.pg 5.2 A practical example: CONSIGHT-1
The CONSIGHT-1 [6] system was developed by General Motors as a vision system
to permit some of their robots to cope with 2.5-D parts arriving at the
robot's work cell on a conveyor belt, at random intervals and in random
orientations. The essential ingredients are a linear array camera looking down
onto the conveyor belt, viewing a slice across the belt, and two light
projectors each projecting a plane of light. The two planes of light and the
viewing plane of the camera all intersect at the surface of the belt, in a
line across the belt. The side view of the arrangement therefore looks like
this: .sp 16 .br Normally, when there is nothing on the belt, the camera sees
the line of light caused by the two projectors. When an object approaches,
just before it reaches the 'viewing line', it obstructs one of the planes of
light but not both, so that the camera still sees a line of light straight
across the belt. When the object is crossing the 'viewing line', it prevents
both of the planes of light from reaching the surface of the belt, and so the
camera only sees part of a line of light across the belt. The dark portion of
the line is the width of the object there: 
.nf

     
     
     View from above

     
     
     
     
     
     
     
     Side view

     
     
     
     
     
     
     
     
     What the camera sees

     
     
     
.fi
This series of slices of the objects formed as the conveyor belt moves the 
object across the 'viewing line' is passed to a computer, which puts the 
information together with details of the belt's position and velocity provided
by transducers on the belt to form a run-length coded image of the object. The
computer can identify the object and calculate the centre of area and the
orientation. Knowing the belt's velocity, it can predict where the object will 
be at later moments in time, and can pass this information to a robot 
controller so that a robot further down the belt can eventually pick the 
object up. The orientation allows the robot to rotate its gripper to a 
suitable angle first. Many robots do not have a full 360 degrees of 
permissiblerotation in their wrist; it is thus possible that some objects may
be so oriented as to be ungraspable. Such objects would be ignored, and
continue down the conveyor belt to be recycled round the system in due course.
.ti 4
One possible snag concerns objects that have holes in them. A hole that is so 
deep and/or narrow that the light from the projectors cannot reach the bottom 
will never show on the image. It could also happen that the hole was narrow
enough to disappear from the image briefly and then reappear:
.br
.sp 8
One way to get round this is to lie to the system, telling it to expect mis-
shapen holes. Only holes that never show up cannot be dealt with in this 
manner.