International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004 
  
  
  
Figure 4: Differential grid of model 116_117 (ISAE), positive 
errors are bright, negative errors dark 
Another effect, influencing the geomorphological correctness of 
the DTM, is the smoothing in the Image Station measurements 
caused by interpolation. Missing terrain information like 
marked in figure 5 is the consequence. 
From the achieved results for Image Station, no trend for AZ 
can be derived for the single matched points because the 
generated DTMs are affected by the ascertained smoothing 
effect. 
To reduce the smoothing effect, the DTMs were generated with 
different filter settings (low, medium and high), but the 
accuracy did not increase significantly. Compared to earlier 
investigations (Grün et al., 2000), where an RMS error of 0.7m 
for Palpa imagery was achieved with ISAE with DTM 
interpolation of the matched points done with DTMZ, this 
indicates that the interpolation module of ISAE determines the 
main part of the height errors. Terrain elevations and sinks of a 
size up to 100 by 50 metres are modeled as flat terrain and 
cause blunders in a magnitude of 10 to 20 metres in model 
116 117 (figure 5). 
  
Figure 5: Manually measured DTM (left) and automatically 
derived DTM using ISAE (right). Marked red: Loss 
of geomorphological information due to strong 
smoothing. 
3.3 Virtuozo™ 3.1/3.3 
Unlike Image Station, Virtuozo uses epipolar images for 
automatic DTM generation, the algorithm consists of 
crosscorrelation and a global relaxation matching technique for 
blunder detection (Baltsavias et al., 1996). The matching points 
are extracted based on a regular grid in image space, and 
therefore the calculated DTM has to be interpolated 
subsequently. For the comparison of the generated DTMs, 
digital terrain models were calculated using both versions 
available at our institute, 3.1 and 3.3, to investigate whether an 
improvement in matching algorithms affects the obtained 
results. In Virtuozo, like in Image Station, different strategies 
for DTM calculation can be chosen by the user. It distinguishes 
between flat and mountainous areas, further parameters that can 
be influenced are the patch width and height. The default value 
of 15 pixels for both width and height is used in this test. An 
integration of manually measured breaklines to refine the 
matched points is also possible. After matching, a visual quality 
control can be accomplished. The matched points are visualised 
in the stereo model, divided into 3 classes (red, yellow, green) 
where the yellow and the green classes contain the more or less 
reliable points and the red class consists of the unreliable 
matching results. Unfortunately, there is no detailed description 
of the point classification in the user manual (Supresoft Inc., 
1999). The points resulting from the matching process still have 
to be interpolated to a grid, in this case with 5m mesh size using 
DTMZ, to compare them with the reference dataset. 
Table 2 shows the results obtained with Virtuozo 3.1 compared 
to the manual measurements, while table 3 contains the results 
for the DTMs generated using Virtuozo 3.3. 
Table 2: Results of the DTM-comparison Virtuozo 3.1 vs. the 
manual measurements 
. Model |, AZ Std.Deviation | Min. - Max. 
116 117}. 17m | 2.95m -17.2m — 18.4m 
1210 211 | -0.03m = 1.00m -9.5m — 6.7m 
| 211 2124... 008m. |. 0.74m _-4.5m — 5.5m 
212 213; -0.06m | 1.35m -24.9m — 11.9m 
213 214 | -0.03m 0.98m |. -10.8m - 8.3m 
223 224 |. 0.02m u Om 5 -5.2m — 4.3m 
Table 3: DTM generated with Virtuozo 3.3 compared to the 
reference DTM 
Model | AZ Std. Deviation Min. — Max. 
116_117 0.35m 2.66m -15.9m — 16.8m 
210. 211.|. 004m. | 0.81m Lu Om =. Im 
211.212 0.13m 0.73m li 004m = 42m 
212.213 | 0.13m 1.20m | -5.3m - 18.6m 
1213 214 | 0.02m 1.03m -10.7m — 7.4m 
223 224 -0.0lm 0.65m -5.2m — 4.3m 
Generally, the standard deviations of the measurements 
produced by Virtuozo 3.3 are marginally smaller than the ones 
from version 3.1. Also the minimum-maximum ranges become 
more narrow in the most models. For both automatically 
generated datasets, the mean height errors are homogeneously 
close to 0m, having similar standard deviations between 0.65m 
and 1.35m. Model 116 117 shows a significant difference: the 
standard deviation of 2.66m and the larger minimum-maximum 
range for the DTM generated by Virtuozo 3.1 compared to the 
smaller values of version 3.3 is a sign of larger blunders. 
524 
  
Interne 
In mo 
Statior 
height 
both s: 
Unlike 
dataset 
surface 
appear 
The re 
results 
cases, 
minim 
that Vi 
3.1 bu 
two pr 
Figure 
34 C 
Figure 
acquire 
hand, 
on the 
3.1/3.3 
The ac 
DTMs 
system 
deviati 
produc 
betwee 
terrain 
Virtuo 
Station 
causes 
rough 
manua 
by the 
Concei 
good r 
Subset