High Reprojection Error despite Correct Translation and Rotation Matrices: Understanding cv2.stereoCalibrate

Question

I've been working on stereo calibration using OpenCV's stereoCalibrate function.

I'm performing stereo calibration for a simple stereo case, where the rotation between the two cameras is 0 and translation only occurs along one axis. I've successfully computed the translation and rotation matrices for the two cameras, and they turned out as expected.

However, despite these matrices appearing correct, the resulting reprojection error is unexpectedly high, measuring at 18.9.

Could someone explain in detail how the reprojection error calculated by cv2.stereoCalibrate is derived? I want to understand the factors contributing to this error and how they relate to my translation and rotation matrices. Additionally, any insights or suggestions on improving my calibration process to reduce this error would be greatly appreciated.

I'm currently using 26 images for stereo calibration, having previously conducted individual calibration for each camera before the stereo calibration. The reprojection error for single calibration is less than 1. When attempting stereo calibration with a larger set of images (50 to 100), it appears to exacerbate the reprojection error.

Thank you!