An ML-supported Pipeline for the Mapping of Urban Densification Potentials

Abstract. Using densification potentials for urban development is one of the current challenges in urban planning. An important prerequisite is an up-to-date GIS data basis, which contains potential densification sites. Existing data is often based on manual acquisition, which is time-consuming and costly. Therefore, the value of such data depends on the effort of creating and updating it, which can be improved by automatic data collection. We developed an ML-supported pipeline containing a deep learning model and 3D point cloud processing. This pipeline addresses two important densification potentials: infill sites and vertical extension. Integrating the automated processing into QGIS as a plugin provides users an easy-to-use tool to generate a data basis for densification potentials for their decision-making process. We evaluate the effectiveness of our approach using real-world data from two different cities. The results show its usability in practice and the possibility of transferring it to different locations.