, 9-11 Nov. 1999
esolve step, some points
solution, may become iso-
y re-pruning.
jally have pinholes where
ice within another surface.
joint problem. The final
| on their 8-neighborhood.
ation is:
ns,
ce patches.
ta and real range imagery
| to demonstrate the per-
l.
of these experiments. The
region with smooth tran-
gion such that depth and
ere (Figure 5(a)). Noise
> a noisy set of range data
th joins are more difficult
resses this case reasonably
the segmentation bound-
nd (b) surface contaminated
rs
International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999
a qum Ar uU
Figure 6: Recovered surfaces and segmentation (a) Planel, (b)
Plane2, (c) all three surfaces, (d) segmentation boundaries
6 Discussion and conclusions
Airborne laser scanning is an increasingly popular data acqui-
sition method for generating DTMs. It samples the surface
at high a density and the range measurements are very ac-
curate. However, the raw laser points are not a meaningful
description of the surface because the major surface charac-
teristic, such as breaklines, formlines, smooth surface patches,
and surface roughness, are not explicitly encoded. Surface
segmentation attempts to extract this information from the
cloud of 3-D laser points. We have proposed a region grow-
ing segmentation method that takes the stochastic nature of
laser points into account and it is robust regarding blunders.
We are currently extending the approach to process directly
irregularly distributed data sets in order to avoid gridding as
a pre-process.
Segmenting laser points into meaningful surface patches
would greatly benefit from additional information. Usually,
laser data sets consist of a (huge) list of 3-D points. lt is
conceivable to record additional information, for example the
entire waveform of the returning laser signal, or the scene
brightness. Waveform analysis and analyzing scene bright-
ness could be incorporated into the segmentation process.
For region-growing segmentation to be successful, a suffi-
cient number of points per surface patch is required. This is
usually the case for airborne laser data sets. Since the seg-
mentation identifies smooth surface patches, breaklines and
formlines are defined by region boundaries. Sometimes, the
region boundaries do not determine breaklines well, however.
This is an inherent problem with laser data sets; the spatial
distribution of footprints may be considered a random sam-
pling as far as object boundaries are concerned. It would be
sheer coincidence if a footprint would coincide with an ob-
ject boundary, say a building outline. Even if it did we would
not know. If object boundaries are important for a particular
application, other sources but laser ranging may be required.
For example, stereo photogrammetry allows direct determi-
nation of 3-D object boundaries. Therefore, it makes sense
to combine the strengths of different sensors.
7 Acknowledgements
The example with the Greenland ice sheet is from the ICESAT
project, funded by the NASA [Csathó et al., 1995]. Other
examples used in this paper are from the Ocean City test
site, established by ISPRS WG I11/5 [Csathó et al., 1998].
REFERENCES
[Axelsson, 1999] Axelsson, 1999. Processing of laser scanner
data — algorithms and applications. ISPRS Journal of Pho-
togrammetry and Remote Sensing, Vol. 54, Nos. 2-3, pp.
138-147.
[Besl, 1988] Besl, P. J., 1988. Surfaces in Range Image Un-
derstanding. Springer-Verlag, New York, 340 pages.
[Baltsavias, 1999] Baltsavias, E. P., 1999. A comparison be-
tween photogrammetry and laser scanning. ISPRS Journal
of Photogrammetry and Remote Sensing, Vol. 54, Nos.
2-3, pp. 83-94.
[Boyer et al., 1994] Boyer, K. L., M. J. Mirza, and G. Gan-
guly, 1994. The robust sequential estimator: a general ap-
proach and its application to surface organization in range
data. IEEE Transactions on Pattern Analysis and Machine
Intelligence, Vol. 16, No. 10, pp. 987-1001.
[Csathó et al., 1995] Csathó, B., A. F. Schenk, R. H.
Thomas, and W. B. Krabill, 1995. Topographic mapping
by laser altimetry. Proceedings of SPIE, Vol. 2572, pp. 10-
20.
[Csathó et al., 1998] Csathó, B., W. Krabill, J. Lucas, and
T. Schenk, 1998. A multisensor data set of an urban and