Extraction of buildings from high resolution satellite data

This paper describes the methods that can be used to detect building objects under the circumstances of high object density and scene complexity seen in 1-metre high resolution satellite imagery. In order to reduce the building hypothesis space in early processes, a local Fourier analysis does an analysis of the dominant orientation angle in a building cluster. This serves to extract the focused perceptual cues. A method to enhance or infer insufficient cues is described with a polygon space generated by binary space partitioning (BSP) tree. The paper introduced the concept of the Building Unit Shape (BUS) and shows how this can be used to extract building outlines. The useful characteristics of this method to detect objects that consist of rectilinear lines are demonstrated in examples and future work is discussed.