LYSIS 
-oimbra, 
he spatial 
temporary 
‘this work 
anding of 
e of some 
a LiDAR 
TIN, two 
have been 
side these 
of cluster 
, and say, 
mapping, 
IS vector- 
le use of 
nework is 
few basic 
). Indeed, 
n, 1992) 
uable and 
ucture of 
hich, after 
in a GIS 
) adds that 
ard notion 
1 different 
ynnections 
e starting 
ms of the 
s possible 
(a set and, 
he spatial 
to test the 
nplement. 
International Archives of the Photogrammetry, Remote Sensing 
It is an unstructured data set without even providing any 
imagery of the surveyed area. Range data is a random collection 
of a considerable number of 3D points, with no pattern pre- 
defined, which are typically used for the generation of TINs 
which basically, in terms of GIS analysis, translate into what we 
define as a set of first order connections in vector domain, Le., 
spatial relationships between objects in direct contact. Using a 
graph-based approach, we are planning to build up networks of 
connectivity through these data sets that may allow the 
performance of what we call higher order connections analysis, 
ie, to investigate and understand the spatial relationships 
between objects within the context of the whole scene rather 
then within the context of their own neighbourhood. 
1.4 An urban scene 
Carrying out this sort of analysis in the context of an urban 
scene is particularly challenging given its relatively small 
component elements (such as, buildings, roads and intra-urban 
open spaces) and their generally complex spatial pattern. In 
fact, according to some authors (including Eyton, 1993, and 
Barr & Barnsley, 1996, both cited in Barnsley and Barr, 1997) 
the classification process of spatial information to produce land- 
cover maps (maps of forms) for urban areas can be considered 
fairly straight forward if we compare it with the process of 
deriving information from those maps on urban land-use (maps 
of functions). This is normally much more problematic namely 
because land-use is an abstract concept: an amalgam of 
economic, social and cultural factors defined in terms of 
functions rather than in physical forms (Barnsley and Barr, 
1997). 
> 
2. DESIGNING THE GRAPH-BASED APPROACH 
2.1 First order information retrieval 
The LiDAR data set being used has got 3metre point spacing 
and it contains both ground points and objects points reflected 
from trees, buildings and other small objects above ground 
level. The data set refers to a surveyed area (1470x1530m^) in 
Southwest London (Kew), including the Public Records Office 
and its neighbourhood, comprising a total of 169819 laser 
points (vd. Figure 1). 
    
SILiDAR points heights (im) 
© 0.340000 - 2, 280000 
2.280001 - 4.010000 
4.010001 - 5, 100000 
5.100001 - 6.610000 
6.610001 - 8.430000 
8.430001 - 10.760000 
10.760001 - 13.390000 
13.390001 - 16.740000 
r 16.740001 - 21.470000 
E e 21.470001 - 26.120000 
Figure 1. LiDAR data set being used 
(Kew, Southwest London). 
* 999 0909000 
  
As explained, our starting situation is an unstructured data set of 
3D points, meaningless in terms of urban scene. To start 
Structuring information and make it more explicit, some 
topological information was brought in by establishing a 
and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
triangulated irregular network (TIN) through the given data set 
(vd. Figure 2). In fact, the generation of the TIN was based 
upon the Delaunay triangulation which, given the fact that it is 
a maximal planar description of the given point set internal 
structure (Kirkpatrick and Radke, 1983), expresses proximities 
and neighbourhoods between the LiDAR points. 
    
Elevations (m) 
- 32.030 
- 28.502 
24,974 
1 
   
dE SE SA 
Figure 2. TIN generated from the LiDAR point set. 
2.2 TIN facets classification 
After the generation of a TIN from the cloud of LiDAR points, 
which translates a set of first order information, two different 
binary classifications (based on two different TIN facets slope 
threshold) were applied to the TIN facets: one uses 60° slope 
value; the other is an equal-interval classification using 45? 
slope value. With the first classification, polygons of steep 
facets (60°-90° slopes) were expected to outline buildings but, 
as we can see on the left hand side of Figure 3, building entities 
are not well defined. In order to obtain a better shape of these 
entities, the second classification was carried out and its results 
are shown on the right side of Figure 3. 
  
    
  
Y LEGEND 
“4 LEGEND ne | 
: Slope | ^Y 
% ie A | t Slope 
RAS DO 0.60 P, Oo 0.4 
r i: 
S 3 31 EN! 60-90 <X nl AA iÁ BE: 45-90 
wow : “ ^ 24 > m A ; 
Figure 3. Two different classifications for the same area. 
(60? vs. 45? degree slope thresholds). 
As the range data available constitutes a very large data set, two 
case studies were chosen amongst the total LiDAR data set: one 
of which includes the Public Records Office and its surrounding 
area, corresponding to a relatively simple urban scene (it is the 
one showed in Figure 3); the other one corresponds to a much 
more complex scene given the higher density of small size 
urban features, buildings and trees. 
To start with, the two binary classifications obtained for the 
simple urban scene (Public Records Office and its surrounding 
area) were compared and contrasted. 
All the operations described above for the TIN facets 
classification and the respective generation of polygonal 
1049