Inferring state transition from bystander to participant in free-style conversational interaction

We propose a novel method for inferring the state transition from bystander to participant in free-style conversational interactions, using physical behaviors acquired from cameras and a microphone. Although existing methods address participants and a presenter, these methods do not consider bystanders, who play an important role in the interaction. In the research field of cognitive science, the existing model considers psychological aspects of changing from bystander to participant. However, this model is difficult to implement because inferring the psychological aspects of bystanders is a challenging task. Instead of using psychological aspects, our method exploits physical behaviors such as standing position, facial direction, and voice direction. We analyzed the suitable parameters of the behavior to increase performance in inferring state transitions, using datasets collected from poster presentations.