I have two cameras with very different resolutions. One camera is the RGB camera of the microsoft kinect sensor with a resolution of 1920x1080 and the other one is a very specific camera with a resolution of 480x320 (not the IR sensor of the kinect).
Calibrating the intrinsics of the cameras seperately works out just fine. I am getting reprojection errors of around 0.1-0.15 pixels. But when I try a stereo calibration on the sensors, I get very high errors of 1.0 pixels and higher.
For calibration I used OpenCV. First I was trying to use Matlabs stereo calibrator, but it didn't work for cameras with different resolutions.
Do you have any hints on how to improve my results? Is there anything special that I should do different when working with very different cameras?

  • $\begingroup$ Hm, not even sure this is something that can be changed – your second camera has one fourth of the resolution of the larger one, so if there's a projection error that would amount to half a pixel for that camera, it'd be two in the other, assuming equidistance. $\endgroup$ Commented Feb 17, 2020 at 18:17
  • $\begingroup$ @MarcusMüller I think in this case an equidistance assumption doesn't apply. The field of view of the low resolution sensor is way smaller. But what you are saying is that the projection error might be acceptable? When I use just a subset of 30/45 images of the calibration data set, I can get the error down to like 0.45px. $\endgroup$
    – Max
    Commented Feb 18, 2020 at 8:57


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.