Abstract
The collective behavior of human crowds is thought to emerge from the local interactions between individual pedestrians, such as the coordination of walking speed and direction (heading). This coordination is guided by visual information about the motion of one’s neighbors. A critical interaction assumed by many models of collective behavior is the alignment of heading with neighbors. Using a behavioral dynamics framework, we simulated experimental data on pairs of pedestrians to formulate a dynamical model of heading alignment. On each trial, a “follower” either walked behind or beside a “leader” for 20m, while the leader made two unpredictable turns. The initial distance and position, turn sequence, magnitude, and timing varied. Head trajectories were recorded with a 16-camera motion capture system. Results show that the follower’s heading direction was closely coordinated with that of the leader (mean cross-correlation r = 0.92, mean delay between pedestrians = 984 ms for behind; r = 0.93, delay = 219 ms for side-by-side). We tested four models against this data to characterize the follower’s change in heading during alignment, using a least-squares criterion. In the simplest model, the follower's angular acceleration is proportional to the sine of the heading difference (mean r = 0.72, gain k = 0.81 for behind; r = 0.70, k = 0.69 for side-by-side); additional terms did not improve fit. An alternative model that incorporates the mean time delays yielded similar gains (mean r = 0.71, k = 1.17 for behind; r = 0.71, k = 1.21 for side-by-side), thus generalizing across different leader-follower positions. The results show heading alignment is controlled by nulling the difference in heading with a neighbor. This provides a cognitively-grounded model of heading alignment that may be generalized to larger-scale crowd dynamics. We are currently investigating the optical variables that control these interactions.
Meeting abstract presented at VSS 2015