Answers are grounded in this paper and cite their sources.
Building Information Modelling (BIM) is central to digital twin applications in the Architecture, Engineering, and Construction (AEC) sector, yet the automated generation of accurate as-built BIM models from point cloud data—commonly referred to as Scan-to-BIM—remains a major challenge. This paper presents a systematic review of automated Scan-to-BIM research published between 2022 and 2026, analysing 25 selected studies through a proposed four-layer functional framework comprising Data Preparation, Segmentation, Geometric Abstraction, and BIM Object Creation. For each layer, representative automation strategies are identified, levels of human intervention are assessed using explicit coding criteria, and key bottlenecks are discussed. Results show that segmentation, geometric abstraction, and BIM object generation are frequently automated, whereas data preparation remains the least automated and least reported layer; only 4 of the 25 studies reported high automation in all four layers. Critical barriers include sensitivity to occluded or noisy data, dependence on predefined family libraries, insufficient extraction of non-geometric attributes, and the loss of spatial topology during instance-level processing. Based on these findings, four priority research directions are proposed: semantics-aware data preparation, learning-based point cloud completion, multi-sensor fusion for multi-dimensional building information, and graph- or Large Language Model (LLM)-assisted topology reconstruction. The review provides a structured lens for developing practical, end-to-end automated Scan-to-BIM pipelines for geospatial digital twins.
No takes yet. Share an insight, caveat, or question.
Yuan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: