International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
discussed in section 5 and conclusions and discussions of
future works are outlined in section 6.
2. MULTI-SENSOR DATA FUSION
Multi-sensor data fusion refers to, in our context, the
techniques for the combination of datasets from the 3D point
cloud data and the 2D image (i.e. intrinsic parameter of the
CCD camera) to create a new dataset. The input to this process
is 3D data from 3D laser scanner, the 2D intensity image from
independent CCD camera, and the intrinsic camera parameters.
These sensors are not calibrated. Detailed description of the
coordinate systems of the 2D and the 3D sensors, data capture
and processing is available in Forkuo and King (2004).
2.14 3D Point Cloud
Cyrax 2500 Laser Scanner was used to carry out the laser
scanning to acquire a discrete representation of the object.
More description of the laser scanner used in this experiment
can be found in (Forkuo and King, 2004; CYRA, 2004). Figure
| shows a screen capture of pseudo-colored 3D point cloud
data of the test area. The 3D point cloud allows for the
construction of a 3D surface Model of the scene. The resolution
of the scan, which controls the number of points recorded in a
scene and the level of detail visible in a scan (Barber and
Bryan, 2001), is simply the smallest change in distance that the
scanner is capable of detecting.
T
3
Figure | A screen capture of Pseudo-colored 3D point cloud
2.1.2 2D Intensity Images
A series of images were taken at different direction and
position (as depicted in figure 2.2) by the digital CCD camera
(Nikon DI1x), which produces an image of at the size of
23.7mm x 15.6mm width. These images are called real camera
images (RCI) and one of these images is represented in figure
2. This camera provides a digital image with the resolution of
3008 by 1960 pixels at true color mode.
2.2 Backprojection of Laser Scanning Data
Multi-sensor mathematical model is a physical model that
describes the integration of 3D laser range camera and the
CCD camera (Forkuo and King, 2004) We use the
photogrammetric principles of collinearity condition with no
systematic correction parameters as the basis for the
implementation of the transformation of 3D point cloud data to
suitable 2D shape information. For details on the collinearity
model and the subsequent steps in fusing the dataset, see
Forkuo and King (2004).
Figure 2 Real Camera Image (RCI)
2.3 The Synthetic Camera Image
This task is to represent the results of the collinearity equation
(discrete x, y) points as image, which could be used in the
image-to-image registration task discussed in section 3. A more
detailed description can be found in Forkuo and King (2004). .
By way of definition, interpolating the backprojected laser
point (which contains irregular point spacing) into a regular
grid at an even spacing using the intensity values generates
what is termed “the Synthetic Camera Image" (SCI). There are
two options related to this interpolation. First option is to
generate the SCI by keeping the original resolution of the point
cloud data and the compute a new pixel size. The second
option, on the other hand, is to keep the pixel size of the real
camera image and then compute the number of pixels or the
resolution. In this paper, this option is used to generate the SCL
The interpolated data was then modeled by f (x, y) = I, where
(x, y) is the pixel position and the / , the corresponding
intensity value which is mapped to a grayscale. Conventional
image processing techniques such contrast stretching and
image enhancement were then used to produce the final image
in figure 3. It is obvious that the geometric features in the SCI
are easier to detect than those in the laser range data. This
image offers a major advantage to interactively (controlled by
human operator) or automatically conjugate matching with the
intensity images produced by digital camera.
Inte.
———
in 1
com
diff
diffi
desc
sam
diff:
pap:
and
ima;
tech
autc
man
rela
soft
For]
nec
qual
synt
3.1
Onc
poir
Syn
(RC
regi
a px
the
not
prol
terr
ima;
ofn
are |
as ]
thes
thei
Furt
com
and
One
prol
(Diz
or ¢
gen
dete
esta
simi
corr
al:
cons
corr
3.1.
The
(edg
For
Step
ima
shox