Measuring Visual Social Engagement from Proxemics and Gaze in the Real World

Abstract

This article presents the progress in measuring visual social engagement in the real-world, relevant for use by social robots in order to increase their social situational awareness. This study acts as an extension to the work by Webb et al. [10] on the visual social engagement metric by showcasing its applicability in real-world contexts. The original visual social engagement metric was developed using online game data, inherently limited by pre-programmed behaviours. In this research, we extend the metric's utility by applying the same set of social signals to real-world scenarios. To tailor the previously established interaction profiles for real-world data, human annotators were asked to pinpoint interaction initiation, removing the use of arbitrary distance values.