VW NN € -—-— 07 
m4 x24. 
NE c X 
      
  
ISPRS Commission III, Vol.34, Part 3A »Photogrammetric Computer Vision“, Graz, 2002 
Road network) 
; connects 
Road link }= == :--- > Junction ) 
Road segment 
     
       
     
     
        
   
       
  
unction 
Complex junction Simple 
Fine scale 
is parallel or... 
orthogonal 
    
  
t borders or 
“is painted on” (Pavement 
| Long Short ) Colored 
| colored line colored ine] symbols 
  
  
   
  
Elongated, flat Compact 
  
       
  
concrete or concrete or 
  
asphalt region | (asphalt region 
  
  
  
  
Part-of relation 
— Specialization 
————- Concrete relation 
---*- General relation between objects 
Coarse scale 
Context (Vehicle ) 
ehicle 
Object TET 
s Acc Tat m 
m casts shadow om occludes | 
m HEP mom n" 
Y 
Geometry | Road Object 
Material v 
s m. 
edes set 
ird e divides : 
  
Long Short Bright À Elongated Compact 
bright line bright line ai) bright region bright region 
Bright homo- 
' | geneous ribbo 
  
[ Bright Image =. 
n L blob Substructure 
Figure 1: (a) Road model (left). (b) Context model (right). 
similar way, we model the integration of GIS-axes and relations 
to sub-structures. Figure 1 b) summarizes the relations between 
road objects, context objects, and sub-structures by using the con- 
cepts "Lane segment" and "Junction" as the basic entities of a 
road network. 
3.2 Extraction Strategy: 
In a very general sense, the extraction strategy inheres knowledge 
about how and when certain parts of the road and context model 
are optimally exploited, thereby being the basic control mecha- 
nism of the extraction process. It is subdivided into three levels 
(see also Fig. 2): Context-based data analysis (Level 1) comprises 
the segmentation of the scene into the urban, rural, and forest area 
and the analysis of context relations. While road extraction in for- 
est areas seems hardly possible without using additional sensors, 
e.g., infrared or LIDAR sensors, the extraction in rural areas may 
be performed with the system of (Baumgartner et al., 1999). In 
urban areas, extraction of salient roads (Level 2) includes the de- 
tection of homogeneous ribbons in coarse scale, collinear group- 
ing thin bright lines, i.e. road markings, and the construction of 
lane segments from groups of road markings, road sides, and 
detected vehicles. The lane segments are further grouped into 
lanes, road segments, and roads. During road network comple- 
tion (Level 3), finally, gaps in the extraction are iteratively closed 
by hypothesizing and verifying connections between previously 
extracted roads. Similar to (Wiedemann and Ebner, 2000), local 
as well as global criteria exploiting the network characteristics are 
used. Figure 3 illustrates some intermediate steps and Figs. 11, 12 
show typical results. In the next section, we turn our focus on the 
integrated models for extraction and internal evaluation. 
4 EXTRACTION AND EVALUATION MODELS 
As (Tónjes et al., 1999) our approach utilizes a semantic net for 
modeling. However, our methodology of internal evaluation dur- 
ing extraction complements other work as we split the model of 
an object into components used for extraction and components 
used for internal evaluation. The model components used for ex- 
traction typically consist of quite generic geometric criteria which 
are more robust against illumination changes, shadows, noise, 
etc., whereas those used for evaluation are mostly object specific. 
In so doing, both extraction and evaluation may be performed in 
a flexible rather than monolithic fashion and can adapt to the re- 
spective contextual situation. The extraction of markings, for in- 
stance, is based on line detection while their evaluation relies on 
  
is approximately : 
Building GIS road axis 
sa. A 
parallel to is close to 
rarae 1..2..--- 
‘defines end of 
E wed 
Orthogonal marking 
- SisAjeue ejep 
peseq-3xa3u07 
  
Analysis of context relations: 
shadow, occlusion, 
building outlines 
approach for rural areas 
Road extraction using ) 
-> Road links, junctions 
  
  
  
Focus on initial Regions of Interest: 
Fusion of hypotheses for road axes and road sides 
-» Homogeneous ribbons 
  
  
  
Construction of lane segments: 
Extraction of marking groups 
Detection of vehicle (convoy) outlines 
  
  
  
Fusion based on lane segments: 
Merging of lane segments 
  
Detection and removal of inconsistencies 
-> Road segments 
| 
Generation of connection hypotheses 
  
  
Local connections and junctions 
Global network connectivity 
| 
Verification of connection hypotheses 
  
  
Road extraction tools 
  
speoi jueies jo uonoex3 
uonejduioo x1o^jau peo 
  
  
  
Context relations 
No further hypotheses 
Road network 
  
  
  
  
Figure 2: Extraction strategy for urban areas. 
A - 165