The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
the quality is almost identical. The performance of DG for the 
GSD 14cm block is slightly worse. This can also be seen from 
the sigmaO which is the worst of all three blocks. For the 
vertical accuracy DG delivers less accurate results compared to 
the ground control point based AT. This might happen due to 
the extrapolation character of DG compared to interpolating AT, 
which is less sensitive to any small and non-corrected system 
calibration parameters. But even though the vertical accuracy is 
worse for DG it still stays within 1.5 - 2.7 GSD. 
If one looks for the AT cases based on control points the 
sigmaO comes down to 1/5 of a pixel especially when 
introducing additional self-calibration. Comparing the two self 
calibrating cases (i.e. the fairly complex Griin 44 parameter 
model versus the 3 interior orientation parameter corrections) 
the finally obtained absolute accuracy is very close in all cases. 
This shows that in this case the correction of interior orientation 
is already sufficient to obtain high performance for object point 
determination. Additionally it was tried to add additional 
corrections for radial lens distortion, which in this case does not 
lead to any improvement in object accuracy. This shows that 
after lens calibration the data is almost free of non compensated 
radial distortion effects. 
What also should be pointed out is the fact, that the interior 
orientation corrections have been significantly estimated from 
bundle adjustment using control points exclusively. No 
additional exterior orientation (EO) parameters from GPS/IMU 
have been introduced. Typically such adjustment is not possible 
mainly due to the large correlations between the estimated EO 
parameters and the estimations for focal length and principal 
point correction. In case of this dual head camera installation 
the tilt between the two camera axes and the additional high 
overlap of images allowed for the de-correlation of interior 
orientation (10) parameters and exterior orientations. From this 
the 10 parameters are determined even without use of directly 
observed GPS/IMU exterior orientations. 
For the GSD 20cm block one final additional adjustment variant 
is given, which only relies on the use of 4 control points located 
in the four block comer. This is only 5% of the number of 
control points used for the previous classical AT. Additionally 
the camera perspective centre coordinates from GPS/IMU are 
introduced as weighted observations with std.dev. of 5cm for all 
three coordinate components (so-called GPS-supported AT). 
Note, that no additional unknowns are introduced during 
processing, neither any additional parameters for self- 
calibration, nor any offset, shift or drift corrections as they are 
often used in GPS-supported AT. Thus the obtained accuracy 
has to be compared to the “AT no” case of the same block. If 
one compares the absolute accuracy the horizontal performance 
is slightly worse compared to the standard AT with 77 GCPs. 
The east component is in the range of 1/2 pixel GSD, the north 
component reaches 3/4 pixel GSD. The height accuracy is 
identical to the classical AT case, which shows the larger 
influence of perspective centre coordinate observations on 
height accuracy mainly. 
If one compares the absolute accuracy from check point 
analyses to the GSD of each of the flights, the accuracy in east 
component is between 20-30% of a pixel, the accuracy in north 
is between 35-60% of a pixel. This is clearly within the sub 
pixel range. For the vertical axis the accuracy is about 0.8-1.2 
pixel GSD. 
If finally the empirical accuracy from check point analysis is 
compared to the theoretical precision values, one also can see 
high agreements, especially for GSD 20cm block and also for 
the GSD 7cm block. This in general proves that after use of 
additional parameters systematic errors are effectively 
eliminated. As mentioned earlier, the correction of interior 
orientation (i.e. only three additional parameters for each 
camera head) in this case seems to be fully sufficient. 
The block GSD 14cm performs slightly worse and shows larger 
differences especially in the vertical component. Here the 
precision values cannot be reached by absolute accuracy. 
Still there are some larger differences between precision and 
accuracy values in the horizontal (especially east component), 
but one has to keep in mind, that the accuracy of our reference 
points is of the same order (1cm) than the precision of this 
coordinate. Thus the term “reference coordinates” is not fully 
valid, at least for this high quality, 1 -2cm accuracy requirement. 
6. CONCLUSION 
The results of an accuracy test of a Dual-DigiCAM-H/39 are 
reported. From the data obtained during a flight over a well 
controlled test field, two topics were investigated. 
1. The GPS/IMU trajectory was processed using a local GPS 
base station, a virtual base station and using no base station 
but only precise 'orbits and precise clock information 
(“PPP”). The differences in attitude between the 
trajectories are well below the accuracy of the used 
GPS/IMU system. The maximum difference in position 
between the solutions with the different base stations was 
below 2cm. The max. position difference between the base 
station solution and the PPP solution was about 7cm for 
the horizontal and about 21cm for the vertical component. 
It was shown, that for many sensor orientation tasks, it is 
not necessary any more to use data of a local base station 
or of a virtual base station. The “PPP” method delivers 
results with a sufficient accuracy. 
2. The geometric accuracy of the dual camera was evaluated 
using independent check points. The position of the check 
points was determined by direct georeferencing and by 
classical AT using different sets of self-calibration 
parameters. The use of the focal length and the principal 
point as free parameters showed very similar results to the 
use of a full set of 44 self-calibration parameters. The 
absolute accuracy of the checkpoint coordinates was 0.2 to 
0.3 pixel GSD in east direction (flight direction) and 0.35 
to 0.6 in north direction (cross flight direction). The 
vertical accuracy was about 0.8 to 1.2 pixel GSD. For the 
case of direct georeferencing the horizontal accuracy was 
in the range of half a pixel for east and about one pixel for 
north. The vertical accuracy was about 1.5 to 2.7 pixel 
GSD. 
The investigations described in this paper confirmed, that the 
data from the tested dual camera installation can be processed 
successfully with standard software packages noted above. 
The calculation of the geometric accuracy showed very 
satisfactory results for data processing with traditional AT as 
well as with direct georeferencing. Further tests will be done to 
point out the performance of integrated sensor orientation with 
low number of control points. 
1056