Automatically enhancing CityGML LOD2 models with a corresponding indoor geometry

The international standard CityGML defines five levels of detail (LODs) for 3D city models, but only the highest of them (LOD4) supports modelling the indoor geometry of a building, which must be acquired in correspondingly high detail and therefore at a high cost. Whereas simple 3D city models of the exterior of buildings (e.g. CityGML LOD2) can be generated largely automatically, and are thus now widely available and have a great variety of applications, similarly simple models containing their indoor geometries are rare. In this paper we present two contributions: (i) the definition of a level of detail LOD2+, which extends the CityGML LOD2 specification with indoor building geometries of comparable complexity to their exterior geometries in LOD2; and more importantly (ii) a method for automatically generating such indoor geometries based on existing CityGML LOD2 exterior geometries. We validate our method by generating LOD2+models for a subset of the Rotterdam3Ddataset and visually comparing thesemodels to their real counterparts in building blueprints and imagery fromGoogle Street View and BingMaps. Furthermore, we have used the LOD2+ models to compute the net internal area of each dwelling and validated our results by comparing these values to the ones registered in official government datasets.