Would-be Groups of
nets would-be legs
1 It 1h
V+VII + VIII
2 III
3 IV
4 VI
Figure 10. Network structure
In order to solve for the competitions (ambiguities), we use
the information at hand:
® the total lenght of each (would-be) net;
e the number of groups involved in each net;
e the number of competitions in each net.
At the current stage of our work, for each line with an
ambiguity, we evaluate the above informations for the
competing nets; we take only the net with highest score in
all three items, discarding the other candidates. Figure 11
shows the extracted road network for the original image.
Figure 11. The extracted road network
S. PERSPECTIVES
We are just at the beginning of our experience in road
extraction, so be many aspect of our strategy will need a
revision or to be studied in more depth.
As far as the choice of the partition is concerned, a possible
improvement may be to use a small number of partitions,
but to set a procedure capable to further segment the region
within, recognizing e.g. a long smooth edge which would be
poorly approximated by a line segment only. This seems to
be better than increasing the number of partitions, when the
outcome may be sensitive to local disturbances of the
gradient. In the same line of thought, we plan also to use
on RM EEE EEE
206
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
polynomials or splines, rather than line segments only.
The solution of the ambiguities need to. be refined, giving
different weights to each information. More important we
will have to develop additional likelihood measures to
complement the current road extraction strategy in its last
stage.
6. REFERENCES
Barzhoar M., D.B. Cooper, 1995.New geometric stochastic
technology for finding and recognizing roads and
their features in aerial images. ln: Automatic
Extraction of Man-Made Objects from Aerial and
Space Images. Edited by A. Gruen, O. Kuebler, P.
Agouris. Birkhaeuser Verlag, Basel.
Gruen A., H. Li, 1994.Semi-automatic road extraction by
dynamic programming. In: International Archives of
Photogrammetry and Remote Sensing (IAPRS). Vol.
30 Part 3/1, pp. 324-332.
Ballad D. H., C. M. Brown, 1982.Computer Vision.
Prentice Hall, Englewood Cliffs NJ.
Burns J.B., A.R. Hanson, E.M. Riseman, 1986. Extracting
straight lines. In: IEEE Transactions on Pattern
Analysis and Machine Intelligence, Vol. PAMI-8 N.4,
pp. 425-455.
Lowe O.G., 1985. Perceptual Organization and Visual
Recognition. Kluwer, Boston MA.
Nevatia R., K. R. Babu, 1980. Linear feature extraction and
description. In: Computer Graphics and Image
Processing. Vol 13, pp. 257-269.
Sarkar S., Bayer Kim L., 1993. Perceptual Organization in
Computer Vision: A Review and a Proposal for a
Classificatory Structure. In: IEEE Transaction on
System, Man and Cybernetics, Vol.23,N.2, pp.382-
399.
Steger C., C. Glock, W. Eckstein, H. Mayer, B. Radig,
1995. Model-based road extraction from images. In:
Automatic Extraction of Man-Made Objects from
Aerial and Space Images. Edited by A. Gruen, O.
Kuebler, P. Agouris. Birkhaeuser Verlag, Basel.
Vosselman G., J. de Knecht, 1995. Road tracing by profile
matching and Kalman filtering. In Automatic
Extraction of Man-Made Objects from Aerial and
Space Images. Edited by A. Gruen, O. Kuebler, P.
Agouris. Birkhaeuser Verlag, Basel.
KEY WOR
ABSTRAC
The thi
provide
imaging
110 km
testbed
the es!
require
accurac
is pro
control
interpo
discuss
Wich €
Modular
Stereo
German
Space
communi
opportu
potenti
resolut
MOMS - 02
generat
data,
extract
througt
orthoin
geometr
specif]
larger
Seige,
Simulat
had
triangi
Sms in
should
three-:
obtaine
provide
predic!
level.
One of
sequen
mission
covere:
north
Austra
arrang
