Semantically-Rich Floorplans from Indoor Point Clouds

Maps form a common basis with regard to the components required for indoor location-based services (ILBS). Without these, location-based information cannot be made available to the user. In Architecture, Engineering and Construction sector, the steadily growing use of BIM in new projects will mean that a corresponding digitalised basis of the building will be available for every building in the future. Depending on the responsible building owner’s effort, this model will correspond to the actual environments. However, if one considers a large number of existing buildings with often insufficient information or digitised plans, ILBS cannot be provided in these buildings. Furthermore, interior spaces, for example, in the area of trade fairs, shopping centres or office buildings, are often affected by renovation measures. Updating this information and providing navigation in the form of maps and routing graphs for ILBS is a time-consuming task. The following paper presents an approach to generate semantically-enhanced floor plans and models from 3D point clouds from laser scanning systems for ILBS applications, such as Facility Management and optimised person navigation. With a test dataset from HafenCity University, a geometric-semantic segmentation method is developed using weak machine learning techniques. This paper also presents a method for creating floor plans from a 3D point cloud, which mainly uses 2D image processing algorithms.