International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanhul 2004
they are used as a reference in the text without further details
about their direction. In the example of Fig. 2, fences are quite
straight. However, there are other examples of maps where
fences show more irregularities in their shape. Creeks do have
windings that are explicitly mentioned in the text, but no further
information about the direction of these windings is given in the
description. Within a digitized map, the irregular shapes of
fences and creeks are represented by polylines. If necessary, e.g.
in cases of rapid changes in direction along the boundary path, a
high number of intermediate points are marked, thus, not only
the endpoints of a boundary part, but also its precise course
between these endpoints is known and represented at the
geometrical level.
4.5 First results
Although the project is an ongoing work and still rather at its
beginning, some of the question of section 4.3 will be addressed
in this section.
The questions concerning the different used levels of abstraction
and a common symbolic level was already briefly examined in
the previous section.
Boundary descriptions in cadastre are supposed to be rigorously
redundant free. Within the available examples, the information
in the descriptive texts is in such compact form that some
knowledge even has to be concluded by logical inference or
mere presumption. Besides the already mentioned missing
information about the direction of fences or the windings of
creeks, table 1 reveals more ambiguous attribute values of
boundary parts.
No. type length in m | adjacent owner | direction of
turning
1 line 301. 00 ?] left
2 creek 68.70 A (formerly Z) continue
3 creek 158.60 B (formerly Z) left
4 line 351.00 D left
S line 190.20 A slightly left
6 fence 86.00 A 22
Table 1. Boundary parts with their associated attributes, listed in
the sequence of their appearance in the textural description in
section 4.2. Some of the information is missing or ambiguous
(marked by ?1, ?2).
Right at the first boundary part, the text mentions that the
descriptions starts at a point of conjunction between two
properties, both belonging to owner A, but one of them formerly
belonged to owner Z. The text does not mention, to which of
these two properties the first line is ad;^^ent. However, after
reading the complete text, it is possible to conclude on the
owner of the neighbouring property: The last boundary part (no.
6) is connected to the first part and has as adjacent owner only
A, thus it can be deduced that ?1 = A (formerly Z). The
direction of turning after the last boundary part - in order to
continue with the first boundary part again - is completely
missing. Only if it is assumed that the description starts at a
turning point and that the bending at the starting point is no
exception from the typical walking direction then it can be
inferred that 22 = left. However, there are examples of other
properties where the starting point is no turning point at all,
therefore ?2 — left is only a hypothesis. An automatic extraction
and disambiguation of the knowledge given in the descriptive
texts (as it is manually prepared in table 1) is a task within the
field of information extraction (Cowie and Lehnert, 1996). This
is not part of the presented project, yet.
The maps of the cadastre contain redundancy on purpose since
they should provide human interpreters with a quick overview
of the situation and precise information. While the length and
direction of the boundary parts is already implicitly represented
by its drawing and the scale of the map, only the annotated
numbers allow a quick reference on the precise distances and
directions. However, the annotated numbers are directly taken
from the textual description of the properties and are not always
consistent with the actually drawn boundary parts. This leads to
the question of consistency. The redundancy of the map is
useful to match boundary parts of the map and of the associated
text which can be checked further on consistency in other
attributes (owner of adjacent properties, type). In the given
example of section 4.2, there are already some inconsistencies
in the type. Boundary parts no. 1 and no. 5 in table 1 are of type
“line” while the corresponding boundary parts in the map are of
type "fence". Either the description is simply not specific
enough for these boundary parts or the map is wrong at these
positions.
For the production of a reverse description, not only the
sequence of appearance of the boundary parts has to be properly
recorded within the chosen knowledge representation. Although
the textual descriptions of the cadastre refer to geographic
objects (boundary parts) that can be identified in both directions
(in contrast to the problem in many navigation tasks where
landmarks are only visible in one direction), the information
about the turning direction has to be suitably adapted. If they are
e.g. given as a certain degree of azimuth, they have to be
computed considering +180 degree. In the example in section
4.2 only relative spatial descriptions such as "left" are used that
need to be exchanged by their opposite term. Composed terms,
such as “slightly left”, can be easily transformed by altering the
spatial term ("slightly right").
5. EXAMPLE II: ANALYSING MESSAGES IN
DISASTER MANAGEMENT SYSTEMS
Although the scenario of the Brazilian cadastre is already
challenging, this second application is of an even higher
complexity. Persons in charge in a disaster management centre
depend on up-to-date information about events and conditions at
different locations. Usually, such information is plotted on a
map and regularly updated with the information of incoming
messages. In case of a catastrophic event, a large number of
such messages in verbal or written form are received and need
to be interpreted and evaluated by staff members in the centre.
Therefore, the aim in this application is to analyse spatial
information in these messages, thus allowing a more efficient
update of these maps. Incoming messages in the context of an
example scenario “thunderstorm” might be:
e Train derailed at level-crossing A and driven into a
magazine of fertilizers at B, fire and a high number of
injured persons at B
e Scveral people in district C complain about shortness
of breath due to strong smoke emission at B
e Main road D is blocked between street E and street F
because of fallen trees and branches
All these messages contain spatial information, but on different
scales (compare Fig. 5). Here, the previously 1D-problem of the
cadastre is expanded to all 4 dimensions. To all three possible
220
Interna
spatial
additio:
that oc
place d
a fire-«
Additio
more
incomit
redund:
order t
semant
future v
Figure
resoluti
features
This pa
textual
represer
applicat
their re
results
symboli
(here as
knowlec
concern
in sectic
message
Allen, J
and Ste
Proceed
Convers
Bähr, H
Fernerla
Bähr, H
Photogr:
Sympos
http://wA
pa
