Inferring State Transition from Bystander to Participant in Free-style Conversational Interaction

We use cameras and a microphone to measure standing positions, face orientations, and voice directions. These physical behaviors are used to infer when a bystander transitions into participating in a conversation.

At a poster presentation, someone watching from a short distance may later join the conversation. We study how to detect this change from bystander to participant using behavior recorded by cameras and a microphone. Earlier methods mainly focused on the presenter and people who were already participating.

A machine cannot easily infer a person's intention to join a discussion. Instead of trying to measure these psychological states directly, our method uses observable cues: where people stand, which way their faces point, and the direction from which their voices are heard.

Using data collected at poster presentations, we analyzed which values of these behavioral measurements are suitable for detecting a change in participation. The aim is to use position, face orientation, and voice direction to infer how someone becomes involved in a conversation, rather than to directly measure what they are thinking.

• 神原 隆宏, 西山 正志, 岩井 儀雄,
  対話型プレゼンテーションにおける非参与者から参与者への状態遷移推定,
  ヒューマンインタフェース学会論文誌, Vol. 22, No. 1, pp. 13 - 24, February 2020.
• Tatsuya Era, Hiroki Yoshimura, Masashi Nishiyama, Yoshio Iwai,
  Inferring State Transition from Bystander to Participant in Free-style Conversational Interaction, Proceedings of 14th IEEE International Conference on Advanced Video and Signal based Surveillance (AVSS), pp. 1 - 6, August 2017.

• Bystander Inference Results