Abstract
When perceiving groups of features, we may rely on ensemble codes of pooled feature signals that precisely describe the average (i.e., the gist) but limit access to the details (Haberman & Whitney, 2007). Here we demonstrate the use of an ensemble code to perceive the average direction of a walking crowd of people. We briefly presented crowds of twelve non-overlapping point-light walkers containing different amounts of direction variability (i.e., walkers had identical or increasingly divergent headings). Observers then estimated the average direction of the crowd. We used an equivalent noise model (Dakin, 2001) to measure response variability as a function of direction variability. This allowed us to estimate the number of walkers pooled to perceive crowd direction and unambiguously reveal the use of ensemble coding. To separately measure the contributions of human form and local motion to perceiving a group's direction, we also presented crowds of static walkers (form without local motion) and point-scrambled walkers (local motion without form). Results were consistent with ensemble coding. On average, observers pooled the directions of five walkers to estimate the direction of a crowd. Direction estimates for crowds with moderate variability were only as noisy as estimates with a homogeneous crowd for all observers, and only as noisy as estimates with a single walker for most observers. Increases in crowd variability only raised response variability when crowd variability was already high. This pattern suggests that observers involuntarily perceived the average direction of the crowd. The ensemble code incorporated both form and local motion, as direction estimates were worse with form or local motion alone than with both combined. In summary, the visual system creates an ensemble code for perceiving the average direction of a crowd. This summary representation is likely generated in high-level visual areas after the convergence of form and motion signals.
Training grant in vision science (NIH NEI T32 EY007043).