A neural paradigm for motion understanding

The main aim of this paper is to propose a new neural algorithm to perform a segmentation of an observed scene in regions corresponding to different m oving objects, by analysing a time-varying image sequence. The method consi sts of a classification step, where the motion of small patches is recovere d through an optimisation approach, and a segmentation step merging neighbo uring patches characterised by the same motion. Classification of motion is performed without optical flow computation. Three-dimensional motion param eter estimates are obtained directly from the spatial and temporal image gr adients by minimising an appropriate energy function with a Hopfield-like n eural network. Network convergence is accelerated by integrating the quanti tative estimation of the motion parameters with a qualitative estimate of d ominant motion using the geometric theory of differential equations.