Semi-automated indoor geometry reconstruction for daylight simulation

This study presents a semi-automatic pipeline for reconstructing indoor geometries from point cloud data for daylight simulation. The pipeline generates watertight models of permanent architectural surfaces with window boundaries through three steps: (1) preprocessing, (2) permanent structure reconstruction, and (3) window boundary extraction. The pipeline was evaluated in four rooms of varying complexity against manually reconstructed models using visual and geometrical comparisons, and by comparing daylight availability and glare metrics calculated from Radiance results. Geometric deviations in terms of Chamfer Distance (CD) were found to be smaller than one centimetre for regular rooms and up to 51 cm for complex rooms. Across annual daylight metrics, sDA and ASE showed less than 6 % bias, while exhibited spatial biases ranging from approximately to , and the mean TAI bias remained below in all rooms. The Daylight Glare Probability (DGP) error remained under 4 %, and the modelling time did not exceed 5 min for any scenario. The approach enables rapid generation of simulation-ready models with acceptable accuracy for Climate-based daylight modeling (CBDM). • Semi-automated pipeline for window-containing watertight indoor surface models from point clouds • Reduced user interaction to inform reconstruction algorithm • Mean TAI bias of less than across all rooms • sDA bias of less than and ASE bias of less than • Errors within 4 % in glare prediction