Large-Scale Mapping of Urban Parking from Aerial Images: A Case Study in Berlin, Germany

Abstract. Existing nationwide spatial datasets for Germany’s traffic infrastructure, particularly parking areas, are fragmented and incomplete, hindering effective traffic management and urban planning amidst growing demands for mobility transition and livable cities. This paper presents a novel approach to create a comprehensive parking area inventory for Berlin using aerial imagery. The methodology integrates AI-based traffic area segmentation and DINO-based vehicle detection with cadastral data. A key innovation is a workflow that classifies parking areas by their orientation and accessibility for refined capacity calculation. The resulting Berlin-wide inventory comprises 1,333,953 parking spots. Our method significantly contributes by mapping private (19 %) and semi-private (21 %) parking areas, which are largely missing from existing inventories, alongside publicly accessible (60 %) spaces. Vehicle detection identified 1,039,155 vehicles (1,019,690 LDV, 19,465 HDV). Initial classification shows 36 % parallel, 27 % diagonal, 20 % vertical, and 17 % unclassified parking spots, with notable variability across districts. This comprehensive inventory addresses a critical data gap, providing a more accurate understanding of urban parking resources. The highly automated and repeatable nature of this aerial imagery-based approach offers significant potential for large-scale applications and temporal change analysis. Future work will focus on developing correction factors for capacities in partially occluded areas and integrating information on underground parking facilities to further enhance completeness.