Crowd Detection Camera System for Virus Prevention

This research proposes an advanced framework for monitoring social distancing in public spaces using computer vision techniques. The system leverages the YOLO v3 (You Only Look Once, Version 3) object detection algorithm, optimized for high accuracy in identifying individuals in diverse and crowded environments. YOLO v3 is chosen for its real-time processing and precise detection capabilities. By fine-tuning the model with a diverse dataset, the system improves performance in challenging conditions, such as varying scales and occlusions. Once individuals are detected, the DeepSORT (Simple Online and Realtime Tracking) algorithm is used to track them across consecutive video frames. DeepSORT effectively combines motion and appearance data, ensuring accurate tracking of individuals, even in dense crowds. This allows the system to maintain consistent identities over time. The integrated YOLO v3 and DeepSORT system measures the distances between people and detects violations of social distancing guidelines in realtime. By setting a threshold distance, instances of noncompliance can be flagged for further action. Tested on various public datasets and real-world video feeds, the system demonstrates high detection accuracy and robust tracking performance. The approach provides a practical solution for monitoring social distancing, helping enforce public health guidelines with real-time alerts and statistical insights. Index Terms—YOLO v3, DeepSORT, Fine-tuning, Computer Vision