I hoped camera calibration/alignment would improve with the 1.7 release. But I can’t get the new April Tags detected at all. Not a single one. I’ve printed them in A4 A5 A6 and held them at different distances, angles and orientations and also varied lighting in the room. Still not a single april tag recognized, after many retries.
System: Windows 10 PC with Lightburn v1.7.03 and a Gweike Cloude Pro laser with 5 megapixel fisheye camera mounted in the center of the glass lid.
I am following the instructions and advice on
and its embedded video
Calibrating with the old circles image still works. It is very fiddly and thus takes up to 15 minutes to complete, but I can eventually get 0.35 or lower on all steps in the calibration wizard. Alignment can then be completed without issue. The end result is not super great though. In the center of the laserable area the difference between camera and actual lasered position is around 1mm. But in the outer corners the difference is up to 8mm. I has hopes that would improve with the April Tags method.
Questions:
is use of april tags for calibration incompatible with fisheye cameras? If that’s the case the documentation should be updated to say so.
is a 8mm difference between camera position and actual lasered position in outer corners to be expected with a 5 megapixel fisheye camera? Or do other users get much better results with similar cameras? If yes what is your camera resolution and other camera specs?
Would you be able to post a screenshot or two of what your camera is capturing when you try to use the AprilTags? It might provide clues for us to determine how you can get a good capture.
When you attempt to calibrate the AprilTags, it is not necessary to mount the camera in the machine to perform the lens calibration. This process depends only on the camera’s lens, not its placement in your machine — as long as the camera and calibration pattern are perfectly still, calibration can be performed anywhere. Have you tried it outside of your laser bed?
My understanding is that all of LightBurn’s cameras have a fisheye lens and AprilTags are now recommended over the circle dot pattern.
My camera is a 120w degree 8mp camera and it’s mounted 18" above my bed. When picking out your camera, did you use LightBurn’s camera selection helper to determine your minimum mounting height for your bed size?
Thanks for the helpful suggestions. I will update with screenshots later. But the image looks quite similar to those on the doc page and in its youtube video. The calibration attempts I described (successful with circles, unsuccesful with april tags) were performed with the lid open and the camera pointed out into the room. Calibrating with april tags inside the machine placed on the honeycomb bed also fails, as does circles on the honeycomb.
I will double check the camera mounting height but my memory says it was checked to be ok after purchase. The camera came bundled with the laser in a fixed mounting position in the lid.
Here is a product image of the machine where the camera’s position is visible
It’s likely that this particular laser requires the lid to be closed before running lens calibration and camera alignment, assuming this aspect of the laser holds the same to the Omtech Polar. It’s an extreme fisheye lens mounted perpendicular to the lid.
Can you say more why you think that is the case? The official video https://www.youtube.com/watch?v=A1ctQdbyn7A shows a fisheye lens camera calibrated with lid open. As I wrote we have performed calibration with circles while the lid is open and gotten useful result and have used it like that for around 2 years now. I’m only trying to improve calibration so the outer corners are less than 8 mm off. With circles placed inside on the honeycomb I can get them to detect in the middle but with bad results but have never managed to get though the corner image calibration steps.
Let me tag on one more question, perhaps only answerable by a LightBurn developer: I read somewhere that LightBurn uses OpenCV under the hood in the camera calibration code. Is there any documentation of exactly what the calibration steps does or what the output values represent? And is there some other way to use OpenCV standalone, with some kinds of tags or checkerboard image or what not, to undistort the camera image and then import those calibration values back into LightBurn?
I may have misunderstood what step you were attempting and overstated the limitation. If you are only attempting the lens calibration then that can be done with the lid open if you are hand holding the pattern card. However, the camera alignment step likely requires the lid to be closed assuming the camera on this laser is perpendicular to the lid as it is on the Omtech Polar.
This video from Gweike on Camera setup seems to confirm this. The lid is opened during lens calibration and closed during camera alignment. Gweike Cloud Camera Calibration with Lightburn
In systems where the lens is perpendicular to the bed of the laser with the lid open, then open would be the required position.
Are you placing the pattern on the bed only or have you tried holding and positioning so that the pattern is visible at the requested positions?
I’ve done the camera calibration on an Omtech Polar and it was extremely fussy due to the extreme fisheye as well as the light source from the machine. You may want to cut light from the machine and rely on good ambient light.
I’ve not seen this documented on the LightBurn side but OpenCV has much documentation on it.
I’ve not seen it done but I see no reason this couldn’t be done externally and manually input the camera parameters back into LightBurn. Right-click on the camera window and export the camera settings to get an idea of what the camera matrix and corrective values look like.
There are a number of forum entries where OpenCV is discussed but I’m not familiar with any that have gone as far as anyone doing their own calibration externally. I’d still suggest a preliminary search to get familiar.
Yes, alignment is done with lid closed. I think the issue I’m experiencing is wholly due to the calibration step, so I’ll set aligment aside here.
I think I’ve tried it all. The first time I spent around 5 hours of repeated attempts with variations in circles size, position, angle and lighting, before discovering that calibration could be performed with the lid open and camera pointed out into the room.
Do you recall what megapixel value its camera has and what end result you got from that in the outer corners? Would help me put my 8mm issue in perspective.
As a general comment I wish the Lightburn camera calibration/alignment documentation gave more real world use case information. Like some examples of what results, in millimeters of deviation in the center and outer regions, can be realistically be expected from carefully performed calibration/alignment. I get that camera quality, fisheye degree and laserable area varies between laser machine setups, but a few different examples could still be helpful to adjust the users expectations. (Will post that wish to the feature request site later, unless someone from the LightBurn teams picks it up from here.)
I don’t recall specifically. Omtech’s site lists 5MP for their current offering. This was done in January of this year so unless they’ve changed the camera I have no reason to believe it was any different.
In terms of accuracy I found the results fairly poor. I attributed it to the difficulty in lens calibration and the extreme fisheye. I did not have a chance to do any post-calibration tuning using the scale and shift fields in the Camera window but initial accuracy near the home position was in the sub-3 or 4 mm range if I’m remembering correctly. I did not test elsewhere. In any case, it was large enough where it was obviously off.
I haven’t seen anything official on this but from anecdotal references on this forum you could reasonably expect to get sub-2mm accuracy across a 700x500mm bed using an 8MP LB camera with non-wide lens fairly repeatably. With careful tuning accuracy in the sub 1mm range seems achievable. At that scale the repeatability of camera position becomes increasingly important. I haven’t seen anything to suggest that accuracy at the sub 0.1mm range is achievable with the current fixed camera scheme.
Head-mounted cameras are not nearly as common on the forum and likely not nearly as well exercised in LightBurn so difficult to gauge performance there. However, I suspect the headroom for those in terms of potential accuracy is much higher even if not currently achieved.
Thank you for that useful feedback. Reading more on the Omtech Polar it seems very similar to Gweike Cloud Pro so perhaps they even use the same model of 5 MPX camera.
I’ll have time to experiment some more with calibration next week. I will then also try to borrow a different fisheye camera to test with, to get a better grip on if the gweike camera is the cause for the issue I’m experiencing.
Not sure if anything’s changed on this but at the time the Polar was released it was fairly well understood to have been OEMed from Gweike. There are some implementation differences I assume either by specification or for cost cutting but they are clearly very related machines.
Please do share your findings. I’m sure it would benefit a good number of people considering the relative popularity of these machines.
