The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
1328 
Fig. 11: colour coded height model extracted from aerial 
images (left) and from Cartosat-1 images (right) 
In Fig. 11 the extracted height models are shown; the software 
used for the DEM generation are respectively ERDAS v. 9.1 for 
the aerial block and RPCDEM for Cartosat-1 stereo pairs. The 
two software packages respond in different ways in the lake 
zone, in fact ERDAS applies an automatic filtering function so 
that it is able to assign an elevation even in the lake areas, on 
the contrary the matching achieved by DPLX did not recognize 
homologous points for the two images, due to the fact that no 
contrast was available on the lakes; therefore the elevation of 
those points was not extracted by RPCDEM. 
Furthermore the accuracy of Cartosat-1 DSM has been checked, 
with respect to the reference DSM over different terrain types: 
open areas and urban areas; thus the overall scene has been 
divided in several selected regions. The analyses have been 
performed both for the digital surface models directly obtained 
from the images, and for the digital elevation models, obtained 
filtering the original DSMs with the Hannover software 
RASCOR. 
The selected open regions are not completely flat zones, 
because they still contain sparse buildings and groups of trees, 
nontheless the accuracy obtained was in the range of the 
standard deviation of the height. 
Regarding the urban areas the discrepancies with the reference 
DSM are bigger; in these areas the smoothing effect of 
Cartosat-1 is more evident, in fact, while the DSM results 
higher than the reference over the streets, it tends to smooth the 
edges of the buildings, as it is is shown by the profiles in Fig 12; 
on the contrary after filtering the profiles are similar and the 
smoothing effects disappear (Fig. 13). 
Fig. 12: profiles through a Cartosat-1 DSM and the reference 
DSM from aerial images an urban area 
Fig. 13: profiles through a Cartosat-1 DEM and the reference 
DEM from aerial images an urban area 
In the Fig. 14 the differential DSM of an open area and an 
urban area are shown; the differential DSM is obtained by the 
comparison beetwen the DSM reference and the Cartosat-1 
DSM generated with RPCDEM using the SISAR RPC (Fig. 14). 
.00 RED 
.00 BRÖBN 
.00 YELLOW 
.00 GREEN 
.00 GREEN-BL OE 
.00 BLUE 
.00 VIOLETT 
Fig. 14: Differential DSM of open area (left) and urban area 
(right) 
4. CONCLUSION 
Results stemming from the geometric reconstruction of 
Cartosat-1 imagery shows that accuracy is at GSD level in 
horizontal components and at about 2-3 m in vertical one. 
The SISAR results demonstrate that RPCs, generated by SISAR 
software, permit high performance for image orientation, 
achieving results close to the geometric reconstruction. A 
particular core was devoted into the estimable coefficient 
selection; by now the usual strategy is the mainly based on 
Tichonov regularization, on the contrary, in this work a 
different innovative method was used, based on the Singular 
Value Decomposition (SVD) and QR that can reliably handle 
the rank deficient matrices, without require an empirical 
evaluation of the regularization parameter. This selection allows 
to use a number of coefficients lower than the standard number 
used in the third order polynomial (78 RPC). 
Digital height models have been generated by means of 
Cartosat-1 stereo pairs. The orientation of the models in any 
case was possible with sub-pixel accuracy by bias corrected 
RPC-solution. In any case for flat terrain the height accuracy is 
better than 1 GSD for the x-parallax or 4m. After filtering for 
elements not belonging to the bare ground the vertical accuracy 
for flat terrain is not less than 3.2m corresponding to x-parallax