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

Abstract As urban areas expand, the importance of crowd control and population management increases significantly. Major events are organized and structured to guarantee participant safety and mitigate disruptions and emergencies. Analyzing the essential dynamics of congregational and crowd-dispersed flow presents significant challenges. This study employed a case analysis of religious events in Saudi Arabia, where the crowd varies annually between 1.5 million and 4 million participants in congregational rituals (Hajj and Umrah). The study employed a methodology for addressing prediction challenges and implemented two distinct algorithms: YOLOv4 and Deepsort. YOLOv4 focuses on object detection in images, while DeepSORT is utilized for tracking detected objects across frames. We derived the image dataset from a compilation of pilgrims’ image recordings, captured from various angles and locations during Hajj in 2019. The dataset underwent training utilizing YOLOv4 and Deepsort methodologies. Using YOLOv4 and DeepSORT to predict congregational and crowd spread-out flow led to the development of a high-performing system that can handle the complex problems that come with managing crowds during Hajj and Umrah. The research shows that it is possible to find and follow people in busy and changing environments. For YOLOv4, the metrics are 95.30% accuracy, 94.80% precision, 95.60% recall, and an F1-score of 95.20%. For DeepSORT, the metrics are 91.50% accuracy and 92.30% recall. These results highlight the potential of this approach to revolutionize crowd management strategies for one of the largest religious gatherings in the world. The model identified individuals and counted those entering or exiting a specific location, comparing these numbers to the venue’s capacity. It subsequently classified the crowd into three categories: low, medium, and high.