Semantically Enriched Voxels as a Common Representation for Comparison and Evaluation of 3D Building Models

3D city models are fast becoming a key factor for planning, design and analysis in the AEC industry, and urban informatics. Data are being acquired from a variety of sources and each consequent representation contains inherent information about its location, geometry, structure and usability. These digital representations can range from unstructured models with very little semantic data such as 3D point clouds, and 3D mesh models or they can be highly structured and semantically enriched models following CityGML and IFC/BIM standards. Matching these diverse representations against one another can facilitate information flow and eventually lead to a coherent interconnected Digital twin. Consistency measures would be needed to compare one representation against another. For comparison of any kind, a common baseline needs to be established. A tentative approach, as outlined in this paper, is to convert all the model types into a common representation such as ‘volumetric pixels’ or ‘voxels’ but the concept of ‘voxel’ as we know it, is not enough to deal with the rich semantic and organizational structure of modelling representations such as IFC and CityGML. Voxelisation is a complex process and conventional conversion methods concentrate on translating the geometry between representation types but the semantic and class hierarchy information is usually not translated. This paper assesses the need for matching 3D models using a common representation (voxels), discusses the challenges of the voxelisation process and proposes the concept of a ‘RichVoxel’.     «

3D city models are fast becoming a key factor for planning, design and analysis in the AEC industry, and urban informatics. Data are being acquired from a variety of sources and each consequent representation contains inherent information about its location, geometry, structure and usability. These digital representations can range from unstructured models with very little semantic data such as 3D point clouds, and 3D mesh models or they can be highly structured and semantically enriched models...     »