Universal Power Law Governing Pedestrian Interactions

We address a long standing hypothesis as to what is the fundamental nature of the laws that drive human interactions in a crowd. Here, we take a "big-data" approach to resolve this question, performing the largest meta-study of existing crowd data to date. The resulting analysis reveals a simple, universal power law governing pedestrian behavior. Applications of this law include simulating crowd flows and predicting pedestrian behaviors, which could potentially lead to safer buildings, improved pedestrian tracking, and ultimately new ways to understand what drives human behaviors.

Main Text (4 pages, 2MB)

Supplemental Material (7 pages, 1MB)

Agents egress through a 5m wide bottleneck [Download]

Self-directed agents form collective patterns [Download]

Bi-directional flow through a corridor [Download]

Two perpendicularly crossing flows [Download]

Agents walk to antipodal points in a circle [Download]

Agents walk across large concentric circles. [Download]

Agents exit through a narrow doorway [Download]

Two animated pedestrians swap positions [Download]
C++ implementation of the derived simulation model [Download]

Python implementation of the derived simulation model (depends on NumPy) [Download]

Bottleneck and Outdoor crowd datasets used in our study [Download]

The authors can be reached via ioannis@cs.umn.edu, bskinner@anl.gov, and sjguy@cs.umn.edu respectively.