ReprojectionError

The reprojection error of the 3D chessboard dots projected from the left into the right image or vice versa (depending on the Direction parameter). The error is computed as L2-norm of the individual point deviations in x and y.

Format

Number

\(E_{rep,PnP}\)

Details

Assuming a projection from the left to the right image, the reprojection error is computed as follows. Let \((Rp_{i,x}, Rp_{i,y})\) be the 2D positions of the \(i_{th}\) chessboard dot in the right image and \((Rp^{'}_{i,x}, Rp^{'}_{i,y})\) the coordinates of the projection of the i th chessboard dot from the PatternPose into the right image. The (asymmetric) reprojection error \(E_{rep,PnP}\) is then computed as:

\(E_{rep,PnP} = \sqrt{\sum[(Rp_{i,x}' - Rp^{'}_{i,x})^{2}+(Rp_{i,y} - Rp^{'}_{i,y})^{2}]}\)

For a right to left projection, the computation is similar, but uses the pattern points from the left camera.