A column-space approach to projective reconstruction

The problem of projective reconstruction for multiple views is considered using a factorization method. A common difficulty of existing formulations of the factorization problem is that they do not adequately constrain the depth parameters thus allowing the algorithm to converge to ‘view-deficient’ solutions with entire views being suppressed. We propose to include a variance measure with an adaptive weighting parameter in the formulation of the factorization problem to overcome this difficulty. Algorithmic solutions with guaranteed convergence are provided to perform factorization under the condition that there may be missing data in the images.