Deteksi Rompi dan Helm Keselamatan Menggunakan Metode YOLO dan CNN

Abstract

Construction workers face a high risk of injury and are prone to accidents while performing their duties. Several factors increase the chance of accidents, including open and heated environmental conditions, building heights, sharp objects, and others. The use of personal protective equipment (PPE) is essential to anticipate or reduce the risk of accidents that may occur. However, it is not uncommon for construction workers to forget or purposefully disregard personal protective equipment. To address these problems, a system capable of detecting personal protective equipment for construction workers is required. This study used the You Only Look Once (YOLO) method to detect the head and body parts of the inputted image. The detected body parts were then cut and processed using the Convolutional Neural Network (CNN) method with the ResNet50 model for classification. The training process with the ResNet50 model was modified on hyperparameters including learning rate, epoch, dense layer, dropout layer, data augmentation, and freeze layer to compare its performance with the model before modification. The results showed that the YOLO model has a very high level of detection speed with good accuracy. Meanwhile, the modified CNN model performed well with an average accuracy value of 96%