scanning patch by orientation 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
thematic —— patch with CCD ered point 
map camera location 
] 
+ 
topological line 
structure —] arc vectorization | «— following 
generation 
  
  
  
  
  
| 
i 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
vector data rectification; geometric 
compression — | mosaicing»inputing [—— measuring 
attribute codes 
eo ra 
+ + 4 } 
plotting GIS statistical 
output data base processing 
Fig.2. Processing diagram 
3. MAIN PROCESSING LINKS 
3.1 Orientation point location 
  
In each patch, there are four cross shaped 
orientation points on the corners. They 
are the control points for rectification 
and mosaicing, so they need to be located 
with high accuracy. Approximately locating 
the orientation point by manual pointing, 
the system detects many feature points and 
locates the point accurately by least 
square fitting. The location accuracy 
reaches to subpixel level, The algorithm 
can prevents the disturbance of other 
lines because it uses the structural 
information of the cross. 
3.2 Line following 
Line following is a sequential line 
extraction method. It is performed in 
the multi-grey levels image. Compar ing 
with the binarization method, its significant 
advantage is that it can use the 
information gotten in last following step. 
So it is easy to take the local feature 
into account and this prevents the effect of 
uneven light and of the change in line 
width and darkness. Also, its location 
precision is higher than that of 
238 
binarization method. 
The main procedures of line following are 
detecting start points and following lines 
sequentially according to some searching 
schemes and rules, 
3.2.1 Detecting start points Start points 
are selected among the most reliable 
points on the line, When a start point is 
found on a continuous line, the whole line 
will be followed in two directions. So the 
criterion to distinguish start points 
should be more strict than that in line 
following process. According to the "from 
coarse to fine" principle, a number of 
candidate points are extracted based on the 
pixel grey value, then more strict criteria 
are added to detect start points. These 
criteria are local sum of grey values, the 
density of candidate points and so on. The 
start following direction is detected by 
some template matching operators. 
  
3.2.2 Line following algorithm Following 
algorithm, namely the searching scheme 
and decision criterion, is the core of the 
line following procedure. A lot of 
algorithms can be used in line following. 
Groch ( Groch 1982) proposed a local and a 
regional method for line following based on 
a profile analysis operater (PAO) . The 
local method detect points of the line 
by PAO sequentially in a arc shaped sample 
line with a variable length of the step. 
This method is mainly used in continuous 
line following. For every step in the 
regional method a rectagle area of interest 
(AOI) is defined straight in front of the 
last line segment increment. The AOI 
contains n samples which are perpendicular 
to the direction. In every sample line hint 
points are selected according to a 
confidence measure. A collinear. measure is 
calculated to chose line points from the 
hint points. This method is mainly used in 
dashed line following. These two methods are 
combined to follow arbitrary lines. 
Joseph ( Joseph 1989 ) used two thresholds to 
follow lines of uneven density. One is 
called fatal threshold at which a single 
pixel on the line terminates extension 
and another some what blacker threshold 
called provisional threshold at which the 
provisional end point of the line is marked. 
In addition, edge following (tracing) 
  
  
  
  
  
  
alg: 
Zam] 
can 
modi 
The 
rest 
adv: 
alg: 
poir 
mant 
-eul 
foll 
line 
diff 
Fig. 
met! 
38 t 
SOME 
3d). 
sche 
gre) 
locs 
leac 
thet 
whic 
3. 2, 
of 1 
thin 
fol] 
inf« 
line 
inf« 
valu