An elevation difference model for building height extraction from stereo-image-derived DSMs

This article investigates the scale issue of inaccurate elevation around buildings in digital surface models (DSMs) and its application in building height estimation. DSMs derived from a single pair of optical stereo images are affected by occlusions and shadows, which lead to indistinct building borders in the DSM. To explore the parameters of how elevation changes in such inaccurate DSMs around buildings, a 'building-ground elevation difference model' (EDM) has been designed in this study. This model describes the trend of elevation differences between a building and its neighbours in order to find a stable ground elevation and to estimate actual building height. The EDM is discussed in application to both flat and sloped ground situations. Experiments on two study sites using the proposed model demonstrate that the estimated height at rooftop points can be comparable to light detection and ranging data with respect to rooftop height estimation, which outperforms the conventional filtering method. Furthermore, the proposed semi-variogram model also sheds light on the scale issue of features in DSMs of different spatial resolutions.