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 project is to evaluating the extraction of various object
classes from digital imagery platforms such as digital aerial
images and Laser Scanning data. The focus of the project was
the registration of the aerial image on the Laser Scanning data
for developing and reconstructing a 3D model. Therefore a set
of digital aerial images and Laser Scanner data have been
provided for this project. To accomplish the project's goals, the
author proposed to implement a modified approach which was
originally explained by Homainejad (2011a). The modified
approach embraces following steps.
l. Interested objects will be extracted from the point
clouds. In this step an operator which has been
developed for this purpose, will apply on the point
clouds data.
2. The extracted objects from the point clouds data will
be converted to a raster format and will be registered
on the image for assisting the process of object
detection and extraction from the image.
3. The extracted objects from the image will be
transformed and registered on a 3D model or a DTM
which were developed from the point clouds for
reconstructing a new 3D model.
The modification has been carried out on the process of the
object extraction from the point clouds in order to improve the
final result and speed up the process. In this proposal the
process will not started from splitting the image to small areas.
Instead, interested objects will be extracted from the point
clouds. Then the extracted object will be transformed to the
image for assisting segmentation and object extraction from the
image. Basically in this study, two first steps of object
extraction from the point clouds and transforming the extracted
objects to the image space for improving the segmentation are
called reverse registration. Figure 1 shows the process of
proposal for this project. The object extraction can be
implemented on the DTM, DSM, 3D model, or a point clouds.
Since a point clouds data has been provided for this project, the
focus of this study is to extract the objects from the point clouds
data and consequently the mathematical model was developed
in order to extract objects from the point clouds data. Most of
studies in object extraction from Laser Scanning data have been
focused on the signal processing or mathematical modelling
which has been developed based on parameters of orientation of
Laser Scanner beam with the surface of the object. For example,
Silvan-Cardenas and Wang (2006) implemented Multiscale
Hermit Transform (MHT) due to decompose signal for profiling
the surface of terrain, or Kirchhof et al. (2008) and Bae et al.
(2009) independently developed a mathematical model for
object extraction from the point clouds based on parameters of
orientation of Laser Scanner beam with the terrain. Since the
provided point clouds from the terrain for this project did not
include any pre-knowledge regarding to parameters of Laser
Scanner orientation and received waveform, a few operators
were developed and implemented for detecting and extracting
interest objects from the point clouds or DTM which was
developed from the point clouds, but only two of them will be
discussed here.
Figure 1. This figure shows the process of the proposal for this
research study. Figure (1a) shows a point clouds from an urban
area. Figure (1b) shows the extracted of four building in this
area form the point clouds, and Figure (1c) shows the bare point
clouds after subtracting all extracted object from the point
clouds. Figure (1d) shows extracted buildings from the image
after transforming and registering extracted building from the
point clouds on the image. Figure (1e) shows an isometric view
from reconstruction of the 3D model after transforming and
registering extracted object from the image on the 3D model
space or DTM developed from the point clouds.
The first operator has been developed based on curvature (K)
and signed curvature (K) of the terrain at point p. The curvature
of a surface at point p is:
k(s) - ||T'(s)]| (Eq. 1)
Where T'(s) = K(s)N(s) is derivative of unit tangent vector,
k(s) is curvature of surface at point p, N(s) is the unit normal
vector. The signed curvature K(s) indicates the direction in
which the unit tangent vector rotate, as a function of parameter
along the curve. Negative singe indicates the rotation is
clockwise, and positive singe indicates rotation is
counterclockwise. With extension of above equation in a 3D
Cartesian space, the curvature will be:
k 2 SEY e FxY x xyy (Eq. 2)
(x^ ey" ez 3/2
180
se
an
alg
CO!
po
ch:
exi
Th
po
wh
the
tan
anc
cha
anc
2.1
Fir:
fra
clo!
rec
exti
20€
hav
thai
froi
at
ope
equ
be
bui
buil
det:
tran
con
on
orie
the
geo
usir
buil
mo
indi
22
For
foc
Las
way
tree
hit |
