International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004 
   
60 60 
Z 40 40 
= 
£20 + 20 
= ~ 
* 0 j 5e ) 0 
= : y f c GS / 
SA EN ; A -20 
ee Tl 
s -40 -40 
2 
-60 -60 
-60 -40 -20 0 20 40 60 80 -50 -40 -20 0 20 40 60 
relative L track (meters) relative L track (meters) 
Before modification After modification 
Figure 4 : improvement of HRS relative location 
(ellipse is the 3 x std boundary) 
As for HRS absolute location performance, the analysis showed 
an important variation of the performance depending with the 
latitude of the acquisition. This variation was firstly thought to 
be related to a specific thermoelastic effect affecting the whole 
payload as it,also touched HRG's location performance. The 
explanation of this variation finally appeared to be a bad 
initialisation of the star tracker relativist compensation process. 
After on board correction of this problem, no more variation of 
the performance with latitude has been observed and HRS 
absolute system location performance without using any ground 
model went down from 50 m rms to 20 m rms (see figure 5). 
       
— 120 120 
$ N 
$ 80 80 
40 \ 40 
2 \ 
= 0 | 0 
2 40 TS} -40 
$ -80 p Ed -80 
= en 
-120 -120 
30-40 0 40 80 480. 00005 40: 80 
absolute L track (meters) absolute L track (meters) 
Before modification After modification 
Figure 5 : improvement of HRS absolute location 
(ellipse is the 3 x std boundary) 
An analytic model has been calibrated in order to improve on 
ground the location performance before the modification (see 
[Bouillon, 2003]). This model is not to use for images acquired 
after the modification. Moreover, it should be replaced by a 
rigorous description of the effect as known for a correct 
description of the impact on images acquired before the 
modification. 
Finally, taking into account all this knowledge, the analysis of 
remaining location errors for all images shows a temporal trend, 
combination of ageing, seasonal stability and discontinuity due 
to the on board modification of HRS' foot thermal control (see 
figure 6). This trend can be modelled and taken into account on 
ground for further improvement of the location performance. 
60 
40 
20 
0 N T T 3 
-20 
-40 
-60 
06-02 09-02 12-02 03-03 06-03 09-03 12-03 03-04 
60 
40 
20 
0 rp Tree TT 
-20 
-40 
-60 
06-02 09-02 12-02 03-03 06-03 09-03 12-03 03-04 
120 
80 
40 
0 S XX E 
-40 
-80 
-120 
06-02 09-02 12-02 03-03 06-03 09-03 12-03 03-04 
Month-Year 
Du p —] 
r 
Pitch (microrads) Roll (microrads) 
Yaw (microrads) 
Figure 6 : Satellite Roll, Pitch and Yaw temporal trend 
computed out of HRS' absolute location performance 
3.3 Stereoscopic and Reference3D performances analysis 
Analysis of the horizontal accuracy of HRS stercopairs, and 
blocks of HRS strips (Reference3D performances) have been 
assessed in collaboration with French National Geographic 
Institute. Main results are presented in [Airault, 2003] and 
[Airault, 2004]. 
First of all, the processing of HRS blocks takes into account a 
specific model for images acquired before removal of the on 
board relativist error, and a temporal model for all images 
including remaining effects described above. These are 
necessary to obtain the best location performance possible for 
all available images. 
Remaining errors are measured through bundle block 
adjustment by taking into account tie points between two 
images of an HRS stereopair, and between different HRS strips. 
The analysis of the process showed out that considering HRS 
instrument as rigid was a necessary hypothesis to manage it. It 
is therefore admitted that HRS' internal parameters do not vary 
with time : this reduces the degrees of freedom and allows a 
good solution. 
Moreover, some possible improvement of HRS inner 
orientation parameters were pointed out. In particular, among 
Reference3D production, a 3000 km wide block is available 
which allows a very precise measure of camera's focal length. 
Finally, this work was carried out on every location site 
available, and also on some Reference3D production sites (total 
of more than 20 sites all over the world). It showed out that 
availability of precise ground control points is the key for a 
good interpretation of errors. As a matter of fact, priority was 
given to sites with GPS points in order to get rid of some 
possible biases of local cartographic systems. 
     
  
  
    
   
   
   
  
   
   
  
   
   
  
  
   
  
  
   
  
  
   
  
  
   
  
   
  
  
   
   
  
  
  
   
  
  
  
  
  
  
    
lox 
Wi 
so 
by 0m [no sr IN 
y M gen tM quest