Extracting building features from high resolution aerial imagery for natural hazards risk assessment

Natural hazards risk assessment requires data on the built environment. This paper reports an image analysis method that can extract building features, mainly roof plan areas, for potential vulnerability analysis. Both pixel- and object-based image processing methods are adopted. First, red/green/blue colour bands and image textures are incorporated in a supervised artificial neural network classifier to achieve good classification results of individual roofs. Second, within object-based methods a hybrid of region and edge segmentations using colours and shapes is employed to extract useful spatial information of salient ground objects. Finally, the extracted spatial information is used to refine the pre-classified image of building roofs. An AUSIMAGE™ digital aerial image with a spatial resolution of 0.2 m is tested. The directly extracted data include roof locations, plan areas, and perimeters. Derived data can include distances from building centroids to street centre lines, and distances between adjacent buildings. Such extracted data can greatly assist detailed bushfire, hail, tornado, and flood risk assessment