-Reality.
, medical, and
ymbined with a
1 many of the
> virtual world
n the accurate
ility and cost,
ard geometric
s, or manual
lel graphically
rts and time
id , digitizing
s an excellent
It saves time
lel. Real-world
omplement or
ors. Currently,
ns an obstacle
tion to the
ications. Other
d trackers, the
y (either head
rate of all the
splay at 20 Hz
magnetic
[3mm/ 0.19]
acoustic
[3mm/ 0.19]
mechanical
[4mm/ 0.19]
optical
[varies]
system
Models suitable for virtual environments differ from those
designed for other graphical environments, such as CAD
systems (Phillips-Mahoney, 1995). The main constraints
are the real-time requirement and how fast the rendering
engine is capable of producing an image on the display.
These constraints will decide on the maximum number of
polygons in the model, thus the level of details that can be
handled. The model must also include information on object
behavior (response to user action), a script, the hierarchical
relationship between objects, and other attributes such as
texture and sound. The model must also include a multi-
resolution hierarchical representation of all its components.
For example, when objects are further than a given distance
from the user, it is sufficient to display them at a decreased
resolution up to a distance where the objects are not
displayed at all. Figure 2 shows the elements of a virtual-
world data base (model) and how this model relates to the
other processing components of the system.
max number of polygons for
graphic engine to get > 20Hz
Take inputs from:
Trackers, mouse,
keyboard, voice.
Include:
* Objects and environment
geometric data (at differen
resolution). - — .— Zl. cM.
* Hierarchy (parent, child,
sibling).
* Bounding volume(s).
* Script describing action t
objects(object behaviour).
* Lighting (shading.)
* Viewing points.
* Program controls / User
Interface.
* Hardware device support
(ports for devices, logical
connections ,..).
akes inputs,
objects, script,
imulates physica
laws. Outputs the
; world status.
Select resolution — akes world
Remove far objects —
Add texture, lighting —
(Real Time Display Update )
Figure 2: Software for VE: Models and Processes
1.2. Sensors for Creating the VE World
Since VE interacts with 3-D environments, relaying on video
images to generate a full non-structured environment is
usually not sufficient to capture all surfaces and features.
Sensors that directly produce 3-D images are better suited for
the creation of the geometric model. However, video images
are useful for adding texture (Haeberli and Segal, 1993.) In
addition, photogrammetric techniques can give accurate
coordinates for certain features and assist in the registration
of the 3-D images generated by laser range cameras.
The construction of most environment models will likely
require a large number of views, due to the limited field of
view of sensors as well as the complexity of the scene itself.
141
Each view taken with a range camera provides an ordered set
of 3D points in a camera-centered Cartesian coordinate
system. In order to reconstruct the model, the rigid
transformation linking the different views to a common
Cartesian coordinate system must be recovered. This process
is known as registration. The position and orientation of the
camera at each acquisition station could be accurately
measured with extrinsic means (such as inertial systems),
which are costly and may be impractical, or by intrinsic
means from the data itself.
1.3. Tracking in VE
Photogrammetry, when performed in real time, can be used
for tracking the position of the viewer's head and hands.
Currently, tracking is based on magnetometer sensors,
ultrasonic sensors, or mechanical devices. The first loses
accuracy in the presence of conductive metals and, to a lesser
extent, stainless steel, and the second is affected by
interference sound sources which make these technologies
not suitable for applications in some environments such as
factory floors. The mechanical devices are bulky and not
convenient for most applications. The current accuracy
levels of these sensors and devices are 3 - 4 mm in position
and 0.19 in orientation and get worse as the distance or noise
/ interference signal level increases.
1.4. Objectives and Scope of the Paper
Creating VE models from laser range cameras is the main
objective of this paper. Another objective is to explore the
potential of photogrammetry for this emerging technology.
The paper deals only with the visual part of VE. It does not
get into other senses such as hearing and sensation of force.
We first describe two laser range cameras developed at the
National Research Council of Canada which are useful for
creating 3-D environment models. Then, two important
issues are addressed; the calibration of the sensors in order to
produce accurate data, and the registration of multiple views
in order to arrive at one unique model of the environment and
objects. The paper then describes the use of photogrammetry
for head tracking followed by a description of the VE facility
at our laboratory and finally some concluding remarks.
2. DESCRIPTION OF THE RANGE CAMERAS
Range cameras based on structured light can generate
complete image data of visible surfaces that are rather
featureless to the human eye or a video camera. Among the
advantages of using a laser source one finds larger depth of
fields, compared with what is achievable with incoherent
light, and better ambient light rejection for accurate
measurement.
2.1. The Autosynchronized Laser Scanner
In the basic geometrical principle of optical triangulation,
the light beam generated by the laser is deflected by a mirror
and scanned on the object. A camera, composed of a lens and
a position sensitive photodetector, measures the location of
the image of the illuminated point on the object. The X, Z
coordinates of the illuminated point on the object are
calculated by simple trigonometry . The error in the estimate
of Z is inversely proportional to both the separation
between the laser and the position detector and the effective
focal length of the lens, but directly proportional to the
square of the distance. Unfortunately, the separation cannot
be made as large as desired. It is limited mainly by the
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
