Drowsiness Detection in the Advanced Driver-Assistance System using YOLO V5 Detection Model

Abstract

The development of Artificial Intelligence (AI) in the automobile industry, particularly in the Advanced Driver-Assistance System (ADAS), has been increasing rapidly in recent years. One of the essential features of ADAS is the drowsiness alert system, which monitors the state of the driver's eyes. This article presents a study on the development of ADAS that focuses on drowsiness detection using deep learning through a Convolutional Neural Network (CNN) approach. The study utilized the YOLO V5 model for object detection, which was trained using custom datasets containing images annotated with two labels: "Mengantuk" (Drowsy) and "Bangun" (Awake) The goal was to recognize whether the driver was drowsy or awake. The results of the study showed that the modified YOLO V5 CNN Model had high accuracy based on evaluation metrics in terms of accuracy, precision, and recall in detecting drowsiness around the area of the eyes, with a Mean Average Precision (mAP) of 99.5% and an F1-Score of 99.8%. For better understanding and visualization, the model was tested using real-time detection through a web camera, using Jetson Nano as the inference device. The model detected drowsiness in real time, with a confidence rate of 80% to 97%