International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
Digital video cameras are used to augment video image
streams. A camera based ARS is easier to implement and
experiment with than the combination of human and HMD
because the calibration of head mounted displays is more
complicated and time consuming than camera calibration
(Leebmann, 2003). The results of calibration and track-
ing are better controllable by measurements than the visual
impression of a human wearing a HMD which cannot be
measured objectively.
The orientation of head or camera movement is tracked
with a low-cost, straped down inertial measurement unit
(IMU) from Xsens, model MT-9B, that provides static ori-
entation accuracy « 1? with a resolution of 0.05° at 512Hz
output rate. Typically such sensors make use of additional
magnetic field sensors together with inclinometers to pro-
vide absolute orientation values without the need for ini-
tialisation at startup but the magnetic sensors make the
IMUs vulnerable to magnetic distortions. The small size
and low weight makes the sensor suitable for head mounted
operation.
For positioning, a differential GPS solution with high 2-
frequency receivers is deployed. In real time kinematic
(RTK) mode the system provides an accuracy of lcm for
the location at an update rate of 1Hz.
The development of new, wearable augmented reality sys-
tems is beyond the scope of our projects that use the tech-
nology. All hardware components are off-the-shelf, con-
nected over USB or serial lines to a notebook. All com-
ponents are mounted on a backpack with a fixed frame to-
gether with the notebook (fig. 2). IMU(s) are mounted
on the display solutions currently deployed, HMD or cam-
era. The system can be controlled with a wrist worn key-
board. The notebook's display can be exported over blue-
tooth based network to an handheld computer for debug-
ging tasks, but during development and testing it is not
fixed to the backpack for more comfortable work.
3 PROBLEMS AND SOLUTIONS
The real time requirement causes the biggest error for aug-
mented reality systems (Holloway, 1995). The system de-
lay (latency) can cause a misalignment of real and virtual
objects. The impression of augmentation is lost, virtual
and real part of the vision are showing different scenes.
Moderate head movements of 50?/s can cause a misalign-
ment of Imm every ms in an indoor environment (Hol-
loway, 1995). Head movements can reach a rate of up to
300°/s, about 1° with 4ms. The rotation of 1° causes an
alignment error of about 11 Pixel in a calibrated video im-
age with a camera resolution of 640x480 pixel (You et al.,
1999). Generally, an augmented reality system has to be
able to update display or video image information at least
with 15Hz for the impression of movement in human vi-
sion.
The low (1Hz-10Hz) update rate of the GPS position causes
problems, too. A pedestrian, e.g. is able to move on 1,7m
within one second at a given speed of 6km/h, making the
1050
—_
Head
Xsens MT-9B
HMD Nomad
Videocamera
M.
(P6£L 333l) 9JweitJ
| VGA T
Notebook
(Dell Inspiron
8200)
Backpack
PDA | L3 WristPC
Figure 2: Backpack AR Setup
high accuracy of 1cm=1ppm of our receiver useless during
the user’s movement. Another serious problem is the fact
that GPS isn’t reliable in urban environments due to shad-
owing effects of tall buildings that cause outages due to
satellite signal loss. One solution to this problem is a com-
bination of the complementary features of INS and GPS:
The double integration of acceleration values together with
the orientation will theoretically give the covered route.
But the error will increase exponentially over time due
to the incremental integration (drift). This effect can be
compensated by GPS measurements. GPS/INS combina-
tion as a flight navigation solution has been introduced in
late 1980s and is offered for sale since 1996 (Scherzinger,
2001).
In the case of pedestrian navigation the situation is differ-
ent. The human movement causes accelerations in all di-
rections. The vertical and longitudinal accelerations (see
fig. 3 for raw vertical and longitudinal accelerations with
an Xsens MT9-B fixed at the lower backside) can be used
for an analysis of a person’s step acceleration patterns in
order to extract the covered distance (Talkenberg, 1999).
DGPS can be used to calibrate the step pattern in situ.
Interi
Diff
ent s
slope
suit
dista
ing r
2002
pass
inaf
is ca
up a
path
com
using
flictis
A we
order
e.g. c
tion t
with
of lo
3.1
An ul
oron
alisti
With:
real c
build
In orc
ics w
be di:
proje
objec
clude
In au
virtua
these
video
geom
to gei
