Smart point cloud–guided 3D gaussian splatting for urban modeling and data-driven analysis
Yingwen Yu (TU Delft - Architecture and the Built Environment)
Zhuoyue Wang (Student TU Delft)
Guanting Zhang (Nanjing Tech University)
Peter van Oosterom (TU Delft - Architecture and the Built Environment)
Edward Verbree (TU Delft - Architecture and the Built Environment)
Steffen Nijhuis (TU Delft - Architecture and the Built Environment)
Yuyang Peng (TU Delft - Architecture and the Built Environment)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Urban analysis often relies on separate geospatial mapping, street-level image analysis, point-cloud analysis and 3D urban modeling workflows, which can lead to semantic and spatial inconsistencies when structural, visual and volumetric results are interpreted together. This paper develops an SPC-guided Smart 3D Gaussian Splatting (S3DGS) analytical framework to address this fragmentation. Using the Aula–library block in Delft as a case study, we construct a CityGML-inspired Smart Point Cloud from laser-scanning data, generate street-level and aerial 3DGS components from multi-view imagery, align and fuse them into a common coordinate frame, and propagate SPC-derived semantics to Gaussian primitives. The resulting S3DGS base supports projection-based semantic composition, observer-centered visual exposure, volumetric spatial-presence aggregation and layered scene typing within the same reference. Cross-layer interpretation links volumetric and visual-exposure types in the same spatial units, revealing conditions such as physically present but visually recessive building mass. This layered reading allows surface composition, visual experience and three-dimensional spatial presence to be traced within the same urban unit. Evaluation results indicate a stable workflow, with an alignment RMSE of 0.0439 m and a kNN label consistency of 92.24%. By turning Gaussian scenes into semantically traceable and spatially co-referenced analytical bases, S3DGS offers a practical pathway from isolated urban measurements toward integrated, viewpoint-aware and three-dimensional interpretation of the built environment.
help_outline