The International Archives of the Photogrammetry. Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
area 3 is inside the damage site, then the fire is also inside the
damage site (cf. Figure 5).
A striking disadvantage of the ontology based reasoning is the
restriction of reasonable relations because the set of identifiable
relations is static and predefined. As can be seen in Figure 5,
direct links between the districts and the operational areas are
missing. According to that, identifying the relations between
both unambiguously by the compositions it is not possible. To
solve this problem and to make reasoning process possible, the
intersections between the regions have to be determined
geometrically (dotted lines in Figure 6). After that all
topological relations can be solved by the possible compositions
(cf. Figure 6 with respect to clearness only an extract of all
possible relations is shown).
3. NEIGHBORHOOD ASPECTS
Besides the topological relations between regions,
neighborhood relations are essential for describing spatial
scenes with respect to the mental model (Gold, 1992). Moreover
these relations are for example important for analyzing potential
risk for objects and plausibility of reports.
Neighborhoods are often geometrically defined by the
Euclidean distance di, which generates a circumcircle around
the reference object. All objects inside this circumcircle are
neighbors of the reference object. This concept of defining
neighborhood relations by a static distance is not very elegant
and discriminates non-standard situations. Such situations are
for example given by the different housing density of a region
with an irregular settlement. In Figure 7, (A,B) as well as (A,D)
are neighbors if the distance di is two units. However (A,C) are
not neighbors, although this neighborhood is desirable. A
problem arises when di is enlarged to three units. Then (A,C)
are neighbors but also (A,E), what leads to an unsatisfactory
result.
Figure 7. Example for a situation of an irregular settlement (one
square represents one object (filled) as well as one length unit)
Therefore a neighborhood graph is needed which is both scale
and orientation invariant and complies with the natural neighbor
definition. Natural neighbors are objects which are linked by
one and only one edge of a triangle. The respective generation n
of neighborhoods is defined according to the graph theory, by
the minimal number n of edges between them (Koch, 2007).
The correct neighborhood relation with respect to the natural
neighborhood for the example of Figure 7 is represented by:
generation one neighbors (A,B), (A,C) and (A,D), generation
two neighbors (A,E). This concept is based on a triangulation
defined by a set of points with a maximum of edges in between,
which have a minimal length and do not cross each other.
Adequate to this definitions are the minimal weight
triangulation (edge remove method) as well as the Delaunay
triangulation, which are quite different.
The minimal weight triangulation on the one hand starts with
the complete graph between all objects. Based on this graph the
shortest edge is selected and all crossing edges are removed
until no edge crosses any other edge (Hlavaty and Skala, 2004).
The Delaunay triangulation on the other hand is based on the
so-called empty circumcircle criterion. Three points form a
triangle when the circumcircle of these three points is empty
(does not include other points). Both concepts of triangulation
have different end functions. The minimal weight triangulation
is focused on minimizing the edge lengths and the Delaunay
triangulation is focused on equal edge lengths. That way the
result of the Delaunay triangulation is a more ideal and
homogenous meshed graph.
An example is shown in Figure 8 for a typical situation of a
residential estate. The solid lines represent the respective
Delaunay neighborhood graph between the objects, here
buildings. This way the determination of natural neighbors of
objects in terms of graph theory is ensured.
Figure 8. Delaunay triangulation between buildings based on
their centre point
4. ORIENTATION AND DISTANCE ASPECTS
Modeling orientation aspects for characterizing a spatial scene
is quite complex because the orientation of objects depends on
the reference frame. Thus the types of reference differ as
follows: intrinsic (orientation is given by an inherent property),
extrinsic (external objects impose an orientation) and deictic
(orientation is imposed by the point of view) (Hernández,
1994). According to that, the intrinsic reference is given by an
inherent property of the object like the front or back side of a
building. Such knowledge is a priori available and can be
included in the domain ontology as a specific feature. Extrinsic
and deictic references are mutable, that is why they require a
reference in time, like the speakers point of view during the
observation time. The basis for analyzing such references within
the DM 2 is already given by the explicit modeling of time by
tuple of object-time-location (Lucas et al., 2007). Further
aspects of orientation descriptions are the canonical identifier
like in front of, to the right of or cardinal points like north of
which also exist in free-form text reports. But solving such