1 2004 International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004 
  
The standard deviations used for the weight matrix were as 
follows (see table 1): 
  
  
  
  
  
j. 
image coordinates + 6 um 
flight object space coords. of x 0.01 m in X,Y and Z 
of the GCP 
of the GPS/IMU position t0.lm 
arious GPS/IMU roll, pitch + 5 deg * 10° 
re, the GPS/IMU yaw + 8 deg * 10° 
riptor 
as the Table 1: Standard deviations of observations used for 
ine. It sensor calibration 
ig the 
|, and All observations were considered as uncorrelated, because no 
it, the other information was available. The values for the position 
from and attitude data are those reported by Applanix. 
hown. 
large In the adjustment, all seven parameters could be determined 
Cen. with high significance. Whereas the results for the six 
conventional parameters were relatively small, the time offset 
was found to be 5.3 msec with standard deviation of 0.3 
msec, and thus approximately one cycle of the 200 Hz data 
set. As mentioned before, however, this value must be 
interpreted as a combination of time offset and correction to 
the principal point in flight direction. 
x 5 3D point determination 
1 5.1 Point determination with a single baseline 
| In the next step, we computed object space coordinates for 
| those GCP which had not been used in the calibration. We 
| used the images from the project flight and only two rays per 
| point and computed the 3D coordinates via forward 
| intersection. In this step the calibration parameters were used 
e as constant values'. The project flight was covered by 
approximately 50 stereo models. All computations were 
tation carried out for each of the available five reference stations. 
The results are shown in table 2. For each baseline the mean 
accuracy of the GPS/IMU observations are given. The 
r part position values come from differential GPS solutions and 
data represent more or less the accuracy of the GPS geometric 
r. to configuration (satellite visibility and length of baseline). The 
eriod attitude values had to be taken from the company information 
GCP since no other information was available. In the right column 
the RMS differences between the computed object space 
coordinates and the known values for the 41 independent 
control points (most of them were not introduced in the 
calibration phase) are given. 
© 
  
ation 
ation The results are in the range of 8 cm in planimetry and 15 cm 
ector in height at independent check points for short baseline (until 
; Was 60 km). For the longer baseline at Stavanger (approximately 
ional 300 km) we obtained RMS values in the range of 10 cm in X, 
arm Y and 19 em in Z. The differences represent the weaker 
geometric GPS configuration and possibly different 
atmospheric conditions between the test field and Stavanger, 
e six which can not be compensated by differential GPS strategies. 
yffset It should be mentioned that these results, obtained with 
time imagery of scale 1:10.000 compare very favourable to the 
luced results obtained in the OEEPE test (Heipke et al. 2002b). 
nage 
>ssed 
Pons ! Note that separate sets of calibration parameters were 
computed for each reference station, and these were used in 
all following computations. 
  
  
  
   
  
   
   
  
   
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Mean accuracy of 
GPS/IMU position and RMS 
Ref. station attitude differences at 
position attitude 
[em] [deg * 107] 
Xo Zo roll, yaw | X,Y Z 
Yo pitch 
Raade 2 S 6,7 -$ =8 | 80 14,5 
(8-30 km) 
Moss 3.1 ga ~ 5 ~ 8 8,1 14,8 
(15-38 km) 
Torp 3.5 8.2 ~5 -8 8,3 IS] 
(20-50 km) 
Soer 3.3 8,1 ~35 ~8 1383 15.4 
(25-60 km) 
Stavanger 4,5 lis s ~8 10,2 19,1 
(307 km) 
Table 2: Results of direct 3D point determination at 
independent check points (ICP) using two-ray 
points and single reference stations, image scale 
1:10.000 
  
5.2 GPS network solution 
We now turn to the GPS network solution. In order to better 
demonstrate the effects of this approach we select only two 
strips, one with good overall GPS data, and another one with 
somewhat worse data. 
Figure 2 shows the same information for the reference station 
Torp as figure 1 does for Raade. It can be seen that strip 1 has 
a small and thus a good PDOP value for both reference 
stations. Strip 2 has a small PDOP for Torp, but a higher one 
for Raade. The graphs for the reference stations Moss and 
Soer are similar to those for Raade. 
We therefore expect, that the results of strip 1 will be good 
overall, and will not be effected by the network solution. 
Strip 2, on the other hand, should show good results when 
computed with station Torp, but worse results when 
computed from Raade. One of the questions was in how far a 
network solution would be able to improve the results 
obtained with the Raade station. 
The results are shown in table 3. The expected tendency can 
be observed when inspecting the values in the table, 
  
  
  
  
Strip Stri 
3: 3. | ; 
J 
1 À L L- 1 
378500 379000 379500 380000 360500 581000 
GPS-Time (sec.) 
T T T 
s 
= 
5 
a 
a 01} 4 
e | ; 
a s 1 i | 
SE | { 
a i i 
= { ! 
> i i 
= 005 n 
3 | 
3 m i EH Hd IU lii i 
ü 1 
  
  
    
Figure 3: — Accuracy of GPS position, reference station Tory