Building detection and regularisation using DSM and imagery information

An automatic method for the regularisation of building outlines is presented, utilising a combination of data- and model-driven approaches to provide a robust solution. The core part of the method includes a novel data-driven approach to generate approximate building polygons from a list of given boundary points. The algorithm iteratively calculates and stores likelihood values between an arbitrary starting boundary point and each of the following boundary points using a function derived from the geometrical properties of a building. As a preprocessing step, building segments have to be identified using a robust algorithm for the extraction of a digital elevation model. Evaluation results on a challenging dataset achieved an average correctness of 96 center dot 3% and 95 center dot 7% for building detection and regularisation, respectively.