4. CASESTUDY
In this section we illustrate the presented concepts by
an example. In particular we describe how explicit
knowledge may guide the segmentation.
We wish to verify if the roads in the road database are
still present in the scanned aerial photograph. This is
only a minor part of the procedure for automatized
updating, but it illustrates very well the complexity of
the problem. The strategy followed arises from the
nature of the object (a network), and the nature of the
task (updating). This task results in an extra know-
ledge source: a road database.
In fig. 1 the present state of the road database is
shown, representing the "old" situation in object space.
Fig. 2 shows an aerial image of the present situation in
object space. The image is the blue part of a true
colour aerial photograph, scanned in blue, green and
red with pixel size 1.60 m.. The present situation
shows a crossing instead of the T-junction in the old
situation. It connects a new road to the network, which
crosses the highway by a fly-over. In addition a new
exit and a new slip-road are present.
The high degree of complexity will be illustrated by
considering just a single road segment, taken from the
database. Its axis represents an arc of the road network
in the database. The road segment is superimposed on
the image (see fig. 2).
Since the location and orientation of the road is
approximately known, only a small region of interest
(ROI) around the road segment can be examined.
Using the real-world knowledge that roads are linear
and have parallel boundaries, a region of interest is
defined, such that the ROI rows are perpendicular to
the local direction of the axis of the road. The road
boundaries in the region of interest form straight lines
parallel to the ROI columns. In this way the ROI can
handle any type of curvature. The width of the ROI
depends on the known road width and the distance to
adjacent parallel roads, stored in the database.
Fig. 3 shows the region of interest of the road segment
in fig. 2.
4.1 High level: hypotheses generation
All possible changes that a road segment may
undergo, should be defined to verify if the road
segment is still present. With respect to the presence
of the road segment, at least the following hypotheses
should be tested:
* The road segment is still present;
* The road segment does not exist.
If the road segment is still present, we should test
whether:
* Its properties have remained entirely unchanged;
* The width of the road segment has changed;
* The curvature of the road has changed;
* The road is crossed by a new road. The new node
is a level crossing;
* The road is crossed by a new road. The new node
is a fly-over, where the new road is at the high
level;
* The road crosses a new road. The new node is a
fly-over, where the new road is at the low level.
These hypotheses are not mutually exclusive. For
example the width of the road may have changed,
while the road is simultaneously covered by a fly-over.
Hypotheses can even be highly correlated. For
example, if the curvature of the road has changed, it is
also likely that the width of the road has changed.
Fig. 1 Part of the road database expressing the old
situation.
480