Detecting Driver Fatigue With Eye Blink Behavior

Akin, Ali; Kalkan, Habil · 2024 · arXiv

archive: archived pipeline: cataloged verified

Get this paper ↗ (full text — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the critical safety issue of driver fatigue, a leading cause of traffic accidents resulting in millions of deaths and billions in economic losses annually. Motivated by the need for non-intrusive, camera-based Driver Fatigue Detection Systems (DFDS) that comply with emerging regulations like the EU’s mandatory driver alert systems, the authors propose a novel approach focusing on eye blink behavior. Unlike previous methods that rely on static thresholds or physical sensors (e.g., ECG, EEG), which can be impractical or uncomfortable, this study introduces a driver-adaptive system. The core hypothesis is that the specific characteristics of eye blinks, rather than just blink frequency, carry significant information for detecting fatigue, and that adaptive thresholds are necessary to account for individual physical differences and camera positioning. The methodology utilizes the UTA-RLLD dataset, comprising videos from 60 individuals labeled as non-drowsy, mildly drowsy, or drowsy. The system employs facial landmark detection to calculate the Eye Aspect Ratio (EAR) for both eyes. To ensure robustness, the authors establish driver-specific adaptive thresholds for blink detection by analyzing the first two minutes of each video, assumed to be a non-drowsy baseline. Upper and lower thresholds are set at 20% above and below the reference EAR, respectively. The study extracts 13 features per blink cycle, divided into two sets: *blink_set1* (statistical features of open and closed eye regions) and *blink_set2* (behavioral features describing the dynamics of closing and opening, such as frame counts for transitions). Binary classification between non-drowsy and drowsy states was performed using five classifiers: Decision Tree, Random Forest, SVM, k-NN, and Logistic Regression, evaluated via five-fold cross-validation. The results demonstrate that incorporating behavioral blink features significantly improves detection performance. When using only *blink_set1*, the Random Forest classifier achieved the highest accuracy of 93.22% and an F1-Score of 92.88%, while SVM performed the lowest at 77.89% accuracy. However, adding *blink_set2* boosted the Random Forest’s accuracy to 98.37% and F1-Score to 98.41%, representing an approximate 10% improvement. This confirms that the dynamics of eye closure and opening are strong indicators of fatigue. Instance-level analysis further showed that the system correctly identified non-fatigue states 73.23% of the time in non-drowsy videos and fatigue states 88.15% of the time in drowsy videos. The significance of this work lies in its validation of adaptive, behavior-based features for fatigue detection, offering a more accurate and personalized solution than static threshold methods. The study concludes that eye blink characteristics are highly effective for DFDS and suggests that future iterations should integrate head movement features to further enhance detection capabilities. This approach supports the development of modular, non-contact systems suitable for retrofitting older vehicles and improving road safety compliance.

Key finding

Deep learning models trained on facial and eye movement features achieve high accuracy in detecting driver fatigue states, with periorbital features showing the strongest correlation with fatigue severity.

Methodology

lab_experiment

Sample size: 34

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via discover_arxiv on 2026-05-04 (4 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-07; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• drowsiness detection algorithms
• microsleep
• drowsiness
• gaze based attention detection
• dms validation

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: physiological data
• Methodological Resource: tool software, measurement protocol