Adaptive Finite Element Method for Total Variation-Based Motion Estimation

This paper extends a primal-dual semi-smooth Newton method for minimizing a general L1-L2-TV functional over the space of functions of bounded variations to an adaptive finite element setting. The proposed adaptive finite element method is based on a-posteriori error estimates derived using the strong convexity of the functional. The adaptive scheme is applied to the problem of estimating optical flow in image sequences, incorporating an adaptive coarse-to-fine warping scheme to resolve large displacements.