Figure 3: 3D building hypotheses based on the digital cadastral map
roof, the buildings are classified (e.g. garage, residential
building, office block, industrial building, church, tower)
by analysing the text information contained in the dig-
ital cadastral map. Frequently industrial buildings and
garages have a flat roof. Since the shape of a desk or
flat roof will follow the shape of the ground plan for most
cases, the 2D contour line of the roof is assumed to be
identical to the given ground plan for these types of build-
ings. In contrast to that saddle roofs were expected for
residential buildings. For theses buildings the generaliza-
tion process described in the preceding section was applied
to define the contour lines of the roof. At the moment the
generation of 3D hypotheses and the verification process
described below can only handle models of saddle and flat
roofs. Even though desk, flat or hipped roofs can be con-
sidered as special cases of saddle roofs, other models which
are able to cope with modifications of the used rectangular
shape will have to be implemented. The ridge lines of the
saddle. roofs are defined by the longer middle axes of the
rectangles, which were generated by the simplification and
combination of the given ground plans.
Generally the direction of the ridge of a roof also follows
rules of town planning, adjoining buildings e.g. have the
same ridge direction, or rows of buildings parallel to a
street have ridge lines that are adjusted to the direction of
the adjacent street. To make use of this knowledge future
work will have to perform further analysis of the given 2D
data e.g. by the detection of neighbourhood relations for
buildings or by additionally extracting information about
streets, which is also contained in the digital cadastral map
in form of text symbols.
Hypotheses about other parameters like the height of
à building can only be generated by using of meta-
information that is not given in the digital cadastral map.
Such kind of information is the location of the buildings
in the city (e.g. city center, outskirts, industrial area) and
the knowledge about typical heights or numbers of floors
of different kinds of buildings (industrial building, church,
288
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
garage, office block) dependent on the location in the city.
The 3D building hypotheses shown in figure 3 were gen-
erated using the generalized ground plans; to define the
unknown parameters in height knowledge on the usage of
buildings (e.g. garage, residential building, office block, in-
dustrial building, church, tower) which was extracted from
the digital cadastral map was utilized. To perform the 3D
construction the following assumptions on the height val-
ues were made:
e Garages: 3m
e hesidential building and office block : 9 m for the
ridge, 6 m for the eaves
e Industrial building : 15 m
e Church : 12m
e Tower : 25 m
e Kindergarten : 5 m
e others : 7.5 m
Because the constructed buildings are only based on quite
weak assumptions, a verification process including addi-
tional information is necessary. For that purpose aerial
images are used.
4 VERIFICATION
To verify the generated 3D building hypotheses and to
determine the unknown parameters of the object model,
additional information is required. In our approach linear
features are extracted from an aerial image and matched
against lines of the object model to provide that missing
information.
4.1.
Most ex
use of t
of straig
ear edg
cess. W
by an à
(1986) !
primiti
Figu
For the
ses anc
ues, tl
against
order :
and mec
in a gl
by the
can be
object
be tras
specti
lines c
exteric
match
dition
to acq
Model
oped c
procec
up are
height
ing te
a app!
during
ficient
of Bac
accure
