A human data-driven interaction estimation using IoT sensors for workplace design

Abstract

Accurate face-to-face interaction estimation is required for a successful data-driven design in workplaces. In previous studies, various sensor-based interaction estimation methods which use proximity and speaking data have been developed. However, these data alone cannot confirm the presence of interactions because non-interacting users also engage in speaking activities. This study aims to develop a novel turn-taking pattern-based interaction estimation (i.e., TIE) framework that integrates turn-taking with location data. The framework estimates interactions in three steps: 1) co-location estimation using a Bluetooth Low Energy beacon; 2) speaking-turn ascertainment through volume-based speaker identification; and 3) interaction group recognition based on turn-taking pattern analysis. Using three different experimental scenarios, the interaction estimation accuracy of the framework was demonstrated to be 77.7%. In the absence of co-location estimation errors, the interaction estimation accuracy increases to 95.5%. The demonstration results indicate that the TIE framework has potential for accurate interaction estimation in workplaces.