Object Detection for Safe Working Environments using YOLOv4 Deep Learning Model

Abstract

The health and safety of employees in workplaces maintains its importance since the concept of production emerged. Recent developments in computer vision and deep learning have made it widespread to be used in work environments as a secondary tool in ensuring occupational safety from surveillance videos. Thus, an important performance is achieved by minimizing human-induced errors in working environments. In this study, a method based on the YOLOv4 deep learning model is proposed to control the use of personal protective equipment from videos and to detect unsafe movements in the working environments of facilities operating in the field of industrial production. In the study, a dataset is created with videos collected from different working environments. In the study, later, on the prepared video dataset, the detection of personal protective equipment such as helmets, vests, masks, gloves, eyeglasses used by workers in factories operating in industrial areas and whether they use the appropriate equipment correctly is determined using the YOLOv4 framework. In the experimental studies conducted within the scope of the study, the mean average precision (mAP) value is achieved as 91.18% as a result of the training performed in the YOLOv4 network. In addition, results of 0.89, 0.91, 0.90, 70.35 and 1.1147 are obtained for other measurement metrics such as precision, recall, F1-score, intersection over union (IoU), and average loss, respectively. As a result, in the proposed study, instant inspection of the videos collected from the cameras installed in the factories, the meaning of the scene and the control of safe working environments are successfully achieved.