CPGAN: Data-Driven Crowd Simulation Reduces Collisions
A novel crowd simulation framework, known as CPGAN (Collision-Penalized Generative Adversarial Network), has been introduced to tackle the issue of elevated collision frequencies in bidirectional pedestrian movements. This model employs a collision loss function based on lateral acceleration and a motion feature extraction technique grounded in Voronoi diagrams within the GAN structure. In tests involving bidirectional situations, CPGAN markedly decreases collisions between pedestrians moving in opposite directions compared to current models. The objective of this research is to bolster pedestrian safety management and optimize facility layouts by improving the accuracy of trajectory predictions while reducing collision hazards.
Key facts
- CPGAN stands for Collision-Penalized Generative Adversarial Network.
- The model incorporates a lateral-acceleration-based collision loss function.
- A Voronoi-based motion feature extraction approach is used.
- CPGAN is evaluated in bidirectional flow scenarios.
- The collision loss significantly reduces opposite-direction pedestrian collisions.
- The research addresses high collision rates in data-driven crowd simulation.
- The work is published on arXiv with ID 2605.31210.
- The study focuses on pedestrian safety management and facility layout optimization.
Entities
Institutions
- arXiv