AbstractVision is an active process. Humans actively sample their sensory environment via saccades, head turns, and body movements. Yet, little is known about active visual processing in real-world environments. Here, we exploited recent advances in immersive virtual reality (VR) and in-headset eye-tracking to show that active viewing conditions impact how humans process complex, real-world scenes. Specifically, we used quantitative, model-based analyses to compare which visual features participants prioritize over others while encoding a novel environment in two experimental conditions: active and passive. In the active condition, participants used head-mounted VR displays to explore 360º scenes from a first-person perspective via self-directed motion (saccades and head turns). In the passive condition, 360º scenes were passively displayed to participants within the VR headset while they were head-restricted. Our results show that signatures of top-down attentional guidance increase in active viewing conditions: active viewers disproportionately allocate their attention to semantically relevant scene features, as compared with passive viewers. We also observed increased signatures of exploratory behavior in eye movements, such as quicker, more entropic fixations during active as compared with passive viewing conditions. These results have broad implications for studies of visual cognition, suggesting that active viewing influences every aspect of gaze behavior – from the way we move our eyes to what we choose to attend to – as we construct a sense of place in a real-world environment.Significance StatementEye-tracking in immersive virtual reality offers an unprecedented opportunity to study human gaze behavior under naturalistic viewing conditions without sacrificing experimental control. Here, we advanced this new technique to show how humans deploy attention as they encode a diverse set of 360º, real-world scenes, actively explored from a first-person perspective using head turns and saccades. Our results build on classic studies in psychology, showing that active, as compared with passive, viewing conditions fundamentally alter perceptual processing. Specifically, active viewing conditions increase information-seeking behavior in humans, producing faster, more entropic fixations, which are disproportionately deployed to scene areas that are rich in semantic meaning. In addition, our results offer key benchmark measurements of gaze behavior in 360°, naturalistic environments.
Quantifying Gaze Behavior During Real-World Interactions Using Automated Object, Face, and Fixation Detection