Stereo Evaluation of CARTOSAT-1 Data on Test Site 5 - First DLR Results

Kurzfassung

DLR's Remote Sensing Technology Institute has more than 20 years of history in developing spaceborne stereo scanners (MEOSS, MOMS) and the corresponding stereo evaluation software systems. The institute takes part in CARTOSAT-1 Scientific Assessment Program (C-SAP) as a principal investigator for German (Southeast Bavaria, test site not yet included in the C-SAP list) and Spanish (Catalonia, TS10) test sites for which also PI evaluations for SPOT-5 HRS SAP had been done in 2003-4. As CARTOSAT-1 data of Catalonia (test site 10) did not correspond to the available ground truth, negotiations with ICC brought forward a new fitting set of ground truth. Unfortunately, this new reference data came too late for phase I of C-SAP. Thus, for phase I of C-SAP participation as a CoI in the evaluation of CARTOSAT-1 data for test site TS5 (Mausanne-les-Alpilles, France) has been agreed upon. For phase I of C-SAP no explicit exterior and interior orientation data of CARTOSAT-1 have been given. Instead, rational polynomial functions (RPC) are provided by the distributing Indian agency as a universal sensor model for each scene. Thus, only the inherent orientation accuracy of the RPC models is established by comparison to the available ground truth. Ground control points are used to correct the RPC (bias correction and also affine transformations). The resulting various residuals are assessed and commented. From these first investigations it can be seen that the offset of the original RPC is in the order of kilometres which was not expected. Bias correction ends up with residuals at ground control points (GCP) in the order of several pixels showing also systematic behaviour. This leads to the conclusion that RPC have to be corrected with affine transformations. The latter lead to residuals in the order of 1 pixel which is satisfactory for the start of investigations. DSM accuracies are assessed via residuals in forward intersection (tie point cloud from matching) and through calculation of 3D shifts between reference DEM and calculated DSM by least squares adjustment (full DEM/DSM comparison). Using affine transformation correction of RPC a standard deviation of the DEM/DSM height differences of 3-4 m is achieved which is very good when taking the inherent DEM/DSM differences into account.