International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part BI. Istanbul 2004 
image coordinates were 4-12 um. In Figure 5 an example of the 
effects in a 5x5 point grid are shown as a line plot for optics 
7183; blocks 2119 and 2120 were consecutive, block 2124 was 
flown 2 days after block 2120 and block 2137 4 months later 
than the others. Corrections of the two consecutive blocks were 
similar. Even the first and the last calibrations with 4 months 
difference had some similarity. Stability of the corrections was 
different for various optics. Further conclusions are not drawn 
from these results yet, because the analysis is unfinished. 
4.4 Accuracy of DG 
The main purpose of the DG accuracy evaluation was to 
compare different block structures. The boresight, interior 
orientation, k1 and global shift (if 20 GCPs) parameters were 
determined with the different block structures, and the 
obtained values were then applied in DG. Also a case with 
only boresight parameters was evaluated (I-block without 
GCPs). The RMSEs in image space are shown in Figure 6. 
Based on the previous results, it could be expected that at least 
the effects of yO-correction and the principal distance 
correction with optics 7183 should be visible. The importance 
of yO-correction is clear, when comparing the results with and 
without yO-corrections (Figure 6). 
An interesting observation was that the accuracy in x-direction 
was clearly better than in the y-direction. Various additional 
parameter models were tested, but they did not eliminate the 
difference. Feasible explanation for the difference in many 
cases is that the roll accuracy appeared to be worse. than the 
pitch accuracy, especially in the blocks 2120, 2124 and 2137. 
The accuracy of I-blocks with and without GCPs were quite 
similar. An exception is optics 7183 where GCPs were 
essential due to the large principal distance correction. 
These results are consistent with the expected accuracy. 4 
mgon accuracy of orientation angles indicates approximately 9 
um resection accuracy with the 150 mm optics and 13 pm 
resection accuracy with the 214 mm optics. With 153 mm 
optics and optics 7163 the accuracy in x and y coordinates was 
about 10-15 um, if principal point error was corrected. In 
optics 7183 the height error deteriorated the results. The 
accuracy of block 2120 was poorer than the others; a good 
explanation for this is the worse attitude accuracy that has been 
detected in earlier studies. 
  
x-direction (micrometers) y-direction (micrometers) 
^ 
^ HAN 
M 
PAS m ^ 
1m d 
TT Per Yun e 3 
v^ 
D^ OE NDA e ape Det - 
x - 
| 
| 
m 
        
  
  
  
   
——2119x —2--2120x | 
2124 x 2137 x | 
10 4-4 | rr m : rrr] 1° 
PFPA 
| -100 | -50 9 {50 | 100 | | | 
  
Figure 5. Image coordinate corrections in 5x5 grid caused by 
Ebner’s parameters for four calibration flights of optics 7183. 
5. ANALYSIS 
  
The results of this investigation are consistent with the 
previously presented results of DG. The importance of the 
determination of interior orientation parameters along with the 
boresight parameters has been observed widely, e.g. Cramer ef 
al. (2002), Heipke ef al. (2001), Wegmann (2002), Jacobsen 
(2003) and Honkavaara et al. (2003). 
Cramer et al. (2002) preferred the mathematical models for 
additional parameters instead of physical models. The above 
results are in accordance with Cramer's results. With Ebner's 
parameters the image corrections were less than 12 jum, and 
with most of the optics less than 5 jum. The use of the full set 
of physical parameters resulted in high correlations between 
various parameters (see also Jacobsen 2003), which in turn 
affected instability to the calibration parameters. The best 
approach is probably to use some central physical parameters 
and then use mathematical parameters to further extend this 
parameter set. It is still questionable, what are the best 
additional parameter models. 
It appeared that the use of the additional parameters, other than 
interior orientation corrections, would probably not be essential 
in DG with the examined optics. The accuracy improvement 
due to more accurate imaging model did not become visible, 
because in the evaluated cases the quality of the orientation ob- 
servations was the bottleneck. In order to verify this conclu- 
sion, additional tests have to be made and calibration results 
from a longer time period should be analysed. Accuracy of DG 
was consistent with the angular accuracy (10-15 um in image if 
significant interior orientation corrections were made). 
Many results of system calibration have been obtained using 
comprehensive block structures. As it seems that the calibra- 
tion with film cameras have to be carried out frequently (Baron 
et al. 2003, Schroth 2003), procedures should be developed for 
minimal block structures. In the empirical study the principal 
  
DG accuracy in image in micormeters 2 
oh ; ce u 
  
  
  
  
  
  
  
| | 13153, cz153 mm 13026, c=153 mm | 
ul 
o T 
= | | | | 
c f er | I |bo| f 1| 1j|bo |f | I | jbo| f | | 
| | | | 
eile re | 12 | 2 0 |re 12| 2 0 re 12/2 0 re 12/0 |re 
pr 1. PISE 2128 | 42134 2119 
  
    
bo| f | | {bol f | 1 
  
12/20 
re|12| 2 | 0 |re|12] 2 
Hal | 
| 
2120 | 2124 2137 | 2129 2135 
  
7183, cz214 mm 7163, c=214 mm 
Figure 6. Accuracy of direct georeferencing for 10 blocks. Symbols below the graphs indicate block structure (f=full, I), number of 
GCPs (12, 2, 0), block name (2121, etc.) and optics (13153 etc.). Calibration parameters: full and I-block with 2GCPs: boresight, 
dc, dx0, dy0, k1 and global shift; I-block with 0 GCPs: boresight, dx0, dy0, k1, I-block with 0 GCPs: boresight (abbreviation bore). 
170 
Interi 
  
point 
with 
no G 
defor 
appli 
Thes 
Singl 
accul 
singl 
GCP 
a sin 
are s 
The 
exter 
know 
numl 
corre 
indic 
data 
quali 
In th 
les 1 
à COI 
12 € 
GCP 
Calil 
The 
entat 
math 
are : 
imag 
of D 
quali 
Thes 
mete 
phys 
radia 
they 
treat 
corre 
comi 
Prin 
accu 
fligh 
adva 
princ 
Baro 
Inert 
Thec 
entat