Perspectives in Vision: On Minnesota Fats and Machine Vision
February 23, 2010
Have you ever watched the performances of any of the truly great billiard players? It is pure artistry; the strategy, the angles, the spin…all the tricks of the trade under the precise guidance of deft and nimble hands. One of the greatest of all time was Minnesota Fats, and truly he was one of the most entertaining to watch.
Shot after shot would fall in rapid succession as Fats approached the table nonchalantly and, with barely a pause, let fly with his cue. Nearly every shot would look easy, and it left you thinking “Well gee, I could do that, it was straight in!” Few understand that a good pool player doesn’t just make hard shots; rather, a good pool player makes sure the next shot is easy. Therein lies the skill in billiards.
In our business as machine vision integrators, I’m very often approached by vision distributors and potential customers who will send us images of a product and ask “can you locate and inspect the [whatever] part in this picture?” And sometimes these images look like mud soup, with poor contrast, poor perspective and so on. Usually this happens after another integrator has given up on the problem and the customer is desperate.
Obviously vision algorithms and other math-intensive approaches are extremely important in machine vision integration; they can sometimes look like magic to the uninitiated, and can in many cases solve these very difficult challenges. But no amount of mathematical “voodoo” (as one of our customers calls it) will save you if the image is irreparably poor.
This goes far beyond choosing the right camera and lighting – it encompasses technique as well. 2-D or 3-D? Laser scatter? Dark field illumination? Some combination of these? The goal is always to generate contrast between the features you care about and the features you don’t. If you set up your shots correctly, just like Minnesota Fats, you too can win the game with relative ease.