Science Service System
You are here : Home : Proposals_Summary

Summary of Proposal LAN0955

Title3D reconstruction of building from Very High Resolution SAR and optical data by using Object-Oriented Image Analysis technique
Investigator Bijker, Wietske - Twente University, Faculty ITC, Earth Observation Science (EOS)
Team Member
Mr. Kim, Namhyun - Twente University, Faculty ITC, Earth Observation Science (EOS)
dr. Tolpekin, Valentyn - Twente University, Faculty ITC, Earth Observation Science (EOS)
SummaryResearch objectives The main objective of this research is to develop a method for reconstruction of 2.5D building model from VHR SAR and VHR Ortho-image using OOA technique. Sub-objectives: 1.To get accurate building height information from InSAR 2.To improve the method of the building footprint detection from VHR Ortho-image by OOA technique, using the height information from InSAR 3.To combine the detected building footprint with the estimated building height from InSAR for a 2.5D building model Method Building height estimation from InSAR Two VHR SAR (Single Look Complex: SLC) images recorded over the same area will be used. First, baseline will be estimated, followed by co-registration of the two SLC images. Phase difference calculation, flat earth removal and phase unwrapping will be applied to extract the phase difference. An InSAR DEM will be generated from the InSAR phase data and geo-coded with GCP from the additional data i.e. topographic map. The terrain height will be subtracted from InSAR DEM to achieve only building height.The height accuracy will be assessed using a reference dataset Extraction of building footprint from VHR optical imagery using OOA technique The visual interpretation of Ortho-image will give an insight in the image, feature properties and their relationships to be used for further analysis. The VHR Ortho-image will be segmented to create homogeneous regions in the image, called image object, object primitive or segment. Analysis of the feature properties of the image objects will be used to develop knowledge about the objects in the image and translated into a rule set for object extraction. External knowledge, in this case the estimated height information from InSAR, will be included in the rule set. The iterative process of rule-based image segmentation and classification will produce enhanced object shapes and more accurate classification results. After classification of the segments, building footprint will be extracted. Refinement of the extracted building footprint will be needed since not all extracted building footprints may show a smooth shape. For example, chessboard segmentation which is based on the top down region splitting principle can be used for the refinement of the shape and size of the extracted building footprints [6]. The accuracy of the extracted building footprint will be assessed using reference dataset i.e. manual building extraction from Ortho-image. 2.5D building model generation The 2.5D building will be modeled by combining the detected building footprint in OOA with the estimated height information from InSAR. For instance, the result as 2.5D building model will be visualized in ArcScene software. Data requirements -SAR data (TerraSAR-X) for the Delft, The Netherlands, area, 1.1m azimuth resolution in high resolution spotlight mode 2 SLC images (interferometric pair) of Delft. VV polarization. -Optical data (QuickBird): already available at ITC for the Delft, The Netherlands area. Reference height data are available (AHN) and where needed, additional data will be collected in the field. Deliverables Method for building height estimation, building footprint extraction and 2.5D building modeling based on TerraSAR-X InSAR data and VHR optical data, described in an MSc thesis. Case study demonstrating the method for a study area in Delft, The Netherlands. Funding Salaries of staff, costs for equipment and fieldwork are funded by ITC faculty.

Back to list of proposals

DLR 2004-2016