Head pose estimation and movement analysis for speech scene

Abstract

Public speaking is an essential skill for a large variety of professions and in everyday life. However, it is difficult to master the skills. This paper focuses on the automatic assessment of nonverbal facial behavior of public speaking and proposes simple and efficient method of head pose estimation and motion analysis. We collected nine speech scenes of the recitation contest in a Japanese high school, and applied the proposed method to evaluate the performance. As for the head pose estimation, our method was obtained acceptable accuracy for the speech scene. Proposed motion analysis method can be calculated frequencies and moving ranges of head motion. As the result, it was found that there is correlation between the moving range and eye contact score.