Fig 1.
Illustration of the relevant causes and events before the crowd crush during the Seoul Halloween.
The evolution of crowd dynamics was influenced by a multitude of endogenous and exogenous factors. Exogenous factors included elements such as flawed crowd management strategies, poor visibility at night, and hazardous geometry near the alley. Conversely, endogenous factors encompassed aspects such as large crowd size, heightened panic, and intense pressure.
Fig 2.
Evolution of the situation in the Itaewon neighborhood on October 29, 2022.
(a) Evolution of the actual populations in the three areas. (b) A condensed timeline and locations of police calls received at Itaewon-Dong.
Fig 3.
Model and simulation of the Seoul Halloween crowd-crush disaster.
(a) Simplified geometry setting for simulation. (b) Directions of the six pedestrian streams, among which the two bidirectional streams in the alley (highlighted in red) are crucial during the reproduction of the crowd crush. (c) Distribution of (ρs − ρn)/(ρs + ρn) at t = 230 min, where ρn represents the overall density from the street north of the alley and ρs represents the overall density from the street south of the alley. (d) Heatmap of overall density in the critical area at t = 230 min (unit: ped/m2). (e) Heatmap of crowd pressure in the critical area at t = 230 min (unit: N/m). (f) Time evolution of the three kinds of forces that were exerted on each pedestrian in the alley (y ∈ [15, 55] m). (g) Time evolution of the average density and pressure in the alley. (h) Time evolution of the VE in the alley.
Fig 4.
Time evolution of several indicators of dangerous level in the three scenarios.
(a) The original scenario. (b) Crowd management strategy assigned. (c) Increasing pedestrian flow in the geometry based on scenario (b). (d) The average overall density in the alley. (e) The averaged crowd pressure P2 in the alley. (f) The VE [28] in the alley.
Table 1.
Comparison of OD estimation in Scenario (a) and Scenario (c).
The dynamic OD estimation is implemented in both scenarios to evaluate the network capacity. A considerable increase in the total inflow rate is demonstrated by PCT (percentage).
Fig 5.
Representation of crowd state prediction and estimation.
(a) The dynamic OD estimation provides an inflow rate for each pedestrian stream, determined through the introduction of a congestion detector. (b) The boundary conditions include the OD information and solid boundaries. (c) The model parameters are derived from an extensive review of empirical studies. (d) Utilizing the input data, a hydrodynamic model is employed that explicitly takes into account the route strategy and aggregated pressure. (e) The dynamic evolution of key crowd states and indicators facilitates the identification of potential crowd risks.
Table 2.
Summary of the inputs and outputs of the model.