3. Istanbul 2004 
lion of terrain 
- scanner data. 
ote Sensing 53. 
ited methods of 
ency estimation 
19th Biennial 
hy in Resource 
SA 
ze, B. 2001. 
LIDAR remote 
jan Symposium 
2001. 
ion to Modern 
tree height and 
1ser in a boreal 
79, 105-115 
Detecting and 
laser scanner. 
2, 68, 925-932. 
' Crowns from 
. International 
sing - ISPRS 
ber 9-13, 2002. 
on models in 
measured with 
ci University of 
n, H., Hyyppä, 
digital images 
metric Journal 
' forest canopy 
il photography 
haracterization 
A, October 22- 
Rónnholm, P., 
Forest Growth 
es using Laser 
Workshop on 
, September 2- 
THE AUTOMATIC EXTRACTION OF ROADS FROM LIDAR DATA 
Simon CLODE*“* Peter KOOTSOOKOS*, Franz ROTTENSTEINER® 
* Intelligent Real-Time Imaging and Sensing Group, 
The University of Queensland, Brisbane, QLD 4072, AUSTRALIA 
sclode,kootsoop@itee.uq.edu.au 
* School of Surveying and Spatial Information Systems, 
The University of New South Wales, Sydney, NSW 2052, AUSTRALIA 
f.rottensteiner@unsw.edu.au 
Working Group III/3 
KEY WORDS: Road, Detection, Extraction, LIDAR, DEM/DTM. 
ABSTRACT 
A method for the automatic detection of roads from airborne laser scanner data is presented. Traditionally, intensity 
information has not been used in feature extraction from LIDAR data because the data is too noisy. This article deals 
with using as much of the recorded laser information as possible thus both height and intensity are used. To extract roads 
from a LIDAR point cloud, a hierarchical classification technique is used to classify the LIDAR points progressively into 
road or non-road. Initially, an accurate digital terrain model (DTM) model is created by using successive morphological 
openings with different structural element sizes. Individual laser points are checked for both a valid intensity range and 
height difference from the subsequent DTM. A series of filters are then passed over the road candidate image to improve 
the accuracy of the classification. The success rate of road detection and the level of detail of the resulting road image both 
depend on the resolution of the laser scanner data and the types of roads expected to be found. The presence of road-like 
features within the survey area such as private roads and car parks is discussed and methods to remove this information 
are entertained. All algorithms used are described and applied to an example urban test site. 
1 INTRODUCTION 
1.1 Motivation 
Road extraction from remotely sensed data is a challeng- 
ing issue and has been approached in many different ways 
by Photogrammetrists and digital image processors. Some 
of the methods are quite complex and require the fusion of 
several data sources or different scale space images. The 
goal of this paper is to suggest an extraction method that 
will provide results equivalent to other methods but re- 
lying solely on the acquired LIght Detection And Rang- 
ing (LIDAR) data. Research on automated road extraction 
has been fuelled in recent years by the increasing use of 
geographic information systems (GIS), and the need for 
data acquisition and update for GIS (Hinz and Baumgart- 
ner, 2003). Existing road detection techniques, often re- 
quire existing data and or semi-automatic techniques (Hat- 
ger and Brenner, 2003) and produce quite poor detection 
rates. 
This paper presents a simple and accurate method for the 
automatic detection of roads from LIDAR data or what 
is sometimes referred to as airborne laser scanner (ALS) 
data. Section 2 describes the background of road extrac- 
tion including previous road detection methods from pho- 
togrammetry and satellite data. Section 3 describes how 
our new hierarchical classification technique is used to pro- 
gressively classify the LIDAR points into a road image. 
The technique uses as much of the LIDAR information as 
possible, such as height and intensity. A full description 
of the steps involved is given and the algorithm is applied 
  
"Corresponding Author. 
  
to an actual urban data set. Results from the data set are 
discussed in Section 4 whilst conclusions and future work 
are examined in Section 5. 
1.3 The Test Data 
LIDAR data from Fairfield in Sydney, Australia, was ini- 
tially collected with an approximate point density of 1 point 
per 1.3 m2. 
  
Figure 1: The Fairfield Test Area. 
The Fairfield data set is very interesting because of its di- 
verse nature. Within the 2km x 2km area, the land usage