Real Time Head Gesture Recognition in Affective Interfaces

In this paper we present the affective message box, a dialog box that employs a real time head gesture recognition system as its input modality. Head nods and shakes correspond to “Yes/No” options on the dialog box. In addition, a confidence measure is inferred from a number of parameters extracted from gesture’s temporal patterns. While in current dialog boxes the input is either a definite yes / no, confidence measures provide applications with additional input from which an appropriate action can be selected. The head gesture recognition system we present here employs feature point displacements to describe basic head motions, which are in turn used with a state-space model to identify the head gesture. Our system detects head nods (95% accuracy) and shakes (91.67% accuracy) in real time (30 frames/sec) without the need for manual pre-processing or prior training.