| road
ifference
eatures
nted here.
"he matrix
matched.
nctions of
“the latter
to a clear
Tal Abramovich
d =12.2m da = 3.5° dw = 5m (=38m match = 0.7
\ il
ce m
a
5 ros
4
\
0 1 0 40 0
Figure 3: An example for applying the fuzzy rules to a set of input parameters.
Membership functions were selected based on common knowledge about roads. Thus, for example, the function
for the declaration “d is small” (top left in Figure 3) drops sharply, since it is not reasonable that the road entity
from the GIS will be far apart from the road hypothesized in the image. Another example is the statement “dw
is small” (third row, third column). In this case, the function drops gradually as it is expected that the width
attribute of the road in the GIS is not precise. Currently, a refinement of these functions was performed
empirically, based on a relatively small number of experiments. Nevertheless, the experiments show that
applying the rules to real data clearly point to the correct results.
The system was applied to real data, which contain an area in central Israel. The GIS data were taken from the
Israeli National GIS. The image features were extracted automatically from an orthophoto of the area. Figure 4
shows two parts of the orthophoto, overlaid by the vector data. The particular features that are shown later in the
results are annotated on the image.
The experiments were performed in the following way:
Selecting a road entity in the GIS.
Selecting all hypothesized road segments from the image that are within a certain buffer around the
selected segment (based on the common coordinate system).
* Applying the set of rules to the GIS road entity and each of the hypothesized road segments, and
calculating the resulted similarity value.
* Presenting the results of the segments that yielded the highest similarity values.
The results for two particular segments (as annotated on Figure 4) are shown in Figure 5.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 21
