Full text: Proceedings, XXth congress (Part 1)

   
  
   
    
    
   
       
    
   
   
     
  
   
   
   
   
   
   
    
  
  
  
   
   
   
   
   
    
   
  
  
  
  
   
  
  
   
  
inbul 2004 
' and with 
1 RPCs as 
an affinity 
rection (2 
1dividually 
significant 
ustment. If 
of missing 
ken out of 
RMSY 
0.98 m 
0.86 m 
0.82 m 
0.80 m 
| points of 
key 
ation with 
ry similar. 
rection. In 
I. The root 
ons is just 
  
  
  
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004 
  
the Saudi Arabia scene, based on 20 control points with 
differences in the elevation up to 338m, the azimuth has 
changed 2.0°, reducing the root mean square differences from 
1.39m to 1.25m. In another area in Turkey the RMS-values of 
two scenes have not been improved with the nominal collection 
elevation and azimuth as unknown, but the 6 affinity parameters 
have been significant and required. 
  
RMSX RMSY 
  
CORIKON, 8 unknowns (affine + 1.45 m 0.97 m 
view direction) 
  
CORIKON, 6 unknowns (affine) 1.51 m 1.00 m 
  
CORIKON, 2 unknowns (shift X, Y) 1.61 m 1.09 m 
  
  
CORIKON, only terrain correction 2.3] m 4.04 m 
  
  
Table 2. root mean square discrepancies at 33 control points of 
I 
IKONOS orientation, New Jersey pixel size Im 
In the case of the New Jersey data set the program CORIKON 
has removed the nominal collection elevation and azimuth as 
not required from the adjustment. This decision is based on a 
Student test and it takes also care about correlations of the 
unknowns. Also an affinity transformation of the terrain 
corrected IKONOS positions to the control points has had a 
limited influence, only a shift in X and Y was required. With 
just the terrain correction, the result is mainly depending upon 
the direct sensor orientation which is causing a shift in X of 
1.7m and in Y 3.9m. In several countries the accuracy of the 
direct sensor orientation is quite better than the problems with 
the datum of the national coordinate system, requiring a fit to 
the actual used coordinate system at least by a shift in X and Y. 
3.2 QuickBird 
  
of the terrain corrected scene positions to control points) was 
not much better like an adjustment with just the 3 significant 
unknowns (shift in X and Y together with nadir). Without an 
improvement of the nadir direction the discrepancies have been 
the same, so in addition to the relief displacement just a shift in 
X and Y was required, that means by theory only one control 
point could be used. The shift values of 77.7m for X and 9.2m 
for Y do present more the datum problems of the national net 
than the accuracy of the direct sensor orientation. The 
rectification to the GTOPO30-DEM is reducing the height 
differences between the actual terrain surface and the reference 
plain. This has the negative effect of a poor convergence of an 
adjustment with the view direction, but it is still possible to 
improve the view direction. On the other hand an improvement 
of the view direction has only limited influence caused by the 
smaller differences in height between the actual DEM and the 
GTOPO30-DEM. 
3.3 SPOT 
  
RMSX RMSZ 
  
CORIKON 8 unknowns (affine + 4.88 m 3.51.m 
  
  
view direction) 5.40m 4.4] m 
CORIKON 3 unknowns (shift X, Y, 6.27 m 4.05 m 
nadir) 6.72 m 4.38 m 
CORIKON 2 unknowns (shift X,Y) 14.97 m 15.29 m 
24.99 m 21.78 m 
  
  
  
  
Table 4. root mean square discrepancies at 39 control points of 
2 SPOT S-scenes in Zonguldak pixel size 5m 
  
  
RMSX RMSZ 
  
CORIKON 8 unknowns (affine + 0.58 m 1.27 m 
view direction) 
  
  
  
onding to 
nowns 
RIKON 8 
e, so with 
zimuth has 
uracy. The 
nge of the 
significant 
collection 
ificant but 
he affinity 
out of the 
ies at the 
1 has been 
er areas. In 
CORIKON 3 unknowns (shift X, Y, 0.58 m 1.36 m 
nadir) 
  
  
  
CORIKON 2 unknowns (shift X,Y) 0.58 m 1.36 m 
  
  
Table 3. root mean square discrepancies at 45 control points of 
the QuickBird Standard image Saudi Arabia 
pixel size 0.61m 
  
  
  
The geometric situation of QuickBird OrthoReady images is 
exactly the same like for IKONOS Geo. From QuickBird with 
the Basic Imagery also the original data only corrected by the 
internal calibration are available together with the full sensor 
information, but a full scene has to be bought. With QuickBird 
Basic images a standard deviation of 388 control points of 
SX=0.55m and SY=0.64m corresponding to approximately 1.0 
pixel has been reached (Passini, Jacobsen 2004). This accuracy 
is usually limited by the identification and the ground 
coordinate accuracy of the control points. This accuracy of | 
pixel is an operational accuracy which can be reached with 
most of the space sensors. It is totally sufficient because the 
registration of object elements is in most cases limited by the 
object definition itself. 
In an area in Saudi Arabia the control point definition in the 
images and the ground coordinate accuracy was not so good, so 
only an RMSX of 0.58m and an RMSY of 1.27m has been 
reached in the QuickBird Standard image. This is still sufficient 
for operational applications. The orientation with 8 unknowns 
(azimuth and elevation together with an affinity transformation 
SPOT scenes are available as original sensor images just 
improved by the internal calibration as level 1A and as rectified 
images, the level 1B. The level 1B-product is more close to the 
map geometry, but it has lost the original geometric relation. 
The average of the root mean square discrepancies (RMS) at the 
control points of 2 level 1B-orientations in the area of 
Zonguldak is with 4.6m below the pixel size of 5m. Such an 
operational accuracy will be reached also with the level 1A- 
scenes. So the reconstruction of the imaging geometry with 
program CORIKON does not lead to a loss of accuracy in 
relation to level 1A-scenes like sometimes reported and it is 
also not necessary to reconstruct the original image geometry. 
With just the shift parameters and a correction of the nadir 
angle with 5.47m the average RMS values have been raised just 
from 0.9 to 1.1 pixels, but without an improvement of the nadir 
angle the result could not be accepted. 
3.4 IRS-1C 
  
RMSX RMSZ 
  
CORIKON $8 unknowns (affine + 9.55 m 7.04 m 
view direction) 
  
CORIKON 3 unknowns (shift X, Y, 9.40 m 7.17m 
nadir) 
  
  
  
CORIKON 2 unknowns (shift X,Y) 10.53 m 8.50m 
  
  
Table 5. root mean square discrepancies at 44 control points of 
an IRS-1C-scene in Zonguldak — pixel size-5.8m 
  
Like SPOT, also for IRS-1C the original (level 1A) and the 
rectified (level 1B) scenes are available. For nadir images IRS- 
IC has a pixel size of 5.8m on the ground. With just a nadir 
angle of 2° the available scene in the area of Zonguldak is close 
to this. The image quality was not optimal - this may be caused 
   
  
  
  
   
  
  
  
  
  
  
  
  
  
  
   
  
  
   
   
  
   
  
   
  
   
  
  
  
   
  
  
  
    
  
  
  
  
   
  
   
  
  
   
     
 
	        
Waiting...

Note to user

Dear user,

In response to current developments in the web technology used by the Goobi viewer, the software no longer supports your browser.

Please use one of the following browsers to display this page correctly.

Thank you.