Tarkin camera upgraded

Just got in a new varifocal lens for the 16MP camera.
This did a couple of great things!
One, the original lens was a wide angle but did not follow the normal formula for fisheye lenses, thus OpenCV’s algorithms were imperfect at flatting it. The ad for the original lens said it was a special “flat” lens that did not fisheye, but it looks like instead that tried to flatten the fisheye with some counter-distortion at the lens that OpenCV was not designed for.

Second, the original wide-angle lens was TOO wide. Much of the FOV went outside the bed, and those pixels were not part of the effective resolution. The varifocal has a separate zoom and focus barrel so I adjusted the FOV to span the bed and nothing else, concentrating the resolution.

And, it’s kind of amazing. The camera can almost read printed text on the bed. It’s much more accurate across the bed as a whole. I’m seeing about 1mm and I think I can improve on that too.

The camera is a separate feature than “Print and Cut”. P&C is a very accurate but more manual 2-step process that does NOT use the camera.

The camera is used to take a shot of the current bed and project it as a background on LB’s workspace. The most frequent application is to adjust your work to the available stock.

e.g., say you have a piece of leather. Leather is expensive and often is an irregular shape as well. You can put it on the bed, clamp it, and take a snapshot which is now your LB workspace background. You have a design of a bracelet and, without the camera, you could put 3 copies arranged parallel with much of the leather stock’s area unused, but it would be impractical to try to fit more copies on that irregular unique shape. The regular Frame only shows the rectangular bounding box, which is not always clear if the parts being cut will actually occupy that space- for example, if the bracelet is oriented diagonally then it touches 2 corners of the rectangular bounding box, but the other 2 corners are far from what will actually be cut and Frame is not that helpful. The Rubber Band Frame is better but has the same problem- a “T” shape will have a lot of space inside the RB Frame that is not actually cut.

With the camera, you can potentially “nest” in more copies- rotate it, offset some copies to reach the right edge, place some side-by-side where there is room, rotate some copies 90 deg to scrunch more into some free space, and end up fitting 6 copies on the same piece, using much more of the stock without any copy going off the edge of the stock. All pretty easy, this is just manually making copies of the bracelet on the workspace and moving/rotating them in LB to fit on a virtual background of the snapshot of the leather on the bed.

1 Like

Danny, this was a helpful post because it described a use case for the camera that didn’t rely on a steady lid. Extreme accuracy isn’t required in your example. My primary use has been etching on batches of inconsistent small parts. For instance, I may be etching names on 28 pens and each requires precise placement of a name on a different pen of varying lengths (due to different kits). If opening and closing the lid each time requires new calibration due to wavering lid positions, then it’s not very efficient.

Admittedly in the use case of my pens I could just as easily used red dot framing if the alignment was accurate. However, to your point, if the shapes and/or stock were irregular then it’d be an issue. An unsteady lid requires excessive calibration in use cases requiring precise placement, especially when the distances are as huge as on Tarkin.

The only practical solution I’ve heard is to install a piano hinge … but I’m not suggesting we do that if we can get the Laguna up and running well. James tells me that the Laguna camera is on the head so is much closer to work and not affected by the lid motion. The two cameras couldn’t do exactly the same things as well as each other but each position seems to have its benefits.

A piano hinge is not a high accuracy solution, actually. It can bind and buckle with any planarity error in either surface.

I looked over LB’s alignment method, then came up with notably different method with a MUCH easier, quicker, more accurate result.

Whatever you think about Tarkin’s lid, it’s looking to be a myth. I opened and closed it 5 times and the worst I got was a +1mm error. And actually that was only once, then I just made a point to press the lid lightly at the top center and let it go. Then I got 3 times with straight up “nothing” for variation with lid (if it was different, it was less than a pixel). So with a basic technique, repeatability can be better than measurable limits.

The resolution is now about 3 pixels per mm in the center. That’s CRAZY. I can read any of the bolder text of the Tarkin “getting started” guide, just can’t quite make out the regular text font. The 16MP camera with varifocal zoom paid off in spades!

The lens flattening algorithm is not perfect. That is, while the camera alignment is very repeatable, the location of an arbitrary point on a perfectly aligned 4 point camera image may be perfect at the 4 points, but will be off slightly elsewhere due to warpage. But the worst I saw was about +/-1mm in X or Y. So, yeah, it’s not “totally” perfect but it’s much closer to perfect than any camera config has ever been.

The camera swivel mount could be something fixed tighter. You can bump it and make it point in the wrong direction, but then you just need to get a live image, swivel it back to see the whole bed, and run the 4-point alignment again which takes about 2 minutes.

Believable. That’s how I lock mine in for repeatability too.
With this precision it should serve in almost all use cases.
Can’t wait to get back and reintroduce myself to your beast!