Science Service System

Summary of Proposal MTH0104

TitleUrban 3D – High Resolution Information Extraction with TerraSAR-X
Investigator Bamler, Richard - Deutsches Zentrum für Luft- und Raumfahrt e.V. (DLR), Institut für Methodik der Fernerkundung
Team Members
Prof. Dr. Stilla, Uwe - TU Muenchen, Photogrammetry and Remote Sensing
Prof. Dr. Heipke, Christian - University of Hannover, Institut für Photogrammetrie und GeoInformation
Prof. Dr. Hellwich, Olaf - Technical University Berlin, Computer Vision and Remote Sensing
Prof. Dr.-Ing. Hinz, Stefan - University of Karlsruhe, Karlsruher Institut für Technologie

1) Persistent Scatterer Interferometry

  • Push the PSI technique to new limits by exploiting TerraSAR's new modes, resolution, incidence angles and polarimetric capabilities
  • Compare TS-X time series with independently generated Envisat/ASAR series. Analysis of PS density, APS signal, non-linear behaviour over time and the SCR statistics of the detected scatterers.
  • Obtain experiences and knowledge on the scattering mechanisms and the point scatterers. The close location of the Munich test site to DLR make ground truth data readily available. Detected persistent scatterers will be investigated on ground
  • Implement advanced estimation algorithms to fully exploit the TerraSAR-X data. The joined estimation of independent stacks and advanced techniques such as tomography are desirable. A precondition for the development of these algorithms is the availability of various data stacks
  • Develop and implement new algorithms taking advantage of different scattering effects and the polarimetric capabilities of the sensor
  • Generate subsidence maps from the test sites demonstrating the advantage of TS-X
  • Generate example products for the newly developed algorithms demonstrating the improved accuracy or new physical interpretation

2) Multi-Aspect Exploitation

  • Adapt speckle filters to the signal statistics of TerraSAR-X data.
  • Adapt segmentation algorithms to the new data
  • Co-register TerraSAR-X data sets which were taken from different viewing direction
  • Optimize the timing for co-registration by developing tools for an automatic coregistration
  • Implement algorithms to reconstruct urban structures in TerraSAR-X data
  • Process the test site Munich
  • Validate results by comparing them wit ground truth.
  • Adapt algorithms to improve the reliability
  • Process other test sites

3) Object Extraction and Categorization

  • Carry out suitable pre-processing with respect to the subsequent image processing techniques
  • Investigate which kind of control structures can be used to achieve an initial coarse co-registration of the TerraSAR-X data and the optical imagery
  • Develop an automated approach for this coarse co-registration
  • Study object features of buildings and bridges in the coarse registered data for different cases (optical image + amplitude SAR or InSAR or PolSAR or PolInSAR)
  • segmentation of suitable object features
  • Object recognition considering the different constellations mentioned above resulting in a 3D description with subsequent improved co-registration
  • Evaluation of developed approaches in a test bed
Important questions which will be answered are:
  • How can salient image patches be segmented into object and neighbourhood/background regions in the formulation of object hypotheses?
  • How can an object class be learned from training images in the presence of background clutter?
  • How can effective and robust learning be achieved in the case of small training sets?
  • Can previously learned object models be of use in the process of learning a new one?
  • How are model complexity and algorithm performance correlated?
  • How effective are object category specific features in comparison to generic features?
  • Should features other than salient regions be used? If so, how is it possible to maintain the scale space concept and achieve scale invariance?
The funding of the project work is based on exisiting DLR and university funding and on the submitted DFG proposal. Envisat/ASAR data are provided by ESA in the CAT-1 framework.

Back to list of proposals

© DLR 2004-2016