help 
an be 
arity 
h the 
sions 
| the 
ation 
nore 
s of 
edge 
The 
ectly 
any 
a has 
es to 
r and 
was 
idow 
ident 
; the 
gital 
nain 
f the 
Is to 
tric 
n of 
ig of 
and 
ition 
s. for 
used 
The 
> are 
ictly 
| as 
thod 
isily 
ined 
ular 
oors 
nge 
ved. 
| on 
ple 
for 
n of 
the 
be 
seen in Fig. 7.2. Lines symbolize the edge segments. The 
underlying grey regions were generated by the layer 
growing algorithm in order to find neighbouring 
regions. Region neighbourhoods are indicated by dashed 
lines. The pixels with dots surrounding the edge segments 
are the edge-support regions from which the edge 
parameters are derived. 
The final matches are displayed in Fig 7.3. The 
match numbers enable the alert eye to identify matched 
edge segments. The unlabeled edge segments are 
unmatched. 
Another image pair can be seen in Fig. 7.4 while 
Fig. 7.5 shows the resulting matches on the object of 
interest. 
Table 7.1 gives the computation times for each 
processing step on a Sun sparc IPX workstation. 
Unfortunately, the times are influenced by partially 
uncontrollable parameters of network performance, as 
some data had to be transferred from a remote station. 
However, it is obvious that the matching of graphs is 
extremely fast in comparision to the preceeding 
processing of the single images. 
Intensive tests of the matching method resulted in 
about 80 to 90% correct matches among the matches 
found on the objects of interest. 
8. Conclusions and Recommendations 
A feature-based stereo matching method was 
developed using a graph structure to control the search 
for matches and to exploit relational properties for the 
identification of matches. The features used are edges of 
variable shapes. 
  
  
  
2 
Fig. 7.2. Edge neighbourhood graph. 
Special features of the matching method are 
1. the use of curved edge segments 
2. an efficient method for the generation of edge 
neighbourhood graphs and the analysis of edge- 
support regions 
3. the avoidance of the epipolar constraint for objects 
with distinctive edges 
  
i 
Fig. 7.3. Matched segments 
  
315