CrowdShield: A Real-Time YOLOv11-Based System for Proactive Detection of Panic and Stampede in Crowded Spaces

Distance monitoring in crowded settings is important in preventing accidents and securing the safety of the population by detecting stampedes or a state of panic beforehand. In the paper, the author presents a novel crowding behavior monitoring system, CrowdShield that uses the YOLOv11object detector model to detect and classify human activities in highly populated areas. In contrast to traditional surveillance systems, which can be cumbersome to monitor manually, CrowdShield uses deep learning to identify key movements that are critical to the crowd like running, conflicting movement directions, and nonpedestrian objects. The system is divided into such behaviors as Running, Sitting, Standing, and Moving in Different Directions and allows the evaluation of the risk of stampede. CrowdShield offers real-time detection of objects and an interactive user interface to give immediate feedback to live video feeds or uploaded images. The results of the experiment indicate the potential of the system to detect panic-inducing behaviors with high accuracy, which is a new application of YOLOv11to the task of crowd safety. The solution will provide a scalable, automated method of crowd management in the open areas in advance, which also benefits the safety measures.