No announcement yet.

Calibration tribulations

  • Filter
  • Time
  • Show
Clear All
new posts

  • Calibration tribulations

    Hi there,

    I've built a revision of my facial scanner using 10 4K USB machine vision cameras. These now need to be calibrated.
    I have some experience with this, and know about the calibration feature in 3DZephyr.

    My question is this: Most texts about camera calibration recommend tilting the camera 90 degrees for at least some of the images.
    This is however often highly impractical since:
    1. your are supposed to fill as much of the sensor area as possible with the calibration pattern (the sensor's "x-axis").
    2. but since the sensor area isn't square, it is impossible to fit the whole calibration pattern in the smaller sensor direction ("y-axis") if it is optimized for the larger direction.
    3. and focal length and focus should be fixed, so you can't back up or zoom in.

    So, would you recommend including tilted images?

    I am currently using a laptop screen to show the pattern. Would screen quality play a role (LCD or OLED)? Would I be better off printing?


  • #2
    Hi Hans,

    I assume you are referring to the internal calibration only (and not about the fixed rig itself) of your devices.

    In this case, yes, I would add random shots from different angles and heights that see the pattern. It won't hurt

    If you use the screen, just make sure to minimize if possible reflections and highlights.

    The calibration is done regardless of "how much" of the pattern fills the screen. Just as long as you take enough pictures, the calibration will be good, so don't worry about x or y % fill