International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
3 BUILDING INTERRELATIONS
As mentioned in the introduction, in many cases two or more
representations of the same object may exist in different data
sets. While attempting to define which object from the two data
sets are the same, we have to define
correspondence can be found in the two data sets. Since spatial
objects can be characterised by their spatial and thematic
characteristic, the differences and similarities can be related to
theme and geometry. It may appear that one object from one
data set can be linked to zero, one or many objects from the
other data set. Such a multiplicity can occur in both the
geometric and the thematic domain.
The following cases can be distinguished:
1. One object corresponds to only one object from the
second data set. The relationship between the
representations is then 1:1. In terms of spatial
relationships this will mean ‘objectl equals to
object2’
2. An object exists only in one of two datasets but can be
assumed that is part of conglomerate in the second
data set. For example, the object ‘bridge’ cannot be
found in the second data set because it is complete
integrated within another object, e.g. ‘road’. In such
cases, we define relationship 1:0. Translated into
spatial relationships, this case has to result in ‘object1
inside/covered by object2'.
3. An object is an aggregation of several objects from
the second data set. For example, the object ‘road’ in
one data set may be confronted with ‘primary road’
and ‘secondary road’ in the second data. In this case,
we define relationship 1:m. The spatial relationship
will be *objectl contains/covers object2'
4. The last most complex relationship is when two
different themes are considered and the object cannot
be thematically matched. For example, the object
‘riverside’ from a large-scale data set, which is only
of interest for an organisation dealing with large-scale
mapping and therefore not modelled in small-scale
applications, cannot be related to a similar object in
the small-scale data set. Instead it will overlap with
other thematic objects in the small-scale data set, such
as ‘grass’ or ‘road’. In this case the spatial
relationship will be “object! overlap/intersect
object2'.
Considering only the theme of the objects, making the
correspondence between two data sets looks a straightforward
approach to be implemented at DBMS level (using the spatial
operations, e.g. SDO RELATE in Oracle Spatial 9i). However,
scale and geometric resolutions (closely related to the scale)
also influence the process of object referencing. When the scale
is different, most commonly the outer rings of the polygons (in
case of area objects) differ significantly. Although smaller,
variations can be observed even in case of equal scales due to
diverse data sources, data productions procedures and resolution
(detail) used.
— hs
Th
& ie
ol T
eo. ee
S dne
A Des
SA NL
M as
N Lacs
Ce das
x = s e
t 2 AL
SN
rie ps,
M [
~ ~~ T
what kind of
224
Figure 1: Scale differences (fragments): 1:1000 (up) and 1:5000
(down)
Bearing these geometric considerations, we have to expect that
the boundaries of two objects will never completely *coincide'.
Thus, it will be rather problematic to execute spatial functions
'equals to’ or ‘covers’ and obtain the unambiguous result
‘TRUE’. In most of the cases, checking particular spatial
relationships between two objects will be influenced by
differences in the outer rings of the polygons. To avoid this
problem, we have developed an algorithm, which does more
than simply applying the topological operators.
The algorithm assumes that if two objects can be related, their
geometries must intersect. The three general steps of the
algorithm can be specified as follows:
e For each object from the first data set it is checked
which are the objects from the second data set that
‘interact’. The type of interaction is not important.
Thus all the non-interacting objects are filtered out.
* For each object (from the second data set) that has
been detected to interact with the object of interest
(from data set 1), a new object ‘intersection’ is
composed that represents the common (overlapping)
area between the two. The area of the new object is
computed.
e The areas of the object of interest and the
‘intersection’ object are compared. A decision if the
two objects can be considered the same is taken using
a threshold. For example, if the areas are more than
90% the same, the two objects are considered the
same (see Section 5 for more details).
This algorithm was implemented in Oracle Spatial 9i, using the
high-level scripting language PL/SQL and the spatial functions
SDO RELATE with mask “anyinteract’ (for the first step),
SDO_INTERSECTION (to compose the geometry of the
‘intersection’ object) and SDO AREA (to compute the area).
4 DERIVING NEW REPRESENTATIONS
Creating new representations from existing ones always
requires classification. (based on theme) and aggregation
algorithms (based on geometry). The issue can become
extremely complicated. For example, considering only
simplification of geometry (without theme differences), no
ultimate set of algorithms exists at this moment, which can
generate a dataset at a required small scale based on a data set at
a larger scale.
Here, we follow a different approach: we make use of two
existing data sets to derive a new data set. The underlying
motivation is the existence of many different data sets (in
topographic offices, municipalities, cadastre) that do have the
same (or similar) themes or that contain slightly different
themes but map the same resolution. An appropriate
combination of such data sets may significantly reduce the
efforts and time for delivering the requested data. Depending on
Inter.
what
deve
gene!
resol
Fort
(see
(high
them
resol
the s
at ce
inter
opere
the tl
comr
algor
follo:
The
PL/S(
the ag
The t
DSI (
with t
More
Zlatar
The 1
differ:
Public
This c
Spatia
the N
Spatia
data t
specif
We te
DSI&
