Camera calibration - the more circles or squares, the harder to detect?

Hi there,

Simply, I’m trying to calibrate images taken from fisheye lens camera. They are of very decent quality, I’d like to say, but there is a problem with it when trying to detect circles.

For example, the original image has got 28 x 28 circles in it, but findCirclesGrid() wouldn’t detect it. However, when I crop the image for its centre and make it contain 14 x 14 circles, the method would detect them all.

I’m just wondering, is it a norm(?) to expect the detector not to find circles if the number of it increases?

If so, would there be any way to resolve that issue?

could you upload such an image?

This is cropped one for the circle part only. The original image has got bigger size

thanks for the picture.

hmmmm that’s some severe lens distortion. maybe the circles detection discards them because they become severely elliptical, i.e. quite flat, in the corners. just wondering.

no promises but I might spend a bit more time on this tomorrow.

@crackwitz Thanks for your consideration.

Currently, my colleagues are using MATLAB and MATLAB’s algorithm somehow detect some of the circles in it and undistort the image quite decently, which is a bit of shame for me as a python lover :sob:

Anyway, I’ll try to solve this issue, if the forum community would help me on that journey…