Analyse des Geokodierungsfehlers für die gemeinsame Bearbeitung von SAR und optischen Daten

Abstract

In this thesis the geocoding error for the combined use of SAR (Synthetic Aperture Radar) data and optical data is analysed. Images of islands in the coastal region near Stockholm are the basis for this evaluation. Radar images of the TerraSAR-X satellite are compared to optical images from bing and google. The coast lines are automatically detected from binary images via variable thresholding. Afterwards two procedures are executed to analyse the offset between the coastlines of two images: On the one hand it is assumed that only a simple affine transformation is needed. This is why the coast line is represented by an ellipse that is determined through fourier transformation and subsequently the ellipse transformation is calculated. On the other hand a more general approach is used and a vector field of the offset is computed. For this purpose the points of one coast line are matched to the points of the other coast line using cosine similarity in the feature space (features position, normal vector and curvature). Then the offset is calculated. The vectors are analysed by various graphical tools. As a result of this work a constant offset could be detected: A shift of the islands in the optical images compared to the islands in SAR-images to west/north west. The modelling of the offset as vector field turns out to be very prone to errors. The principle of ellipse approximation appears to be more helpful for a first approximation of the geocoding error.