Answers are grounded in this paper and cite their sources.
Drowsy driving is a significant road-safety concern that requires reliable methods for identifying visual indicators of driver fatigue. This study proposes an Enhanced Ant Colony Optimization (EACO)-based YOLOv8 framework for facial-part detection, integrated with an image-level rule-based driver-state inference module. The detection framework identifies four facial-part categories: open eye, closed eye, mouth open (yawn), and mouth closed (no yawn). To improve image quality and feature visibility, a preprocessing pipeline incorporating anisotropic diffusion filtering, gamma correction, Contrast Limited Adaptive Histogram Equalization (CLAHE), and min--max normalization is employed. EACO optimizes nine YOLOv8 hyperparameters using adaptive pheromone control, dynamic heuristic weighting, and elite-ant reinforcement. Experiments are conducted using an image-level dataset split of 70% for training, 15% for validation, and 15% for independent testing. On the held-out test set, the proposed EACO-YOLOv8 detector achieves 98.94% precision, 98.99% recall, a 98.97% F1-score, 98.85% mAP@0.5, and 96.42% mAP@0.5:0.95. Compared with baseline YOLOv8 and conventional Ant Colony Optimization (ACO)-based YOLOv8, the proposed approach achieves higher detection performance across the reported evaluation metrics. In addition, the rule-based inference module maps the detected facial conditions to three operational driver states---Awake, Drowsy, and Yawning---and achieves an image-level classification accuracy of 95.16% against manually labelled ground-truth states. These findings demonstrate the potential of combining metaheuristic hyperparameter optimization, facial-part detection, and interpretable rule-based inference for image-based driver monitoring. However, subject-independent evaluation was not possible because reliable subject identifiers were unavailable in the original dataset. Consequently, generalization to unseen drivers, cross-dataset validation, temporal analysis, and real-time deployment remain important directions for future investigation.
No takes yet. Share an insight, caveat, or question.
Prasad et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: