Improving accuracy in fixed multi-camera rigs...

[HansWellens]

One question I have been struggling with, is what the best approach is to improve accuracy in fixed multi-camera rigs.

I understand that external camera parameters of the various may be estimated perfectly from multiple views of the same "calibration target" (object with targets added for improved accuracy).
A bundle adjustment after generating the sparse point cloud should improve accuracy even further.

But wouldn't it be better to calibrate the cameras individually using Zephyr's calibration tool, in order to solve for the internal camera parameters (14 separate calibrations in my case)?
This would then be augmented by solving for the external parameters using the calibration target with markers/bundle adjustment.

It might be that I underestimate the capabilities of the modern calibrations algorithms. I am certainly no specialist. I just do not understand how one could solve for internal camera parameters without having multiple views taken using the same camera, and without rolled camera images.

One little extra question: would the calibration tool also work when taking multiple pictures of the (non-repeating) random pattern in the enclosed photo (without the markers)?

Thanks!
Hans

[Andrea Alessi]

Hi Hans,

yes, if you wish you can absolutetly do 14 separate calibrations, one for each one of your physical devices. It won't hurt, although if you have all your devices using the same settings the autocalibration is probably safe to use already.

Yes, the random pattern shown from zephyr is just a way to ensure a lot of features are seen in every photo you take. You could even do the calibration photos with any other photo set (provided is shot correctly, so a decent amount, translation and rotation, and enough features seen).

Please also note our markers are simply a way to easily let zephyr recognize control points automatically - they do not help the calibration in any way.