Connecting digital and physical representations through semantics and geometry

Abstract: The fields of geodesy and building information modeling (BIM) meet each other in the intersection between the physical and the digital world. Within the construction industry, the role of geodesy has typically been to describe the position of assets and to transform the geometries of those assets between coordinate systems suitable for design and coordinate systems with a known relation to the Earth. This is not changed by the introduction of BIM but rather emphasized by it, as higher degrees of automation and prefabrication increases the need for strict and non-distorting transformations. The objectoriented aspects of BIM require that captured geodata can be semantically classified and that objects can be reconstructed and extracted from the geodata. In this landscape, geodesy is the bridge between model and reality, connecting the two worlds through both semantics and geometry. This thesis is a comprehensive summary of three papers within these two topics. The first paper describes the geometric transformations required throughout the life cycle of a built asset and assesses the georeferencing capabilities of the open BIM standard Industry Foundation Classes (IFC). The second and third paper propose and showcase a methodology where image-based deep learning is used to extract roadside objects from mobile mapping data. The findings of the first paper include suggestions for how IFC can be improved in order to facilitate better georeferencing, and the second and third paper show that the proposed methodology performs well in comparison to a manual classification.