I am using the HALCON Calibration Assistant (HDevelop) to perform a full camera calibration with a dot-grid calibration plate, following the standard workflow of acquiring 10-20 images with the plate in different positions and orientations.
The official HALCON manual (hdevelop_users_guide) includes this warning (repeated multiple times):
ATTENTION: Remember that it is essential to keep your camera setup (aperture, focus, pose) fixed, once you have chosen it! This applies to the calibration process itself as well as to the subsequent application.
The disclaimer explicitly names aperture, focus, and pose, but is silent on shutter and gain.
Separately, the manual's Quality Issues table suggests adjusting "the lens aperture or the shutter" to fix overexposure problems during acquisition.
My understanding so far:
- Aperture and focus must never change, because they directly alter the optical geometry (focal length, lens distortion coefficients, vignetting), the very parameters being calibrated.
- Shutter speed and gain are purely exposure/electronic parameters that do not affect the optical geometry, so the intrinsic camera parameters (focal length, principal point, distortion) should be unaffected by changes to them.
- The only risk from inconsistent shutter/gain across images is indirect: bad exposure -> inaccurate mark center detection -> noisy or failed calibration, which is a data quality issue, not a geometric bias.
My question:
Is the above reasoning correct? Specifically:
1. Does changing shutter speed or gain between calibration images introduce any geometric bias into the calibration result, assuming all images are well exposed and marks are detected accurately?
2. Is there any subtler effect I am missing, affecting subpixel mark localization?
3. Does the answer differ between checkerboard-based calibration (e.g., OpenCV) and dot-grid / circular mark calibration (as used by HALCON)?
I am aware this question is HALCON-specific in origin, but the underlying math is general, so answers from OpenCV or any other calibration framework perspective are very welcome.