Journal of Information Technology in Construction
Journal of Information Technology in Construction
ITcon Vol. 30, pg. 1099-1122, http://www.itcon.org/2025/45
Automated Steel Structure Model Reconstruction through Point Cloud Instance Segmentation and Parametric Shape Fitting
| DOI: | 10.36680/j.itcon.2025.045 | |
| submitted: | March 2025 | |
| revised: | June 2025 | |
| published: | July 2025 | |
| editor(s): | Robert Amor | |
| authors: | Florian Noichl, M.Sc.
Chair of Computing in Civil and Building Engineering, TUM Georg Nemetschek Institute Technical University of Munich, Germany ORCID: https://orcid.org/0000-0001-6553-9806 florian.noichl@tum.de Yuandong Pan, Dr.-Ing. Department of Engineering, University of Cambridge, United Kingdom ORCID: https://orcid.org/0000-0002-5331-6901 yp296@cam.ac.uk André Borrmann, Prof. Dr.-Ing. Chair of Computing in Civil and Building Engineering, TUM Georg Nemetschek Institute, Technical University of Munich, Germany ORCID: https://orcid.org/0000-0003-2088-7254 andre.borrmann@tum.de | |
| summary: | This paper presents an automated method for converting laser-scanned point cloud data of steel structures into accurate digital 3D models. Point cloud data from laser scanning in such environments typically contains gaps, occlusions, and noise that complicate precise digital reconstruction of complex steel frameworks. Our approach addresses these limitations by combining geometric feature analysis with skeleton-based topology preservation. The method identifies individual steel beam instances within the point cloud, determines precise beam orientations through iterative model-based algorithms, and reconstructs occluded sections using skeletal representations. Cross-sectional profiles are matched to standardized catalogs through multi-objective optimization, generating complete 3D models in IFC format. Validation on industrial point cloud data of I-beam steel structures demonstrates high accuracy, achieving a mean angular precision of 0.1° for beam orientation and a mean geometric deviation of 4.1 mm between source data and reconstructed models. The method maintains robust performance across varying point densities and partial occlusions. This technology addresses critical needs in the construction and manufacturing industries. Potential applications include automated as-built documentation, construction quality control, retrofit planning for existing structures, and generation of digital models for infrastructure management. The automated processing eliminates manual interpretation bottlenecks in point cloud workflows, reducing processing time significantly. | |
| keywords: | Scan-to-BIM, point cloud, material inventory, instance segmentation, multi-objective optimization, procedural geometry, model reconstruction | |
| full text: | (PDF file, 1.686 MB) | |
| citation: | Noichl F, Pan Y, Borrmann A (2025). Automated Steel Structure Model Reconstruction through Point Cloud Instance Segmentation and Parametric Shape Fitting, ITcon Vol. 30, pg. 1099-1122, https://doi.org/10.36680/j.itcon.2025.045 | |
| statistics: |