International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999 
     
THE USE OF LASER SCANNER DATA FOR THE EXTRACTION OF BUILDING ROOF DETAIL USING STANDARD 
ELEVATION DERIVED PARAMETERS 
T.D. Chilton, J. Jaafar and G. Priestnall 
School of Geography 
University of Nottingham 
tchilton@wardrobe.u-net.com 
jaafar@ geography.nottingham.ac.uk 
Gary.Priestnall@nottingham.ac.uk 
KEY WORDS: Laser Scanning, LIDAR, Roof Detail, 3D City Models. 
ABSTRACT 
3D city modelling is a rapidly growing research area in the field of feature extraction. As the demand for 3D data increases, so does the 
necessity for higher detail building models. The use of aerial imagery and photogrammetric processing has been dominant in the field of 
feature extraction for several decades. Recently this dominance has been challenged by laser scanning techniques which offer direct 3D 
data capture at high resolutions. This paper investigates the use of laser scanning data for roof detail extraction. This has already been 
considered by Maas and Vosselman (1999) who used data with a sample spacing of more than five points per m?. The data set in this 
paper was created using an Optech Airborne Laser Terrain Mapper 1020 LIDAR sensor with 2.5m point spacing which was interpolated 
onto a 2m regular grid. The aim of this paper is to explore the extent to which useful roof detail can be extracted using this relatively low 
resolution LIDAR data. A 2D building outline database was used to define the roof boundaries. 
The ability of the LIDAR data to extract roof detail was tested using algorithms based on LIDAR elevation and derived slope and aspect 
parameters. An industrial area and a residential area were selected as test sites. Results show that the LIDAR aspect parameter was the 
most successful at reconstructing building roofs. Using LIDAR aspect data, the results show a percentage correspondence of derived 
roof ridges to actual ridges of around 33% for the residential area (86% for the industrial area). The amount of roof detail extracted is 
greater for the industrial study area due to the simpler, larger roof structures. The report concludes that relatively low resolution laser 
scanning data can be used to extract roof detail, but that results are only significant for large, simple roof buildings. 
1 INTRODUCTION 
In recent years there has been an increasing interest in the 
availability of 3D descriptions of real world scenes, especially 
within urban areas. These 3D city models are being sought 
after for use in a wide range of applications including urban 
planning and virtual reality applications (Newton, 1996). As 
the technology behind 3D modelling matures, there is an 
increasing requirement for greater building detail. This paper 
looks at a possible data source for one aspect of building 
description, that of roof detail. 
The field of feature extraction has been dominated for several 
decades by the use of aerial photography and photogrammetric 
processing. Building recognition is one research area within 
this field which focuses mainly on the extraction of building 
outlines using edge detection techniques (Frere et al., 1997). 
Attempts have been made using aerial imagery to extract 3D 
roof detail supplementary to the ground plan information. 
Griien and Dan (1997) attempted to match extracted roof and 
building line segments derived from aerial imagery to an a 
priori building classification. A semi-automated approach was 
tested by Lang and Forstner (1996) by creating building 
models based on building examples from the actual study area. 
It is likely that this procedure would be more time consuming 
than the method adopted by Griien and Dan (1997), but may 
    
end up being more accurate due to the site specific nature of 
the building models. 
Photogrammetry for 3D modelling has limitations. Braun et al. 
(1995) state that aerial imagery contains too much information 
in addition to the building information that would enable easy 
extraction of building outlines and roof information. 3D 
information is imposed onto a 2D image format making 3D 
reconstruction harder. Lengthy stereo-matching procedures are 
required to extract height information using user defined 
sampling methodologies. Low contrast edges and poor image 
perspectives make building extraction more complex still. 
Laser scanning is a relatively new technology that is 
challenging the dominance of photogrammetry in the field of 
feature extraction, especially for 3D city modelling. Laser 
scanning offers direct 3D data capture at high resolutions and 
the possibility to extract roof detail. Hug (1997), Maas and 
Vosselmann (1999) and Weidner (1997) have looked at the 
extraction of buildings using laser scanner data. Maas and 
Vosselmann (1999) have looked in particular at the extraction 
of roof detail. They use raw laser scanner point data which for 
their study has a density of over five points per square metre. 
Triangulation of the point data and other processing stages 
were required to produce high detail roof structures. Authors 
"such as Jaafar et al. (1999) and Haala and Brenner (1999), 
however, have used the more common form of laser scanner