I have a sequence of video frames coming from a wide-angle lens and I need an end-to-end procedure that corrects the fisheye distortion and reports the lens parameters as well as the camera pitch/yaw—without ever showing the lens a checkerboard or any other special calibration target. The final use is straight, reliable computer-vision processing on the rectified frames, so accuracy matters more than pretty visuals. What I already have: raw video and a rough idea of the lens field-of-view. What I do not want: a workflow that depends on printing patterns, or re-shooting footage. Ideally the correction runs automatically on the existing frames, estimates distortion coefficients, and lets me pass a single undistort/rectify call to the rest of the CV pipeline. I would like the final package to include: • Source code (Python ) that ingests the video, recovers the intrinsic & extrinsic parameters, and outputs rectified frames in real time or near real time. • A short note on the maths or references you used so I can justify the approach internally. • A way to feed in new clips from the same camera and obtain calibrated results immediately. If you have worked with self-calibration, structure-from-motion, vanishing-point fitting or similar techniques, this should be straightforward. Please share one example of a project where you removed lens distortion without targets so I can gauge suitability.