A, 9-11 Nov. 1999
le is 480m x420m.
ccording to these points is
ences owns a forest enter-
of Vienna, Austria. For an
ta of a first pulse flight by
e point density is 9 points
lights were performed dur-
As it can be seen in fig. 4
are planted in the area.
iginal data. The vegetation
are still in the data. (Note
border.) This DTM was
ummer flight. No filtering
a perspective view with the
ain. The width of the area
hown. A filtering with our
in the ground model. The
our alorithm was a winter,
| Is shown, no filtering was
e seen, while the deciduous
of the lack of foliage. In
ed over the terrain. The
om line of the valley) are
ildings are eliminated. The
can be seen in the figure,
completely. As mentioned
innot penetrate dense bush
ring would have resulted in
tails.
ucted for this area as well.
re measured manually. The
co-ordinate is =5cm. The
s 4096. The r.m.s.e. of the
ese check points is H66cm.
~ording to terrain slope, the
uracy can be derived. For
is 20cm, growing linearly
ore for even steeper areas.
International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999
Figure 4: Summer first pulse model with an ortho photo draped over it
f
5
I5,
k des EN
Figure 6: Winter last pulse model, no filtering was performed
for this model
This result can be improved significantly, if a systematic er-
ror is removed from the DTM. For this end we compared
the break lines of the DTM ([Rieger et al., 1999b]) with the
check points, which were measured manually along the same
break lines. À systematic shift between the check points and
the detected lines in the xy-plane could be observed. Ac-
cording to this investigation the error in the geo-referencing
is about 2m in the xy-plane. After shifting the DTM the ac-
curacy results are significantly better: for 0%—-10% slope an
accuracy of +16cm, for 65% slope an accuracy of +50cm.
On one hand, this indicates how important a correct geo-
referencing of the data is. On the other hand, it shows that
it is not enough to have only one GPS reference station, as
it was the case in this example.
4.3 Conclusion from the examples
There are a number of conclusions to be drawn from the
examples presented and the overall experience we gained with
the processing of laser scanner data.
e The quality of the ground model can be increased sig-
nificantly by applying a qualified filtering method. The
filtering performed by the companies can be seen as a
pre-filtering. There is still a considerable amount of
manual work in the processing.
The height accuracy of laser scanner data in flat ter-
rain is around 20cm. It deteriorates with increasing ter-
rain slope. The accuracy of £0.5m is reached between
60% and 70% terrain slope. If the geo-referencing is
not correct, the accuracy becomes worse, especially
for mountainous terrain. This result is slightly bet-
ter than our earlier investigations [Kraus et al., 1997]
and [Kraus and Pfeifer, 1998].
It is not enough to use only one GPS reference station
to perform the geo-referencing of the data. One solu-
tion, which does not need more GPS reference stations,
is a block adjustment with height and plane control
points of the local co-ordinate system. Original laser
points (no grid-points) need to be identified, which
match these control points. The parameters of a spa-
tial similarity transformation can be obtained by per-
forming the block adjustment. These parameters can
be used to either transform the original laser points or
the grid points into the local co-ordinate system.
A higher point density does not automatically yield a
higher accuracy of the DTM. However, geomorpholog-
ical detail can be captured much better with a higher
point density. This results in an increase of accuracy
at break lines or other features. Nevertheless, the po-
sition of these features can be shifted because of the
filtering process.
5 Outlook
In order to increase the amount of automation in the
processing and evaluation of laser scanner data the ex-
traction of break lines will be necessary. There are