A Preconditioned Alternating Direction Method of Multipliers for the TV-$L^1$ Optical Flow via Dual Approach

This work introduces a preconditioned dual optimization framework with the alternating direction method of multipliers (ADMM) to the optical flow estimates. By introducing efficient preconditioners with the multiscale pyramid, our preconditioned algorithms give competitive optical flow estimates under appropriate variational functional frameworks. We propose a novel preconditioned alternating direction methods of multipliers (ADMM) with convergenceguarantee for the total variation regularized optical flow problem through optimizing the dual problems. The numerical tests show the proposed preconditioned ADMM algorithms are very efficient for the total variation regularized optical flow estimates.