Real-time Personal Protective Equipment (PPE) Detection Using YOLOv4 and TensorFlow
Adban Akib Protik, Amzad Hossain Rafi, Shahnewaz Siddique
Abstract
The major cause for the continuous spreading of the COVID-19 disease is the human to human transmission of the virus. The virus enters the human body through exposed areas such as the mouth, nose and eyes through touching, coughing, sneezing or close proximity with infected people. So it is imperative people cover specific body parts to avoid contracting the disease and stop the transmission of the virus. Personal Protective Equipment (PPE) such as face mask, face shield, gloves help in this cause. Unfortunately some people are reluctant to wear PPE. In such situations a real-time monitoring system for PPE is needed. In this study we have developed a detector which can detect in real time if people are wearing PPE or not. The detector is developed using YOLOv4 computer vision model which specially performs well in real time object detection. We have prepared a combined dataset consisting of collected images and our captured images. We have annotated our images and image augmentation is done on all the training set images. We have also converted the detector weight to TensorFlow format to check live detection performance and added features like live object count and record keeping. In this paper we propose a method using YOLOv4 model with combined dataset and other techniques to detect four classes of objects. We get promising results from the tests and mAP of the object detector is 79%.