Linking Conversation Analysis and Motion Capturing: How to robustly track multiple participants?

Abstract

If we want to model the dynamic and contingent nature of human social interaction (e.g. for the design of human-robot-interaction), analysis and description of natural interaction is required that combines different methodologies and research tools (qualitative/quantitative; manual/automated). In this paper, we pinpoint the requirements and technical challenges for constituting and managing multimodal corpora that arise when linking Conversation Analysis with novel 3D motion capture technologies: i.e. to robustly track multiple participants over an extended period of time. We present and evaluate a solution to by-pass the limits of the current standard Vicon system (using rigid bodies) and ways of mapping the obtained coordinates to a human skeleton model (inverse kinematics) and to export the data into a format that is supported by standard annotation tools (such as ANVIL). 1. Introduction: Detecting