Camera Motion Estimation by Geometric AIC for Factorization with Missing Data

Estimating missing data of feature point tracking is one of the important task in computer vision. For this, many methods have been proposed. Many of them fit an affine space to feature point trajectories, or factorize the input data with missing data into a low-rank matrix. We focus on the fact that the dimension of an affine space and the rank of a factorized matrix varies according to camera motions. We propose a method for estimating camera motions by using the geometric AIC and optimally estimate missing data according to camera motions.