Differential scaling

[pbourke]

A question about markers and scale bars.

I have markers and known lengths along two roughly orthogonal axes. I assumed (perhaps incorrectly) that the model would be scaled differently in order to meet these scale bar values. Rather than just scaling uniformly.

But it seems that isn't happening, or I don't know to enable that ability. Obviously, one would like to think that all axes are being reconstructed with a constant scale, but in this case that does not seem to be the case.

For example, if I set the scale bar values on the horizontal plane, the vertical distances are incorrect, while other measurements on the horizontal plane are correct. Similarly, if I set the scale values for the vertical plane the distances on the horizontal plane aren't correct, while other measurements on the vertical are correct.

[JMR]

Hi, Paul:
Can you describe how this object has been photographed? It looks like it has been photoscanned through the upper hole, making it difficult to obtain a good cameras structure. How are base lineas aligned with respect to the optical axis? it seems to me that there is a problem with your intrinsic parameters calibration. I've seen cases where the vertical axis (along optical axis) is badly scaled with respect to horizontal (perp to optical axis) and I've found they are linked to a badly estimated focal lengh. My suggestion would be to use the same lens+camera on an object that offers favorable shotting conditions for a good autocalibration and then save interior params to be reused And work on this problematic case as precalibrated or at least the lens calibration params you saved, keyed in as initials before alignment.
I'm curious, what camera and lens combo has been used? do photos come with right focal lenght written on EXIF or are they null and have been guessed by Metashape?
Best regards

José Martínez

[pbourke]

It's a downward pointing fisheye (220 degrees) lowered into the manhole, please see attached.
Video is being recorded, I extract every 1/10 or 1/5 video frame for the reconstruction.
No exif data.

ps: I totally accept this is not how I would design a system to do this, what we are hoping is to use the existing manhole inspection system to perform reconstructions.

[JMR]

Hi Paul: Yes this is an intrinsically weak use case. So the way to prevent the strong correlation between interior and exterior params, would be to precalibrate and fix lens calibration. I have worked with a company whose specialty is manhole inspections, and they also had weird issues with fisheyes.
Feel free to contact if you need help for calibration.
Regards


José Martínez

[pbourke]

I'm still surprised, if true, that scale markers only result in a uniform scaling.