Mean field approximation to multimodal motion estimation problem

The 2D Markov Random Field (MRF) model, combined with the Bayesian estimation framework, has proved to be an efficient and reliable computing tool to the optical flow estimation problem. Specifically, we are investigating the multimodal approach, where complementary constraints are imposed on the optical flow model. However, this approach suffers from expensive computational requirements, which is the direct consequence of the large dimensions of the optimization problem. Recently, a deterministic optimization technique, namely the mean field approximation has been proposed, which not only provides satisfactory estimation result, but also reduces the computational cost drastically. Here we apply this new technique to the above mentioned multimodal motion estimation problem.