Real time implementation of subpixel motion estimation for broadcast applications

Future development in areas such as digital television transmission using data compression techniques, television standards conversion or video noise reduction depends highly on the inclusion of accurate motion vector information in the process. The motion vector field resulting from estimation with a translation model are horizontal and vertical displacement vectors which describe how each pixel has moved between the current and the previous picture. Techniques to generate the motion vector field include pel recursive gradient methods, phase correlation in the transform domain and block correlation techniques. The paper describes an implementation of an extended block correlation method which is robust, gives accurate motion vector estimates and also is suitable for compact implementation for high bandwidth signals