Optical flow recognition from the power spectrum of a single blurred image

In this paper a new technique for calculation of the optical flow is presented. When there is motion in the observed scene, an image taken will be motion blurred (to a degree depending on the exposure time). Up to now most of the algorithms for estimating the motion in a scene ignored motion blur and treated it as noise. On the contrary, motion blur is structured information and in certain cases can be used to infer the velocities locally. This new approach uses the information of the motion blur in the frequency domain to extract the orientation and the magnitude of the velocity-optical flow.