ibul 2004
JA is the
| MRDB
ocedures.
te
'alisation
ta for an
nking by
provides
9 of data
data sets.
or morc
and if no
into an
therefore
; between
t kind of
zeometric
ocedures.
, because
.g. object
rway, Or
ry (shape
compares
agregated
ination of
vhich are
bject. Fer
ects was
e for area
itic filters
lidates to
»cedure is
possible
] for the
y linking
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
building ground plans with settlement areas of the ATKIS
BaseDLM (Figure 3).
BINNEN
zr
3
Mida igni
S «fon F
C
-Dxmr dH. |
Dan NH
«all other
D CMS hacer
Figure 3. Ground plans linked with an ATKIS settlement area.
3.3 Linking by generalisation
The third possibility to generate MRDB data is the generation
of object links during the automatic creation of another object
representation from a reference object. This is needed if a
completely new data set has to be derived form a base data set
or in the case of automatic updating. As an example for the
medium to low resolution model generalisation we studied the
automatic derivation of an ATKIS DLMS0 (1:50.000) database
from an ATKIS BaseDLM (1:10.000-1:25.000) database. The
ATKIS DLM50 object catalogue describes the object types
which should be captured in the DLMS0 and which geometries
types should be used to represent the real world objects.
Therefore in general we have to deal with area, line, and point
generalisation, but for our first studies we concentrated on the
subject of area generalisation. These operations are
implemented for the derivation of the DLMS50, they are
however generic enough to be transformed to representations of
arbitrary scales. The area generalisation process consist of three
steps:
Reclassification of the object types
Aggregation of adjacent areas with equal object type
Shape generalisation
uv N —
The first step is needed because some object types of the
ATKIS BaseDLM do not exist anymore in the ATKIS DLM50
landscape model. The second step is needed because in the
ATKIS DLM50 the minimum size criteria for capturing of
certain object types as area objects has increased. The third step
handles all cases of area objects which are still not big enough
after the second step to be captured as arcas in the ATKIS
DLM50 model. In such cases according to the ATKIS DLM50
object catalogue these areas has to be represented by a point or
have to be eliminated completely. In both cases one has to
establish a reclassification and an additional aggregation step to
these areas. This additional reclassification and aggregation step
can be done in different ways. Four possible solutions are
shown in Figure 4. The replacement of an area can be done “by
definition” which means that a priority list of new object types
is given that describes which new object type has to be used to
replace the old object type relative to the adjacent areas (Figure
4a).
453
Definition
Number of
Neighbours
Size
Medial axis
J
i
J
i
a)
5)
€)
d)
Figure 4. Possible solutions for aggregation of area objects.
E.g., if an area with object type farmland has to be replaced
then a possible priority list can be: 1. grassland, 2. garden area,
3. area without vegetation, and so on. That means if an adjacent
area has the type grassland then the area will be reclassified as
grassland. If no adjacent grassland can be found then one has to
look for an area without vegetation. If there is no such area then
may be there is an area of type indefinable area, etc. The
priority list has to make sure that always a new object type can
be found. Another way is to choose the most frequent object
type of the adjacent neighbour areas (Figure 4b) or to choose
the object type of the largest adjacent neighbour area (Figure
4c). A more sophisticated approach is to compute the skeleton
(e.g., medial axis, straight skeleton) of the area which has to be
replaced and to increase all adjacent areas according to the
computed skeleton (Figure 4d) (Bader, 1997). The maximum
number of equal neighbours and the maximum size approach
have the same drawback, that this maximum number must not
be unique. E.g., an area can have as many neighbours of object
type X as of object type Y. It is also possible that an area has
more than one adjacent neighbour areas with the same size but
different object types. These ambiguities can be solved in the
most cases by combining the criteria's (number and size of), but
in general there can be still ambiguous cases. The approach
with the skeletons takes all neighbour areas in account and
increases all neighbour areas relative to the shape of the area
which has to be replaced. The drawback of this approach is that
it is more complicated to be implemented and it can produce
artificial shapes, like the two small triangles in figure Figure 4d.
We have implemented the first approach using priority lists
from the NMA's and a straight skeleton approach to aggregate
areas without prior knowledge. Figure 5a and 5b shows an
example for this type of area aggregation. During this
aggregation process the links between the original areas of
medium resolution and the new areas of low resolution are
stored in the link table of the MRDB federation layer.
Figure 5a. Example: Before rule based aggregation of areas.
