Implementation YOLOv5 Method for Detecting Safety Equipment Completeness Images on Site Tower (Case Study: PT. Nexwave Surabaya)

Abstract

The safety equipment completeness detection system is vital for workplace accident prevention. One effective method for this system is YOLOv5, known for its speed and accuracy due to its optimized deep neural network architecture. In this research, we developed a system titled "Implementation of YOLOv5 for Detecting Safety Equipment Completeness on Site Towers (Case Study: PT. Nexwave Surabaya)." We trained this system with a custom dataset from PT. Nexwave Surabaya, comprising 380 images of 5 detection classes: helmets, gloves, safety shoes, vests, and harnesses. The system is built on a Raspberry Pi 4B and connected to a USB camera for real-time safety equipment detection. Testing involved all safety equipment in use, with vests and harnesses alternated for ground and elevated workers. Object detection results showed confidence values ranging from 0.52 to 0.95. The highest confidence value, 0.90, was achieved at a light intensity of 27,360 lux and a distance of 4 meters. To ensure successful results, an average confidence value of >= 0.70 is required for uploading to Google Drive, with detected results stored as backup on the SD card. This system significantly enhances workplace safety by effectively detecting safety equipment completeness using YOLOv5 and Raspberry Pi technology.