Motion estimation and tracking using multiresolution affine models

Publisher Website
Google Scholar
Abstract
As in many areas of image analysis, a key issue when tackling motion is determining the appropriate size of local analysis window to be used. Although the use of a small window can lead to high resolution estimates of the underlying motion field, it is more likely to lead to est imates which are biased by the effects of noise and aperture effects due to insufficient gradient variation within the window. On the other hand, although larger windows can lead to more robust estimators in these respects, they bring with them the complication that the variation of underlying motion field is likely to be m ore complex than a simple translation, ie linear or quadratic (3). Moreover, the optimal window size is unlikely to be the same across a given frame or across the sequence of frames - objects in a scene are general ly of different size, occur at different depths and have different motions, all of which effects the choice of window. The work described in this paper attempts to address this pro blem. It is based on two key elements: the use of an affine model to describe local motion variation with in a sequence; and the incorporation of this local model into a multiresolution framework to describe the global motion field. Affine models provide greater flexibility in modelling local motion, being able to represent rotation, dilation and shear as well as translation, enabling larger window sizes to be used and hence the potential for more robust estimation. The use of a multiresolution framework allows the window size to be adapted to the underlying motions present in the scene, providing a mechanism for selecting the best set of local affine motion descriptors. The result is a piecewise-linear representation of the motion field in t erms of a 'mosaic' of 'patches' of varying size moving with affine motion, corresponding to planar facets mo ving with rigid motion in 3-D - the so-called 2.5-D model (4). The approach has led to the development of algorithms for estimating motion between frames and tracking region elements over time. The algorithms are efficient and l ead directly to concise descriptions of the motion field in terms of a relatively small number of mosaic el ements. Experiments have shown that the estimator and tracking mechanism perform well on both synthetic and natural sequences. An overview of both algorithms is given in the following sections. The reader is referred to the relevant references for more

This publication has 0 references indexed in Scilit: