Quasi-perspective structure factorization with missing data

The paper focuses on the problem of structure and motion recovery from a monocular image sequence under quasi-perspective projection model. Previous study on this problem adopts singular value decomposition (SVD) to the tracking matrix with rank constraint. The method is time consuming and does not work for incomplete data. In this paper, we propose to adopt power factorization to the problem. The proposed algorithm overcomes the limitations of previous SVD-based counterpart. It is easy to implement and can deal with missing data in the tracking matrix. The algorithm can also be applied to nonrigid factorization. Extensive tests on synthetic and real images validate the proposed method and show its improvements over existing solutions.