33. Istanbul 2004 
‚ the grass lands 
pen areas. First, 
"d and coated by 
the pixels of the 
eas it is high and 
d the grass lands 
saturation areas. 
e can obtain the 
ing areas. 
  
Transform 
etry. We used a 
ect the candidate 
lata — the binary 
d for extracting 
ght line with the 
can be used for 
he Hough space 
e straight line as 
(1) 
the x-axis; DO is 
nsfer space, we 
ure 4. To detect 
ne the primary 
ons and ribbons 
extracted. The 
gmented binary 
ribbon, and the 
h (the difference 
rm. In each step 
se in the Hough 
moved from the 
" multiple peaks 
rminated by the 
1e length of the 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
  
  
(b) Detected road stripes displayed on the image 
Figure 4. Road stripe detection by Hough 
transform from segmented lidar data 
3.3 Verification of road stripes and parking areas 
The detected primary streets by Hough transform are possible 
streets and just straight line equations (parameters). To form a 
real street *grid', we should identify the candidates and remove 
some wrong segments. The first step is to overlay the straight 
lines onto the binary image. For each line, break it to be 
segments where it transverses building areas. It can be fulfilled 
simply by the binary image. Thereafter, each verified line 
segment is adjusted by geometric correction — to move it to be 
in the street centre where the dual distance between it and the 
building edge is equivalent. 
We judge that the short segments going through the big open 
areas are with low possibility of being a part of the street and 
high possibility of being a parking area. To verify a parking area, 
we employ the vehicle clue to confirm the area. The vehicles are 
extracted by a pixel based classification method. Some samples 
of vehicles are provided by manual digitization, and they are 
used for learning the pixel intensity value of the vehicles. The 
possible pixels of the vehicles in the road and parking areas are 
shown in green and blue colours in the Figure 5 (a). In the 
study, the open areas contain roads and parking lots. We assume 
a region with nearly squared shape and big area has high 
possibility of being parking lots. A morphologic operation is 
applied to the binary image to detect the big open areas. In 
Figure 5 (b), the highlighted areas are possible open areas rather 
than roads, but the roads could go though the area. Combining 
the analysis result of shape and vehicle clue from lidar data and 
the optical imagery, we compute the ‘score’ of an open area of 
being a parking lot. The high score indicates the high possibility 
of being parking lot. By computing the length of the segment 
which goes through the parking area, the segments mostly lie in 
323 
the parking areas are removed. As shown in the Figure 5 (b), the 
short segments circled will be removed because they are most 
likely gong through the parking areas. 
  
  
  
  
  
(b) Verified road stripes displayed on the image 
Figure 5. Road stripes and parking areas 
verification 
3.4 Road Topology 
The road topology is formed by intersecting the road segments 
extracted from the previous steps, as shown in Figure 6. 
  
Figure 6. Road grid formation