International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B4, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
the process of detecting and extracting the corresponding
objects from the image. Finally, the transforming and
registering objects from the image on the 3D model is the
another consideration in this research study.
4.1 Detecting and extracting buildings
For extracting objects from the point clouds data, a polygon was
initially defined that enclosed the study area. Then an operator
which was developed based on the Equation 2 or Equation 4
was applied on the polygon for extracting objects. Since the
results of two operators are similar, only the result of the first
operator will be discussed here. At the first step the operator
was searching the buildings inside the polygon according to the
knowledge that initially introduced. The following pre-
knowledge have been initialised for the building extraction from
the point clouds: (i) the elevation of building which usually
more than 2 meters from ground, (ii) the edge of the roof which
usually stays in one level, and (iii) the roof has a regular
geometrical shape such as rectangular, pentagon, and so on. As
mentioned earlier, no attempt was made to introduce a template
to the operator for assisting in building detection and extraction.
Usually, some studies focused on detecting and extracting
buildings which have a rectangular shape. Basically, most of
buildings have different shapes; however, their shapes are
geometrical. According to the initial knowledge, the operator
tested the all points inside the defined polygon against the
knowledge and the mathematical modelling. The extracted
building were saved in a database and then captured in the
computer in a 3D space. Each building has a particular shape,
and operator is able to detect and extract any buildings with any
type of shape. Figure 2 shows the extracted buildings inside the
study area in a 3D model. There are some issues in the process
of buildings extraction. For example, in a few cases a large tree
partially obstructed a building, and the operator will extract only
a part of buildings which wasn't covered by the tree. In this
situation, the extracted building will be captured in a 3D CAD
environment and the missed part will be estimated and then a
correction approach will be implemented to fix and fill the
missing part. Both extracted buildings and their profiles were
checked for finding any missing part regarding to the obstructed
by an object. The pre-controlling from the result shows the
operator extracted the buildings precisely. The roof of some
buildings had a very complex shape but the operator was able to
extract all parts of the roof.
4.2 Detecting and extracting trees
For detecting and extracting trees from the point clouds data, a
grid has been defined. The size of the cell of the grid is roughly
equal to the size of the footprint of the Laser Scanner. The grid
has been oriented with the point clouds. Then the operator
searched the defined area for detecting and extracting trees
according to the introduced knowledge. The knowledge was
defined based on the behaviour of the returned waveform from
the tree. Since the new Laser Scanners are able to record up to
four returned waveforms of a transmitter signal from woods
area, and also the returned waveforms scatter in the grid, the
operator will assess the density of the point clouds in each cell
of grid. If the operator recognises the density of points in a cell
of grid is different with the defined density, then it will assess
the curvature of surface for each point. In this case, the operator
will recognise a tree from background, if the curvature is
changing rapidly. It has to be noted that the operator is not able
to recognise the type of a tree because this is beyond the scope
of this study.
4.3 Detecting and extracting roads
In this step, the start, the end, the changing direction points, and
the maximum and the minimum width of the road will be
introduced to the operator as initial knowledge. Then operator
will extract roads inside the defined area. Actually, an
international standard available for road construction, but some
roads and streets have own characteristics that they have to be
initially considered for the road extraction. In most of cases, the
variation of curvature along the cross section of the roads
always is very smooth and when reaches the edge of the road
the curvature will be changed rapidly. In this study the operator
will extract the road according to the initial knowledge and
operator will not stop when the curvature changed rapidly
because the point clouds data includes running vehicles on the
road as well as parked vehicles. These unwanted data will
interfere the process of road detection and extraction. Therefore,
it was decided to extract the road according to initialisation and
later a cleanup process was implemented in order to omit
unwanted data.
Figure 2. This is the result from extracting buildings inside the
study area. As the figure shows each building has a specific
shape.
Figure 3. This figure shows the profile of the buildings that
shown in Figure 2. This figure shows how the curvature at
edges of roof is changing rapidly.
4.4 Extracting crown land
In this step, all extracted data will be subtracted from the
original point clouds data and the result will be a bare terrain
that is considered as crown land. The crown land includes a
number of gaps that can be filled using interpolation approach,
but it has to be aware that some area was cut and removed
during a development and an interpolation will changed the
182
ex
rec
Fo
ext
po
thi
col
ini
of |
(ii)
hax
bui
obj
det
fro
cor
fro
par
opc
inte
4.6
Thi
obj
mo
exp
anı
the
pro
will
pix
cor
effe
rem
exte
rem
Ho
dist
be 1
dev
Wh
and
poir
defi
Wh
Scal
calc
fron
and
In t
regi
Sinc
the -
