Michael Sokolov

Abstract

This paper presents a suite of algorithms for use in motion analysis. A brief survey of previous work in motion is undertaken. The relationship of the gradient based technique presented here to other motion analysis techniques is explored. A simple single-scale algorithm is developed from a consideration of image warping. The underconstrained two-dimensional motion equations are solved by an implicit smoothness constraint. The response of the algorithm in the frequency domain is explored. The limited spatiotemporal range of the single-scale algorithm is extended by the use of a multiscale image pyramid and a hierarchical control structure. Iterative refinement at each scale is used to eliminate the propagation of any error to other scales. Finally applications of the algorithn to motion compensated noise reduction and motion amplification are demonstrated.