Pedestrian Flow Model Based on Cellular Automata Under Visual Trajectory and Multi-Scenario Evacuation Simulation...

Area of Science:

• Computational Social Science
• Computer Vision
• Traffic Engineering

Background:

• Accurate pedestrian flow modeling is vital for public safety and emergency planning.
• Existing models often lack real-world data integration for nuanced crowd behavior.

Purpose of the Study:

• To develop an interdisciplinary method combining computer vision and cellular automata (CA) for precise pedestrian flow simulation.
• To enhance evacuation models with data-driven transition probabilities and behavioral factors.

Main Methods:

• Collected and analyzed unidirectional pedestrian flow video data using YOLOv11 and DeepSORT for high-precision trajectory tracking.
• Converted trajectory data into data-driven transition probabilities for the CA model's Moore neighborhood.
• Developed an improved evacuation model incorporating real-data transition probabilities, speed-density, panic, life value, and hazard dynamics.

Main Results:

• Simulations revealed moderate panic can shorten evacuation times, but excessive panic leads to disordered behavior.
• Group movement is limited by the slowest individual, and faster hazard sources decrease the number of safe pedestrians.
• The model provides insights into crowd dynamics under various emergency scenarios.

Conclusions:

• The integrated computer vision and CA approach offers a novel method for refined pedestrian evacuation simulation.
• Findings support improved safety design and emergency management strategies in public spaces.
• Data-driven transition probabilities significantly enhance the realism of CA-based evacuation models.