Extraction of Buildings from QuickBird Imagery for Municipal Use – the Relevance of Urban Context and Heterogeneity

A spatial component is associated with the majority of municipal activities, namely in urban planning and management. The high frequency and scope of spatial changes in cities demands ways of expediting the production and updating of large-scale geographic information, as required by Portuguese legislation. For that purpose, current and future very high spatial resolution satellite imagery (VHR) may be an advantageous alternative to classical data sources and methods, i.e., aerial photography and photogrammetry. Therefore the nature of this recent data source, volume of data, and expanding range of applications has been driving the development of advanced semi-automated object-based image analysis methods for efficient feature extraction. At the same time, the urban environment is becoming more complex and heterogeneous, possibly turning the feature extraction process more challenging. While much research has focused on developing, adapting and applying these approaches, less attention has been devoted to the interplay of data source (imagery), feature extraction methods, and geographic characteristics of the area under analysis.

The work presented in this paper takes place in the context of the exploration of VHR satellite imagery and new methods as an alternative source of geospatial information for large scale mapping to assist urban planning in Lisbon, Portugal. Lisbon is both a historical and modern city having a dynamic and complex landscape. This effort tests the semi-automated extraction of buildings from areas with different characteristics, and analyzes the impact of the heterogeneity of these features in the extraction process. For this purpose, three study areas having the same size but diverse character were selected: one in the more homogenous and slow-changing old historical district, another in a heterogeneous residential and old industrial area, and a third in a new residential area under development. Buildings were characterized with respect to intrinsic and contextual features, namely roof type (color and material), size, shape, and density. Multispectral and panchromatic QuickBird images were orthorectified and subjected to pansharpening, and feature extraction software was employed to digitize buildings. Extraction performance regarding both quality of extraction and relative workload (time by building) was analyzed in a geographic information system considering the building’s characteristics.

Quality assessment was exhaustive and used reference datasets independently collected from the imagery by visual interpretation. GPS-assisted field work was also conducted to characterize the areas and document the buildings. Results indicate that the geographical context is relevant for the success of semi-automated feature extraction and that the process becomes more challenging with the increasing heterogeneity and complexity of the built environment.