Investigation of the potential of multiscopic VHR satellite imagery for the production of 3D models of complex urban areas

Improved radiometric quality and geometric accuracy of high resolution sensors like Ikonos and Quickbird allow a more detailed mapping of urban areas. Due to this new generation of sensors the potential for producing 3D models of such complex areas has substantially increased.
The objectives of the MAMUD research project is to investigate how earth observation can contribute to a better monitoring, modeling and understanding of urban dynamics, and its impact on the urban and suburban environment. The outcome of the research can lead to the development of a useful tool for urban planning approaches based on sustainable development.
While most remote sensing based studies on urban dynamics only consider two-dimensional structure, in this research also stereoscopic and multiscopic imagery will be used to derive the third dimension, which is obviously important in describing urban morphology. The fast growing urban area of Istanbul is chosen as the study area of our research.
The subject of this treatise mainly deals with the potential of multi-angle image acquisition for improving the accuracy of the stereo model and reducing the impact of shadow and occluded areas. For the VHR multiscopic study an area extending from the centre of Istanbul to the urban fringe is chosen.
As image triplets might not be available, a triplet might be constructed from multi-orbit images. A configuration consistent of one Ikonos stereo pair combined with a nearly vertical Ikonos image from more or less the same orbit will be used. The Ikonos stereo pair is captured in March 2002. The images of the stereo pair have an elevation angle of respectively 67.59° and 75.59°. The collection azimuth is respectively 1.6° and 214.1°. The third image, captured in May 2005, has an elevation angle of 80.93° and a collection azimuth of 23.5°.
Research will concentrate, in first phase on the development of methods to optimize the extraction of 3D models and orthophotos from the bundled Ikonos triplet. Optimal methods need to be found to improve the preprocessing of the imagery, to improve the semi-automatically derivation of photogrammetric products and to improve the postprocessing of the products. What concerns the postprocessing, we can point at the load of manual editing that is necessary to get an acceptable quality for automatically derived 3D models from VHR satellite imagery of a dense urban area. Secondly we will also investigate the possibilities of creating stereo models out of images from the same sensor taken on a different date, e.g. one image of the stereo pair combined with the third image. Finally the photogrammetric products derived from the Ikonos stereo pair as well as the products created out of the triplet and the constructed stereo models will be investigated by comparison with a 3D reference. This evaluation should show the increase of accuracy when multi-imagery is used instead of stereo pairs.