In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Voi. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009
A SEMI-AUTOMATIC APPROACH TO OBJECT EXTRACTION FROM A
COMBINATION OF IMAGE AND LASER DATA
S. A. Mumtaz 3 '*, K. Mooney 3
3 Dept, of Spatial Information Sciences, The Dublin Institute of Technology, Bolton Street, Dublin 1, Ireland
(salman.mumtaz, kevin.mooney)@dit.ie
Commission III, WG III/4
KEY WORDS: LiDAR, Object Extraction, Data fusion, Buildings, Trees, Roads
ABSTRACT:
The aim of the authors’ research is to develop an automated or semi-automated workflow for the extraction of objects such as
buildings, trees and roads for noise mapping and road safety purposes. The workflow must utilise national airborne spatial data
available throughout the country and be capable of robust incorporation in the noise modelling systems of a national roads authority.
This paper focuses on the extraction of multiple objects by fusing data captured by two independent sensors, namely the Leica
ADS40 aerial camera and the Leica ALS50 airborne laser scanner (LiDAR). A workflow has been developed for the extraction of
objects utilizing height values from a normalised DSM generated using LiDAR or aerial images, multiple LiDAR echo data and
NDVI (Normalized Difference Vegetation Index) data computed from multispectral ADS40 data.
Major tasks include LiDAR data classification, segmentation and its integration with the information extracted from aerial images.
Buildings are extracted first and this facilitates the extraction of other objects. Preliminary results of this semi-automated process
indicate high completeness rates for buildings trees and roads but 60% quality rates (e.g. buildings). Quality may be improved by
manual extraction of small objects but continuing research is focussed on reducing reliance on such manual intervention.
1. INTRODUCTION
The National Roads Authority (NRA) in Ireland is responsible
for generating noise maps in the environment of roads used by
more than 8220 vehicles per day. According to the EU noise
directive this exercise must be repeated every five years. Inputs
for generating the noise maps include terrain model, location
and dimension of buildings, trees, noise barriers and the
geometric properties of roads. Capturing this data using field
surveys or digital images is time consuming and expensive,
especially if the same exercise must be repeated every five
years. It is the intention of this work that all required objects be
extracted using automatic or semiautomatic techniques from
LiDAR and aerial image data of the type available from the
National Mapping Agency of Ireland, OSi (Ordnance Survey of
Ireland). Later the extracted information can be easily
combined and analyzed along with noise data in a GIS system.
For noise mapping, building detail or tree models are not
required. Buildings or trees boundaries with height information
are sufficient.
High resolution image and LiDAR sensors (ADS40 & ALS50)
were used to capture the data for a part of County Sligo in the
northwest of Ireland. Digital images were captured in April
2007 with a ground resolution of 15 cm. LiDAR data were
captured separately in May 2007 at a flying height of 1241 m
with a swath width of 800 m, resulting in an average point
density of approximately 2 points/m 2 . The ALS50 sensor
recorded position, multiple echoes and intensity of the returning
pulse.
The area selected for processing is about 3 km 2 and is covered
by a single image strip and four LiDAR strips. This eliminates
the necessity for bundle block adjustment and ground control
point acquisition. The reason for relying completely on direct
geo-referencing in this research is the fact that in many
situations ground control points may not be available. Strip
adjustment of the LiDAR data was performed using the Terra
Match application from TerraSolid.
1.1 Motivation
In recent years, research on automated object extraction has
increased because of the increased use of GIS (Geographical
Information Systems) with the consequential need for data
acquisition and update.
Digital Photogrammetry is considered to be one of the most
precise methods for capturing large scale data for GIS analysis
from high resolution aerial images. However, it requires
significant resources to digitize all objects of interest. As
detailed high resolution digital images are regularly acquired as
part of the national programme of OSi, it is considered
important to develop automatic or semi-automatic techniques
to exploit their potential for applications such as noise
modelling involving the extraction of objects such as buildings,
trees and roads.
LiDAR can provide high density 3D point clouds in a very
short time with acceptable horizontal and high vertical
accuracy. OSi also acquires national LiDAR data using the
ALS50 sensor from Leica Geosystems. The availability to the
national roads authority of Ireland of both of these high
resolution data sources provides the impetus for this research.
However, the development in sensor technology is far more
rapid than the advancements in automatic or semi automatic
object extraction. Moreover there is still a large gap between
* Salman Ali Mumtaz
