Calibrating Lenses for scan perfection

[Subunderground]

Hello!

I'm trying to get the very best results possible from our 86 camera setup. It's a fixed studio and the lenses are always set the same. Cameras are D600's with the kit EFS18-55mm lens that came with them.

So I've got two questions really.

Before I dismantle the studio and use Agisoft Lens to calibrate all of the cameras individually, should this get better results from Scans or is photoscans inbuilt calibration just as good anyhow? ie. does it even need doing?

and...

Do the lenses focal length have to be fixed in stone for it to work? ie. gaffa taped to a position, or are they adjustable if I take calibration for 4 or 5 different focal lengths per camera? I'm not entirely sure how flexible it is once I've calibrated a lens.

Many many thanks in advance as always!!

[maddin]

Not sure whether upfront calibration will improve things. I once read you might be better off placing some well-structured object in your recording volume and calibrate the cameras from that.

Do the lenses focal length have to be fixed in stone for it to work? ie. gaffa taped to a position, or are they adjustable if I take calibration for 4 or 5 different focal lengths per camera? I'm not entirely sure how flexible it is once I've calibrated a lens.

I am fairly certain that changing the focal length requires new calibration, I don't think you can accurately inter-/extrapolate from a few focal length measurements. I also don't think all your cameras have to use the same focal length, but changing focal length after calibration sounds like a no-no.

Taping them down once you have set them to what you need seems like a good idea to me.

Martin

[tommyboy]

Not sure whether upfront calibration will improve things. I once read you might be better off placing some well-structured object in your recording volume and calibrate the cameras from that.

Do you remember where you read this?  I am very interested in setting up something like this to:

1) initially focus the cameras
2) using coded targets, establish world units as well as orientation
3) get a rock-solid solve on camera positions, which can then be re-loaded to other sets of images (particularly uncooperative smooth/featureless surfaces that don't align nicely otherwise)

I'm thinking a mannequin with noisy spraypaint applied, plus coded targets on the ground and a vertical bar, could produce this result.  Curious if anyone has attempted this already?

[Infinite]

Not sure whether upfront calibration will improve things. I once read you might be better off placing some well-structured object in your recording volume and calibrate the cameras from that.

Do you remember where you read this?  I am very interested in setting up something like this to:

1) initially focus the cameras
2) using coded targets, establish world units as well as orientation
3) get a rock-solid solve on camera positions, which can then be re-loaded to other sets of images (particularly uncooperative smooth/featureless surfaces that don't align nicely otherwise)

I'm thinking a mannequin with noisy spraypaint applied, plus coded targets on the ground and a vertical bar, could produce this result.  Curious if anyone has attempted this already?

This can work but I can say from experience, over time from micro vibrations and other factors, imported camera alignment will fail over the course of a shoot. Say if you shoot over a couple of hours and keep re-importing from an original calibration set from the start. If you take continual update shots as you go of the reference object, it will help.

Otherwise re-using the start reference capture set, you will start to notice noise and misalignment creep in.

[tommyboy]

This can work but I can say from experience, over time from micro vibrations and other factors, imported camera alignment will fail over the course of a shoot. Say if you shoot over a couple of hours and keep re-importing from an original calibration set from the start. If you take continual update shots as you go of the reference object, it will help.

Otherwise re-using the start reference capture set, you will start to notice noise and misalignment creep in.

Thanks Lee. I have certainly noticed this too, when poor subject matter (black suit pants for example) resulted in large blind spots in the solve, as about 10% of cameras failed to align.  I loaded in cameras from another solve from photos taken about 30 minutes prior.  Even though all cameras are now "aligned" and the poorly-solved areas are vastly improved, others areas of the model now have more noise than before.

Is there any other way to deal with this?  Can the cameras be re-solved with this initial location being a 'suggestion' of where to start?  If I re-solve all cameras, they seem to ignore their previous location/orientation, but if I try to right-click "Align Selected Camera" on the NA cameras after the 90% of cameras have aligned, the remaining cameras still won't align.  It looks like the use of coded targets helps specifically to prevent this issue, I'm certainly keen to try that.

[Infinite]

This can work but I can say from experience, over time from micro vibrations and other factors, imported camera alignment will fail over the course of a shoot. Say if you shoot over a couple of hours and keep re-importing from an original calibration set from the start. If you take continual update shots as you go of the reference object, it will help.

Otherwise re-using the start reference capture set, you will start to notice noise and misalignment creep in.

Is there any other way to deal with this?  Can the cameras be re-solved with this initial location being a 'suggestion' of where to start?  If I re-solve all cameras, they seem to ignore their previous location/orientation, but if I try to right-click "Align Selected Camera" on the NA cameras after the 90% of cameras have aligned, the remaining cameras still won't align.  It looks like the use of coded targets helps specifically to prevent this issue, I'm certainly keen to try that.

I mentioned this a few years back. It would be a good feature to have for sure but I think difficult to implement, perhaps ask in the feature section of the forum? I will +1 it

[tommyboy]

I mentioned this a few years back. It would be a good feature to have for sure but I think difficult to implement, perhaps ask in the feature section of the forum? I will +1 it

This seems to be possible by hacking Ground Control, have you tried this?  It looks like one could hand-code a CSV file with camera positions from a cameras.xml export.  Or if you add a few coded targets and distance measurements into a calibration scan (and ideally included some coded targets in their regular scan), one could load marker/camera positions from the calibration solve, then "optimize alignment" to solve them exactly.  I'm going to try this, can you see anything obviously wrong with this approach?