Detection and Modeling of Pedestrian Groups Using Laser Sensor Trajectories

Abstract. This study proposes a pedestrian behavior modeling framework that explicitly incorporates the existence and dynamics of pedestrian groups. Using high-resolution laser sensor data collected in a hospital atrium, spatiotemporal features describing interpersonal distance, relative speed, and walking direction are extracted from pedestrian trajectories. Based on these features, machine learning techniques are applied to identify pedestrian groups, with Support Vector Machines (SVM) and Random Forests used as baseline models. The results show that the SVM achieves stable and accurate group identification under complex walking conditions. Building on the detected group structures, the pedestrian behavior model is extended to represent group-related psychological stress, including individual stress, stress induced by other pedestrians, and stress arising from group dispersion. Model parameters are calibrated using laser-derived trajectory data with individual attributes such as staff roles and mobility aid usage. The proposed model reproduces observed walking trajectories with high fidelity, achieving position errors within 80 cm for approximately 80% of pedestrians. Finally, the model is applied to spatial evaluation of pedestrian environments by mapping distributions of estimated psychological stress. The results reveal elevated stress levels in waiting areas such as reception zones, while group dispersion stress is more prominent in low-density regions where groups tend to spread out. These findings demonstrate that incorporating pedestrian group behavior enhances the interpretability and applicability of pedestrian models for evaluating and designing public spaces.