Know your enemy

It’s not enough to know about machine vision. Trust me on this: no one cares how well you grasp the intricacies of normalized correlation or OCR. For the technology to have any valve it has to do something useful, so what matters is your ability to apply machine vision.

Successful application demands that you understand the product and/or process that vision is being used on. You need to understand the nature of the surfaces that will be illuminated, and how they will vary. You cannot look at a dozen samples and think you know what to expect because you don’t. Many processes can throw a whole lot of variety at you, yet still be “in control.” Only by knowing what the process is and how it runs can you get any sense of what variation your vision system will have to address.

To me this is a good argument for developing your machine vision solutions in-house as opposed to contracting with an integrator. No one knows the manufacturing processes as well as the folks who’ve lived with them for years. But unfortunately, many companies lack the resources to do this and have no choice but to hire an integrator. If you’re in that position, here’s a little advice:

If the guy you’re working with is happy with just a dozen samples, or if he has no interest in the process that produces your parts, ask yourself if he will really be able to deliver a robust system.