This paper presents a new multimodal system for group dynamics and interaction analysis. The framework is composed of a mic array and multiview video cameras placed on a digital signage display which serves as a support for interaction. We show that visual information processing can be used to localize nonverbal communication events and synchronized with audio information. Our contribution is twofold: 1) we present a scalable portable system for multiple people multimodal interaction sensing, and 2) we propose a general framework to model A/V multimodal interaction that employs speaker diarization for audio processing and hybrid dynamical systems (HDS) for video processing. HDS are used to represent communication dynamics between multiple people by capturing the characteristics of temporal structures in head motions. Experimental results show real-world situations of group communication processing for joint attention estimation. We believe the proposed framework is very promising for further research.
展开▼
