Motion estimation relaxing the constancy brightness constraint

In the field of motion estimation, the constancy luminance principle has always been used as hypothesis to constrain the problem of optical flow extraction. When real conditions do not satisfy this assumption, classical motion estimation techniques fail. On the basis of models of image representation and image acquisition, a general model that considers brightness changes by means of a multiplicative function and an additive function, has been proposed. In this paper we discuss the problems that such a model poses for the extraction of the optical flow. We show that, in some conditions, the parameters of the model can alias the motion parameters. One consequence is that reliable estimates of the parameters have to be obtained by avoiding aliasing conditions. Another consequence is that some classical techniques, which use multiresolution approaches to solve the problem of local minima for motion estimation, cannot be used anymore. Finally an algorithm that estimates separately brightness change parameters and motion parameters avoiding aliased conditions is presented with simulation results.<<ETX>>