Model-based image analysis for unconstrained human upper-body motion

Recent research on model-based image coding for videotelephone and videoconferencing applications has mostly been concerned with head motion tracking and typically represents the human head as a 3D wire-frame model with texture-mapped surface features. However, the movements of the arms and hands are also important, particularly in sign language communication, and therefore should be included in the overall model. The paper describes a system which uses an articulated generalised cylindrical human model to track limb movements in a sequence of images. It outlines the closed-loop strategy developed to recognise and track human body motion and presents initial results for a complete implementation of the system