Imagine you have the original LEGO blueprint for a building (the BIM model) and a detailed 3D scan of what was actually built (the point cloud). This research builds a smart “snap-and-align” system that automatically lines up the real-world 3D scan with the digital blueprint so they match perfectly inside a construction digital twin.
On large projects, teams struggle to accurately compare as-designed BIM models with as-built reality from 3D laser scans. Manual alignment of point clouds to BIM is slow, expert-heavy, and error-prone, which delays progress tracking, quality control, and clash detection. This work proposes an automated, robust registration method that reliably aligns local point cloud segments to BIM models, enabling practical, up-to-date construction digital twins.
Domain-specific registration algorithm that leverages hybrid visibility map encoding for robust alignment of real-world scans to BIM; hard to replicate without access to construction datasets and geometric expertise.
Classical-ML (Scikit/XGBoost)
Unknown
High (Custom Models/Infra)
Processing very large, noisy point clouds at building or site scale in a reasonable time while maintaining accuracy of alignment to complex BIM models.
Early Adopters
Focuses specifically on robust local registration between point clouds and BIM using hybrid visibility map encoding, tailored for construction digital twin workflows rather than generic 3D registration, making it better suited for real project conditions with occlusions and partial scans.