(b) 
  
Figure 6: Real data: (a) left stereomate; (b) 3-D edges in 
orthogonal projection 
Figure 7: Results of applying the split-and-merge method 
to the synthetic data 
  
Figure 8: Results of applying the split-and-merge method 
to the real data (orthogonal projection) 
were extracted completely, and the circular arcs were seg- 
mented into small straight lines. The results are shown in 
figure 7 
Real data: The results of the split-and-merge segmen- 
tation for the real data are shown in figure 8 as an orthog- 
onal projection of the 3-D segments received. These results 
are very encouraging. Many straight segments were de- 
tected. The noisy parts of the curves, which are interpreted 
as such also by a human observer, remained unchanged. A 
comparison between the results and the image shows corre- 
spondence between straight lines and man-made features. 
3.2 1 — s results 
The y — s segmentation algorithm was implemented ac- 
cording to the description in section 2.2. The selection of 
threshold values is more crucial than it is for the split-and- 
merge case. The main reason for this problem is the fact 
that we deal with angular parameters, while the real physi- 
cal perturbations are linear. Therefore, the threshold value 
for a certain line length will not necessarily be suitable for 
other lengths. Despite this, we used values which are ac- 
ceptable for the synthetic data, as described below. The 
offset threshold for the  — s curve was set to 10. We 
also limited the accepted circular arcs radii to the interval 
2 — 200 m. We have not limited the arc angle and arrow at 
this stage. 
Synthetic data: The results of executing the y — s 
algorithm with the synthetic data are presented in a 3-D 
view in figure 9. The results need some explanations. 
1. Longer segments (either straight lines or circular arcs) 
were segmented into shorter ones. However, it can be 
seen that most of the segments were classified cor- 
rectly. 
2. Small segments, which were characterized as noise ef- 
fects were created near the discontinuity points. 
The phenomenon of breaking an expected segment into 
a small number of shorter segments can be resolved in the 
Figu 
synt 
lines 
pres: 
spat 
thet; 
segn 
com! 
justi 
fects 
elim 
bori: 
duce 
also 
tions 
are ( 
appl 
shou 
The 
Alth 
not 
the ı 
tion 
strai 
of sh 
redu. 
and-: 
even 
circu 
tion 
Howe 
issue 
lines 
prop 
blurs 
Free: 
objec 
Cause 
sions