3D calibration

[HansWellens]

Hello there,

In my facial scanning setup, I include four markers which are located on a small flat plate affixed to a headband.
Would the calibration be more accurate if I included vertical and/or antero-posterior offsets in the positions of these markers?
Are measurements on the depth axis (Z-axis) expected to be as accurate as those in the XY plane?

Best regards,
Hans

[Andrea Alessi]

Hi Hans,

the marker themselves are not used for the calibration, in fact you don't even need them.

Just to clarify, are you meaning the autmoated camera calibration done by zephyr (intrinsics parameter) only, correct?

Or do you mean the full SfM process, including using control points (eventually automatically detected by zephyr) as 3D constraint e.g. with a a total station?

[HansWellens]

Hi Andrea,

My apologies. I realize now I did not express myself very clearly.

Indeed, I was referring to the latter: the full Sfm process, whereby the markers are used as 3D constraints (scaling the 3D world as opposed to arbitrary scaling). The question arised from a research project where I assess the accuracy when digitally repositioning my own 3D scanned upper and lower dentition into my 3D scanned face. When doing so, I consistently notice a slightly larger error in the Z-direction (depth direction). The reference for comparison is a 3D CBCT reconstruction of my own head. One explanation (of many possible ones) is that I currently only use vertically and horizontally offset markers (so two of the four side-by-side markers are positioned a little higher), but no Z-offset. So, the question then becomes whether including markers in the Z (i.e. depth) direction could possibly improve measurement accuracy in this direction...

[Andrea Alessi]

It is indeed better to have markers around the scene, rather than very close togheter. Please also note that the marker per se have no effect, but only the 3D constraint itself.

It is also very important to note that regardless of the position of the constraint, how the dataset is taken is also important. Make sure to maximize parallax in all directions for better results.