4.2 Info Cloud Merge
In case of airborne (non-oblique) imagery, the merge of info
clouds can be carried out similar to the disparity merge, based
on the rasterization of individual results as proposed by Hirsch-
müller (2008). With gap filling and consistency checks com-
parable to section 2, the final geometric result will be a gridded
2.5D DSM. A corresponding true ortho-image could be derived
from the info cloud's color data.
Especially 1f the info cloud is a product in itself, it is desired to
merge the three-dimensional results, which also increases point
density. This idea 1s illustrated in Figure 4. Note that the exam-
ple has been combined in a straightforward way, without geo-
metric or radiometric adaptation.
A more sophisticated merge of info clouds should be carried out
as an integrated geometric and radiometric adjustment — based
on the entire information provided by the info cloud, similar to
the combined geometric/radiometric point cloud matching ap-
proach used by Gehrke (2012) for ADS quality control.
5. EXAMPLES FROM DIFFERENT SENSORS
Our SGM implementation has been used to process very large
amounts of ADS data sets, at North West Geomatics’ (North
West), Leica Geosystems, their customers and other institutions.
The multi-sensor extension was run with DMC/DMC-II and
RCD30 frame data so far. Info cloud examples from all these
sensor types are discussed in the following; Table 2 gives an
overview on the data sets and performance numbers.
Figure 5: RGB colored info cloud of an ADS80 job in Down-
town Sacramento (top) and enlarged TIN view of the California
State Capitol (bottom).
5.1 Sacramento (ADS80)
The Sacramento, California, data set was captured with an
ADS80 for North West production in 2011. It covers the entire
metro area in a GSD of 15 cm. The SGM-based image cloud
was derived from all panchromatic views of the ADS, i.e. nadir/
backward and nadir/forward stereo pairs. The results of which
were merged in disparity space as described in section 4.1. RGB
and near-infrared color was assigned to the info cloud based on
the multispectral nadir views. The RGB colored info cloud from
an SGM job in Downtown Sacramento is shown in Figure 5.
The TIN model of the California State Capitol (Figure 5, bot-
tom) shows a good representation of building edges, with only
minor issues (e.g., above the entrance). Shadow areas next to
the building and the cupola are fully and correctly covered by
points, even though such areas are most challenging for our
SGM implementation and image matching in general (cp., e.g.,
Legat, 2012). Points located on building walls and below trees
were eliminated during processing.
5.2 Bregenz (RCD30)
This small RCD30 block in Bregenz, Austria, consists of three
images with a nominal GSD of 15 cm; the stereo overlap is
about 60%. The data was provided by Leica Geosystems for the
very initial testing of our multi-sensor SGM. Info cloud results
from one of these stereo pairs are shown in Figure 6.
Figure 6: RGB colored info cloud derived from RCD30 data in
Bregenz, Austria (top) and enlarged TIN view (bottom).
Com
in Fi,
pavec
no re
for c
due tt
es, vi
easily
more
edge
recog
Figur
lap in
ding
that b
