Research Information System
in: Advances in Geoinformation Sciences, Alias Abdul-Rahman,Ivin A. Musliman,Imzan Hassan,Ainn Zamzuri, Editor, Springer, London/Berlin , Basel, pp.41-49, 2025
This paper presents a methodology for enhancing the segmentation of 3D building models using the BuildingGNN method and converting these segmented outputs to CityGML, the standardised format for 3D city models. The process involves leveraging a pre-trained BuildingGNN model to segment various components of 3D building models, such as walls, roofs, windows, and doors. Key challenges addressed include georeferencing the segmented models using affine transformations and ground control points (GCPs) to ensure real-world positioning and geometry validation to comply with CityGML specifications. The segmented models were converted into 3D tiles to visualise their integration within a geographic context, demonstrating their location and orientation on a map. This approach provides a practical means for urban planning and geographic information systems by delivering precise and semantically rich 3D building models. The paper discusses the methods used, the challenges encountered, and the solutions implemented to enhance the conversion process.