Improved visual-information-driven model for crowd simulation and its modular application

TL;DR

This work advances data-driven crowd simulation by introducing an improved visual-information-driven (IVID) framework that emphasizes full-vision cues and exit information to achieve strong generalizability across bottleneck, corridor, corner, and T-junction modules. The model combines feature extraction (velocity, social interactions via Radar-NN, full-vision visual rays, and exits) with a three-layer temporal convolutional VPNN and a rolling forecast to generate continuous trajectories. A modular approach enables assembling complex, composite scenarios from simple modules, with simulations in composite settings producing fundamental diagrams that align closely with real-world data and outperform the classical SF baseline. The findings highlight visual information as a critical, generalizable driver of pedestrian navigation and demonstrate practical potential for designing pedestrian facilities and safety systems using data-driven, modular crowd simulation. $w=8$, $q=8$, dilations $1,2,4$, $D_e \\in \\{20,100\}$ m, $\beta \\in \\{5^\circ,10^\circ,15^\circ,18^\circ\}$, $R=1.2$ m, $\alpha=18^\circ$ were used in the experiments.

Abstract

Data-driven crowd simulation models offer advantages in enhancing the accuracy and realism of simulations, and improving their generalizability is essential for promoting application. Current data-driven approaches are primarily designed for a single scenario, with very few models validated across more than two scenarios. It is still an open question to develop data-driven crowd simulation models with strong generalizibility. We notice that the key to addressing this challenge lies in effectively and accurately capturing the core common influential features that govern pedestrians' navigation across diverse scenarios. Particularly, we believe that visual information is one of the most dominant influencing features. In light of this, this paper proposes a data-driven model incorporating a refined visual information extraction method and exit cues to enhance generalizability. The proposed model is examined on four common fundamental modules: bottleneck, corridor, corner and T-junction. The evaluation results demonstrate that our model performs excellently across these scenarios, aligning with pedestrian movement in real-world experiments, and significantly outperforms the classical knowledge-driven model. Furthermore, we introduce a modular approach to apply our proposed model in composite scenarios, and the results regarding trajectories and fundamental diagrams indicate that our simulations closely match real-world patterns in the composite scenario. The research outcomes can provide inspiration for the development of data-driven crowd simulation models with high generalizability and advance the application of data-driven approaches.This work has been submitted to Elsevier for possible publication.

Figures (10)

• Figure 1: Overall framework of the improved visual-information-driven model. (a) Feature extraction. (b) Velocity-prediction neural network. (c) Rolling forecast.
• Figure 3: Schematic of the velocity-prediction neural network. (a) Architecture of the velocity-prediction neural network. (b) Architecture of the TCN layer bai2018empirical.
• Figure 4: Schematic of the modular approach. (a) An example scenario comprising a bottleneck and a corner, with the bottleneck module represented by the green area and the corner module represented by the blue area. (b) Schematic illustrating the extraction of visual and exit information for a pedestrian located within the bottleneck module. (c) Schematic illustrating the extraction of visual and exit information for a pedestrian situated within the corner module.
• Figure 5: Sketches and snapshots of the controlled experiments (https://ped.fz-juelich.de/da/doku.php). (a) bottleneck. (b) corridor. (c) corner. (d) T-junction.
• Figure 6: Mean ADE, FDE and TTE in various parameter combinations. The x-axis denotes the combinations of $D_e$ and $\beta$.