ISPRS Commission III, Vol.34, Part 3A ,,Photogrammetric Computer Vision“, Graz, 2002 
  
  
  
     
     
  
% Road Mask + Road Seeds 
  
  
  
  
(a) The road line is broken down by the control point 
  
PP 
     
(b) The road line is modified and extended by the control point 
Figure 11. Guiding a road line tracing using a control point. 
  
  
  
  
  
  
Extending road line from q is conducted in the same way as 
addressed in previous section, while tracing the road line from p 
to q is different at the following two places. 
1) Ineach iteration, the extensional direction of road terminal 
point p, is assigned by Did: 
2) Matching cost of the road template at an image point p 
with a rotation angle a is defined as follows. 
À (6) 
f Ap, - q 
f[r'20,*f,-909,* f, (7) 
S. EXPERIMENTS AND DISCUSSIONS 
In this research, an IKONOS image nearby KAWAGOE City, 
Japan is used to test the validity and efficiency of the algorithm. 
Pixel size of each band is 1 meter. There are four bands, i.e. red, 
green, blue and near infrared. A road mask is generated using a 
commercial remote sensing software IDL/ENVI, where road 
pixels are discriminated from others using maximum likelihood 
method. Road seeds are extracted by tracing edge pixels. For 
easier operation, interface of the software is designed as follows. 
The IKONOS image is overlaid on both road mask and road 
seeds. The operator directly work on the IKONOS image, while 
road lines are extracted from road mask and road seeds. In order 
to examine the accuracy of the result, a 1:25,000 road map that 
produced for car navigation system by Tokyo Cartographic Co. 
Ltd is exploited. 
5.1 Extracting a road line 
A result of extracting a main road is shown in Figure 12. The 
road line is extracted by assigning a starting point and two 
directional control points subsequently by the operator (see 
Figure 12 (a-c)). The road line ran to wrong directions at points 
“A” and “B”, where multiple roads conjunct together. Points 
“A” are enlarged at Figure 12(d-f), where road line extraction, 
IKONOS image, the road mask (grey) and road seeds (black) 
are listed. The operator examined the result of road line 
extraction, and assigned directional control points to guide the 
road line extending to right ways. In this case, the road line 
extends 2772 pixels at the starting point, 450 pixels at the 1*' 
directional control point, and 783 pixels at the 2™ directional 
control point. The road line lasts totally 4005 pixels (24005 
meters). It sounds that the method is rather efficient in 
extracting long road lines. 
5.2 Generating road map of a specific area 
Generating road maps of a dense building area is shown in 
Figure 13(a). The skeleton of most of the main roads can be 
easily identified from road mask that generated (Figure 13(a), 
Left #1, grey), whereas road mask on small roads are 
intermittent and imperceptible. On the other hand, road seeds 
on both main and small roads are not as clear as that shown in 
Figure 12(b) and Figure 13(b), as the road boundaries are mixed 
with the surrounding buildings, trees and shadows. Road line 
extraction of main roads relies most on road mask, whereas 
most of the small roads failed in extraction. 
Generating road maps of a countryside area is shown in Figure 
13(b). It can also be found that most of the main roads is 
sketched by road masks, whereas small roads got lost. On the 
other hand, road seeds reflect not only main roads and small 
roads, but also other rapid changes of photometric features. By 
properly selecting starting and directional control points, road 
lines succeeded in both main and small roads. 
6. CONCLUSION 
In this research, a method is proposed to create and/or update 
road maps in urban/suburban area using high-resolution satellite 
images. A road mask is generated by discriminate road pixels 
from others using a commercial remote sensing software. Road 
seeds are extracted by tracing edge pixels. Road line extraction 
is conducted on both road mask and road seeds. Experiments 
are conducted using IKONOS images with a ground resolution 
of 1 meter. Experimental results show that the method is valid 
in extracting main roads in high dense building area and all 
roads in countryside efficiently. 
REFERENCE 
[1] Barzohar, M., D.B.Cooper, Automatic finding of main 
roads in aerial images by using geometric-stochastic 
models and estimation, IEEE Trans. PAMI, vol.18, no.7, 
pp.707-721, July, 1996. 
[2] Fiset, R., F.Cavayas, M.C.Mouchot, B.Solaiman, Map- 
image matching using a multi-layer perceptron: the case 
of the road network, ISPRS Journal of Photogrammetry 
and Remote Sensing 53 (1998) 76-84. 
[3] Geman, D., B.Jedynak, An active testing model for 
tracking roads in satellite images, IEEE Trans. PAMI, 
vol.18, no.1, pp.1-14, January, 1996. 
[4] Gruen, A., H.Li, Linear feature extraction with 3-D LSB- 
Snake, Automatic Extraction of Man-Made Objects from 
Aerial and Space Images, pp.287-297, 1997. 
[5] Park S.R, TXKim, Semi-automatic road extraction 
algorithm from IKONOS images using template matching, 
Proc. 22" Asian Conference on Remote Sensing, 
pp.1209-1213, 2001. 
A- 410