영상 데이터 기반 안전 사고 예방을 위한 위험구역 접근탐지 시스템에 관한 연구

Abstract

According to the Ministry of Employment and Labor's recent announcement of the status of fatal accidents, the number of fatal accidents in 2023 decreased by 46 from 644 to 598 com- pared to 2022, occurring in the 500s for the first time ever. However, most of these fatal acci- dents are caused by collisions with heavy equipment, and due to the nature of heavy equipment, accidents are likely to lead to serious accidents. In addition, heavy equipment is essential equip- ment for large-scale sites such as construction and shipyards. Therefore, this study aims to pre- vent accidents by estimating the distance of a person approaching heavy equipment. We propose a system that detects objects (people) using the deep learning algorithm YOLOv8, estimates the distance of the detected person using the OpenCV library solvePnP, and adjusts the estimated distance value when the object approaches below a certain distance, while giving an immediate alarm to the heavy equipment operator. Here, solvePnP does not utilize deep learning among the algorithms that can estimate distances, and can estimate distances by utilizing the minimum number of points, such as the Bounding Box, which is the result of object detection.