Algorithm Evaluation and Error Analysis

(i)

Suppose that \(N\) 2D points are measured with independent standard deviations of \(\sigma\) in each coordinate. The RMS residual error (5.3) for the best fit line using orthogonal regression is \(\sigma \left(1 - 2 / n \right)^{1 / 2}\) because \(d = 2\) and \(N = n\) [Ebe00].

(ii)

Revisit once chapter 18 is finished. The answer seems to be along the lines of 2D homography – error in one image (5.5) and Error in both images (5.6).

References

Ebe00

David Eberly. Least squares fitting of data. Chapel Hill, NC: Magic Software, 2000.