bul 2004 
edge fea- 
[o do so, 
re gener- 
resulting 
ight than 
ions with 
points on 
1 optimal 
perty, we 
d Ebner, 
network 
othetical 
To form 
is calcu- 
(3) 
/potheses 
accepted, 
Due to 
es has to 
potheses 
inal road 
ned IRS- 
'ere used. 
different 
est areas. 
ve evalu- 
by visual 
e sample 
le shows, 
>xtracted, 
arms, for 
ere is the 
"ders of a 
road net- 
"different 
correctly 
ricultural 
expected, 
umber of 
through a 
iedemann 
| allowed 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
Eus 
Figure 7: Result of road extraction in agricultural area 
| Figure 8: Image data 
Figure 9: Result of road extraction in agricultural area 
  
  
proach for Semi-Automated Road Extraction from Medium- and 
Figure 11: Result of road extraction in mountainous area 
S CONCLUSIONS 
We have proposed a new approach for the extraction of roads in 
agricultural areas in 5 m resolution IRS data. The approach is 
based on the fact, that roads in these areas are often not directly 
visible, but they run along field borders and form elongated struc- 
tures. Starting with the extraction of linear features, connection 
hypotheses are generated. The evaluated connections are opti- 
mized by means of ziplock snakes. The verification of the gener- 
ated road sections is performed by checking the path of the snake. 
Using the verified road sections a road network is constructed by 
means of global grouping. Experiments have shown, that not only 
the main roads but also smaller roads can be extracted from the 
used IRS imagery. By means of the approach of (Wiedemann et 
al., 1998) we are also able to extract roads in desert and moun- 
tainous areas from IRS data. 
6 ACKNOWLEDGMENT 
We gratefully acknowledge the support by Bundeswehr Geoin- 
formation Office (AGeoBw), Euskirchen. 
REFERENCES 
Dal Poz, A. and do Vale, G., 2003. Dynamic Programming Ap-