Samantha A Montoya¹ (samantha.montoya@pennmedicine.upenn.edu), Zachary J Kelly¹, Sophia Mirabal¹, Ruby S Bouhassira¹, Carlyn Patterson Gentile², Alan A Stocker¹, Geoffrey K Aguirre¹; ¹University of Pennsylvania, ²Children's Hospital of Philadelphia

In contrast to spatial vision, few studies have examined the temporal statistics of natural vision, especially having accounted for eye movements. We measured temporal variation in natural, gaze-centered vision, extending to the nearly full operational range of human temporal sensitivity. Five participants performed a set of 10 scripted, 4 minute long, real-world activities (ambulatory and seated, indoor and outdoor) while wearing custom-built spectacles equipped with a high-speed camera (120 FPS, 150° FOV) and synchronous eyetracking. Here, we discuss temporal statistics for walking indoors and working at a desktop computer. Videos were transformed to the perspective of the observer’s visual field prior to calculating temporal spectral power densities (SPDs) of luminance contrast over 5s intervals with 3° spatial resolution. Analyses were conducted with and without correction for eye movements by performing virtual foveation. We summarized the data as the alpha exponent of a 1/f^alpha model fit to the SPD, along with the total temporal contrast variance. In analyses omitting adjustment for eye movements, temporal SPDs were well described by 1/f^alpha, with exponents close to 2. With virtual foveation, we observed a whitening of the power spectrum at higher temporal frequencies. For walking indoors, during which tracking eye movements are prevalent, SPD whitening occurred above ~35 Hz. While working at a computer, the whitening began at ~20 Hz. Accounting for eye movements also introduced spatial structure to the pattern of temporal variation, with systematic variation between central and peripheral field locations. The particular properties of temporal variation in natural vision are important for efficient coding models, and studies of the features of stimuli that produce visual discomfort. We find that temporal variation systematically departs from an assumed 1/f^2 distribution in the presence of eye movements, and that these effects vary with behavior and visual field location.

Acknowledgements: Funding was provided by NIH 1R01EY036255