Dynamic Bayesian networks for visual recognition of dynamic gestures

Abstract

Dynamic Bayesian networks are a powerful representation to describe processes that vary over time inside a stochastic framework. This paper describes an online visual recognition system to recognize a set of five dynamic gestures executed with the user's right hand using dynamic Bayesian networks for recognition. Gestures are oriented to command mobile robots. The system employs a radial scan segmentation algorithm combined with a statistical-based skin detection method to find the candidate face of the user and to track his right-hand. It uses four simple features to describe the user's right-hand movement. Our system is able to recognize these five gestures in real-time with an average recognition rate of 84.01%, better result than using hidden Markov models for recognition.