Motion field computation with a continuation algorithm

THe problem of dense motion field estimation usually comes down to the minimization of a non-convex energy function. This minimization is achieved thanks to multigrid methods or more often using the first order approximation of the structural model. In this paper we propose a different approach based on the estimation of the image properties which lead to a convex hamiltonian. These properties are used to define a continuation algorithm: approximations of original images are computed using these conditions and used for the first stage of our continuation algorithm. The modeled images are then gradually transformed back to the original ones while the process evolves. This method is tested on synthetic sequences and its results are compared with those of other methods.© (1997) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.