'ET LOCATION
equential images in real-
processing. However, in
play of dynamic target
red. À suitable on-line
s requirement.
r automatic target image
target recognition and
ymputational time in this
and target recognition
rithm based on a prior
bsequent images in the
y the time necessary to
targets and to compute
cessive images can be
tional cost of the target
Pentium-90 PC running
an be seen that a lot of
tions. This is because in
1 target image pixels is
5%. For example, in a
ly 5603 pixels represent
ls in total for the 768 x
at 600ms are required to
s for a 400 target image,
background scanning in
procedures. Only 44ms
n of all 400 target co-
ude the matching and
uccessive images.
300 | 400 | 500
500 | 600 | 660
267 | 356 | 404
33 44 56
target location
omputed from centrifuge
ina 1996
In a centrifuge experiment, the movement of targets between
any two successive images is small such that all target co-
ordinate calculations can be reliably based on the target co-
ordinate information of the previous image. In this way it is
only necessary to access and process image data from a small
area surrounding each target. The prior knowledge based target
location algorithm calculates target positions from the first
image using the general algorithm and then uses these as search
centres for targets in the next image in the sequence. The
algorithm is implemented using a stack technique. All the target
co-ordinates in the first image are computed and then pushed
onto a stack. With the centrifuge test underway, the x, y co-
ordinates of each target are sequentially popped from the stack
to provide an initial value for the target centre in each
successive image. In this way a considerable number of “empty
operations” and target recognition procedures are saved.
Target matches between any two successive images are not
necessary since each target will have the same label as its seed
point from the previous image. Deformation vectors between
any pair of images may then be drawn on the computer monitor
as required. Furthermore, appropriate mathematical models can
be fitted to the data during the course of the experiment to
obtain geotechnical parameters. Figure 14 illustrates a typical
result from a dynamic view of a geotechnical experiment. At the
time of writing, all of these processes can be completed under
one second on a P90 PC. The information provided offers the
potential to supply feedback to monitor and ultimately control
the progress of the geotechnical experiment.
5. CONCLUSIONS
The S-VHS Video recorder test results, presented in figures 3 to
6 lead to two conclusions. Firstly that retro targets can achieve
much better results than conventional targets in noisy situations;
secondly that the S-VHS Video recorder is a particularly
unsuitable storage device for small conventional targets of low
contrast.
The JPEG image compression method has been developed for
use with continuous tone photographic colour images and is
capable of achieving very high compression ratios. It is
optimised according to human visualisation requirements.
However, it can offer very promising compression ratios in
retro-targeted photogrammetric situations. The amount of JPEG
compression which can be tolerated in images for
photogrammetric measurement must be decided according to
the desired target co-ordinate precision, on the quality of the
imaged targets and on the performance and design of the
photogrammetric imaging system. On the basis of the above
experiments, it can be concluded that targeted images can be
‚compressed using JPEG at a ratio of about 50:1 if 1/10th target
location precision is sufficient and 10:1 if target location
precision of the order 1/20th pixel are required. Furthermore the
JPEG software has been easily combined with the TIFF format
into a general purpose tool using third party software libraries.
The prior knowledge based target location algorithm has proven
suitable for deformation analysis of sequential targeted images
where target movement in any successive image is reasonably
small. The computational cost of target co-ordinate
measurement is much less than general target location
algorithm, typically ranging from 10 to 60 ms depending on the
number and size of the targets in use. This means that target
image measurement in nearly real-time is possible without the
assistance of any other hardware. More research is required to
extend the method to include situations where ambiguities and
target occlusions are present.
A feasibility test for the use of MPEG in the centrifuge imaging
environment is required, but it is anticipated that where it is
necessary to store centrifuge image sequences, a hardware based
method will be used to compress and store images in real time
at a compression ratio of between 10 and 20 times with less
than 1/10th of a pixel measurement error. Direct co-ordinate
extraction using the prior knowledge based target location
algorithm will be used where higher precision is required and
image sequences are not needed for subsequent visual analysis.
ACKNOWLEDGEMENTS
This research is funded by EPSRC Grant No. GR/J74022.
Discussions concerning target location with Dr. T.A.Clarke and
the photogrammetric data supplied by Prof. M.R. Shortis are
gratefully acknowledged.
REFERENCES
Beyer, H.A., 1992, Geometric and Radiometric Analysis of a
CCD - Camera Based Photogrammetric Close-Range
Systems, Dissertation, ETH Nr.9701, 185 pages.
Chen, J., Clarke, T.A., 1992, The Automatic Recognition,
Location and Labelling of Targets in Digital Photogrammetric
Engineering Measurement, Int. Archives of Photogrammetric
and Remote Sensing, Vol.29(5), pp.686-693.
Chen, J., 1995, The use of Multiple Cameras and Geometric
Constraints for 3-D Measurement, PhD Thesis, City
University, UK, 245 pages.
Clarke, T.A., Robson, S., Qu, D.N., Wang, X., Cooper, M.A.R.,
Taylor, R.N., 1995, The Sequential Tracking of Targets in a
Remote Experimental Environment, ISPRS Intercommision
Workshop: From Pixels to Sequences - Sensors, Algorithms
and Systems, Vol.30, 5/W1, pp.80-85.
Hoflinger, W., Beyer, H.A., 1993, Evaluation of the Geometric
Performance of a Standard S-VHS Camcorder, SPIE Optical
Tools for Manufacturing and Advanced Automation,
Videometrics-II, Vol.2067, pp.104-114.
Lammi, J., Sarjakoski, T., 1995, Image Compression by the
JPEG Algorithm, Photogrammetric Engineering & Remote
Sensing, Vol. 61, No. 10, pp. 1261-1266.
LeGall, D., 1991, MPEG: A Video Compression Standard for
Multimedia Applications, Communications of the ACM, Vol.
34, No. 4, pp. 47-58.
Shortis, M.R., Snow, W.L., Goad, W.K., 1993, The Influence of
Storage Media on the Accuracy and Repeatability of
Photogrammetric Measurement Using CCD Cameras, SPIE
Optical Tools for Manufacturing and Advanced Automation,
Videometrics-II, Vol.2067, pp.80-92.
Shortis, M.R., Robson, S., Short, T., 1996, Multiple Focus
Calibration of A Still Video Camera, ISPRS XVIII Congress,
Commission V, in press.
Robson, S., Clarke, T.A., and Chen, J., 1993, The Suitability of
the Pulnix TM6CN CCD Camera for Photogrammetric
Measurement, SPIE Optical Tools for Manufacturing and
Advanced Automation Videometrics II, SPIE Vol.2067,
pp.66-77.
Taylor, R.N., 1995, Geotechnical Centrifuge Technology,
Blackie Academic and Professional Publishers, Glasgow, UK.
296 pages.
Wallace, G.K., 1991, The JPEG Still Image Compression
Standard, Communications of the ACM, Vol. 34 No. 4,
pp.30-44.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
