Digital modeling layer: two digital modeling methods

Two digital modelling methods will be discussed in detail: Part 1 describes scan-to-BIM approaches and Part 2 introduces image-to-BIM approaches. For Part 1, to create 3D digital models, scan-to-BIM (i.e., a process of taking 3D scan data as input and outputting as-built building information models (BIMs)) is a popular choice, as the output (BIMs) greatly facilitates communication and information sharing for building facilities and the input (3D scan data) can be acquired rapidly and is capable of providing very accurate 3D measurements of the building environment. However, the scan-to-BIM process (i.e., creating as-built models from point clouds) remains a challenging problem, as it is either a manual process or requires developers to handcraft complex feature descriptors for interpreting point clouds automatically, which is time-consuming and inefficient. Recently, deep learning (DL) has undergone enormous progress and shown promising performance in various 2D and 3D computer vision tasks, such as object classification and semantic segmentation, enabling machines to be able to reason and interpret 3D point clouds more accurately and robustly. Inspired by the great success of DL, this chapter formulates the scan-to-BIM process as a five-step workflow under the new deep learning paradigm. Concretely, our scan-to-BIM workflow consists of five salient subprocesses: (a) scan planning and modelling requirement analysis; (b) reality capture for data collection; (c) point cloud processing to achieve automatic 3D scene understanding on point clouds; (d) digital modelling to achieve BIM models, and (e) BIM auditing to verify information model compliance. Part 1 contributes to automated 3D scene understanding of point clouds for the construction domain, as well as the as-built BIM creation from real-world 3D point clouds. For Part 2, photogrammetry is one of the most widely adopted approaches in geometric modelling (i.e., digital twinning). Despite the emergence of technologies in digital twinning based on photogrammetry, there is still no systematic procedure to elaborate and streamline them step by step. This part will survey state-of-the-art photogrammetry-based approaches and discuss their applications to digital twinning, following the logistics of working in five steps, that is, (a) field data collection; (b) data preprocessing, including 3D point generation; (c) geometric modelling; (d) object recognition, and (e) relationship modelling. In addition, some useful commercial software for digital twinning has been introduced. Finally, we identify and discuss the research gaps waiting to be resolved in the future. This chapter will support the development of a digital modelling layer to create a digital twin, and be beneficial to practitioners when evaluating and selecting suitable approaches and tools for geometric digital twinning.