Gary Priestnall 
  
  
  
  
— 
Base Class — (Known Line) 
Object Class — (Road) 
sa Sub-Class — (Motorway) 
is could Geometric Attribute ——p (Centreline, width) 
ctral Non-geometric attribute ———» (Number of lanes, surface material) 
rk, and also 
topology 
hes to 
combined 
inuity 
egally' 
roads from 
opological 
lude: 
; Or is 
resolution 
context 
ained. They 
ert and other 
important. 
n (inferred 
S). 
' road 
ot only give 
> whole 
- time and 
; new clues 
on geometric 
ated, 
'ometric 
|. This 
ar features of 
se classes art 
il] populate 
- known class 
le for à road 
Figure 3. Object schema and example categories 
Junction features will hold vital network topology details and can form good starting points for searches to complete 
linear segments and begin growing the network. This section discusses the interrelationships between objects and how 
they might help define local contextual rules (which in turn can be influenced or constrained by the regional context in 
which an object finds itself). In addition to the rules for an individual feature in terms of its geometry and other 
properties, the spatial relationships between different feature types can be summarised, as in Table 1. These feature 
interrelationships relate to the context sketches described earlier (Baumgartner ef al, 1997). Furthermore, many of these 
rules for interrelationships benefit from the knowledge of containment within urban or rural context regions as 
highlighted by comments within the matrix, for example, single carriageway roads often display near parallel grouping 
within urban regions. 
  
— 
River 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Railway Motorway / Single Track Tree line Building Hedge Line/ Field 
Dual C-W Road Line Boundary 
Rounded Independent, Independent, Independent, Near parallel Near parallel in Near parallel with 
Railway junctions, low | often near- often near- often near- in urban, urban, feature, particularly 
curvature. normal normal bridging | normal bridging | weaker in weaker in in rural. Forms edge 
Near parallel bridging or or or rural. rural. of field. 'T' junction 
only inurban | tunnel tunnel tunnel more common than 
(goods yards) Cross. 
Often 'Y' Pass under Pass under Vegetation Occasional near | Near parallel, less 
River junctions motorway, motorway, follows river parallel likely to obscure but 
elevation usually normal usually normal often close grouping butset | still possible with 
profiles of and near back unless trees in hedge. 
rivers at juncs parallel, often | channeled in Hedges normal to 
may resolve obscuring urban area river lines common. 
direction river line 
Distinct Largely Few parallel Few parallel Possible near 
Motorway junctions large independent of tree lines but linear groupings | parallel hedge line 
Dual roundabouts single track near parallel of buildings in set back. 
Carriage - with slip roads roads, crossings | forest edge rural; Normal hedges less 
Way often normal set back common. 
Many types of Occasional Near parallel Occasional near 
Single junction but 'T' near parallel groupings of parallel and 
track and '+' junctions | tree lines buildings in proximal hedge line 
road common. Near especially in both rural and in rural regions. 
parallel roads in | rural. Can urban. Normal hedge 
urban obscure road. boundaries common. 
Largely Occasional near | Mixing possible. 
Tree line independent parallel esp. in Quite difficult to 
urban discriminate 
(boulevard) 
Near parallel Largely independent 
Building groupings in although occasional 
line urban. Often near-parallel pairs 
road inferred’ 
between two 
rows of 
buildings 
Topology similar to 
Hedge line roads T and + 
/ Field junctions common. 
Boundary Proximal near 
parallel pairs rare 
  
6 THE PLACE OF CONTEXT IN THE OVERALL STRATEGY 
Table 1. Matrix of linear object inter-relationships 
The role of contextual information is seen to be important at several levels. Firstly, by classification of medium 
resolution multi-spectral imagery, a simple set of context regions based upon broad land cover categories can be 
produced. Polygonised versions of these regions form objects in the hierarchy and can be used to allow containment of 
objects within them to be represented. The degree to which the choice of linear feature extraction algorithm, at least the 
associated parameters of an algorithm, is influenced by the context region, is under investigation. Vector primitives 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
745