Revealing the hidden framework: Reconstructing occluded structural elements from indoor point clouds

This paper presents a hybrid geometry- and knowledge-driven methodology for the detection and reconstruction of structural elements from point clouds in residential buildings, including visible, partially occluded, and completely occluded elements. The approach combines the segmentation of visible and partially occluded elements with geometric fitting and building knowledge to support the inference of completely occluded elements through inter-storey continuity and intra-storey alignment, followed by the reconstruction of all identified and inferred elements using the IFC schema and Swept Solid Representation (SSR). Validation on simulated and real-world case studies showed that directly detectable elements were identified at rates of up to 100% for columns and 83.3% for beams, with cross-section fitting errors below 0.5 cm for columns and 1.8 cm for beams. The inference stage recovered 78% of columns and 51% of beams, with mean localization errors below 4 cm across all case studies.