Hi,
It seems you are a bit confused about how the size of the checkerboard in the image affects the end result. Checkerboard size has several influences on calibration. For feature detection accuracy, larger checkerboard patterns in the image generally mean that the individual squares and their corners (which are critical for calibration) are more easily and accurately detected. Smaller checkerboards are easier to rotate and keep in different orientations, thus helping in geometric diversity and ensuring accurate estimation of intrinsic parameters, especially in case of lens distortion. Also, as mentioned, the guideline is that the target should fill at least 20% of the image, which is a general recommendation to ensure that the calibration process has enough detail to estimate these parameters accurately. In your case, there could be the following reasons for lower reprojection error:
- High-Quality Data: Even if the checkerboard doesn't fill the recommended 20% of the image, if the images are of high quality (sharp, well-lit, low noise) and the checkerboard corners are accurately detected, the calibration process can still converge to a very accurate result.
- Diverse Poses: If you managed to capture the checkerboard in a wide variety of poses (different orientations and positions), it compensates for the smaller size by providing the calibration algorithm with a rich dataset that captures the camera's behaviour across its field of view.
- Your calibration algorithm also plays a significant role.
You can refer to the below documentation to get more information on camera calibration:
