ROBUST PROJECTTVE FACTORIZATION IN THE PRESENCE OF MISSING OR UNCERTAIN DATA

摘要

This paper presents a batch type projective reconstruction method to overcome missing or uncertain data. The key feature of the proposed method is that it formulates the factorization problem as a trilinear minimization problem of covariance weighted reprojection error with respect to the motion and 3D point locations, and their inverse depths, To perform minimization, we employ the resection-intersection like technique, which uses only linear computation in contrast to non-linear method. Additionally, the proposed method can be readily applied even if some features are missing. We show the result of experiments on both synthetic and real data, and demonstrate effectiveness and robustness of our method.