A reduced-order Kalman filter for moving images.

Abstract

An extended Kalman filter is used to estimate the velocity of an object moving across an image frame and to reduce the undesirable effects of noise. The extended Kalman filter is implemented in the spatial frequency domain to reduce the number of computations. The resulting filter structure is a parallel bank of third-order extended Kalman filters. This parallel structure is referred to as the modified extended Kalman filter. The performance of the modified extended Kalman filter is evaluated under a variety of noise conditions using computer simulations. Simulations employed two test objects moving across a reference image in the presence of zero-mean, white, Gaussian noise. The performance of the filter was demonstrated when these objects were moved at integer and noninteger velocities. Performance was also evaluated when a stationary background was included with the white noise.