'. Istanbul 2004
f several pulse
cal analysis of
le to determine
ient, which can
teps.
1 enhance our
ut this comes at
ulses the whole
1d development
algorithms and
rded waveform
oported by the
no. P-15789.
- present status
togrammetry &
999. The Laser
le, digitisation-
vegetation and
try & Remote
to commercial
ig & Remote
altimeter return
ice topographic
Artech House,
emote sensing.
)04. Website
ar returns from
ice and Remote
81. Microwave
Artech House,
Der Laserstrahl
Österreichische
'GI 4/2003, pp.
CHANGE DETECTION FOR UPDATING MEDIUM SCALE MAPS USING LASER
ALTIMETRY
G. Vosselman, B.G.H. Gorte, G. Sithole
Delft University of Technology, Faculty of Aerospace Engineering
Department of Earth Observation and Satellite Systems
P.O. Box 5058, 2600 GB Delft, The Netherlands
{m.g.vosselman, b.g.h.gorte, g.sithole }@lIr.tudelft.nl
Commission III, WG III/3
KEY WORDS: Laser scanning, LIDAR, Classification, Change detection, Mapping, Updating
ABSTRACT:
To increase the update rates of topographical databases, research is performed to automatically detect changes using airborne laser
scanning data. After the determination of the bare-Earth points, the remaining points have been classified as either points on
buildings or points on vegetation. Additional usage was made of registered colour imagery taken during the laser scanning survey.
The results show that buildings can be detected reliably using laser altimetry data sets. However, they also show that mapping rules
(which buildings should be in the map and which can be neglected) need to be implemented accurately. Otherwise, the change
detection procedure would signal a need for map updating for buildings that are not to be mapped.
1. INTRODUCTION
To satisfy the demands for more frequent updates of
topographic databases, mapping agencies are looking into the
possibilities to partially automate their production processes.
Automated mapping still seems to be far out of reach.
However, new technologies like laser scanning can help to
speed up the production process. When revising a
topographical database, much time is currently spent on
checking whether the information is still up to date.
Significant costs savings can be obtained if one would be
able to automatically flag the objects in the database that
need to be updated. In this way an operator would no longer
have to look at map areas where no changes took place. This
paper reports about studies on the usage of laser scanning
data for automated change detection of buildings for the
purpose of updating a medium scale map (1:10.000 scale).
In general, change detection can be performed on multi-
epoch data or by comparing data of a single epoch to a map.
Surface model differences generated from multi-epoch data
of laser scanners immediately show newly constructed or
demolished buildings and roads (Murakami et al. 1998,
1999). In most cases, such data will, however, not be
available. Alternatively, one can compare object extracted
from laser data of a single epoch to the objects of a map. For
this purpose one first needs to segment the laser data and
classify the segments. This approach is followed in this
paper.
In Section 2 related literature on the classification of laser
scanning data and the usage of laser scanning data for change
detection is briefly reviewed. Section 3 discusses the
segmentation and classification of laser scanner point clouds
into bare Earth, building, and vegetation segments. Results of
this classification are presented in section 4. The segments
classified as building segment are compared to the building
objects of a topographical database. The purpose of this
comparison is to detect buildings that are new, changed in
size or shape, or demolished. For this step to be successful, it
is important to implement the same object selection rules as
described in the mapping catalogue used for the production
of the topographic database. Differences caused by
generalisation of the building shapes in the database also
need to be accounted for. The developed procedure for
change detection is described in Section 5. The results are
discussed in Section 6.
2. RELATED LITERATURE
The classification of laser point clouds into points on the bare
Earth surface and other points is of large importance for the
production of digital elevation models with laser scanning.
Many studies have been devoted to this subject. Sithole and
Vosselman (2004) provide an overview on these filter
algorithms together with an experimental comparison.
For the purpose of change detection it is required to further
classify the points that do not belong to the bare Earth
surface. Maas (1999) and Oude Elberink and Maas (2000)
extract texture measures from height co-occurrence matrix.
These texture measures, together with differences between
first and last pulse laser data and the heights of a normalised
digital surface model are used as the input for an
unsupervised K-means classification. Depending on the
number of object classes to be distinguished, 90% to 97%
correct classifications were obtained.
Matikainen et al. (2001, 2003) use a bottom up region
merging algorithm to create segments. For these segments
attributes like texture measures from a co-occurrence matrix,