Tuesday, April 9, 2013

Determining the resolution required – a question for you

When we think resolution most often we mean millimeters viewed per pixel (mm/px), or the raw resolving power of the vision system. Given that it will take 3 to 5 pixels to detect a feature, we can then estimate the size of the smallest defect the system will find. But what mm/px number is enough?

Many of my vision systems are developed for defect detection, usually to replace a human inspector. That means the quality standard was written based on what the unaided eye can see. In turn that means there is no definition of a minimum defect size: it becomes a matter of “if the inspector can see it, it’s bad.”

Leaving aside the question of whether we’re throwing away perfectly functional product, what does that mean for the resolution of our system?

My rule of thumb has been that my system should be able to find a defect of 0.3mm, implying that it needs a raw resolution of 0.1mm/px, or better. But now Andy Wilson has given me a clue that there is a better way.

His chatty blog post, “Fish that spit” (March 26th, 2013,) discusses a method of determining the visual acuity of fish. This is a pretty serious issue, because if the same method could be applied to humans, I’d have a measure of what resolution I need to engineer.

So here’s my question: does anyone have any numbers or data on human visual acuity? Any other comments on this?

1 comment:

Anonymous said...

I remember reading a pretty good presentation regarding the smallest feature detectable by the human eye... Unfortunately, I forgot to bookmark the post. Nonetheless, determining the resolution required relative to operator inspection, I believe, warrants examining a number of factors. Most of these surround the contrast of the features being detected. Obviously, finding small black on white defects will be accomplished with greater ease than black on dark gray. That aside, I believe that we can numerically extrapolate a rough idea as to what is possible for a human inspector to detect (assuming black on white features)...

Assuming someone has 20/20 vision, this would mean that the individual can resolve 1 arc minute (physical limitation is around 20/8 based on cone density in the eye). This works out to about 0.016 degrees (conversion from arc minutes to degrees). We then need to employ a little trigonometry to arrive at the formula that the smallest dot resolvable (i.e. defect) = 2 * Distance * tan(visual resolution / 2). So, for example, at 6" with a person that has 20/20 vision, dot size = ((6*25.4)*2)*tan(0.008) = 42um.

That being said, somehow, at my facility people routinely do better. Makes my life fun.