Assessment of a Photogrammetric Approach for Urban DSM Extraction from Tri-Stereoscopic Satellite Imagery

Abstract

Built-up environments are extremely complex for 3D surface modelling purposes. The main distortions that hamper 3D reconstruction from 2D imagery are image dissimilarities, concealed areas, shadows, height discontinuities and discrepancies between smooth terrain and man-made features. A methodology is proposed to improve automatic photogrammetric extraction of an urban surface model from high resolution satellite imagery with the emphasis on strategies to reduce the effects of the cited distortions and to make image matching more robust. Instead of a standard stereoscopic approach, a digital surface model is derived from tri-stereoscopic satellite imagery. This is based on an extensive multi-image matching strategy that fully benefits from the geometric and radiometric information contained in the three images. The bundled triplet consists of an IKONOS along-track pair and an additional near-nadir IKONOS image. For the tri-stereoscopic study a densely built-up area, extending from the centre of Istanbul to the urban fringe, is selected. The accuracy of the model extracted from the IKONOS triplet, as well as the model extracted from only the along-track stereopair, are assessed by comparison with 3D check points and 3D building vector data.