An analysis of manoeuvring in dense crowds

Abstract

While walking through a crowd, a person balances several desires, such as reaching some goal position, avoiding collisions with others, and conserving energy. Crowd models generally try to mimic this behaviour by planning short paths that avoid collisions. However, when the crowd density increases, choosing a collisionfree path becomes more difficult. In such highdensity crowds, one can observe torso twists; people rotate their upper body to decrease their width perpendicular to the motion path, in order to squeeze through narrow spaces between other crowd members. In this paper we investigate this behaviour, by recording and analysing dense crowds. We show that the paths chosen by the participants can be predicted by generalized Voronoi diagrams, and identify relations between instantaneous speed and look-ahead distance, and between the participants' torso orientations and goal positions.

Torso Crowds

“…These edges form the medial axis between the agents, and thus represent a more or less comfortable path that maximizes local clearance. St€ uvel et al [23] have observed that people in dense crowds indeed follow such paths (also see Fig. 3).…”

“…The edges of such a diagram represent the path of maximum clearance between agents; intuitively this corresponds well with the desire of people to minimize perceived effort when walking [21], [22]. St€ uvel et al [23] showed that in a dense crowd people indeed move along such paths. The Explicit Corridor Map method by Geraerts [24] uses city-scale generalized Voronoi diagrams for path planning.…”

“…Our crowd simulation system considers the torso as the main moving element. The algorithm is based on the findings by St€ uvel et al [23], [27], who observed and recorded dense crowd behaviour. In their experiment, participants were given the task of manoeuvring through the crowd to predefined points.…”

“…The movement of the crowd was recorded using a motion capture system, and these data serve as a ground truth for the behaviour of people actively manoeuvring through dense crowds. We refer to St€ uvel et al [23] for the details of the experiment. Our Torso Crowd model is designed to support the observed motions of the crowd-escaping participants, and believable simulation of essentially stationary people.…”

“…People generally move in the direction of their goal, but divert from the shortest path when it is obstructed or when an alternative path is significantly more comfortable. Averaging at 0.4 m/sec, the traversing speed through a dense crowd is relatively low [27]. This, combined with the dynamic nature of crowds and possibly a lack of overview of the situation, makes long-distance planning of exact paths to the goal impractical.…”