st satisfy
(5)
se vector
tem:
(6)
1age
for every
(7)
| (7) are
tment to
irameters
the two
rspective
(8)
e three-
left and
ector (1,
, AK) of
ordinate
ameters.
( relative
9 control
ion. For
the GPSVan, this method is used to check or
update the relative orientation parameter when
the stereo vision system is reinstalled.
2.3. Rotation Offset Determination
The rotation offset angles are defined as the
rotation difference between the stereo vision
system and the positioning system. In the
GPSVan, they are small angles ( « 5 degrees for
most cases). The only way to determine them is
by an analytical method. The following
constraints are used to compute rotation offsets:
* Same points measured in different image
pairs have the same X, Y, Z coordinates
Points with the same elevation measured
from a single image pair have the same Z
coordinate.
*
Our method is based on the following coordinate
transformation equation of the GPSVan:
X X X X
y zy ¥ t Y ={"Y +
Z global Z), Zi. ps
X
R nori] BR Lien sysrot y +| AY
Z
vision
(9)
This equation defines how a point can be
transferred from the stereo vision system into
the global coordinate system. We want to
determine the rotation offset between the stereo
vision system and the positioning system. These
offsets are small angles and can be expressed
by (da, d$, dK). The first derivative of rotation
angles is used in the rotation matrix:
1 —d& -dKk
R fier =| da 1 T (10)
-dk dé 1
Let :
X X X AX
Y, mE] SER nori] R sysrot y [AY
Zo Z gps zZ vision AZ
(11)
483
Xo X
Ya | Ro Y (12)
20 vision
then
X Xo X9 Xo
Y m|iYo [5 RoeiRonel Yor | = | Xp |
sort (20 20 Zo
—320 0 dé
0" °y, Vey Ride
| Canari
(13)
This is a linear equation with three unknowns and
it is used to form the constraints to derive the
rotation offsets.
A point measured from image pair i and j forms
three constraints:
x
Y = (14)
N ^X X
global global
Constraints (14) determine parameter da, dé.
For the same elevation points m and n, their Z
coordinates are equaled by :
global 7 Zzlobai (15)
Constraint (14) determines dó,dx.
Collecting all the equations in (14) and (15), the
rotation offsets are computed by a least squares
solution.
3. Accuracy Evaluation
3.1 Positioning Accuracy of the
Stereo Vision System
The algorithms described above were
implemented on our post-processing workstation.
To evaluate the positioning accuracy,
calibrations with different camera parameters
and constraints were conducted. All tests were
based on two Kodak DCS digital CCD cameras,
which were part of the vision System on the
GPSVan. They have a resolution of 1280(H) X