System architecture and techniques for gesture recognition in unconstrained environments

Abstract

Controlling appliances in home environments by gestures is a step towards more intuitive and natural human computer interfaces. A brief overview of an existing vision based gesture recognition system and its architecture and details of ergonomic remote control of devices by gestures are clarified. The focus is on motion detection, object normalization and identification, modelling, and prediction of motion using a Kalman filter. The initialization problem of the Kalman filter of a vision based system for human motion tracking differs from initialization for physical systems, where manuals report measurement errors. One main aim was to develop the initialization and adequate Kalman model for human motion. Most aspects mentioned in the report were implemented in the ARGUS prototype.