영상 인식 알고리즘을 이용한 안전 보호구(안전모) 탐지에 관한 연구

Abstract

Safety accidents at work sites are directly related to workers' lives, and the manufacturing industry's interest in safety accidents is increasing every year. Safety accidents at work sites are caused by a variety of factors, and it is difficult to predict when and why they occur. In this research, an intelligent image recognition-based worker safety protection device wearing algorithm that can determine suitability of wearing safety protective devices is developed and the proposed algorithm is sought to be applied to the site. In this study, the You only look once (YOLO) algorithm is applied to analyze the presence of workers wearing safety protection equipment in real time. Accuracy of object detection for safety protection equipment is very important. Thus, this study compared/analyzed the algorithms of two YOLO systems (YOLOv2, YOLOV3) and improved the performance of the model by changing Hyperparameters, Fine-tuning and Dataset of the selected algorithms. In the future, studies will be conducted on how to improve the accuracy of object detection and complement the accuracy of object detection in the proposed YOLO series algorithm.