Wood inspection

It seems to me that natural products must be very difficult to inspect. Take wood flooring for example. As an organic material, every piece will be slightly different, so how do you inspect it for defects?

This is the kind of task people are good at. Yes, they get tired and distracted and make mistakes, but the human brain is an excellent tool for identifying deviations from what is considered acceptable.

German company ATB Blank build machines that automate wood inspection, so when a story about one of their applications popped up on the Vision & Sensors website, (“Seeing Through Defects” June 4^th 2012,) I had to read it carefully to find out how they do it.

The article describes a 5 camera system. (IDS is the vendor.) Four are used to look at the ends and sides, which sounds pretty straightforward to me, but the fifth looks at the grained surface. So what algorithms do they use to distinguish cracks, chips and dents from natural features?

Sadly, though perhaps understandably, they don’t disclose. More frustratingly, there are no photos on the V&S website, so your intrepid blogger did some digging.

In the Press section of the ATB Blank website there’s a link to a pdf describing the system. Sadly, my German isn’t that good, but I found the pictures interesting. There is however some information in English about the Argus Spectra system that performs the inspection. But nothing about the algorithms.

In conclusion then, ATB Blank know how to inspect wood. How they do it they’re keeping secret.

As best I can tell then, they seem to be using some form of neural network learning tool to teach the system what good wood looks like. I’m pretty sure MVTec offer such a tool in their Halcon vision package, and CVB may have a similar capability.