The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
— 
4.2 Need for Ground Control Points: 
Almost all investigators stress the need for using ground 
control points, which vary in number, to realise acceptable 
quality of DSM and orthoimage using Cartosat-1 stereo orthokit 
products. Kocaman 2008 reports that the orientation accuracies 
realisable with 70 GCPs and 6 GCPs are not much different 
with more than one example. 
4.3 Geometric Model Comparisons: 
Nandakumar & Srivastava 2006 report that current 
investigations (reported during Goa 2006) do not clearly bring 
out the advantage of one over the other, with regard to the two 
image-to-ground orientation methodologies, viz., the generic 
satellite-sensor-orbit-attitude model and the user-refined 
rational polynomial function model. Willnejf et al 2008 compare 
three models including a 3D-affine model approximating the 
imaging process with a parallel projection, whose coefficients 
are determined using a minimum of four non-coplanar GCPs. 
This 3D-affme model has variations in terms of the choice of 
coordinate system used to represent ground coordinates. With 9 
GCPs and 60 check points, the RMS errors in object space were 
less than 1.8 m in both planimetry and height when ground 
points were represented in UTM coordinates. With same 
number of GCPs, affme-corrected RPC model yields sub-pixel 
RMS errors in both planimetry and height. With only 3 GCPs 
also, pixel level accuracies in planimetry and height are 
achievable. Results for the generic push-broom scanner model 
also yield pixel level RMS errors using 9 GCPs. 
4.4 DSM Quality Comparisons: 
As can be noted from Table-1, TS-5 Mausanne, TS-10 
Catalonia and TS-9 Warsaw have the best quality reference data 
sets in that sequence. Also there are maximum independent 
evaluations carried out for the Mausanne and Warsaw sites in 
that order. Hence we limit our comparisons of DSM quality 
obtained by independent investigations to these three sites. Kay 
& Zielinski 2006 classifies the Mausanne reference test area into 
different landcover classes and slope categories as explained in 
Table-3 and Figures-3 and 4. 
Landcover 
Classes 
Arable 
Forest 
Urban 
Water 
% of Total 
Area 
64 
30 
5 
1 
Slope 
Classes 
0-10% 
10-20% 
20-40% 
> 40% 
% of Total 
Area 
71 
7 
9 
13 
Table-3: Mausanne Test Site Classification 
Figure-4: Slope category Distribution 
Land cover mask 
Land cover 
Std. Dev. 
MEAN 
Arable 
3.8Qm 
-0.32m 
Forest 
5.12m 
-0.92m 
Urban 
3.77m 
-0.11m 
Slope class 
Std. Dev, 
MEAN 
0-10% 
3.85m 
-0.17m 
10-20% 
4.78m 
-0.80m 
20 - 40% 
4.27m 
-1.94m 
> 40% 
5.81m 
-0.95m 
Table-4: DSM Comparison Results (Kay & Zielinski 2006) 
Table-4 gives the results of DSM generated using LPS V9.0 
using RPC approach with 6 GCPs with a 10 m grid posting as 
compared with the reference DEM. Kay& Zielinski 2007 
compare the results generated from Jan data set and Feb data set 
in the overlap area. The results are comparable in slope 
category-wise comparison. For the landcover categories, the 
Feb. results are slightly inferior (higher SD) to Jan. results in the 
Forest category. Jacobsen 2006 & 2007 report the results as a 
straight line fit to the Z-errors drawn against the slope values. 
Finally the SD of the height errors are expressed in terms of the 
stereo-parallax, thus enabling a direct comparison between the 
performances of different sensors. Quote: The vertical accuracy 
can be expressed like following: 
SZ — h/b * Spx : Formula 1: standard deviation of Z 
h=height b=base Spx = standard deviation of x-parallax [GSD], 
For Cartosat 1 the height to base relation is 1.6. With this 
relation and formula 1, the achieved results can be transformed 
into the standard deviations of the x-parallax, allowing a 
comparison with other sensors. Table-5 summarises the results 
for Mausanne and Warsaw. 
Figure-3: Land cover category distribution 
matched DSM 
filtered 
open 
forest 
open 
forest 
Mausanne. January 
0.98 
0.83 
0.79 
0.73 
Mausanne, February 
0.99 
0.70 
0.80 
0.67 
Warsaw 
0.79 
1.02 
0.60 
0.78 
Warsaw build up area 
0.66 
0.49 
Table-5: Accuracy of x-parallax (computed from constant 
value of function depending upon inclination) [GSD] 
Here, in Table-5 filtering refers to an attempt to convert DSM to 
bare earth DEM. With a standard deviation of the x-parallax 
between 0.49 and 0.80 GSD similar x-parallax accuracies like 
1282