Automatic identification of building types based on topographic databases – a comparison of different data sources

ABSTRACT Data, maps and services of the national mapping and cadastral agencies contain geometric information on buildings, particularly building footprints. However, building type information is often not included. In this paper, we propose a data-driven approach for automatic classification of building footprints that make use of pattern recognition and machine learning techniques. Using a Random Forest Classifier the suitability of five different data sources (e.g. topographic raster maps, cadastral databases or digital landscape models) is investigated with respect to the achieved accuracies. The results of this study show that building footprints obtained from topographic databases such as digital landscape models, cadastral databases or 3D city models can be classified with an accuracy of 90–95%. When classifying building footprints on the basis of topographic maps the accuracy is considerably lower (as of 76–88%). The automatic classification of building footprints provides an important contribution to the acquisition of new small-scale indicators on settlement structure, such as building density, floor space ratio or dwelling/population densities. In addition to its importance for urban research and planning, the results are also relevant for cartographic disciplines, such as map generalization, automated mapping and geovisualization.