. What two categories of parameters does camera calibration recover, according to this lesson?
Calibration recovers the ideal pinhole intrinsics matrix K plus the radial/tangential distortion coefficients that describe how a real lens deviates from that ideal.
. Why does the lesson generate many synthetic checkerboard views from different angles rather than just one?
With noisy corner detections, a single view underconstrains the problem; many diverse views average out noise and better pin down both K and distortion.
. What does k1 and k2 radial distortion do to straight lines in a scene, when captured through a real lens?
Radial distortion terms in the distortion model curve straight lines; the sign of k1/k2 determines whether they bow outward (barrel) or inward (pincushion).
. After applying cv2.undistortPoints with the *estimated* (not true) calibration, why don't the grid lines straighten out perfectly?
Calibration from noisy corner detections only approximates the true K/distortion, so undoing distortion with the estimate leaves a faint residual bow.