"The 3-D orientations of the calibration pattern might be too similar across images" even if it is not

Dear all,

I am trying to calibrtate a camera through the camera calibration app. I keep getting the following error:

"Unable to estimate camera parameters. The 3-D orientations of the calibration pattern might be too similar across images. Remove any similar images and recalibrate the cameras again."

I get this regardless of the calibration images, how much I move the checkerboard, changing camera and so on. No other message are provided in the console.

Is it possible that this error is showing up for other reasons then having similar patterns? I do not have any issue from other view points, even if they are perfectly specular to this. Is it possible to obtain further information or even knowing which are the patterns that matlab see as "too similar"?

Thanks for your help.

[Matt J]

Show us the images you are giving to the calibrator that you consider the "most dissimilar".

[Stefano]

Here are the images:

calibration images

Cam1 doesn't work, while cam2 works perfectly (the one in the center works fine). The setup is perfectly simmetrical:

This is just one of the calibration image sets I took. Whatever I do, tilting the calibration target more, moving it more, or anything doesnt change. The camera on the right doesnt calibrate. I even swopped them , and the one on the right still doesnt calibrate, which indicate that the issue is not related to the file format or anything..

[Matt J]

Are we sure the landmark identification step succeeded? Do you have the extracted chequerboard coordinates?

[Stefano]

Yes, that works fine. I checked the images one by one in the app and the detection works fine. I really dont knwo what's the issue.

[Stefano]

One addition thing I found that is driving me crazy. If I mirror the images left/right then it works. I am speachless.

[Matt J]

Perhaps you have interchanged the images from camera1 and camera2?

[Alessandro]

Hi Stefano, I have tried to calibrate your "cam1" using Matlab R2023a, and it went just fine for me. Also, the same calibration on Matlab R2024b fails... so perhaps it is an issue of compatibility between different versions.

Try to install the 2023a version and let me know if it works for you!

[Matt J]

This is a stereo calibration, not a single camera calibration, correct?

[Stefano]

No, it is a single camera calibration. I am just trying to calibrate with the single camera app and it doesn't work. I confirm what Alessandro said above: 2023a works, 2024b and 2025b don't. There is probably a bug in the newer versions.

[Stefano]

Update: Using 2023a the calibration app works and reprojection errors are low. However the field "Intrinsics" is empty for some reasons. On other cameras all is fine.

Answer 1

[Brizzolara]

I finally "solved" this issue with a workaround: I took the images related to the the non working camera and rotated it upsidedown (180 degrees rotation). Now everything works fine. Even a 3 cameras stereocalibratikon works.