Implementasi Yolov5 Deteksi Mata Lelah Berbasis Android

Abstract

Excessive screen exposure can trigger digital eye strain, reducing visual comfort, attention, and overall productivity. Prior studies in computer vision indicate that deep learning–based object detection, particularly the YOLO family, can recognize facial and eye-related visual patterns efficiently, making it suitable for early-warning systems on mobile devices. This study aims to implement YOLOv5 to detect signs of eye fatigue in real time using the front camera of an Android smartphone. The novelty of this work lies in deploying a lightweight object-detection model on-device through TensorFlow Lite and integrating an automatic notification mechanism as a preventive intervention. The proposed methodology includes collecting and labeling an eye-image dataset into two classes (awake and drowsy), training a YOLOv5 model in Google Colab, optimizing and converting the trained model to TensorFlow Lite, and integrating it into an Android application for live-camera inference. System performance is evaluated using accuracy, precision, recall, and inference speed (FPS). Experimental results show that the system achieves 95.6% accuracy, 94.3% precision, 96.1% recall, and an Average speed of 22 FPS, enabling responsive detection and timely notifications. In conclusion, the Android-based YOLOv5 implementation is feasible as a preventive solution to help users monitor eye-fatigue symptoms and encourage healthier screen-use habits.