Measuring and Modelling Crowd Flows - Fusing Stationary and Tracking Data

TL;DR

This paper introduces a method to detect congestion and model crowd flow during mass-sports events by fusing stationary detector data with athlete tracking data, applicable in real-time or retrospective analysis.

Contribution

It presents a novel data fusion approach combining stationary and floating data sources for crowd flow modeling using a macroscopic model with a triangular fundamental diagram.

Findings

The method effectively detects congestion and jam-fronts.

Minimal data requirements make it practical for real-time use.

Synthetic data validation shows robustness and accuracy.

Abstract

The two main data categories of vehicular traffic flow, stationary detector data and floating-car data, are also available for many Marathons and other mass-sports events: Loop detectors and other stationary data sources find their counterpart in the RFID tags of the athletes recording the split times at several stations during the race. Additionally, more and more athletes use smart-phone apps generating track data points that are the equivalent of floating-car data. We present a methodology to detect congestions and estimate the location of jam-fronts, the delay times, and the spatio-temporal speed and density distribution of the athlete's crowd flow by fusing these two data sources based on a first-order macroscopic model with triangular fundamental diagram. The method can be used in real-time or for analyzing past events. Using synthetic "ground truth" data generated by simulations with the Intelligent-Driver Model, we show that, in a real-time application, the proposed algorithm is robust and effective with minimal data requirements. Generally, two stationary data sources and about ten "floating-athlete" trajectories per hour are sufficient.