The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B4. Beijing 2008
1168
histograms, in particular for the data take on 30 Ih April from
2000m, the peak is more flat with offsets. As the difference of
DSMs was calculated over urban area for every pixel, there are
many reasons causing these differences. As mentioned above,
systematic offset emerge from systematic errors in the D-GPS
coordinates, but could emerge also from systematic differences
between the LIDAR reference DEM and the 3K DSM. One
difference is that the 3K DSM contains the vegetation surface,
whereas in the LIDAR reference DEM the vegetation was
eliminated. Besides, in the 3K DSM generation, many errors are
caused by shading effects, moving objects, surfaces with low
texture, etc.
10
%
5
0
H=2000m B/Z=1:4 ?
17/06/2007
n\=40000
Difference in [m]
Figure 6 Histogram of DSM difference image (3K DSM
minus Reference DEM) for three different data takes
over urban area
Table 3 shows the results of the performance tests of the DSM
algorithms in terms of the point density, the calculation time,
and the number of outliers. From 1000m a. G. the highest point
density at 4.4 points per square meter is reached after region
growing. Hence, the calculation time is also very high at
together 655 minutes for a square kilometre, which is definitely
too long for near real time disaster monitoring. Point density
and processing time reduce with higher flight heights, e.g. from
2000m the reached point density is only 1.35 resp. 0.65 points
per square meter and the calculation time 204 resp. 241 minutes.
The difference between the point densities from the 2000m data
takes is mainly caused by the “image decorrelation” over urban
areas. The decorrelation is related to the base-to-height ratio,
e.g. the data take on 17 th June has a higher base-to-height ratio
(1:4) as the data take on 30 th April (1:17).
H [m]
Hierarchical
matching
Region
growing
p/m 2
t/km 2
p/m 2
t/km 2
%
30/04/2007
2000
0.07
70
1.35
134
4.5
30/04/2007
1000
0.19
250
4.44
405
4.6
17/06/2007
2000
0.02
101
0.65
141
4.5
Table 3 Performance of DSM algorithms in terms of point
density [p/m 2 ], calculation time in minutes [t/km 2 ],
and number of outliers in [%]
4. APPLICATIONS FOR NEAR REALTIME DSM
4.1 Monitoring of slides and avalanches
Land slides, slope failures or other movement of masses are a
big natural threat in montane regions. Facing natural disasters
like this, the BOS and rescue forces need detailed information
about the situation in these regions, which could partly derived
from remote sensing imagery.
2000 2006 2007
Figure 7 The slope failure in Austria in the years 2000, 2006,
and 2007
A study in the years 2006 and 2007 examined the potential of
the 3K camera system to monitor a large area slope failure. Test
area was Vorarlberg in Austria, where a slope failure moving
fast at times during the last 150 years threads human villages. In
this area, a weak layer parallel to the surface is mainly causing
the natural event. The tear-off edge of the slope failure moves
uphill and threads a small village above the slide. Figure 7
shows the change of the slide since the year 2000.
3K images were acquired on the 27 th April 2007 in three flight
strips from 1500m to 2000m a. G. Around 25 Ground control
points were measured with GPS and reference DSMs from the
years 2000, 2003, and 2006 were acquired by the Austrian
cartographic office.
Using the 3K image data, a DSM of the region around the slide
was generated according to the proposed processing scheme.
The absolute accuracy of the DSM was validated with GCP.
Thus, the accuracy of the DSM in position is around 14cm and
in the height around 40cm. Given height variations of several
meters, the accuracy is sufficient for this kind of slide.
Figure 8 shows the difference of the DSM between the 3K
based DSM from the year 2007 and the DSM from the year
2006. In this case, enormous movement of masses between the
two acquisition dates could be detected in the difference image.
The surface height varies up to 30m.
Around 4.5% of all matching points were detected as outliers
during the forward intersection. Outliers detected in the
bidirectional matching before are not included in this number.
The remaining outliers which were not detected are below one
point per million (moving objects not included).
The movement of the masses can be better seen in the cross
section at a profile line through the whole slope (see figure 9).
It could be seen, that the movement of the slide downhill is
separated in different zones of erosion and accumulation, i.e. in