Simulating pedestrian movement in T-junction corridor: A novel vision-driven convolutional graph attention model with a dataset from experiments

Abstract

With the rapid pace of urbanisation, the safety and efficiency of pedestrian traffic face increasingly severe challenges, particularly in densely populated public areas. Optimising pedestrian flow effectively has therefore become a critical issue requiring urgent attention. To address this challenge, this study proposes a vision-driven convolutional graph attention model (VI-CGAM) for simulating pedestrian future movements. The VI-CGAM comprises three components: a visual information-based interaction graph construction module, a graph attention network-based spatial feature extraction module, and a convolutional neural network-based temporal feature extraction module. In addition, this study conducted a series of experiments on pedestrian diverging and merging in T-junction of varying widths, collecting key data using unmanned aerial vehicle to construct a novel T-junction pedestrian movement dataset. The results show that VI-CGAM accurately simulates pedestrian trajectories, as well as the density and flow rate characteristics in key areas. Furthermore, ablation studies were conducted to demonstrate the effectiveness of each component of VI-CGAM. This study provides a robust algorithmic support and valuable data resources for intelligent transportation systems, with the potential to improve pedestrian flow management and safety planning in public spaces.