Towards Automatic Body Language Annotation

Abstract

This paper describes a real-time system developed for the derivation of low-level visual cues targeted at the recognition of simple hand, head and body gestures. A novel, adaptive background subtraction technique is presented together with a tool for monitoring repetitive movements, e.g. fidgeting. To monitor subtle body movements in an unconstrained environment, active cameras with pan, tilt and zoom capabilities must be employed to track an individual's actions more closely. This paper then explores a means of detecting small and large scale human activity within images produced from active cameras that may be reoriented during monitoring