Abstract
To explain the collective behavior of human crowds, we begin by characterizing the local interactions between individual pedestrians. One such interaction is following, which may lead to coherent crowd motion. In previous work, we found that the follower matches the leader's speed (Rio et al., 2014) and heading direction (Dachner & Warren, 2014), and investigated the visual coupling between them. Dachner & Warren (VSS 2016) proposed that the follower's speed and heading are controlled by nulling the leader's optical expansion and change in bearing direction, depending on their relative positions. Here we attempt to simulate Dachner & Warren's (2016) data using a vision-based dynamical model with only these optical variables as input. 12 participants were instructed to follow a virtual pole (40 cm width), which appeared in three initial positions relative to the participant (0°, 30°, 60° from straight ahead). Mid-way through a trial, the pole changed its rate of expansion, bearing direction, or both; these optical variables specified a change in the pole's speed, direction, or both. Participants walked freely wearing a head-mounted display, while their head trajectory was recorded. Based on the same visual input, the model outputs a simulated trajectory for each trial. Preliminary results demonstrate a good fit to the participant data. Incorporating visual thresholds for expansion rate and angular velocity from Regan & Hamstra (1998) further improves the model's fit. The results support the hypothesis that, when the pole is in front of the participant, optical expansion controls speed and change in bearing controls heading, while when the pole is to one side, these relations reverse. Models of crowd behavior are typically not based on visual information. A vision-based model more closely simulates following behavior and provides insight into the visual coupling between pedestrians in a crowd. Supported by NSF BCS-1431406 and NIH T32-EY018080-08.
Meeting abstract presented at VSS 2017