International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004 
  
  
11 strip, O9 RMS AGCPs RMS NGCPs 
  
X Y d X Y Z 
(144 photo) [um] |. (cm) puc (om 
  
"two-step" 9.4 10.91 95.1.8.2.] 8.7 [169 
  
  
CA| CA 
tal bo 
"one-step" 9.5 10.8! 8.4 | 8.2 | 8.7 | 14.0 
  
  
  
  
  
  
  
  
  
  
  
Table 5 - RMS on the artificial (119) and natural (45) GCP for 
the calibration parameters estimated from the whole 
block. 
As it is apparent, no true differences in ground point accuracy 
can be traced between the two methods; only a small 
improvement seem to emerge as far as elevations are 
concerned. With respect to the reference AT, direct 
georeferencing is about half as accurate in terms of sigma 
naught and for the horizontal, while elevation is just 50% worse 
than AT. Differences between AGCPs and NGCPs are now 
smaller. 
Figure 6 - Systematic component (10 cm) in Y direction 
Decomposing the RMS, the standard deviations are much the 
same as for AT, but a systematic component (10 cm in Y, 4 cm 
in X) is now present in the RMS (see Figure 6), while it was 
insignificant in the AT; this applies also to the results of all 
subsequent calibration datasets. We are still investigating why 
this happened. 
  
  
Figure 7 - Calibration block made of 4 strips, 5 GCP 
A possible explanation may arise from inconsistencies between 
the GPS solution for the flight and the GPS solution for the 
ground network. Although the ground reference stations for the 
flight were the same used in the network, maybe processing of 
the kinematic data didn't result in a high accuracy solution. 
Because of the block design, any systematic error in the GPS 
solution cannot be adsorbed by the offset parameters and will 
show up on the ground. 
In a second set of tests, smaller sections of the block have been 
used: 4 strips (two East- West and two Nort-South, flown twice) 
with 51 images and 5 GCP (see Figure 7); two strips (East- 
West, flown twice) with 21 images and 4 GCP (see Figure 8). 
Figure 8 - Calibration block with two strips, 4 GCP 
The picture emerging from this table (table 9) is less clear. 
There are hints that the one-step leads to somehow better results 
in elevation. This may be simply due to the inner strength of the 
one-step solution, which is less sensitive to poor ground control, 
because the pseudo observations of the IMU/GPS data manage 
to prevent excessive block deformations (which are mainly in 
elevations, with the 300 mm lens). The extended two-steps does 
not improve the accuracy of the standard two-step except in one 
case. On the other hand, sigma naught, which is fairly 
independent of block configuration in all cases with the other 
two methods, is always worse. 
  
  
  
  
  
  
  
  
  
  
  
4 cross strips, Oo RMS AGCPs RMS NGCPs | 
(51 photos) [um] (cm) (cm) 
+5 GCP X Y Z X Y Z 
“two-step” 9.8 | 5.1 |410.9 18.81.84 4] 28.7.127.3 
“one-step” 9.9 4-54 1:10.80 14.31:8.1. +87 119.0 
“two-step” w.corr] 19.2 1°53 {12.01184| 84 {941253 
2 strips, 
( 21 photos) 
+4 GCP E 
“two-step” 10.0. 1°3.2::109:3047 3.2.1 3.3 {39.0 
“one-step” 10.1,21:5:2 5410.8 1.9.4 .1,8.3.1.8 7 112.0 
"two-step" w. corr| 13.3 | 5.3 | 12.0/18.4| 8.4 | 9.4 | 25.8 
  
  
  
  
  
  
  
  
  
  
Table 9 - RMS on the artificial (119) and natural (45) GCP for 
the calibration parameters estimated from. 
The last series of test was performed with the one-step method 
only and with the minimum ground control necessary, by 
varying the number of images. As reported in (Pinto et Forlani, 
2002), we found in previous simulations that just one GCP may 
suffice, unlike "two-steps" where a standard control is 
necessary for the AT; result are shown in table 10. 
The results of previous simulations are basically confirmed also 
in this case: while the horizontal accuracy remain the same with 
decreasing block size and block strength (and does not get any 
worse for the same block with more GCP), there is a clear 
deterioration of the accuracy in elevation. 
  
      
  
   
   
   
   
    
    
    
  
    
     
   
   
  
  
  
  
  
     
   
  
  
  
    
    
    
     
   
    
    
    
   
  
  
  
  
      
   
     
    
    
  
Internat 
T strips 
4 cross s 
2 strips 
  
Table 10 
The pre 
IMU/GPS 
testfield « 
stressed | 
the frame 
just at the 
emerged 
minimal « 
accuracy 
model fo 
improver 
hope to ç 
adjust to 
positions. 
discrepan 
and to e 
whether t 
REFERE 
Referenc 
Gruen, A 
and kinei 
Vol. 59,1 
Referenc 
Casella, ' 
project oi 
Worksho 
Sensor ( 
Spain, on 
Cramer, | 
georefere 
and Rem« 
79-84. 
El-Sheim 
applicatic 
photograi 
and Rem 
95-100. 
Forlani € 
adjustmei 
and Renx 
Heipke, € 
results o 
Heipke €