tanbul 2004
> found that
tabase.
rs
tested. This
it, the result
ysis of the
| of airborne
tion for the
his database
10.000. The
n.
h an Optech
1 an average
which many
ntained only
1e laser data.
ogether with
:al database.
nd buildings
(dark) were
id row". For
not be taken
the mapping
[ that were
lding objects
ion, the laser
the building
5 shows the
the original
ositions after
matic pattern
ie laser data
ifts are not
> from 2 to 4
accurate map
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
no systematic offsets were observed. Most likely, the offsets
shown in Figure 5 are caused by the monoplotting procedure
used for the production of the topographical database. After
applying the determined shifts to the database objects, this
error should not lead to errors in the subsequent change
detection step.
i a
C
Figure 5: Systematic offsets between building segments in
the laser data and the contours of the
TOP1Ovector map.
6.4 Analysis of detected changes
The results of the change detection are visualised in Figures 6
and 7. Figure 6 shows the detected new and demolished
buildings. The buildings classified as changed are shown in
Figure 7. All demolished buildings are detected correctly. If a
building was demolished and replace by a completely new
building, this building was classified as changed. Also these
type of demolished and rebuild buildings were detected
correctly.
Li
> * AS
3 ^ a g we jn
c
po
>
Figure 6: Results of detection of new and demolished
buildings. The segments extracted from the laser
data are shown in light grey and green. The map
objects of the TOP10vector map are overlaid in
black and red. The red objects are classified as
demolished buildings. The green segments are
classified as new buildings.
The detection of new buildings appeared to be more difficult.
All new buildings are detected correctly, but in addition some
sheds were detected in the laser data that are not present in
the map. These incorrectly detected “new” buildings were
caused by a different interpretation of the visibility rule in the
mapping catalogue. Whereas our algorithm checks for a line
of sight between the shed and the nearest road, the mapping
agency first determines the main building to which the shed
belongs and then argues whether the shed can be seen from
the road at the front of the main building. Such a rule is
actually quite complex and requires some scene interpretation
that is difficult to implement.
Figure 7: Segments in the laser data classified as changed
buildings.
The buildings that were classified as changed fell into three
categories:
e Several buildings were indeed changed, or demolished
and replaced.
e In a few cases vegetation adjacent to buildings led to
enlarged laser data segments. These were incorrectly
interpreted as building extensions.
e Finally, the change detection revealed several errors in
the topographical database. One example is shown in
Figure 8. Whereas the operator mapped three separate
buildings, the laser data shows that these buildings are
connected by lower parts with flat roofs.
Figure 8: Detected mapping error (see text). Left: laser data
segments (grey) and database objects (green
(dark)) Right: colour image of a three-line
scanner
7. CONCLUSIONS
In this paper a study for automated change detection of
buildings in a medium scale digital map using airborne laser