line direction is the X axis, the azimuth direction is the Y
axis, the optical axis direction is the Z axis. The error
decreased when the B/D ratio increased, so we chose
the images photographed at 0.46 B/D. In order to
construct an optimum control set, one control point, who
showed the largest RMS error was sequentially excluded
from the control points. At last, optimum control points
were obtained except for 2 points.
This optimum control point set provides a little smaller
RMS error with the unknown parameters of DLT
increased. However, the RMS errors of the unknown
points were minimum when the unknown parameters
were 12. Therefore, 12 parameters were adopted to
determine the tracking of targets attached to the moving
car. As shown in Table 2, RMS errors in z direction are
always larger than for the other two directions.
Especially, target 2 provides smaller RMS errors than
control point set in all directions, and its magnitudes in x
and y directions are within 1mm. Generally, the accuracy
between rows(y axis) is much better than that of columns
in the video images. RMS errors in x direction of targets 1
and 2 are about 1.5 - 3 times those in y direction. But
target 2 shows a reverse phenomenon.
At last, we got the trajectory of targets attached to the
car. In Fig. 8(A), we can see the vibration of the car in
positive x direction owing to the impacts of shutting the
drivers door , and the vibration in negative x direction
owing to shutting the passenger door. After these two
impacts compounded, it gradually becomes weakened.
The second half of diagram describes the trajectory when
the car starts to drive ahead. Target 3 was attached to a
little rounded surface, and its narrow intersection angle
between two cameras might cause some large deviation
of the trajectory compared to the other two targets. In y
direction, the settling of the car body when driver and
passenger get into the car, rebound, and final equilibrium
state are recorded very clearly. In z direction, it remains
still while the driver gets into the car and starts ignition.
When the car starts to drive ahead, the targets'
trajectories grow farther.
Root Mean Square Errors of DLT
Z 15
oz mEy
= % 10
X o 5 NEz
ui 0 BEp
0.18 0.37 0.46
Base/Distance Ratio
Fig. 7. RMS Errors of DLT for Different B/D
(Unknown Parameters: 12)
zi .. | Exmm) | Ey(mm) Ez(mm)| Ep(mm)
Control Points 1.4 1.0 4.9 5.2
Target 1 14.8 10.0 73.2 75.3
Target 2 0.4 0.9 4.7 2.0
Target 3 16.2 4.8 37.4 41.2
Table 2. RMS Errors of Object Coordinates
(Unknown Parameters of DLT:12)
6. CONCULSIONS
Two video cameras and VCR combined with frame
grabber were investigated and tested for vibration and
movement monitoring. The quality of digital images
photographed with the video camera, recorded on S-
VHS VCR using S-Video, and digitized with the frame
grabber is good enough to monitor these dynamic
phenomena. However, the TV broadcasting standards
prevent a better high resolution video image. It will be
improved with the advent of HD-TV. In order to secure a
blurless image of a moving object , it is necessary to
take into account the imaging geometry, shutter speed,
and illumination. Although the frame advance function of
the VCR was satisfactory for synchronization in this
experiment, more elaborate devices are indispensable to
monitor high frequency vibrations. Image processing
software for the PC(e.g. IDRISI) was a usable tool for
image preprocessing and target centering. For the real
time photogrammerty system, its macro command must
be improved to manipulate Windows without manual
assistance. DLT was an appropriate model for the non-
metric imagery except its accuracy is lower than for the
self-calibration model. In conclusion, a video system
and appropriate photogrammetric principles provided the
expected results for monitoring of a moving car.
ACKNOWLEDGEMENTS
The authors wish to express thanks to John Webster,
Director, Audio-Visual Services, UNB, for giving
permission to use his video system. The support of the
Korean Science and Engineering Foundation and of the
Natural Science and Engineering Research Council of
Canada is gratefully acknowledged.
REFERENCES
Abdel-Aziz, Y.I. and Karara, HM. (1974).
Photogrammetric Potentials of Non-metric Cameras. Civil
Engineering Studies, Photogrammetry Series No. 36,
University of Illinois, Urbana.
Baltsavias, E.P. & Stallmann, D. (1991). Trinocular Vision
for Automatic and Robust Three-Dimensional
Determination of the Trajectories of Moving Objects.
Photogrammetric Engineering & Remote Sensing, Vol.
57, No. 8, pp. 1079-1086.
Clarke, T.A. (1995). An Analysis of the Prospects for
Digital Close Range Photogrammetry. ISPRS Journal of
Photogrammetry and Remote Sensing, 50(3):4-7.
Curry , S., Baumrind, S. & Anderson, J.M. (1986),
Calibration of an Array Camera. Photogrammetric
Engineering & Remote Sensing, Vol.52, No. 5, pp.627-
636.
Dunbar, P. (1986). Machine Vision. Byte. January 1986,
pp.161-173.
158
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
Faig, W. (197
with Non-metri
Systems, Am
Church, Virgin
Laurin, D. G. (
Space Struc
Photogramme
Lusch, D. P.
CIR Videogre
ASPRS, Terre
Moniwa, H (1
with Self-Ca
Dissertation, |
NB, Canada.
(A)
ge Roo m di a RRA
(COL
