ISPRS Commission III, Vol.34, Part 3A ,Photogrammetric Computer Vision*, Graz, 2002 
  
  
  
  
  
  
     
(b) Gap with road side support 
(a) Gap without road side support 
  
  
Figure 9: Model for grouping lane segments into lanes: 
Components used for extraction | Components used for evaluation 
  
S, OH 
  
¢) Testing ribbon area along s(w) (d) Histogram of height variation in ribbon area 
  
  
  
Figure 6: Model for ribbons: 
Components used for extraction | Components used for evaluation 
  
  
  
> Orientation difference between 
pairs of lane segments: limited 
Gap length: limited 
» Lane segment widths: similar 
> Lane segment heights: similar 
> Gap analysis: see evaluation of 
lane segments 
> Support of road sides: high 
> Overall curvature: low 
> Overall height variation: low 
> Profile of gradient VI along 
$ | : "half-pipe" 
> Width of ribbon w: bounded 
> Length of s(w): lower bound 
> Height variation of s(w): low 
> Intensity of s(w): high 
> Homogeneity of s(w): distinct 
  
  
  
  
  
  
  
  
  
(c) One-sided homogeneous road side 
VI 
$1! 
1 
$1 
(b) Edge model in gradient image 
| Y. 
VH 
(d) Histogram of height gradients 
  
      
    
P. de n 
  
i 
gh ues Papier — | 
(a) Road segment from connected lanes 
  
  
(b) Connection between road segments 
  
Figure 10: Model for road segments and road links: 
  
  
Figure 7: Model for road sides: 
Components used for extraction | Components used for evaluation 
  
  
> Gradient VI: curvature maxi- 
mum along § | 
> Length of s: lower bound 
> Areas beside s: at least one ho- 
mogeneous area required 
> Height gradient V H: low 
Components used for extraction 
of road segments 
Components used for evaluation 
of road segments 
  
  
> Parallelism of neighboring 
lanes: high 
> Gap analysis within road seg- 
ment: see evaluation of lanes 
> Widths of lanes across road 
segments: similar 
> Heights of neighboring lanes: 
similar 
> Fragmentation (missing lane 
segments): low 
  
  
Components used for extraction 
of road links 
Components used for evaluation 
of road links 
  
  
  
  
   
    
(a) Lane segments from markings 
" 
(b) Segment from markings and road side 
  
  
Figure 8: Model for lane segments: 
Components used for extraction | Components used for evaluation 
  
  
> Groups of markings (pair): par- 
allel, width bounded 
> Group of markings (isolated): 
width hypothesized 
> Group of markings and road 
side: parallel, width bounded 
> Intensity: bright (see illustra- 
tion in Fig. 6 "Ribbons") 
> Homogeneity: distinct (see il- 
lustration in Fig. 6 Ribbons”) 
> Height variation: low (see il- 
lustration in Fig. 6 Ribbons”) 
> Orientation difference between 
pairs of road segments: limited 
> Gap length: limited 
> Road segments to connect: 
compatible lane configuration 
> Gap analysis: see extraction 
and evaluation of road segments 
> Length of road link: high 
> Overall curvature: low 
> Overall height variation: low 
  
5 RESULTS AND DISCUSSION 
Figures 11 and 12 show the final result of road extraction. The re- 
sults have been evaluated by matching the extracted road axes to 
manually plotted reference data (Wiedemann and Ebner, 2000). 
As can be seen, major parts of the road networks have been ex- 
tracted (white lines indicate extracted road axes). Expressed in 
numerical values, we achieve a completeness of almost 70 % 
and a correctness of about 95 %. The system is able to detect 
shadowed road sections or road sections with rather dense traffic. 
However, it must be noted that some of the axes’ underlying lane 
segments have been missed. This is most evident at the complex 
road junctions in both scenes, where only spurious features for 
the construction of lanes could be extracted. Thus, not enough 
A - 167