AUTOMATIC DETECTION OF CHANGES FROM LASER SCANNER AND
AERIAL IMAGE DATA FOR UPDATING BUILDING MAPS
Leena Matikainen, Juha Hyyppä, Harri Kaartinen
Department of Remote Sensing and Photogrammetry, Finnish Geodetic Institute,
P.O. Box 15, FIN-02431 Masala, Finland — Leena. Matikainen(@fgi.fi, Juha.Hyyppa@fgi.fi, Harri.Kaartinen@fgi.fi
KEY WORDS: Building, Change Detection, Updating, Mapping, Automation, Laser scanning, Segmentation, Classification
ABSTRACT:
The goal of our study was to develop an automatic change detection method based on laser scanner, aerial image and map data to be
used in updating of building maps. The method was tested in a study area of 2.2 km? near Helsinki. Buildings were first detected by
segmenting a digital surface model (DSM) derived from laser scanner data and classifying the segments as buildings, trees and
ground surface. Height information, aerial image data, shape and size of the segments and neighbourhood information were used in
classification. Detected buildings were then compared with an old building map and classified as new, enlarged and old buildings.
Similarly, buildings of the old map were compared with the building detection result and classified as detected, partly detected and
not detected. Compared with an up-to-date reference map, 88% of all buildings in the study area and 98% of buildings larger than
200 m^ were correctly detected in the building detection stage. Promising results were also obtained in change detection between the
old map and the building detection result, especially in detecting new buildings. Results of the study suggest that automatic building
detection and change detection is possible and could produce useful results for map updating. Further research should include
improvement of the segmentation stage to better distinguish buildings from trees and development of the change detection method.
1. INTRODUCTION
To keep digital map databases as up-to-date as possible,
efficient methods for the data acquisition and updating process
are needed. Remotely sensed data provide plenty of useful and
up-to-date information, and automatic extraction of different
objects and land-use classes from these datasets has thus
become an important research topic (see e.g. Baltsavias, 2004).
During recent years, automated map updating from remotely
sensed data has been studied by e.g. Hoffmann et al. (2000),
Niederóst (2001), Armenakis et al. (2003), Knudsen and Olsen
(2003), Jung (2004) and Walter (2004).
When updating maps from remotely sensed data, the first task is
change detection, which is carried out by interpreting the
imagery and comparing the imagery and/or interpretation results
with the existing map data. After change detection, new objects
can be extracted, changes to existing objects made and the
database updated. Automatic change detection between map
and image data has been studied by e.g. Armenakis et al.
(2003), Knudsen and Olsen (2003) and Walter (2004).
Armenakis et al. (2003) used Landsat imagery, and the object
class under study was lakes. Knudsen and Olsen (2003) used
aerial photos to detect changes in buildings, and Walter (2004)
used airborne digital camera data to detect changes in five land-
use classes. An alternative approach to change detection is to
compare two remotely sensed datasets acquired at different
dates. For example, Murakami et al. (1999) used laser scanner
data and Jung (2004) stereo images to detect changes in
buildings.
The goal of our study is to develop a change detection method
based on laser scanner, aerial image and map data for updating
of building maps. The idea is to produce a preliminary updated
building map that would present approximate building polygons
associated with attribute information showing if the building
has been built, removed or changed after the map was made.
The preliminary map could then be used in further steps of map
updating, e.g. verification of the changes, exact location and
modelling of the buildings, updating of a map database and
finally creation of a 3D city model. These further steps could be
manual, semi-automatic or fully automatic. Laser scanner data
were selected as the primary source of data because they have
proved to be promising for building extraction and modelling
(see e.g. Haala and Brenner, 1999; Maas and Vosselman, 1999;
Morgan and Tempfli, 2000; Vógtle and Steinle, 2000; Fujii and
Arikawa, 2002; Rottensteiner and Briese, 2003). The height
information facilitates building detection and in further steps
allows 3D modelling.
The change detection method discussed in the present article is
based on two steps: building detection and actual change
detection. Buildings are first detected using laser scanner and
aerial image data. In the change detection stage, the building
detection result is then compared with an existing building map
to detect changes. The main parts of the method are similar to
those presented in Matikainen et al. (2003). An aerial ortho
image was now used in addition to laser scanner data and some
changes to the building detection stage were made. The method
was tested in a new study area and with a different laser scanner
dataset. Up to now, the main focus of the study has been in
building detection. Numerical results from comparison of the
building detection results with reference data will be presented
in the article. The quality of change detection results was
evaluated visually.
2. STUDY AREA AND DATA
The study area is located in Espoonlahti, Espoo, about 15-20
km west from the city of Helsinki (see also Ahokas et al., 2004:
Rónnholm, 2004). The total area covered with the laser scanner
and aerial image datasets is about 5 km”. An arca of about 0.4
km? was selected as a training area for developing classification
rules for building detection. Up to now, the building detection
and change detection process has been applied to the training
434
Intern
area à
buildii
which
differe
area a
varyin
The |
FALC
withou
was 4(
points
averag
digital
the las
2004).
first p
determ
first p
detect
Groun
builds
anothe
with a
Classif
terrain
ground
An int
The ir
height
inform
interpa
did nc
detecti
the im:
in inter
Aerial
scanne
2003.
[Images
laser-d
referen
Howev
in the c
COVEre(
the use
Buildin
Survey
updatec
Espoo
develor
of obje
Land S
data so
accurat
present.
ground
of builc
convert
map, p
convers
constru
some si
