best perfor-
hing for flat
correlation"
, 1989] to a
llent perfor-
nent [Day &
[lite images.
gorithm has
1d was used
rorithm finds
] a template
mage allow-
ses the sum
een the two
plate match-
oints placed
ble template
possibilities
ifferent sym-
image itself.
«ture charac-
target image
ure variation
plates.
least squares
passive tar-
oorer results
ed from the
/se produced
template ex-
esults.
ses an itera-
to converge
d size of the
1g. Particu-
e size of the
to provide a
ant measure-
/ eigen value
mplate for a
s were made
he optimum
of iterations
fferent tem-
respectively.
template of
rovides opti-
| of repeata-
se targets in
iracy of this
Figure 9: Laser projector based and testfield based targets.
repeatability measurement of testfield targets was affected
by change in their size due to large depth of the testfield,
while that of the laser dots was affected by speckle [Clarke &
Katsimbris, 1994].
3 PROJECTOR MODEL
Active triangulation based systems have been used a num-
ber of times to solve the correspondence problem for auto-
matic measurements of textureless objects. Light projection
based active triangulation generally involves direct calibra-
tion [Trucco et al., 1994] of the whole system (black box
type). Direct calibration consists of measuring the image co-
ordinates of a grid of known three dimensional points, then
look-up tables are built for the whole image through interpo-
lation. In this calibration process there is no need to model
any phenomena and this suits laboratory based machine vi-
sion applications. However, it is difficult to maintain the
rigidity of the projector with respect to the camera during
Table 1: Deviation of observations of same points in different
frames.
No. Min. Mean Max. | r.m.s. | Std dev
Pixels | Pixels | Pixels | Pixels Pixels
X* 361 0.00 0.027 0.309 0.046 0.042
Y* 361 0.00 0.023 0.251 0.042 0.038
Kon 22 0.00 0.029 0.191 0.039 0.035
y** 22 0.00 00210; |:.0-112 0.031 0.024
* — Active targets (laser dots) & ** — Targets of the testfield
Patch radius 2pixels
— —- Patch radius 4pixels
e Patch radius 6pixels
c — —— Patch radius 8pixels
— - — Patch radius lOpixels
Probability distribution
e
+
0.2
= S hart]
0.0 2.0 4.0 6.0 8.0
Number of iterations
Figure 10: Probability of iterations with patch dimension.
field applications of a direct calibrated system. Such systems
are not flexible enough to accommodate the varying depth
and size of different close range objects. There is no stan-
dard method to calibrate such a system and generally such
calibration requires a three dimensional testfield. Develop-
ment of a standard photogrammetric camera model of the
projector will probably be the best possible solution for these
problems, and this has been attempted here.
3.1 Calibration
Rigid placement of the projector over the telescope of a
geodimeter provided a good opportunity to measure the ori-
entation of different dots of the projector. The use of a fixed
spherical autoreflecting target helped in the precise and easy
vertical placement of the laser dots over the target. Large
numbers of geodimeter observations generated during calibra-
tion of interdot angles of the projector were communicated
to the computer using a Psion organiser as a communication
link. The interdot angles among the dot matrix of the pro-
jector were found to be quite stable over time and space. A
simple statistical (Table 2) analysis of the deviation of two
such observations over a five week period gave a reapitability
of + 0.01 gon.
3.2 Interior orientation parameters
Precise knowledge of focal length, position of principal point
and lens distortion parameters are necessary for accurate three
dimensional measurements of close range objects. In a system
543
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
