Intelligent Systems Laboratory
Task Specification
In order to set up the state-space search for the automatic generation
of a vision algorithm, the task which the vision algorithm is to
accomplish must be precisely specified. In practice specifications
on inspection tasks range from a high of
precise definitions of flaws with numeric tolerances through
a middle level of a few example parts with the flaws circled with
a magic marker and perhaps labeled by type, to a low of a few
descriptive phrases like \lq\lq orange peel defect".
Due to this imprecision, it is not unknown
for industrial vision engineers to end up with a design that works on
the few given initial examples and which falls far short of what the customer
had originally hoped for, but failed to communicate. Even in some cases
where the goals of the system are clear, the environment in which
the system is going to work may not be well specified. In this case,
the system will work perfectly in its lab demonstrations and not
at all on the factory floor.
This project is developing a language and user interface by which
an inspection task may be more precisely and easily specified.
