Stereo Evaluation of CARTOSAT-1 Data Summary of DLR Results During CARTOSAT-1 Scientific Assessment Program

Abstract

The Remote Sensing Technology Institute (IMF) of the German Aerospace Center (DLR) has more than 20 years of history in developing spaceborne stereo scanners and the corresponding stereo evaluation software systems. It takes part in the CARTOSAT-1 Scientific Assessment Program (C-SAP) as a principal investigator for German (Southeast Bavaria) and Spanish (Catalonia) test sites and as a Co-I for a French test site (Mausanne-les-Alpilles). A rich variety of landscapes is present in these three test sites. In all cases ground truth in form of GCP (or orthoimages of high resolution) and DTM/DSM (digital terrain or surface models) of sufficient accuracy have been delivered by the principal investigators.Rational polynomial functions (RPC) are provided by the distributing Indian agency (Space Applications Centre (SAC) of ISRO, Ahmedabad) as a universal sensor model for each scene. The inherent absolute orientation accuracy of the RPC models in the CARTOSAT-1 stereo imagery used here turned out to be around 100 m (normally). Thus, to exploit the high resolution of 2.5 m, RPC have to be corrected via the available ground truth. It is shown that the correction by an affine transformation is necessary in order to achieve sub-pixel accuracy in the stereo evaluation of full scenes. The remaining standard deviations of the residuals in image space during RPC correction are about 0.5-1 pixel in ground control points (GCP). Stereo evaluation is done by DLR processing software. Hierarchical intensity based matching and subsequent region growing are used to automatically derive a dense set of stereo tie points. An effective blunder reduction is based upon bi-directional LSM, quasi-epipolar reprojection of the tie points, and control of residuals in stereo forward intersection. Shifts between aft/fore orthoimages are found to be in sub-pixel range. DSM accuracy assessment is done via the statistics of height differences compiled by the forward intersection software. This is sufficient if accurate GCP for RPC correction are available. For direct comparison of the generated DSM with the reference DTM/DSM a 3D shift is estimated via least squares adjustment and mean and standard deviations of the DTM/DSM differences after shifting are provided. In summary, standard deviations of 2-4 m are achieved.