Evaluation of smartphone photogrammetry for 3D surface roughness computation

Rock surface roughness plays an important role in excavation engineering projects and applications related to surface support structures, such as shotcrete and thin-sprayed liner (TSL). The morphology of the substrate affects the mechanical behaviour at the interface significantly. Therefore, it is necessary to quantify the roughness to investigate the relationship between roughness and mechanical properties. The joint roughness coefficient (JRC) proposed by Barton and Choubey (1977) is widely used in practice. It initially estimated the roughness by comparing it to 10 standard profiles visually. However, this method has lots of limitations, it is not only subjective but also a two-dimensional parameter with a small lab scale (Beer, 2002). The subsequent research on JRC, from visual comparison with Barton’s standard profiles to measuring the roughness with a profilometer (Weissbach, 1978; Alameda-Hernández et al, 2014). Then, 2D profiles or 3D surfaces by different technologies with higher resolution and accuracy, such as 3D laser or optical scanners (Fardin, 2004; Jiang, 2020, 2021). These 3D scanners are expensive and only suitable for laboratory scales. LiDAR scanners (Lato, 2009; Aubertin, 2022) have been applied in industrial size, but it is difficult to achieve a millimetre-level accuracy, which is mostly used to detect the thickness of shotcrete in practice. Thus, finding a cost-effective surface roughness measurement method with high accuracy is necessary. Photogrammetry is another method to digitalise the roughness of the surface. Similar to 3D scanners, photogrammetry also generates the point cloud of the objective surface and then reconstructs the 3D model of the objects. 3D reconstruction technology is a trendy technology in recent years. The structure from motion (SfM) algorithm (Westoby, 2012) to rebuild the 3D model with a flexible workflow, aligning photos, building dense point cloud, mesh and texture, which has been proved that can rebuild an accurate model in variable fields.