International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
6.2 Discussion and Future Work
In the example presented, initialization for one window
leads to successful matching of similar windows once they
are properly extracted. So far, a pose estimate 1s only car-
ried out for matches which yield at least 4 intersection
points. That means that windows are found where there
is at least one edge per side actually extracted from the
range image. It is desirable to allow for a limited amount
of uncertainty so two or three edges per window can be
used to instantiate a model. This will be subject to further
investigations.
The extracted windows can be used to propose a pattern
in which windows are arranged along a façade. Windows
found by the same model with the same constraints param-
eters are used for that. This makes it possible to predict the
presence of windows also for spots where no structures are
found: The hypotheses can then be used to direct another
search step.
Another way of improving the fit is to use weighted esti-
mation of parameters. Possible candidates for weights are
the following:
|. Intersections of edges that are actually present in the
edge image could be assigned a higher weight than in-
tersection points that are calculated by prolongations
of edges.
t2
. The length of the extracted edges could be used in
some way.
7 CONCLUSION AND OUTLOOK
A semi-automatic method for finding multiple occurrence
of a shape in a building's facade has been proposed. In this
paper, we have described how we applied segmentation
of laser scans to produce an edge image and constrained
search to match structures in the edge image.
In the future, there are several applications for this proce-
dure in our research project. We will apply the algorithm
to photos of a building as well in order to find correspon-
dences between the laser scan and the photo. The objective
is to automatically apply textures to 3D models of a build-
ing derived from a laser scan. Models for shapes so far
consist only of straight lines. It is planned to extend the
model library so that models contain parameterized curves
as well.
Once structures are found, properties describing a build-
ing's facade can be defined. For example, one can count
the number of windows which are arranged horizontallay
or vertically. It is even possible to conclude the number
of storeys that a building has. It is also possible to de-
termine the relative size and position of windows or other
structures by comparison to the building's size and geom-
etry and identify a particular building amongst others in
images coming from a different data source.
1084
It is also possible to use this algorithm for registration of
different terrestrial laser scans. From structures found in
every single scan, one could estimate the relative pose of
these scans to each other and calculate a transformation.
This way, structures found in buildings by our algorithm
can replace tie points which are generally used for this task.
REFERENCES
Burns, J. B., Hanson. A. R. and Riseman, E. M., 1986. Ex-
tracting straight lines. IEEE Transactions on Pattern Anal-
ysis and Machine Intelligence 4, pp. 425-455.
Flynn, P. J. and Jain, A. K., 1991. BONSAI: 3-D Ob-
ject Recognition Using Constrained Search. IEEE Trans-
actions on Pattern Analysis and Machine Intelligence 13,
pp. 1066-1075.
Grimson, W. E. L., Lozano-Pérez, T. and Huttenlocher,
D. P, 1990. Object Recognition by Computer. The MIT
Press.
Gülch, E., 1994, Erzeugung digitaler Gelándemodelle
durch automatische Bildzuordnung. PhD thesis, Univer-
sity of Stuttgart, Institute of Photogrammetry.
Niemeier, W., 2001. Ausgleichungsrechnung. Walter de
Gruyter.
Rottensteiner, E, 2001.
buildings based on hybrid adjustment using 3D surface
models and management of building data in a TIS. PhD
thesis, Vienna University of Technology, Faculty of Sci-
ence and Informatics.
Sester, M. and Fôrstner, W., 1989. Object location based
on uncertain models. In: Proc. 11. DAGM Symposium,
Hamburg, Informatik Fachberichte, Vol. 219, Springer
Verlag, pp. 457—464.
Vosselman, G., 1992. Relational Matching. Springer-
Verlag.
Walker, E. L., 1999. Combining geometric invariants with
fuzzy clustering for object recognition. In: Proc. of the An-
nual Conference of the North American Fuzzy Informauon
Processing Society, pp. 571—574.
Wang, X., lotaro, S., Taillandier, E, Hanson, A. R. and
Teller, S., 2002. Recovering façade texture and microstrue-
ture from real-world images. In: Proc. of the ISPRS Com-
mission III Symposium: Photogrammetric Computer Vi-
sion, pp. A-381 ff.
Werner, T. and Zisserman, A., 2002. Model selection for
automated reconstruction from multiple views. In: Proc.
of the British Machine Vision Conference, pp. 53-62.
ACKNOWLEDGEMENT
The presented work has been done within in the scope
of the junior research group “Automatic methods for the
fusion, reduction and consistent combination of complex,
heterogeneous geoinformation”. The project is funded by
the VolkswagenStiftung.
Semi-automatic extraction of
Insti
ABST
There
past. A
in sucl
tion cc
for obi
propos
] IN
When
proble
sensor
to obte
new se
terresti
quire d
nowad
tured a
11 €
What 1
ject? «|
scannii
bound:
consist
proach
origina
reduce:
Usuall:
context
for vari
e F^
an
of
to
je
su
an
vi:
mi
de
po
