Inferring Intent and Action from Gaze in Naturalistic Behavior
We constantly move our gaze to gather acute visual information from our environment. Conversely, as originally shown by Yarbus in his seminal work, the elicited gaze patterns hold information over our changing attentional focus while performing a task. Recently, the proliferation of machine learning algorithms has allowed the research community to test the idea of inferring, or even predicting action and intent from gaze behaviour. The on-going miniaturization of gaze tracking technologies toward pervasive wearable solutions allows studying inference also in everyday activities outside research laboratories. This paper scopes the emerging field and reviews studies focusing on the inference of intent and action in naturalistic behaviour. While the task-specific nature of gaze behavior, and the variability in naturalistic setups present challenges, gaze-based inference holds a clear promise for machine-based understanding of human intent and future interactive solutions.