Pedestrian Traffic: on the Quickest Path

TL;DR

This paper introduces a flood fill dynamic potential field method for pedestrian simulation that encourages agents to choose quickest paths around corners, aligning more closely with real pedestrian behavior by balancing realism and computational efficiency.

Contribution

It presents a novel flood fill approach that accounts for agent occupancy to model quickest path behavior, minimizing artifacts and maintaining low-density movement realism.

Findings

The method effectively models pedestrians avoiding the shortest path in favor of quicker routes.

It reduces artifacts compared to naive flood fill metrics.

The approach is computationally efficient and preserves low-density movement realism.

Abstract

When a large group of pedestrians moves around a corner, most pedestrians do not follow the shortest path, which is to stay as close as possible to the inner wall, but try to minimize the travel time. For this they accept to move on a longer path with some distance to the corner, to avoid large densities and by this succeed in maintaining a comparatively high speed. In many models of pedestrian dynamics the basic rule of motion is often either "move as far as possible toward the destination" or - reformulated - "of all coordinates accessible in this time step move to the one with the smallest distance to the destination". Atop of this rule modifications are placed to make the motion more realistic. These modifications usually focus on local behavior and neglect long-ranged effects. Compared to real pedestrians this leads to agents in a simulation valuing the shortest path a lot better than the quickest. So, in a situation as the movement of a large crowd around a corner, one needs an additional element in a model of pedestrian dynamics that makes the agents deviate from the rule of the shortest path. In this work it is shown, how this can be achieved by using a flood fill dynamic potential field method, where during the filling process the value of a field cell is not increased by 1, but by a larger value, if it is occupied by an agent. This idea may be an obvious one, however, the tricky part - and therefore in a strict sense the contribution of this work - is a) to minimize unrealistic artifacts, as naive flood fill metrics deviate considerably from the Euclidean metric and in this respect yield large errors, b) do this with limited computational effort, and c) keep agents' movement at very low densities unaltered.