mation
ipment
rst is a
d from
es and
les the
nake it
0 make
a. This
rties of
mobile
tructing
icle and
ation is
Global
1 areas,
hielding
mobile
1ism for
mapping
ion that
ring the
structing
| Sensors
captured
| directly
building
, has the
texture
a limited
nent will
ducive to
ldings of
Ir system
ee image
rojection
r with a
(awasaki,
ape to be
designed as desired, but its drawback is that the camera itself
will also be captured in the picture. The lens type, in contrast,
can capture a completely omnidirectional image without
capturing the camera, but its drawback is that design is quite
difficult. In our system, we have adopted a fish-eye lens that can
support the capturing of tall buildings without capturing the
camera itself. It must also be pointed out, however, that spatial
resolution in an omnidirectional camera is low since a wide-area
image must be projected onto a finite pickup device at one time.
There are two approaches to solving this problem: a hardware
approach such as using a high-definition camera, and a software
approach such as processing multiple images to raise resolution.
We have adopted the former approach, using high-definition
cameras in our mobile mapping system.
2. MOBILE MAPPING SYSTEM
2.1 Configuration of the mobile mapping system
The configuration of this system is shown in Fig. 1 and an
external view of our mobile mapping system is shown in Fig. 2.
The system can be broadly divided into an “Imaging Section”
that captures and stores omnidirectional images and a
“Positioning Section” that collects positional information when
capturing images.
The Imaging Section consists of omnidirectional cameras, a
computer for recording captured images, and a shutter controller
for controlling each camera’s shutter. The cameras are placed
on the roof of the vehicle to prevent the vehicle from being
captured in the images.
The Positioning Section consists of a GPS antenna, a
Differential Global Positioning System (D-GPS) device, a
vehicle-speed sensor, a gyro sensor, a GPS interpolation device,
and a computer for recording positional information. The
Imaging Section and Positioning Section are synchronized by a
GPS-based time server.
This system can accurately acquire the data required for
reconstructing building models while moving.
Roof of vehicle
? Positioning section
GPS antenna i : Omnidirectional camera
^ / Imaging section N
Omnidirectional camera
i r i
i i
; ] ;
| D-GPS device | | GPS time server
8
Gyro sensor
) GPS interpolation device
i Position recording PC
Image recording PC
| Shutter controller
Vehicle interior
Figure 1. Mobile mapping system composition
Figure 2. External view of our mobile mapping system
2.2 Omnidirectional cameras adopted for the system
In an urban area, a mobile mapping system must be able to
capture tall buildings. This condition can be met by an
omnidirectional camera that can capture a 360° scene at one
time. As mentioned above, there are two major types of
omnidirectional cameras: a mirror-projection type and a lens
type. Because the camera is captured in the image and forms an
obstruction in the former type, we have adopted an
omnidirectional camera that employs a fish-eye lens for which
the camera will not be picked up. This type of camera, however,
while being able to capture an expansive 360° area, suffers from
low spatial resolution. As a countermeasure to this problem, we
decided to use a digital camera having a 6-million-pixel CCD.
Using a camera with high spatial resolution makes it possible to
reconstruct high-quality texture and high-accuracy building
models.
On the other hand, it takes about two seconds to capture one
image with the type of camera adopted in our system, making
for a low shooting rate. For this reason, the interval between the
shooting positions for each image is large, and variation occurs
in texture quality and model accuracy according to the distance
from the shooting position and target building. To avoid this
problem, we mount two omnidirectional cameras in our system
and incorporate a shutter controller so as to control the shutter
interval for each camera and facilitate flexible control of the
shooting interval. This, in turn, makes it possible to adjust the
baseline distance in omnidirectional stereo according to the
distance to the target building and to therefore eliminate the
variation in accuracy and texture quality among the buildings to
be reconstructed.
Table 1 lists main specifications of our omnidirectional camera
and Fig. 3 shows an actual example of an omnidirectional image
taken with our mobile mapping system. The omnidirectional
images captured by this system are of high quality and can
accommodate tall buildings.
On the basis of these features, the omnidirectional cameras
adopted for our system are considered to be an effective means
of acquiring data in an urban area.
—199—