YOLOv5-M: A Deep Neural Network for Medical Object Detection in Real-time

COVID-19 pandemic is still a global health issue, causing about 684 million cases and 6.84 million deaths around the world. Personal protective equipment (PPE) such as gloves, face masks, face shields, goggles, etc., can be an effective measure to combat COVID-19. In this work, we proposed, YOLOv5-M, a modified version of YOLOv5, for medical object (PPE and face mask) detection tasks. Experiment results on a recent five-class real-time, challenging dataset CPPE-5 (medical PPE) show the effectiveness of YOLOv5-M. YOLOv5-M outperformed four other existing state-of-the-art object detectors: Faster-RCNN, Single shot object detectors, YOLOv3, and YOLOv5 in terms of training speed, and model performance. The proposed model is also tested on the Face mask detection dataset, and it achieves competitive performance. Apart from that, maintaining proper social distancing inside hospitals among healthcare workers and patients is critical in minimizing nosocomial transmission. Despite the commodity of PPE, some individuals may still get infected with COVID- 19. The proposed system also has the feature of calculating the social distance between healthcare workers. Taken together, the proposed system has the potential to be implemented in real-time healthcare settings.