Automated recognition of urban objects

Urban geometry is one of the most important aspects that influence urban microclimatic conditions. Developing and updating databases of urban geometry is, therefore, important for studying climatic aspects of urban form, especially where no town plans or updated surveillance data is available. The current paper presents a parametric model which enables automated recognition of urban objects, in particular open spaces, from very-high spatial resolution remotely sensed data. Once objects are recognized, morphological attributes are automatically extracted. The developed model combines spatial, spectral and context – based recognition. It adopts a different approach to the challenge of automated object recognition, in which Geographic Information Systems (GIS) play a major role in the recognition of objects. While the segmentation is performed with image processing software, the classification is performed in GIS using a rulebased reasoning model. A major challenge in the automated recognition of urban objects is that urban objects often do not adhere to the basic assumptions that automated recognition systems are based on, such as consistent pixel intensity. To overcome this problem, the presented methodology, makes use of (a) the variety of generic context-based relations between objects in urban form, and (b) the ability of GIS to recognize contextual relations. The model was applied to a case study and statistically tested for its accuracy. Results are promising and demonstrate the potential of the model as a quantitative and systematic tool. Being a parametric model, it can be modified and applied on a large number of case studies. Recognized objects and extracted attributes can be used for constructing and updating GIS databases of urban form. * Corresponding author.