In the industrial environment, the production
frequencies are sometimes deviating from the
standard video rate in such a way that solutions with
other image acquisition rates must be employed.
Certain high resolution, non-standard CCD-cameras
also have slower image generating cycle due to the
read-out time of image data.
For the registration of very fast processes, with more
than 500 frames/second, analogue high speed film
cameras are still used to a large extent, even if CCD
cameras with fairly good resolution are becoming
available also for these purposes (fig 4).
O ms
@ MacReflex
e LI MapVision
O = e g Bl TrackEye
1 1 1 1 Fo
7 50 100 500 [Hz]
Multiplex
mode
fig4 Image Acquisition Speed
4.3 Detection and Extraction of Data
The extraction, or measurements, of data can be seen
as an information compression and information
extraction of the parts in the image which are of
interest. This is primarily done by low-level image
processing techniques. Two main types of
information can be defined (fig 5):
- Area based information
- Point based information
- Grey-level correlation techniques
- Thresholding/slicing techniques
Examples of the first category are histogram or
textures of defined regions. None of the illustrating
systems use this type of information.
The point based information are mostly derived from
an area in the image as well, but the purpose is to
compute point coordinates. If the target points have
different reflectance or emittance properties than the
background image, a thresholding may be done to
extract the target areas. This is a simple and fast
technique. The measuring point can be defined by
Reflective markers
Projected laser spot
Light Emitting Diodes, LED’s
Both the systems developed by non-photo-
grammetrists use this type of targets as their only data
source, #1O and n@ System #3 [] uses it
primarily for the system calibration.
The detection and extraction of the target points are
fast with the thresholding technique, but it does not
enable the system to measure on natural object
points, e.g. corners. To be able to do so, grey level
correlation techniques must be used. This is a time
consuming task, but can be speeded up if the location
of the searched pattern is approximately known, as is
the case e.g. when tracking points in a motion
sequence. This method is used by #3 [lana s4 B
oh
2
& O MNS
@ MacReflex
er e [] MapVision
e Bi TrackEye
er O
NI D
wi u
| l
I I
Binary image/ Grey-level
Thresholding operations
fig5 Data Extraction
The precision of the extracted image coordinates are
of course of vital interest for the final result. All of
the illustrating systems claim a high precision in the
measurements of the image coordinates, which
means a 1/20:th - 1/50:th of a pixel.
If more complex image operations are to be done in
real or near real-time, the implementations must be
done in hardware. If the full frame must be processed
even todays hard-ware implementations might not
be enough. The amount of data in a stereo CCD
System requires app. 12.5 Mb/sec (Grün, 1991).
Comments From an error theoretical point of view
the data extraction methods used is mostly un-
satisfactory. Very few, if any, of the systems can
produce error estimates for the image coordinates
which can be used in the further processing of error
estimation.
The extreme difference in speed for system #2@ is
due to the fact that each camera has a dedicated hard-
ware unit capable of measuring 20 pts/50 Hz image.
There is no further analysis of data in real-time as for
the other systems.
The ability of measuring on natural targets requires
grey-level based methods. This reduces the speed of
the point measurements, but if more complex
operations are to be developed in the future they
must anyway be done in the grey-level image. This
would indicate that grey-level based methods are
more in principle more general.
4.4 Analysis of data
Depending on the task, the analysis of data may range
from the computation of single 3D coordinates or
histogram analysis to the advanced reconstruction of
