GRaD: Gesture Recognition and Disambiguation Framework for Unconstrained, Real-Life Scenarios

Automatic gesture recognition, used in humanrobot and human-computer interaction, has been an active area of research for at least three decades, but there are still recurring issues that hinder its application in unconstrained, real-life scenarios. Interpersonal differences in gesture performance present a prominent issue, which limits the gesture recognition process. Furthermore, these differences are in some cases irrelevant with regard to the message to be conveyed. This work presents a framework for gesture recognition and disambiguation. The goal of this framework is to make the mechanisms of gesture recognition itself as abstract as possible, thus lowering the effect of interpersonal differences, in order to be able to use a simple recognition algorithm. Features that are ignored during the recognition can be used in the disambiguation step, in case they are relevant for this particular message.