Best practice for chessboard recognition quality measure

edit

Hi, I am using the cv::findChessboardCorners and cv::cornerSubPix to recognize a chessboard from an undistorted image for camera to marker detection, meaning that I have all camera parameters and are only looking at the transformation between camera and marker. However, I would like to get a measure of how good (or bad) a given chessboard detection is.

So my question is, what would be the best practice to get such a quality measure? - I am thinking of projecting the found corners to a perfect chessboard, and averaging the translational error of each corner compared to the chessboard, but I am not sure if there are any pitfalls here, or if other methods would fit better.

Best Regards, Thomas