Predicting congregational and crowd spread-out flow using YOLOv4 and DeepSORT

Area of Science:

• Computer Vision
• Artificial Intelligence
• Public Safety

Background:

• Urban expansion necessitates advanced crowd control and population management strategies.
• Major events require structured organization for participant safety and emergency mitigation.
• Analyzing congregational and dispersed crowd flow dynamics presents significant challenges.

Purpose of the Study:

• To develop and evaluate a predictive system for managing large crowds during religious events.
• To address challenges in predicting and analyzing crowd flow dynamics.
• To enhance safety and efficiency in managing mass gatherings.

Main Methods:

• A case study analyzing religious events (Hajj and Umrah) in Saudi Arabia with 1.5-4 million participants.
• Implementation of two distinct algorithms: YOLOv4 for object detection and DeepSORT for object tracking.
• Training the YOLOv4 and DeepSORT models on a dataset of pilgrim images captured during Hajj 2019.

Main Results:

• YOLOv4 achieved 95.30% accuracy, 94.80% precision, 95.60% recall, and 95.20% F1-score.
• DeepSORT achieved 91.50% accuracy and 92.30% recall.
• The system successfully identified and tracked individuals, counted entries/exits, and classified crowd density (low, medium, high).

Conclusions:

• The developed system effectively manages complex crowd dynamics during major religious gatherings.
• The YOLOv4 and DeepSORT approach demonstrates the potential to revolutionize crowd management strategies.
• Accurate individual identification and flow analysis are achievable even in crowded, dynamic environments.