Abstract Collective motion in animal groups emerges from local visual interactions, but the network structure underlying human crowd dynamics remains unexplored. Here, we develop a method to reconstruct dynamic visual influence networks from motion-capture data on walking human crowds, using the time-dependent delayed directional correlation with visibility constraints, specifically visual field size and visual occlusion. We introduce two network measures, direct influence (local leadership) and branching influence (global leadership), that reveal distinct spatial gradients. Front-positioned individuals exert greater global influence than rear-positioned ones, and such spatial position explains a substantial proportion of leadership variance in the group. The results show why a small number of individuals can influence the motion of large crowds. These findings establish quantitative principles for crowd networks that parallel leadership hierarchies in animal flocks, while providing evidence-based foundations for human crowd management strategies.
Yoshida et al. (Wed,) studied this question.