data which is a regular grid interpolated from the point data. 
Problems such as mixed pixel effects caused by the grid 
interpolation can reduce the amount of information that can be 
extracted from the data (Axelsson, 1999). 
This paper uses a 2m resolution gridded laser scanner data set 
which represents the terrain, vegetation and buildings as a 
Digital Surface Model. The grid was produced by interpolation 
of the raw point data which was created by an Optech Airborne 
Laser Terrain Mapper (ALTM) 1020 LIght Detection And 
Ranging (LIDAR) sensor. The sensor scans the surface with a 
2.5m point spacing. The point spacing and resolution is much 
lower than that used by Maas and Vosselmann (1999), and is 
more representative of the majority of LIDAR data sets which 
can have point densities of up to one point per 10m°. If roof 
detail can be extracted from the 2m resolution data, it suggests 
that the majority of laser scanner systems with relatively low 
resolution data sets may be able to satisfy the demand for roof 
detail for 3D city models. This would benefit laser scanner 
users who cannot afford or find high resolution laser scanner 
data for their applications. 
The aim of this paper is to investigate whether or not any 
meaningful roof detail can be extracted from the test LIDAR 
data set. 
The roof detail will be extracted from the LIDAR data by 
processing elevation and derived slope and aspect parameters 
using ARC/INFO GIS software. Algorithms will be developed 
within ARC/INFO that will manipulate the parameters to 
extract the maximum amount of roof detail. Each parameter 
will be assessed for its performance alongside the other 
parameters. Ordnance Survey 2D vector building outlines will 
be used to isolate the LIDAR building data so that the 
  
Figure 1 LIDAR representation of industrial area with 
vector building boundaries (boundaries reproduced 
from Ordnance Survey mapping with the permission of 
The Controller of Her Majesty's Stationery Office, 
Crown Copyright. ED 273554). 
   
    
    
  
   
    
    
   
   
   
    
    
   
   
   
     
    
    
    
  
   
    
    
   
  
   
   
   
  
  
  
  
  
  
  
  
  
  
  
  
   
   
   
    
   
International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999 
methodology can concentrate on roof detail extraction rather 
than building recognition. The extraction results will be 
compared with the actual roof structures observed in the field. 
2 STUDY AREA 
Two study areas were chosen that represented alternative 
building scenarios. An industrial area was chosen because of its 
high number of large, simple roofed buildings. If the LIDAR 
data is to derive accurate roof detail for buildings then it will 
most likely be from these. Figure 1 shows the industrial area as 
the LIDAR sensor captured it with building outlines added for 
extra clarity. An example photograph of the industrial area is 
given in Figure 2. It shows the dominant roof structure for this 
area which is a two segment roof split by a central ridge 
running parallel to the building's long axis. 
A residential area was chosen to represent a more challenging 
task for the LIDAR data. The buildings in this area (Figure 3) 
are smaller with more complex roof structures than the 
industrial area buildings. There is also non-building noise from 
objects such as trees, cars and hedges (Figure 4). 
3 METHODOLOGY 
To assess the performance of the LIDAR data parameters for 
extracting roof detail, a field survey was undertaken to create a 
control data set of all roof structures. These were then 
compared qualitatively to the LIDAR algorithm results using 
various comparative statistics. 
3.1 Error Assessment 
Before the algorithms were tested, an error assessment of the 
LIDAR and vector buildings was carried out. This was to 
ensure that any results were put into context with any inherent 
inaccuracies in the data sets. 
  
Figure 2 Example of industrial area buildings. 
  
  
   
Internation 
  
Figure 3 LIDAR 
vector building bo 
Ordnance Survey 
Controller of He 
C 
  
Figure 4 Exan 
A quantitative asses: 
made by comparing | 
spot heights from a 
(RMSE) was derived 
vertical error (Jaafar 
accuracy of the LIDA 
the author's own obse 
The planimetric acc 
extracted from the dat 
3.2 Survey Methode 
The survey data set i 
for every building in 
of each building wa 
including dormers an 
à 1:1250 building ma