Evaluasi Kinerja Model Yolov8 Dalam Deteksi Visual Kantuk Dan Distraksi Pengemudi

Abstract

Traffic accidents caused by human factors such as drowsiness and distraction remain a leading cause of road fatalities globally. Computer vision-based approaches offer promising solutions to identify driver inattention through visual cues. This study aims to evaluate the performance of the YOLOv8 object detection model in recognizing visual indicators of driver drowsiness and distraction, including open eyes, closed eyes, yawning, and mobile phone usage. A custom dataset was manually annotated using the Roboflow platform, and model training was carried out using the Ultralytics YOLOv8 framework on Google Colab. Data augmentation techniques such as horizontal flipping, shear transformation, color jittering, and noise injection were applied to enhance the model’s robustness. Evaluation was performed on the validation set using performance metrics including precision, recall, F1-score, and mean Average Precision (mAP). The results demonstrate that the model achieves high accuracy across all object classes. This study is limited to evaluating model performance and does not cover real-time implementation or system integration.