BROWSE
Detail
Conference Paper
Gaze-based Drowsiness Recognition for Level 3 Automated Driving: Designing Core Features and Temporal Windows
- Authors
- Eun Hye Jang, Mi Chang, Woojin Kim, Jiwoo Han, Daesub Yoon
- Issue Date
- 2026-04
- Citation
- Conference on Human Factors in Computing Systems (CHI) 2026, pp.1-7
- Publisher
- ACM
- Language
- English
- Type
- Conference Paper
- Abstract
- In Level 3 automated driving, drivers must supervise the system and resume control when requested, making drowsiness a critical safety risk. This study investigates gaze-based drowsiness recognition, focusing on selecting core gaze features and configuring temporal windows for real-time monitoring. Eye-tracking data were collected from 100 participants while they experienced automated driving in a simulator equipped with production vehicle components. Drowsiness was assessed using Karolinska Sleepiness Scale ratings every five minutes. Gaze-based features were extracted over multiple windows and refined through correlation analysis, multicollinearity reduction, model-based feature importance, and cross-window reliability analysis. Four core features-pupil variability, mean fixation duration, eyelid openness, and vertical gaze dispersion-were identified, with a 60-second window providing the best reliability and 30-second windows remaining viable when shorter detection latency is required. A Random Forest classifier achieved an AUC of 0.832 using the four features over 60 seconds. These findings offer design-oriented guidance for implementing gaze-based drowsiness recognition in Level 3 automated driving systems.
- Keyword
- Driver drowsiness recognition, Gaze-based monitoring, Supervisory driving, Temporal window selection
This work is distributed under the term of Creative Commons License (CCL)
(CC BY)
(CC BY)
