Extraction of 3D structural data from Virtual Outcrop Models: problems and best practices.

The rapid improvements of computer vision–based photogrammetric image processing pipelines (i.e., Structure from Motion–Multi View Stereophotogrammetry: SfM-MVS), coupled with the availability of various low-cost and portable acquisition tools, such as Digital Single-Lens Reflex (DSLR), mirrorless cameras, Unmanned Aerial Vehicle (UAV) and even smartphones, have revolutionized outcrop studies in structural geology and have brought traditional field geology into the digital age. This has had a transformative impact on Virtual Outcrop Models (VOMs), which have been promoted from mostly visualization media to fully interrogable quantitative objects. Among the several applications of VOMs in structural geology, extraction of near planar features (e.g., fracture and bedding surfaces) is one of the most important. Various procedures aimed at this purpose exist, spanning from fully automated segmentation and best fitting of point clouds to the manual picking of 3D polylines on both point clouds and textured meshes.

Here we illustrate the pros and cons, best practices, and drawbacks of the main procedures for near planar geological data extraction from VOMs. While automated or supervised recognition and subsequent best-fitting of coplanar patches in point clouds has received remarkable attention, its application generally limits to rare case studies. Indeed, most commonly, geological outcrops do not expose patches of near planar surfaces which are large enough to carry out a robust best fitting, and the structural interpretation of the outcrop only permits manual picking procedures. In the latter case, the use of textured meshes must be preferred to point clouds, and during digitization the accuracy of the textured mesh must be considered, as well as the intrinsic roughness of any geological surfaces. The analysis of coplanarity and collinearity of the picked pointsets may help in identifying traces that diverge from idealized (low) collinear and (high) coplanar configurations. However, typically suggested threshold values often produces small datasets. Nonetheless, the goodness of the extraction of data based merely on the visual inspection of the best-fit plane, handling coplanarity and collinearity in real-time through live computation of best-fit planes from picked pointsets, is often acceptable.