Robust Factorisation with Uncertainty Analysis

This paper proposes how the classic factorisation algorithm for affine reconstruction can be extended to be robust to outliers and proposes how the uncertainty analysis can be performed in this case. The robust estimation approach elaborated here is based on the iteratively re-weighted least-squares but the use of other robust methods is also discussed. Moreover, the uncertainty analysis presented in this paper could be similarly used in a RANSAC or LMedS extension of the factorisation algorithm. The experiments verify that the proposed approach is reliable and able to give to consistent estimates and uncertainty measure for affine structure and motion