Automatic parameter adjustment for a machine vision
based inspection system
Machine vision systems
are a complex combination of lighting sources, cameras, lenses, image processing
algorithms and the object of interest to the system, such as e.g. a glass ampoule in a quality inspection
system or a pedestrian in a people counting system. The system will
end on a multitude of system
to solve a problem succesfully and finding optimal system settings are very task specific.
constructing the system, there will generally be many parameters that have to be fine
tuned which can be
or intensive tasks.
Innoscan produce machine vision
based inspection machines of e.g. glass ampoules and vials for the
pharmaceutical industry. The machines automatically analyze the ampoules for impurities using machine
and the ampoules are rejecte
d if they contain impurities
ever, a lot of time is spent on manually
tuning the image processing parameters of the system. For instance, choosing an optimal digital filter
for the image processing and setting various threshold
values. For each se
tting of the system, it will have a
certain rate of true/false rejections and true/false accepts on a test set of ampoules.
The purpose of this project is to develop and analyze methods to perform this fine
Both inputs (ie. parameters
of the system) and outputs (classification into reject/accept) of the system can
be accessed through a software interface, and therefore it
possible to build an automatic method that
can adjust the parameters of the system in an “intelligent” way to op
timize the classification rate. The
project may involve both image processing, computer vision and machine learning techniques.