The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
1185
Camera
RC 30
DMC
DMC
Year of flight
2004
2006
2007
Flying altitude (m)
4700
4500
2000
Focal length [mm]
153
120
120
GSD [cm]
45
45
20
Number of fotos
136
326
234
Number of flight lines
7
8
10
End lap / side lap |%]
60/25
60/25
60/25
Table 4: Block configurations
For the three blocks 3 DEMs were created with Match-T V5.0
setting the grid width approximately to the recommended value
of 30 pixels; i.e. 15m for 45cm GSD and 7.5m for 20cm GSD.
Low smoothing was chosen allowing the modeled surface to be
close to the correlated point cloud. During the evaluation phase
Match-T DSM V5.1 was released, which allows to further
decrease grid width towards values of 10 pixels and below.
With that version additional DEMs were created with a grid
width of 10m and 5m for 45cm GSD and also 5m and 2m for
20cm GSD. For the comparison only “good” points were used,
which were not classified by Match-T to have low redundancy
or bad accuracy.
LIDAR data served as an independent reference, collected in
January 2007 with an OPTECH ALTM 3025 system from
2 250m flying altitude. Parallel flight lines were flown with
20% side lap and a mean point density of 0.33 points/m 2 in the
non-overlapping areas. After ground classification this data is
converted into a regular DTM with 2m grid width. Its accuracy
is in the dm-level.
Test area was a 7.6km x 5.4km wide region close to the city of
Tarragona in Spain (see Figure 2). It includes a 0.8km 2 sized
test-site in a flat area (TSflat), which contains neither
buildings nor higher vegetation and a 2.8km 2 sized test-site in
the city of Tarragona (TSurban).
Figure 2: Test area (7.6 x 5.4 km 2 ) with the two test-sites
TS flat (upper left) and TS urban (lower right)
The grid points of the different versions were compared to the
LIDAR data. The results are graphically represented in Figure 3.
DEMs in different configuration were created for the entire test
site (Match-T V5.0 only) and for the test sites TS flat and
TS urban. The descriptors in Figure 3 indicate whether a digital
(DMC) or an analog camera (RC30) was used, followed by the
GSD in cm and the grid width deduced from that flight data.
An appendix “_v51” is added if Match-T DSM V5.1 was used;
e.g. “ana45_10m_v51” represents the results of a 10m spaced
DEM obtained with Match-T DSM V5.1 from RC30 images
having a ground pixel size of 45cm.
The DEM calculated for the entire test site as well as for
TS flat are compared to the regular spaced 2m DTM LIDAR
data, representing the ground surface without vegetation and
buildings. The DEM calculated for TS urban is compared to
the unclassified LIDAR point cloud representing also
vegetation and buildings. The results for the three different test
sites represent three different levels of accuracy. The
represented standard deviations a are the results after filtering
out all differences exceeding 3a. In case of test site TS flat
1-2%, in the other cases an average of 15% of the differences
was filtered out.
The most accurate results are obtained for test site TS flat,
where the DEM points are compared to the LIDAR DTM on a
bare ground surface. In this flat area the DEM interpolation
error is small and the comparison is more exact and reliable.
Digital and analog images of 45cm ground pixel size result in
more or less the same accuracy (1/2 ground pixel) with slight
advantages for the DMC results. The digital images of 20cm
GSD result in 8 to 9 cm accuracy reaching the accuracy level of
the LIDAR data. The variation of the grid width has no
significant impact on the results.
In test site TS urban the DEM grid points are compared to the
unclassified LIDAR points. Especially in urban areas the
differences are much bigger, since the DEM grid points can not
represent the height discontinuities at the edges of buildings or
other man made objects. Figure 4 shows a profile of the
unclassified LIDAR points (white) and DEM grid points at lm
spacing derived from DMC images of 20cm GSD (red). The
example shows, that some details resolved by LIDAR are not
resolved by Match-T. For the 45cm pixels the accuracies
obtained with analogue images are slightly better. The 10m and
5m spaced DEM, however, show big gaps where most of the
points were classified to have low redundancy or bad accuracy.
With DMC images a 10m and 5m spaced DEM from 45cm
GSD could be derived without problems.