GEOREFERENCE OF TERRASAR-X IMAGES USING NETWORKS OF GROUND CONTROL LINEAR FEATURES

Abstract. One of the main novelties introduced by the contemporary spaceborne SAR sensors is the fine Ground Sample Distance (GSD). The GSD of 1 m, which is available now, is comparable to the resolution offered by optical sensors. It creates the conditions to bring closer SAR and optical data and directs research towards SAR processing methods of comparable accuracy with optical processing methods. This paper focuses on the capability offered by contemporary SAR sensors to identify fine man-made and physical features of earth's surface. 3D road edges are treated as Ground Control Information, in order to compute the projective transformation that relates the 3D features of the object space with their 2D projection in a SAR image. The computation of the projection is done with a novel ICP-based method that matches a network of 3D Free-form Linear Features of the object space with their 2D projections in the image space. The proposed method is tested for the georeference of a whole TerraSAR-X scene. Computed results are evaluated with independent Check Points. The quality of the results are superior to those computed with salient point based approaches.