Two separate CCD camera data acquisition systems have been built
in the section (figure VI-A,B). The first is a stereo monochromatic image
acquisition system in which the cameras are precisely synchronized to
individual pixels. Synchronization is essential for metric processing of
stereo images of dynamic processes. Currently, this system has a single
shot frame grab mode and a slow serial link to a remote computer. It will
be useful for acquiring small format test data.
The second CCD camera unit is more complex and was designed for
field data acquisition of Moire scoliosis screening data. It is capable of
storing about 5 frames per second onto a 10 MB disk cartridge unit. The
cartridges are removable, providing essentially unlimited storage of
images in the field. The system operates like a digital polaroid camera.
The image is previewed on a display then, when the desired frame is
selected, it is recorded on disk. After being recorded, the image can be
reviewed at once without off-line processing. This system is quite robust
and portable even in its prototype form. It has realtime custom image
processing hardware to perform image enhancement and symmetry correlation
analysis. An estimate of symmetry is used to obtain an automatic trigger
for recording the image, thereby performing a rudimentary automated data
acquisition control function.
3.0 SYSTEM FOR REALTIME PHOTOGRAMMETRY : APPLICATION SOFTWARE
The system for realtime photogrammetry is used for two purposes:
the automatic measurement of three-dimensional coordinates of object
points for industrial applications, and the evaluation of the performance
of CCD cameras.
For the automatic measurement of targetted object points, a
program which combines image processing, pattern recognition, and
photogrammetric algorithms has been developed. The program, named OBJECT,
is described in the block diagram [Figure VII]. There are eight processing
steps, the first six of which are performed mostly by dedicated hardware
and in real time. The system requires 10 to 20 milliseconds to perform
these six steps on each target. The results are two sets of image
coordinates, one from each camera, ready for matching and photogrammetric
processing. The last two steps require 50 to 60 millisecond execution time
for each object point. Some of the program operations are described below.
For noise reduction, a sequence of images is taken, instead of
only one, and averaged. Also the bias frame (or the dark-exposure frame)
is subtracted from the scene image frame to remove built-in noise.
The enhancement process is needed to emphasize important features,
such as targets or edges, and suppress the others. Operations required for
this purpose are a variety of linear and nonlinear filters and point-by-
point mapping operations. The system has about fifty library filters and
mapping functions to choose from.
In the image segmentation process, the enhanced image is converted
into a binary image for easier feature extraction. The threshold required
for this operation may be different in different parts of the image,
depending on the light conditions. Therefore, the image is divided into
windows and a threshold is selected for each window separately. This
operation requires histogramming each window and is carried out entirely
by dedicated hardware.
The result of the segmentation process is an image composed of
158
mA a uu res LE