Automatic Interpretation of Remotely Sensed Images for Urban Form Assessment

A system for generating information for an urban inventory by analysing remotely sensed or ground based sensed images is described. The urban inventory contains information about all land parcels in an urban area and the information is stored in a GIS database. The analysis system uses the semi-hierarchical multiresolution MCV image labeling algorithm and ensemble SVM classifiers to detect building footprints, trees and other urban land cover classes. The system has high accuracy for building footprint and tree detection on the data on which it has been tested.