Motion Estimation for Video Bandwidth Compression Using a Heterogeneous Pyramid Image Processing Architecture

In many image processing contexts there is a requirement for some form of motion estimation. A number of strategies exist, differing in the accuracy and granularity of the motion estimate and in their computational requirements. The simplest and most coarse-grained scheme is 'block' motion estimation, where the current frame is divided into blocks of pixels and for each block an estimate of its most likely motion is computed by reference to a neighbourhood in the previous frame. Such coarse motion estimates are of utility in situations where a fairly approximate but fast trajectory estimate is required. Examples are motion-compensated interframe predictive coders and various aerospace applications. The block motion estimator is simple and amenable to parallel computation, making it feasible for implementation at realtime frame rates. This paper describes the mapping of the block motion estimator on to the lower levels of the Wamick Pyramid Machine (WPM), a heterogeneous pyramid architecture for parallel image processing I understanding which is currently under development at the University of Warwick. The paper describes how the required computations are performed and presents simulated timings. The advantages of the multi-SIMD architecture of the WPM are illustrated.