Mary Hayhoe¹, Youjin Oh, Daniel Panfili, Nathaniel Powell; ¹University of Texas Austin

In natural environments it is unclear how attention is directed to information needed to avoid independently moving obstacles that are largely unpredictable. It is often assumed that attention to such objects is attracted by stimulus-driven signals, such as motion. However, retinal motion patterns are actively shaped by gaze behavior, by walkers’ gait, and by the structure of the environment. Little is known about the nature of the time-varying retinal stimulus since it has been difficult to gather this kind of data, and retinal object motion is confounded with the effects of self motion, adding complexity to the signal. In addition, familiarity with the probabilistic structure of natural environments might allow a more pro-active role, implicating both bottom-up and top-down strategies. We recorded gaze, head, and foot movements, together with 3D scene data and depth estimation, while subjects walked on crowded sidewalks. We extracted retina-centered motion statistics and gaze distributions in the scene. Because gaze is stabilized by the vestibular-ocular reflex, motion is always zero at the fovea and increases with eccentricity. We calculated the statistics of these retinal patterns and found obvious systematic changes in velocity distribution with the addition of pedestrians to the scene. This might allow independently moving objects to be detected as statistical outliers, suggesting that peripheral motion contrast might attract gaze. To test this, motion contrast caused by stationary 3D structure must be accounted for. We are currently analyzing the statistics of spatio-temporal contrast patterns caused by these three sources (self-motion, object-motion, and stationary 3D structure) of retinal motion. Preliminary data suggest that the statistical patterns from the different sources are easily discriminable. In addition, examination of gaze patterns suggest that walkers frequently monitor distant regions in the scene, supporting the role of proactive (top-down) strategies for pedestrian detection and avoidance as well.

Acknowledgements: Supported by: NIH EY05729, NSF 2318065