Towards Automated 3D BIM Reconstruction of Existing Industrial Buildings from Point Cloud Data

Patricia González-Cabaleiro, Muataz S. A. Albadri, Antonio Fernández, L. Díaz−Vilariño

Abstract. This paper presents a comprehensive methodology for the automated semantic segmentation and 3D reconstruction of industrial building elements, including roof panels, floor, rafters, purlins, and columns, from unstructured point clouds. The proposed approach integrates orientation-based filtering, projection onto characteristic planes, morphological analysis, and optimization-based I-profile fitting to generate accurate 3D models. The workflow begins with point cloud preprocessing, where the data are aligned with the building axes and cleaned of outliers, followed by subdivision into two subsets based on local surface orientation. Binary projections are then processed to extract element contours, while roof slopes and panel inclinations are automatically estimated to guide the reconstruction of rafters and purlins. The method was validated on a real-case study of 930 m² industrial warehouse scanned with a mobile laser scanner, resulting in a raw dataset of seven million points. The segmentation achieved F1-scores above 0.90 for floors, roof panels, rafters, and columns, and 0.75 for purlins. Profile fitting yielded an average width error of 3.8%, confirming the robustness and reliability of the reconstruction across diverse structural components.