Abstract: With the recent outbreak and rapid transmission of the COVID-19 pandemic, the need for the public to follow social distancing norms and wear masks in public is only increasing. According to the World Health Organization, to follow proper social distancing, people in public places must maintain at least 3ft or 1m distance between each other. This project focuses on a solution to help enforce proper social distancing in public using YOLO object detection on video footage and images in real time. The experimental results shown in this paper infer that the detection of human subjects based on YOLO has stronger robustness and faster detection speed as compared to its competitors. Our proposed object detection model achieved a mean average precision score of 94.75%. The network ensures inference speed capable of delivering real-time results without compromising on accuracy, even in complex setups. The social distancing method proposed also yields promising results in several variable scenarios. The proposed system also successfully demonstrated people and crowd detection with varying degrees of the crowd. The system obtained crowd detection accuracy is around 90% and expected to be readily implemented on real hardware drones and tested in real environments.
Keywords: Human Object Detection, Feature Extraction, Distance Tracking, Machine Learning.
|
DOI:
10.17148/IJARCCE.2021.105140
