Simulation of collision avoidance behavior in crowd movement by data-driven approach
Researchers have developed a new data-driven model called CPGAN (Collision-Penalized GAN) to improve crowd movement simulations. This model integrates a collision mechanism into its loss function, specifically addressing the high collision rates often seen in bidirectional and multidirectional pedestrian flows. By using a novel lateral-acceleration-based collision loss and Voronoi-based motion feature extraction, CPGAN significantly reduces opposite-direction collisions and accurately reproduces lane formation. AI
IMPACT Improves accuracy of crowd simulation models, potentially aiding in urban planning and safety management.