The University of Sussex

Active vision techniques for visually mediated interaction

A. Jonathan Howell, Hilary Buxton

In this paper we introduce adaptive vision techniques used, for example, in video-conferencing applications. First, we present the recognition of identity, expression and head pose using Radial Basis Function (RBF) networks. Second, we address gesture-based communication and attentional focus, using colour/motion cues to direct face detection and capture `attentional frames'. These focus the processing for Visually Mediated Interaction via an appearance-based approach with Gabor filter coefficients used as input to time-delay RBF networks. Third, we present methods for the gesture recognition and behaviour (user-camera) coordination in an integrated system.

Download compressed postscript file