Tal Abramovich 
  
  
  
  
  
  
  
  
Table 1: Rules for evaluating the similarity between road entities from a GIS and hypothesized road ‘ 
segments in an image. 
IF THEN \ 
da is small AND / is large | Similarity is good E 
da is large Similarity is bad ps 
dw is small AND ? is large Similarity is good fo 
dw is large Similarity is bad ETT 
d is small Similarity is good E 
d is large Similarity is bad |. 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
a, 
d A = 
4 —— 
/ I" 
/ 
/ 
s 
, LI. 
/ 9 
/ / 
x / 7 
/ / 
/ € 
a  da@=a-a b 
A n Mem 
go W, for th 
= ^ from 
/ is sm 
j attrib 
/ empii 
apply 
The s 
dw — w,—w Israel 
é 2 show: 
result 
Thee 
Figure 2: Definitions of geometric properties that affect the similarity evaluation — (a) direction difference 
( da ); (b) segment length (/); (c) difference of road width (dw); and (d) distance between the features 
with respect to a common coordinate system (d ). s 
e 
4 EXPERIMENTATION AND RESULTS . 
The proposed approach has been tested by a prototype system, and some preliminary results are presented here. 
Figure 3 shows the set of rules, which were described in the previous section, as a matrix of graphs. The matrix The r 
demonstrates an operation of the rules on a set of parameters that apply to a pair of segments to be matched. 
Each row of the matrix represents one rule. The left-hand side of the row contains the membership functions of 
the inputs. The right-hand side shows the membership function of the output. The selected part of the latter 
(shown as the dark part under the graph) is considered in the process of combining all fuzzy rules to a clear 
response. 
  
20 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.