You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Close-range imaging, long-range vision

S. Ozawa', I. Miyagawa*, K. Wakabayashi’, T. Arikawa”
“NTT Cyber Space Laboratory, NTT Corporation, 1-1 Hikarinooka Yokosuka-Shi Kanagawa, 239-0847, Japan
Commission V, WG V/2
KEY WORDS: Mapping, Vision, Urban, Building, GPS, Virtual Reality, Mobile, Image
We have developed a mobile mapping system for the purpose of acquiring building models having three-dimensional information
such as building shape and position in urban space. The system mounts two omnidirectional cameras and position-sensing equipment
on a vehicle and obtains images and positional information in a synchronous manner. It features three major functions. The first is a
GPS interpolation function that enables positioning even at places in an urban environment where signals cannot be received from
GPS satellites. The second is a synchronization function that synchronizes positional information and omnidirectional images and
enables reconstructed models to be accurately converted to global coordinates. The third is a shutter control function that enables the
shooting intervals of the two omnidirectional cameras to be either synchronized or non-synchronized. These functions make it
possible to select images for all sorts of baselines from multiples images captured by the two omnidirectional cameras and to make
stereo measurements for an optimal baseline according to the height of the target building and its distance from the camera. This
system can perform high-speed data acquisition at a rate of four square kilometers per hour.
Three-dimensional (3D) digital maps that contain building-
related information such as position and shape in urban space
are eagerly awaited in a wide range of fields (Teller, 1999;
Virtual Helsinki, 1999).
In particular, 3D digital maps can be used as an important
information platform in fields like Intelligent Transportation
Systems (ITS) and Geographic Information Systems (GIS), and
they can be applied to the simulation of radio-wave propagation,
disaster countermeasures, etc. They are also expected to find
use in the entertainment field such as in video-games.
In short, there is an extremely broad range of needs for 3D
digital maps. In this regard, not only will the demand for
accurate shape measurement and high-quality texture intensify,
but the acquisition of such information will also have to be
automated and the cost of acquisition lowered.
Against this background, various techniques have been
propesed for acquiring urban models by the airborne approach
using aerial photographs, laser rangefinders, and the like. The
airborne approach has a major advantage in that a wide area can
be efficiently reconstructed. For this reason, even our research
group has been active in this area, researching a technique for
reconstructing building shape from aerial images and a
technique for reconstructing building shape from laser range
data (Miyagawa, 2000; Horiguchi, 1999).
The airborne approach, however, lacks information on the
sides of buildings and cannot support the use of building models
as seen from the ground such as for “walkthroughs.” On the
other hand, there has been much activity in the research of
“mobile mapping” that mounts cameras and range sensors on a
vehicle to reconstruct building shape (Ellum, 2002; GIS
Development, 2000; Tamura, 1998). Not only can mobile
mapping reconstruct the sides of buildings, it can also acquire
data in a relatively short time by utilizing key properties of
With the above in mind, we set out to develop a mobile
mapping system for the purpose of automatically reconstructing
urban building models by installing cameras on a vehicle and
making mobile measurements.
The high-accuracy acquisition of positional information is
essential to mobile mapping, and to this end, the Global
Positioning System (GPS) is often used. In urban areas,
however, the quality of position data may drop due to shielding
by buildings, to multi-path effects, etc, and many mobile
mapping systems have come to incorporate a mechanism for
interpolating GPS position data. For our mobile mapping
system, we have developed and incorporated a function that
interpolates GPS position data in real time while monitoring the
quality of GPS positioning.
Various techniques have been proposed for reconstructing
shape including a sensor-based technique using range sensors
(Zhao, 2000) and an image-based technique using captured
images (Uehara, 2000). The sensor-based technique can directly
acquire the shape of a building but cannot obtain building
texture. The image-based technique, on the other hand, has the
advantage of being able to acquire shape and texture
simultaneously. A conventional camera, however, has a limited
angle of view, which means that the range of measurement will
likewise be limited. Such a camera is therefore not conducive to
an urban area that wills more than likely feature buildings of
various heights. For this reason, we have adopted in our system
an omnidirectional camera that can capture a 360 degree image
at one time.
An omnidirectional camera may be of the mirror-projection
type, which captures an image projected on a mirror with a
camera, or of the lens type that uses a fish-eye lens. (Kawasaki,
2001). The mirror-projection type enables the mirror shape to be
on {