Automatic classification of gestures: A context-dependent approach

Gestures represent an important channel of human communication, and they are “co-expressive” with speech. For this reason, in human-machine interaction automatic gesture classification can be a valuable help in a number of tasks, like for example as a disambiguation aid in automatic speech recognition. Based on the hand gesture categorization proposed by D. McNeill in his reference works on gesture analysis, a new approach is here presented which classifies gestures using both their kinematic characteristics and their morphology stored as parameters of the templates pre-classified during the training phase of the procedure. In the experiment presented in this paper, an average of about 90% of correctly classified gesture types is obtained, by using as templates only about 3% of the total number of gestures produced by the subjects.