patial
ntinu-
»fforts
ligital
2 data
scape
1 data
model
from
ore, a
ision.
1t the
dings
found
al re-
these
from
sup-
cape
yuto-
jlicit
information can be used to support the data revision pro-
cess based on digital aerial photos, and a revision based
on different digital maps.
The following chapters will outline possible fields of inter-
est, which can be supported by an automatic interpreta-
tion of DLM. Our approach designing a knowledge based
shell to be linked with an object oriented database depends
on two key issues
e which data model is used within the DLM ?
e which collection of simple objects with correspond-
ing methods seems to be adequate ?
2 INTERPRETATION OF DIGITAL
LANDSCAPE MODELS
The automatic interpretation of digital landscape mod-
els allows us to deduce information that is not explicitly
stored in the data model itself. For the automatic produc-
tion of thematic maps based on a given DLM (e.g. a map
of industrial areas) a spatial and semantic analysis of the
DLM objects is necessary. In the area of data revision the
interpretation can be used to support :
e a revision based on automatic image process-
ing
e arevision based on different databases of sim-
ilar scale
e a revision based on databases of different
scale
This topics will be described in the next sections.
2.1 Image Processing
Existing digital maps, which have to be updated, offer also
prior information contents for data revision based on dig-
ital aerial photographs. The geometry and thematics of a
digital map can contribute much to solve the image inter-
pretation problem, for instance to extract object attributes
like shape, texture, size etc. from the imagery or to predict
regions of interest for which alterations are to be expected,
e.g. new housing estate. The link of the existing map and
the aerial imagery is bidirectional which means, the sym-
bolic scene description of a map is imported using an E/R
model in the image space. Then questions of image in-
terpretation can be answered, for instance, what type of
features in the image is to be expected (shape, size, texture
etc.). On the contrary extracted image features may rep-
resent update map information, e.g. the boundary line of
a street, the ring polygon of a house etc. This knowledge
can be used to control further steps of processing.
This kind of using spatial data as prior information re-
quires matching techniques between image objects and dig-
ital map objects. In particular the mapping of image ob-
jects, which do not exist in the digital map, allows to de-
duce implicit information like geometric or neighbourhood
relations. With this implicit information we are able to set
up hypotheses in the sense to interpret the image features
in terms of real world objects. Therefore, we need oper-
ations which match image features (lines, pixel regions,
etc.) with existing vectorial map data.
In [Haala & Anders 1996] the large scale database ALK
is used to predict 3D building models for a 3D building
reconstruction in digital aerial photos. Figure 2 shows
constructed building hypotheses based on the shape and
spatial relationships of the ground plans included in the
ALK. Another approach for updating the ATKIS DLM
200 database using satellite images is described in [Klaus-
Jürgen Schilling, Thomas Vägtle, Peter Müfig 1994].
Figure 2: 3D building hypotheses based on the ALK
2.2 Databases of Similar Scale
If one class of spatial objects in a given DLM should be
revised using another DLM, which contains the same kind
of spatial objects in a more or less equal scale, one has to
match both digital maps. The matching of two given vec-
tor databases is also called Conflation, which comes from
the Latin con flare meaning “blow together” [Maureen
Lynch and Alan Saalfeld 1985]. After the matching process
is carried out, it should be possible to detect differences
between the two databases. Two basic problems occur
within the matching of two digital maps
e differences in accuracy of data capture (figure 3)
and
e differences in data modelling of a spatial object
(figure 4)
which can be overcome the better the more implicit infor-
mation can be used. In [Walter & Fritsch 1995] an ap-
proach based on relational matching for the matching of
ATKIS road-objects with GDF road-objects is described.
ovans E ER
HE Hot
a) b) j
Figure 3: a) dataset A, b) dataset B of the same area,
c) differences in the accuracy of data (after [Walter &
Fritsch 1995])
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
