Derivation of urban objects and their attributes for large-scale urban areas based on very high resolution UltraCam true orthophotos and nDSM – a case study Berlin, Germany

The development of automatic extraction methods for urban building and vegetation objects and the realization on a large urban data set have been accomplished within the project ‘Derivation of Buildingand Vegetation Heights and Structures in Berlin’. This project was executed on behalf of the Senate Department for Urban Development and Environment of the City of Berlin. As input an UltraCamX dataset consisting of the true ortho mosaic with four spectral channels and the normalized digital surface model (nDSM) with 30cm resolution were used. The size of the area adds up to about 450 km2. For the delineation of roof tiles additional slope and aspect layers were created. The workflow was developed in eCognition for automatic extraction of elevated urban objects and their height structures. Additionally we used focal statistics in ArcGIS to extend the workflow for the extraction of single tree crowns, since detailed information about the correct position and number of urban trees is relevant and completes the urban geo database. In this way a unique, complete and extensive workflow for an automatic urban objects extraction arises. Within the project methods for robust extraction of buildings with roof tiles as well as vegetation were applied. Greened roofs are automatically extracted and assigned to the class buildings. Furthermore building tiles which are located under trees are extracted by the intersection with the cadastral building data. For the transferability of the complete rule set a multi-layered workflow consisting of automated data import and export, iterating segmentation methods and fuzzy classification as well as object reshaping was developed and applied. The methods are transferable and effectively operate on large data sets. As results complete layers with building and vegetation shapes including single tree tops and crown outlines with comprehensive attributes like height and area are derived. The accuracies are in the range of 85 % (trees) to 95 % (buildings). The shapes primarily serve as input for a complex urban-specific climate model, which also takes the morphology of hinterlands into account. In addition this data can be used as a basis for a variety of urban planning tasks e.g. for modeling of noise pollution, the estimation of urban structure types, for townor green space planning. They are a suitable basis for supplementing and upgrading of the official cadastral database, even though certain restrictions concerning the accuracies exist. Furthermore the object shapes are suitable for visualization tasks.