Giselle R. Urquijo¹, Mark D. Lescroart¹; ¹University of Nevada, Reno

As we move through our environment, visual patterns of motion constantly project onto our retinas. Recently, studies investigating patterns of retinal flow during natural locomotion revealed that, as we walk, we experience patterns of spiral expansion. However, they have dominantly focused on structured walking tasks in one setting. In order to characterize the diversity of patterns of motion that occur in day-to-day life, we analyzed the Visual Experience Dataset which provides hundreds of hours of egocentric video with gaze tracking as humans perform naturalistic behavior. To subdivide the data into tasks, we binned the dataset’s labels into six broad categories of activities: locomotion, manual tasks, passive, screens, social, and sports. We estimated gaze-centered optic flow during fixation epochs using Recurrent All-Pairs Field Transforms (RAFT). From these samples of optic flow we removed full-field slip by factoring out horizontal and vertical vectors. We then performed principal component analysis (PCA) on them. Consistent with past work, we find that the first two principal components (PCs) across all tasks are patterns of expansion and spiral motion. We then computed PCA on the optic flow samples from each task separately. We found that locomotion and sports, which are dominated by self-motion, have lower dimensional patterns of optic flow compared to tasks where there is little to no self-motion i.e. screens, social, passive, and manual tasks. Within locomotion, we find that a pattern of vertical divergence, i.e. leftward and rightward motion away from the vertical meridian, appears within the first three PCs. This suggests that people tend to fixate on vertical boundaries in an environment as they approach them. This is a previously unreported motif in visual motion which may be manifest in tuning of neurons and/or in the organization of the brain.

Acknowledgements: Supported by NSF CAREER #2340895 to M.D.L.