Optic Flow by Scale - SpaceIntegration of Normal

In this paper we will present a method for computation of optical ow elds in which we take advantage of the concept of scale. We believe that it is possible to extract optical ow elds from normal ow elds, by tting the normal components of a local polynomial model of the optic ow to the normal ow eld. This tting is based on an analytically solvable variational optimisation problem, in which an integration scale-space over the normal ow eld regularizes the solution. An automatic local scale selection mechanism is proposed in order to select proper integration scales, which will lead to an improvement of the quality of the produced optical ow eld. The optic ow method proposed in this paper is least committed in the sense that we model the local variation of the optic ow by a simple polynomial model for which the range of validity is determined by the local choice of integration scale. This makes the proposed algorithm a useful tool in e.g. experimental uid dynamics and other kinds of experimental sciences where motion analysis is predominant. The performance proole of the method is compared with that of existing optic ow techniques and we show that the proposed method outperform the leading algorithms on the benchmark image sequences proposed by Barron et al. 3].