Camera based scanner to detect luminescent marker markings on a surface

[obdata]

Hello,

I'm new here and have no idea if this is the right place or the right forum for this, but here goes

I work in a woodworking shop where we cut up maybe $400,000 to $600,000 of lumber per year. We are always looking for ways to reduce waste, since wood is a natural product with defects and we are cutting it into parts of fixed sizes, that must be free of defects, so there is always a lot that is wasted.

So I just got a crazy idea, and I don't really know where to start. Here's the idea:

I'd like to build a scanner for detecting surface defects in the wood. Now there are scanners like this around, that are very advanced, fully automatic and very fast, but very expensive. So I'd like to come up with something kind of DIY (read simpler and less expensive), in which an operator would simply use a luminescent pink crayon, or something such, to outline defects right on the material. Then the material would be conveyed under a camera (or multiple cameras), from which we could get an image of the surface of the board. Then the image needs to be processed to detect the crayon markings, (and thus the defects that we want to avoid ), and also the edge of the board. Then from there the information can be used for computerized calculation of the best part nesting and cutting strategy, just as an aid to the operator in deciding how to get the best usage out of the material when sending it to the saws.

I have no idea if this makes any sense to anyone
I really don't know overly much about the technological part of this idea, I just know that it could be done, and I think it wouldn't need to be terribly expensive if done in a "DIY, opensource" way. Besides, DIY is all the fun.

Anyway, I attached an image of several boards of wood, which shows the type of knots and other defects that we have to cut around to produce the specific parts that we need.

So, if anyone has any good ideas, I'd like to hear them.

Thanks,

Jason Oberholtzer
Bareville Woodcraft Company
www.barevillewoodcraft.com

[Rufus]

So I just got a crazy idea, and I don't really know where to start. Here's the idea:

Your idea isn't really much simpler than the expensive systems. You are using an expensive man to mark defects with paint because a computer vision system will find it a bit easier to recognise paint than defects. Everything else is as complicated.

Given your example planks do you only cut across them avoiding defects or length ways as well?

Do you decide what to do with a plank just before you saw it or do you want to choose which planks to saw depending on what you need and the defects the planks have?

[dannyf]

I would think it would work: you probably need an (optical?) sensor that can sense one line at a time. Based on the result, you can decide if you should reset a marker that indicates the potential beginning of a good plank - there may be some optimization going on based on the desired plank sizes.

The trick is likely to find a sensor. Those used in a photocopier would be a good starting point. Or maybe in a newpaper printing factory. Alternatively, some sort of ultra-violet sensor / "dark light" could work.

[dannyf]

What about using a camera tethered to a PC? One sheet at a time?

You take a picture of the sheet, digitized it to extreme black / white and then decide where the starting point of a plank is. That is probably the least expensive option.

[Kohanbash]

Hi
It might be easier to use a laser scanner for this then a camera (but a bit more expensive for the sensor).

There are many laser scanners that will also provide intensity (in addition to range). Based on the images shown you might be able to detect those defects without using a crayon to mark them.

One popular vendor for these scanners is https://www.mysick.com/partnerPortal/TopFrameSet.aspx?LeftTarget=LeftNav.aspx&RightTarget=HomePage.aspx

[codeboy2k]

If you were trying to DIY this, I would suggest taking some photos and playing around with Imagemagik tools. You can easily take your existing sample image, put it though some equalization, color-balancing (to throw it off balance, actually) then threshold it and you get an image like this, which can easily be used further to select the light and dark areas (perhaps selecting areas that are more than 1 pixel line wide) and use that to guide your cutting software further.

Further processing this to produce bounding boxes of x-y regions to avoid is a matter of scanning from left to right and top to bottom on the image pixel-data here.  The image can be loaded into a small program in GNU Octave to do this (it's also free software), and the output can be a list of bounding boxes covering the defects: [x1,y1; x2, y2]. I didn't go to that step to write that program but it's not that hard.

I realize the image is multiple planks, and if you want to process multiple planks at once in a single image you need to do edge detection as well. But if the image was of a flat sheet of plywood, for example, you could easily floorplan your cutting strategy around the defects.  For straight cut planks you can process them one at a time and cut the defects out.

Caveat: getting it to work over one sample like this is easy. Getting it to work reliably over multiple different samples under different conditions will take a lot of your time, and is really why the "pro" stuff is so expensive

[nctnico]

OpenCV is the magic word here. A couple of years ago I wrote an algorithm to measure wounds. The most difficult ones are those where the wound and the skin around it are red. The timber doesn't seem difficult compared to that.