4.2 Low level: segmentation techniques
We will discuss two segmentation techniques, which
both detect road boundaries. Both do not result in a
final decision at the high level, since they do not
provide enough information to base a decision on. In
addition other segmentation techniques that provide
information about road surfaces have to be applied.
The boundary of the road is characterized by:
Many segmentation techniques can handle boundaries
with the above characteristics. We choose:
Edge enhancement/Thresholding
Based on the assumption that the grey value of the
road contrasts with the adjacent terrain, we use an
edge detector for extraction of the boundary of the
road. The edge responses are thresholded and
skeletonized to one-pixel thick lines.
Dynamic programming in a restricted ROI
(Gerbrands, 1988)
Dynamic programming is a general optimization
technique that searches for a path with a maximum
cost solution. Appropriate costs should indicate the
significance for the presence of an edge of the road.
The edge strength of all pixels in the ROI leads to
appropriate costs. Now the problem can be formu-
lated as tracing a maximum cost path.
Results of both techniques are shown in fig. 4 and 5.
In fig. 4 can visually be observed that three broken
boundaries have been traced. In fig. 5 can be seen that
the path is bendy at the position where the fly-over is
present in the new situation.
4.3 High level: analysis of segmentation results
In order to reach the above conclusions by reasoning
at the high level, we have to include knowledge about
operating characteristics of both segmentation techni-
ques.
We use this knowledge for two purposes:
1. evaluation and clean-up of the segmentation result;
2. testing of hypotheses.
They will be successively described.
J[| K
= ee
||
Fig. 4 Result of edge enhancement/thresholding scheme and subsequent skeletonisation applied on the ROT of fig.3
be m m
AE Mc I Mq"
Em pm, mn a nl nl me n a e i p ^A
T EE
Li “|
Fig. 5 Result of dynamic programming applied on the ROI of fig. 3
