Comparison of Digital Elevation Models Generated from Spot-5 Hrs Stereo Data and Cartosat-1 Stereo Data

SPOT-5 and Cartosat-1 are mapping satellites with high resolution and along-track stereo capability. This paper describes the work carried out towards the qualitative and quantitative comparison of the Digital Elevation Models (DEM) derived from stereo pairs of SPOT-5(HRS1 and HRS-2) and Cartosat-1(Fore and Aft) over the test site located in Bavaria, Germany. This is one of the test areas selected towards SPOT-5 Assessment Programme. DEMs are generated with two different kind of modelisation; one with geometric correction of images using physical sensor model and other one with Rational Polynomial Coefficients (RPC). In the first approach, Ground control points (GCP) are used for bundle adjustment with a self calibration to improve the interior and exterior orientation parameters. In the second approach, RPC’s are derived initially from the grid generated through direct georeferencing using the physical sensor model and then compensated with GCPs. Out of the 47 points provided, 5 are used as GCPs for rectification of SPOT images and 3-D modeling was checked on remaining 42 independent points. Both orientation methods gave RMS errors less than a pixel for both mono and stereo orientation. We tried to transfer these points to Carto-1 fore and aft images through sub-pixel matching. But most of the points could not be precisely transferred to Cartosat-1 as the GCP locations are not clear cut features on the SPOT image. Secondary controls are derived by identifying few well defined road junctions in SPOT scenes and deriving the corresponding ground co-ordinates through SPOT stereo intersection which gave an accuracy of 2-3m.They are manually identified in Cartosat-1 images. Two stereo GCPs are used for modeling Cartosat-1 fore and aft images. Accuracy is evaluated at the remaining points for different subscens. RMS errors of the order 6 to 7m in X, Y, Z are obtained. Increasing the number of GCPs does not improve the accuracy. The DEMs are generated using a combined approach of feature based and areabased multi-scale image matching method and semi automatic editing tools. The quality of the matching is controlled by forward and backward matching of the two stereo partners using the least squares matching method. The “no match” areas are filled with interpolation during the DEM production process and final DEM is generated at regular interval. DEM accuracy is determined by using the available DEM data of superior accuracy. Detailed visual and quantitative comparison results of the DEMs from SPOT-5 and Carto-1 are presented in the paper. Even tough absolute accuracy is poor for Cartosat-1 due to the inaccuracy of secondary controls, quality of the DEM is much superior to SPOT-5 DEM. Statistics of the errors are analyzed and tabulated for the DEMs generated with different global spacing.