From single buildings to cities: accurate LOD modelling from tiled airborne cross-source point clouds.

City-scale 3D building modelling is essential for understanding complex cities, but it remains difficult due to heterogeneous data sources. The process is particularly challenging with cross-sourced point clouds and processing them in tiles. Differences in density and noise between LiDAR and photogrammetry, combined with tile boundaries that cut through buildings, often lead to incomplete or inconsistent models.
In this study, we extend the plane-based reconstruction method originally designed for single buildings to work at a city scale. We propose a workflow that handles tiles intelligently. By using buffered processing and clustering, we ensure that buildings spanning multiple tiles are reconstructed completely. We also introduce a strategy to assign each building to a single tile, which avoids duplicates and keeps the process scalable. We evaluated this approach in dense urban areas with diverse building types. The results show that the method generates consistent models across tile boundaries while maintaining high geometric accuracy. This framework supports automated modelling of large areas and provides a solid foundation for analyzing complex built environments.