Predicting Crowd Movement Trajectories: A Graph Network-Based Approach Informed by Simulation Data

As-built assessment of transport hubs, based on large-scale simulation and passenger big data, has been an active area of research recently. Exploring the direction of crowd flow at future moments within a building is crucial for the rational assessment of the layout within the building. To tackle this problem, this paper proposes a two-stage framework for predicting passenger trajectories. In the first stage, our graph LSTM models the passenger’s trajectory sequence with the aim of capturing the temporal and spatial correlations in the passenger’s historical trajectory. In the second stage, different attentional weights are assigned to the previously embedded information to capture the interaction between pedestrians and the surrounding crowd. Finally, an LSTM decodes the information encoded for the passenger, predicting multiple plausible future trajectories. Experiments show that our model achieves excellent performance in the metrics ADE and FDE compared to the existing baseline of pedestrian trajectory prediction models.