ISPRS Commission III, Vol.34, Part 3A ,,Photogrammetric Computer Vision“, Graz, 2002 
  
Suppose p, is the kth edge pixel of an edge line patch, and it is 
the pn, -th neighbour of its previous pixel p,.,» Where both p, 
and p, , have been removed from the binary image of edge 
pixels. In order to find the k+1th edge pixel of the edge line 
patch, the 8-neighbors of p , are examined, where the 
neighbour having a direction near to n , has a higher priority, 
so that is examined first. If the &'th neighbour of p, 1$ found an 
edge pixel, it is accepted as the &--1th edge pixel of the edge 
line patch. On the other hand, if no edge pixel is found in the 8- 
neighbours of p, , an examination along the extension of 7 " 
from p, is conducted. If the blanks from p, to edge pixel q in 
Figure 5(b) is less than a given threshold, q is accepted as the 
k+1th edge pixel. If no edge pixel is found in all the above 
examinations, the procedure of linking edge pixels of the edge 
line patch stops. 
rod o«a 
[] rj E] 
Ne O's OC 
Ka SM 
m C LU 
Figure 5. Searching in 8-neighbors and along the extension of 
local direction. 
Breaking edge line patches is conducted, where the edge line 
patch is broken down at the pixel of the rapid change of 
direction at either local or/and global level. For any pixel in the 
edge line patch, if the angle from starting pixel to the pixel to 
end pixel is smaller than a given threshold 7, then the pixel is 
regarded as a broken pixel at global level (see Figure 6(a)). In 
addition, for any pixel Æ in the edge line patch, given a value s 
for local length, if the angle from pixel Æ-s to pixel Æ to pixel 
k+s is smaller than the threshold ¢, then pixel Æ is regarded as a 
broken pixel at local level (see Figure 6(b)). 
Broken pixel 
: r if o<t 
Starting pixel End 
(a) Broken pixel at global level 
Broken pixel 
if act 
End 
(b) Broken pixel at local level 
Starting pixel if acf 
Figure 6. Breaking edge line patches 
3.2 Jointing Edge Line Patches 
For any edge line patch e , it is jointed with other edge line 
y P 
patches in an iterative way as follows. Given two thresholds L 
in pixel and X in degree, a searching space is defined at each 
terminal point p as shown in Figure 7(a) where ], is the 
extensional direction of e, at p. For any edge pixel q in the 
A - 408 
search space, where q is on an edge line patch e, > à cost is 
calculated as follows (see Figure 7(b)). 
E(q) = |, |+|o;| (1) 
Where, /, is a line passing through p and q, I, is the tangential 
line of e at q, ot, is the angle between / and l,, and a, is 
the angle between /, and /,. If an edge pixel q yielding the 
minimal cost E(q) is found, e is jointed to e, as shown in 
Figure 7(c). 
  
Figure 7. Jointing edge line patches 
  
Specify a starting 
point by the operator 
Y 
Automatically trace a 
road line by template 
matching 
md 
Assign a directional 
point by the operator 
T y Y 
Guide the road line 
tracing using the 
directional point 
aut 2 nx 
Figure 8. Flow of semi-automatic road line extraction 
  
  
  
  
End 
  
  
  
  
  
  
4. ROAD LINE EXTRACTION 
In this research, road lines are extracted in a semi-automatic 
way by matching a road template with both road mask (M) and 
road seeds (S). The flow of road line extraction is shown in 
Figure 8. At the beginning, a starting point is specified by the 
operator. A road line is extended from the starting point in an 
iterative way, where in each iteration, a rectangular road