Analysis of behaviors by speaker and audience by using time series model

Abstract

In this report, the dominant behaviors of the speaker and audience in the lecture are analyzed by using the time-series model. First, we detect as a feature for behavior the number of skin-colored pixels in face region of the speaker and audience. Next, we construct time series models for these features for the speaker and audience. Furthermore, we analyze the synchronization phenomena in the speaker and audience by comparing time series models.