International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
Figure 9: Result of the existing object analyses step of the
change detection methodology; light grey: building
segments for 2002 DSM, darker grey: new building, dark
grey: demolished building
It can be observed that in the upper right corner a new building
was detected, whereas in the lower image part a tear off was
recognised. In fact, the complex of three buildings in the lower
image part was entirely destructed and (partly) rebuilt. But as
the building at the right part of the complex was almost
identically rebuilt concerning its shape and height, nearly no
changes were detected; only a lower part (and therefore
considered as a segment on its own) originally surrounding the
building in 1997 is lacking now.
4.3 Modified and confirmed buildings
The segments not yet classified in this process are assumed to
be either unchanged or modified. Inside the ground plans of the
remaining segments, i.e. those segments not yet classified as
not-analysable, missing or new built, the according parts of the
DSMs are extracted and the elder data is subtracted from the
newer one. The resulting height difference image is filtered
using the opening operator of mathematical morphology, i.e. an
erosion with a consecutive dilatation is performed (e.g. Haralick
& Shapiro, 1992), regarding occurring height differences as
foreground values. The purpose of applying the opening
methodology is to remove single, isolated pixels or small group
of pixels with height differences. This is because they often
represent small objects on the roofs tops, e.g. chimneys or small
dormers, or differences occurred due to arbitrary measurement
deviations.
The remaining (groups of) pixels are then classified according
to the type of height differences. Three classes are used:
* height difference inside the tolerance range
* positive height difference (heightened)
* negative height difference (decreased)
A
N
Only segments containing pixels classified into the last two
classes are further analysed. the others are considered
unchanged.
For each segment, as well for those linked to t, as those of t;,
the dominating difference type is determined. If both classes
occur in almost the same amount inside the segment, it is not
classified; then only the altered parts are classified accordingly,
i.e. such segments contain different classified parts. In all other
cases, the whole segment is classified according to the
dominating differences type, i.e. either as added-on or reduced
building.
^
Figure 10: Reduction and add-
on at building ensemble; the
large part in the middle was
increased, the others
reduced
Figure 11: Building appearing
heightened
In Figure 10 a result is shown obtained for a building ensemble
(segments of the newer data set are laid beneath in light grey).
The middle part was heightened, whereas the surrounding parts
were decreased. In the lower right part a large building was
partly teared-off; the one in the middle was extended so that it
now covers a part of the former ground plan of the reduced one.
As the change analysis is carried out for segments of both dates
separately, a decrease (inside the ground plan of the elder) as
well as an increase (of the newer) segment can be determined at
this location, visualized by mixed colours at the most right
corner. Such, the kind of alteration can be understand better.
Figure 11 is the representation of a building which was found to
be heightened (by one floor). In fact, at this location a building
was teared off and a new one erected, but exactly inside the old
ground plan and equal shaped. Only a roofed entrance area was
added, which appears as a bar directly beneath the building. The
classification result is as far correct as another result can not be
obtained regarding solely the height data. For a more precise
raüng additional knowledge besides the laser scanning data
would be necessary.
5. CONCLUSIONS
The work presented in this paper is part of a project that stands
in the context of a disaster management tool. The projects task
is to provide fast and reliable information about the damage
state after strong earthquakes. Regarding this, the presented
approach is a first attempt to detect buildings likely to be
damaged.
The methodology is based on laser scanning derived DSMs and
it showed to be capable to reliably rate occurring buildings
changes in some (rough) modification classes solely based on
height data. New built and teared-off buildings were found
correctly in the test area as long as they were not merged with
Intern
other |
step. €
not-alt
Only |
reduce
receiv
segme
indepe
alread
Nevert
necess
extract
indicat
analys
extract
time, |
prefixe
makes
Beside
obtaine
purpos
change
used e.
in urba
Signifi
applyir
segmer
segmet
compai
with th
To imp
areas, t
improv
For ins
several
already
models
Altan,
MOLAI
Istanbu
Remote
Hydera
Bhattac
D. and
study ©
In: JAP
on CD.
CRC 4
Center
and
karlsrut
Haralic]
vision. '
Jung, F
aerial
Remote
