the hypothesized roof lines. Assuming that image coor-
dinates of these roof lines and the corresponding image
lines only differ from each other due to the deviation of
hypothesized roof height and true roof height, the radial
distance (distance in direction to nadir point) between the
corresponding roof and image line can be used to compute
their difference in height. Figure 6 shows the situation of
figure 5 as a 3D view with matched image lines and the
buildings constructed using the 2D ground plans and the
assumed height parameters. After the matched image lines
have been determined in 3D, the hypothesized height and
slope of the roof can be adopted by a least squares ad-
justment. Even though it is possible to use a single image
for the verification task, it is fraught with meaning to add
at least information extracted from a second image to get
results which are more reliable. This second image will
always be available if an aerial image flight was carried
out.
5 CONCLUSION
Within this article we discussed and presented first re-
sults of a procedure aiming on the 3D reconstruction of
buildings by combing preexisting outlines of buildings and
image data. To generate hypotheses for the 3D shape of
buildings the ground plans extracted from a digital cadas-
tral map were generalized the combination of adjoining
ground plans and the simplification of the contour lines.
The information on the use of the buildings also avail-
able from the digital cadastral map was utilized to define
the unknown height parameters of the buildings. The hy-
potheses generated by analysis of the 2D data base then
were verified by matching lines which were extracted from
an aerial image against the lines of the hypothesized build-
ing model. Because the building 3D hypotheses can be
projected into image space using terrain heights, which
are e.g. provided by an existing authoritative DTM, in
principle a single image can be used for the verification
process, like it was carried out in this article. Nevertheless
in our opinion it will be necessary to use at least two over-
lapping images to increase the reliability of the verification
and parameter estimation process.
Image interpretation is far from being solved in complex
290
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
areas like build-up regions. Nevertheless it is very promis-
ing to support this process by additional information like
the used 2D GIS data. Moreover, due to the increasing
number of already existing digital data, in our opinion the
automatic extension and upgrade of these data bases like
was is demonstrated in this article will become a topic of
growing importance for future applications.
References
Braun, C., Kolbe, T., Lang, F., Schickler, W., Stein-
hage, V., Kremers, A., Forstner, W. & Plümmer, L.
(1995), ‘Models for photogrammetric building recon-
struction’, Computer & Graphics 19(1), 109-118.
Burns, J., Hansen, A. & Riseman, E. (1986), ‘Extracting
straight lines’, IEEE Transactions on Pattern Analy-
sis and Machine Intelligence 8(4), 425—443.
Carosio, A. (1995), Three-dimensional synthetic land-
scapes: Data acquisition, modelling and visualiza-
tion, in D. Fritsch & D. Hobbie, eds, ‘Photogrammet-
ric Week '95', Herbert Wichmann Verlag, pp. 293—
302.
Flynn, P. & Jain, A. (1991), ‘CAD-based computer vision:
From CAD models to relational graphs’, IEEE Trans-
actions on Pattern Analysis and Machine Intelligence
13(2), 114-132.
Haala, N. (1995), 3D building reconstruction using linear
edge segments, in D. Fritsch & D. Hobbie, eds, ‘Pho-
togrammetric Week ’95’, Herbert Wichmann Verlag,
Heidelberg, pp. 19-28.
Haala, N. & Hahn, M. (1995), Data fusion for the de-
tection and reconstruction of buildings, in A. Gruen,
O. Kuebler & P. Agouris, eds, ‘Automatic Extraction
of Man-Made Objects from Aerial and Space Images’,
Birkhauser Verlag, Basel, Boston, Berlin, pp. 211-
220.
Illert, A. (1990), Automatische Erfassung von Karten-
schrift, Symbolen und Grundrißobjekten aus der
Deutschen Grundkarte 1:5000, Dissertation, Univer-
sität Hannover.
KEY W
tion, da
ABSTR
The me
cedure:
are nov
becaus
The pa
which is
sureme
taking €
though
An exp
to be n
pectati
obtaine
of the t
The au
cedure
years.
tive ori
in Digit
be obs
dures
these |
ysis ar
promin
based
ment o
the pre
with a
The d
grounc
preser
tem hi
The gt
spond
agesc
corres
