Page 431 - Sensors and Control Systems in Manufacturing

(a) Sensors in Flexible Manufacturing Systems 385

(b)

FIGURE 8.2 (a) Image fl aw detection and analysis. (b) Captured image of protein
like strings measuring maximum length of fi bers in microns. (Curtsey American
SensoRx, inc.)
a matrix of arranged fibers. This string protein structure is used in
growing human tissue of a wounded patient.
A robot-vision system must exercise some judgment in perform-
ing visual tasks—those for which the input is a visual image (normally
obtained from an ordinary video camera). Which visual tasks are
relatively easy for machine vision, and which are hard? The distinc-
tion is not so much that some tasks are hard and others easy; rather, it
is the detail within a task that distinguishes easy problems from hard
ones.
What makes a problem hard? Some of the contributing factors are:

• Objects that vary widely in detail. (Examining stamped or
milled product may be easy, while molded or sculpted items
may be more difficult. Natural objects are by far the hardest
with which to deal.)
• Lighting variations, including reflections and shadows, as
well as fluctuations in brightness (as found in natural sun-
light). These variations may go unnoticed by human inspec-
tors, but they can make otherwise easy problems difficult or
impossible for robot vision.
• In general, ignoring “unimportant” variations in an image while
responding to “significant” ones is very hard. (Most hard
problems can be placed in this category.)