rical 
ions 
n of 
pads 
1 10 
ject 
ime 
hey 
1 be 
ents 
will 
s of 
For 
S to 
ve 
cel 
on 
he 
us 
Annett Faber 
      
$ 4 3g 
: T 
r= (COS @% ‚sin (Y) 
Figure 3: Mapping of the orientation angles of orthogonal edges to the same value 
The idea of our algorithm is to estimate the dominant orientation of extracted line segments in a moving window and build 
connected components of centers of windows with similar orientation. 
The analysis of images regarding to a dominant orthogonal road structure requires two main steps: the Preprocessing in 
connection with the line representation and the Segmentation. This steps we will describe in the next subsections and are 
show in Fig. 2. 
4.2 Preprocessing and line representation 
The Preprocessing of the images and the representation of the roads as lines we realized by: 
Line extraction When using the image data of the MOMS-02 camera, we magnified them by a factor of four using bicubic 
interpolation. This lead to a pixel size of appr. 1.25 m on the ground. 
We use the program for line extraction from C. Steger (Steger, 1996) which is specialized for extracting roads in digital 
images. Output of this algorithm are vectorized line data. 
Evaluation of orientation The orientation $;; of extracted line segments is determined and mapped to the unique four 
fold angle a;;. 
Coding of lines corresponding to the orientation To simplify the procedure we go back to the raster representation. 
Each pixel of a line segment will be set to its fourfold angle. 
4.3 Segmentation 
The segmentation process described in this section is divided in (1) the evaluation of a histogram of the orientations, 
(2) the determination of an estimated dominant orientation and (3) the partitioning. The first two steps are working in a 
moving window covering the complete map. The size of the window has to be adapted to the average density of the road 
network, which up to now is done manually. 
Determination of histogram of orientation Within each window we evaluate the grey value histogram without consid- 
eration of the background value. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 277