n the 
sheds 
1 with 
1ch as 
d the 
] each 
ns are 
ations 
dfills, 
s and 
ms, it 
der to 
ng by 
  
ROAD NETWORK DETECTION FROM SATELLITE IMAGERY 
USING MATHEMATICAL MORPHOLOGY. 
Sabbineni Prasad 
R. K. Gupta 
National Remote Sensing Agency 
Dept. Training & Educational Activities 
Balanagar 
Hyderabad, Andhra Pradesh 500037 
India 
ISPRS Comission VII/Working Group 9 
ABSTRACT 
Road network detection is useful for the planning of urban growth and development. A new algorithm, 
called as "dynamic edge" operator, was conceived and developed for detecting the road network from the 
satellite imagery using mathematical morphology techniques. Near IR band (0.79 - 0.89 um) of SPOT 
MLA image, having 20 meter spatial resolution, of Hyderabad city was used as near IR offers better 
contrast between road and surroundings. Depending upon the local statistics (mean) of the grey levels of 
3x3 structuring element, the edges were taken either as dilated or eroded. Road networks traced from this 
image were compared with the digitized map of the Hyderabad Urban Development Authority (HUDA). 
All first ordered roads matched exactly, with a little mismatch in the second order roads while some of 
the third order (non-metal) roads were not detected in satellite data based analysis. The road network map 
obtained from the eroded and dilated edges (existing technique) were position biased, did not match so 
well with the reference map as with the dynamic edge operator. Out of 278 road segments obtained from 
dynamic edge technique, 11 segments referred to newly (did not appear in HUDA map made in 1989) 
laid road. Thus the technique does not provide any error of commissioning. Due to poor contrast and less 
width, the non-metal roads cannot be detected in satellite imagery. Further, the edge strength was also 
poor in case of eroded and dilated edge as compared to that for dynamic edge operator and gave evidence 
for the superiority of the dynamic edge operator. 
139