired by a highly precise Advanced TOpometric Sensor (ATOS) on a 20x30 cm rock joint sample. Roughness ratio is computed by comparing the noisy and denoised surfaces to the original ATOS surface. The roughness ratio indicates the success of all denoising methods. Besides, it shows that SWT oversmoothes the surface and the performance of the DWT, NLM and PNLM vary with the noise level and data resolution. The noise effect becomes less prominent when data resolution decreases.

Potential and Limitations of Terrestrial Laser Scanning for Discontinuity Roughness Estimation

“…To improve surface roughness estimation, previous researchers investigated a transforming domain filtering method, namely the DWT (Bitenc et al, 2015a(Bitenc et al, , 2015b(Bitenc et al, , 2019Khoshelham et al, 2011), and a spatial (pixel) domain filtering method, namely the Non-Local Mean (NLM) (Bitenc et al, 2016;Smigiel et al, 2011). Prior research has focused on denoising in a direction perpendicular to the best fit plane.…”