The testing of automated driving functions has to be improved in order to allow for a broad usage of automated and autonomous vehicles on public roads. The current approach assumes that a testing field for such cars can and should be a virtual representation of a real-world scene. This implies a need for a reliable, up-to-date, accurate, and semantically rich 3D models of a road space environment. Moreover, arising questions about satisfactory levels of semantics, temporal, and spatial resolution have no definite answers. In order to create 3D maps, point clouds acquired in aerial and mobile mapping campaigns are often utilised. However, available automatic methods for 3D models creation do not completely fulfil demanded requirements. Those models either lack detailed geometry representations or have poor semantics. The models which are manually created cover those gaps but time-consuming modelling process prevents scaling of 3D maps for wider areas. The recent trends in 3D maps creation focus on reconstructing objects without taking into account semantics and geo-contextual information of already created 3D maps. Therefore, the goal of this work was to create a method which allows to automatically enhance the geometry of existing 3D road space models by means of available point clouds. A workflow had to be easy to use and enable user-friendly customisation of an expected refinement level even for a non-expert user in the field. The methodology is based on novel approaches from fields of geoinformatics and photogrammetry which are inevitable to achieve the goals of the project. The FME software serves as a backbone of the project where LASTools, Python scripts and the external software MeshLabServer are integrated. Thanks to that, the whole processing and reconstruction workflow is steered by one software. Validation of the methodology and visualisation of results are performed in the state-of-the-art city models managing tool 3D city Database suite and the game engine Unreal Engine which is used in automated driving simulators like CARLA. Additionally, the possible semantics enrichment of models representing roads is shown. The city centre of Ingolstadt, Bavaria, Germany served as a testing ground for the methodology. The datasets of LoD1, LoD2, LoD3 buildings and roads from HD Map in the CityGML standard area were used.     «

The testing of automated driving functions has to be improved in order to allow for a broad usage of automated and autonomous vehicles on public roads. The current approach assumes that a testing field for such cars can and should be a virtual representation of a real-world scene. This implies a need for a reliable, up-to-date, accurate, and semantically rich 3D models of a road space environment. Moreover, arising questions about satisfactory levels of semantics, temporal, and spatial reso...     »