reprojection error

The distance between a point's observed image location and where a calibrated camera model predicts it should land.

Reprojection error measures a calibration's own quality: given the recovered camera matrix $\mathbf{P}$ , project each known 3D calibration point back into the image and measure the pixel distance between that predicted location and the point's actually observed location. A well-fit calibration keeps this error small and consistent across every calibration photo; a large or uneven error signals a bad correspondence, an overfit distortion model, or a stale calibration.

Examples

A calibration with a $0.3$ -pixel average reprojection error is trustworthy; one with $8$ pixels signals a bad correspondence, an overfit distortion model, or a stale calibration.