Page 433 - Sensors and Control Systems in Manufacturing

FIGURE 8.3 Sensors in Flexible Manufacturing Systems 387
Recognition of a
learned object.

It is important to note that this task actually has two distinct parts:
first, object familiarization—that is, learning what an object looks
like, then object recognition (Fig. 8.3).
There are many ways of learning to recognize objects. Humans
can learn from verbal descriptions of the objects, or they can be shown
one or more typical items. A brief description of a pencil is enough to
help someone identify many unfamiliar items as pencil. Shown a few
pencils, humans can recognize different types of pencils, whether they
look exactly like the samples or not.
Robot-vision systems are not so powerful, but both these
approaches to training still apply to them. Robots can be given descrip-
tions of what they are to recognize, perhaps derived from CAD data to
guide a machine tool (Fig. 8.4) or they can be shown samples, then be
expected to recognize objects more or less like the samples.
Recognizing objects once they have been learned is the second,
and more difficult, part of the task. Several basic questions arise in
virtually every recognition task. Among them are “What are the
choices?” and “What can change?”
Specifying the actual task completely is normally the hardest part
of the application. When this has been accomplished, the basic ques-
tion is, for what particular vision features should the search begin to