Science Service System
You are here : Home : Proposals_Summary

Summary of Proposal MTH0715

TitleHigh-Precision Co-Registration of TerraSAR-X and RapidEye Data
Investigator Hellwich, Olaf - Technische Universitšt Berlin, Computer Vision and Remote Sensing
Team Member
Dipl.-Ing. Bornemann, David - Berlin Institute of Technology, Computer Vision and Remote Sensing
Dr. Koenig, Rolf - GFZ German Research Centre for Geosciences, Geodesy and Remote Sensing
SummaryIn the proposed project, we look into the development of an entire, complete new and automatic approach for a high-precision fusion of TerraSAR-X (TSX) and RapidEye (RE) imagery by means of a geometric modelling. The major idea is to create an accurately fitted RE image close to the ortho-image geometry by using a digital elevation model (DEM) for co-registration. A reliable and appropriate DEM can be directly provided by the data of the SRTM X-Band mission. Instead of using a DEM, a digital surface model (DSM) is more reasonable in order to minimize the effects of displacements during a rectification process. Mainly, the new approach will be implemented by registering the RapidEye data (Level 1b, 6.5†m resolution) on already precisly geocoded TSX data (SSC, Stripmap-mode, 3†m resolution), because precise orbit data of each scene taken from the TSX sensor are available. For a successful treatment, we prefer to fuse both data sets on the feature level than on pixel level. Here image features, like lines or points have to be extracted in both types of imagery. Every feature extracted will be annotated by its coordinates in object space. Assuming the rectified and roughly registered image data, as well as the feature parameters are properly estimated, corresponding features overlay in object space. This enables the geometric modelling of the fusion by means of their positions and their derived distances. Later on, remaining empty image pixels between neighboured features can be interpolated using the information of the DSM, too. Additionally, a stable co-registration requires robust similarity measurements to provide optimally estimated correspondences between extracted features on one hand and on the other hand robust optimization techniques to refine these. Due to the fully geometric modelling of the co-registration problem, information about the physical sensor model, such as sensor orientations and precise sensor orbits, are needed as well. All information together describe an overdetermined problem where the geometric image parameters of the features, the given orbit parameters and the initial height information are the observations. Usually, the observations will be formulated as functions of the unknowns. In our case, possible unknowns are the object space coordinates, refined orbit and sensor parameters, as well as the initial DSM heights. On this basis, we intend to exploit several adjustment calculation algorithms, especially the least squares method to minimize the remaining residuals between corresponding features. As final results, we expect a precisely co-registered and fused RE orthoimage with respect to the TSX data, more accurately estimated sensor orbits in addition to an improvement of the surface model used. The new approach and its expected results will provide an interesting pre-processing step for following applications and analyses and will make it easier to use fused data.

Back to list of proposals

© DLR 2004-2016