2.5d Reconstruction of Building from Very High Resolution Sar and Optical Data by Using Object- Oriented Image Analysis Technique]

DISCLAIMER This document describes work undertaken as part of a programme of study at the Faculty of Geo-Information Science and Earth Observation of the University of Twente. All views and opinions expressed therein remain the sole responsibility of the author, and do not necessarily represent those of the Faculty. i ABSTRACT 2.5D building modelling is of great interest for visualization, simulation and monitoring purposes in different cases like city and military mission planning or rapid disaster assessment. Manual and automatic extraction of building information from SAR interferometry is complicated by layover, foreshortening, shadowing and multi-bounce scattering, especially in dense urban areas with tall buildings. This can be overcome by using optical imagery to provide the shapes of the building footprints whereas SAR data is used for the refinement of the detected footprints and for providing building height information. Two TerraSAR-X images were used in this research, with High Resolution Spotlight Mode (azimuth resolution 1.1m) together with the QuickBird image. The study area is located in Delft, the Netherlands. First, a digital Surface Model (DSM) was estimated from SAR interferometry processing. Different levels of external height information from Shuttle Topography Mission Digital Elevation Model (SRTM), LIDAR Digital Terrain Model (LIDAR DTM) and LIDAR Digital Surface Model (LIDAR DSM) were used for phase flattening and registration to investigate how they influence the overall accuracy of SAR interferometry processing. Building footprints were extracted with an Object Oriented Analysis (OOA) method from the QuickBird image, with input of the InSAR DSM as external knowledge. Due to spectral complexity of urban areas, it is impossible to extract correctly buildings from the QuickBird image only based on spectral information. OOA including the building height information from InSAR DSM improved the extraction of building footprints from the QuickBird image. Then, the 2.5D building model was derived by combining the building footprints and the building height from InSAR DSM. Finally, the vertical accuracy of the model was assessed with a reference from the LIDAR DSM and the field data from a clinometer. The 2π height ambiguity was determined as 45.94 m from the baseline estimation (131.35 m) of the InSAR pair. If the baseline is larger, the height ambiguity is smaller; so it yields more fringes in the interferogram and is more sensitive to topography or structures such as buildings. The SAR interferometry processing with the LIDAR DSM reduced the phase unwrapping error and helped the registration of …