Drowsiness Estimation from Low-Frame-Rate Facial Videos using Eyelid Variability Features

Area of Science:

• Computer Vision
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Drowsiness detection is crucial for safety in various applications.
• Conventional methods often rely on blink-related features, which are sensitive to video frame rates.
• Low-frame-rate video analysis presents challenges for accurately capturing rapid movements like blinks.

Purpose of the Study:

• To propose and validate a novel method for drowsiness estimation using eyelid variability features.
• To demonstrate the robustness of the proposed method across a range of low frame rates.
• To compare the performance of eyelid variability features against conventional blink-related features.

Main Methods:

• Facial videos with frame rates from 3 to 30 frames per second (fps) were analyzed.
• Drowsiness was estimated using proposed eyelid variability features.
• Conventional blink-related features were also used for comparison.
• Correlation between estimated drowsiness and ground-truth labels was calculated.

Main Results:

• The proposed eyelid variability features maintained a high correlation (0.55 to 0.53) across frame rates from 30 fps to 3 fps.
• Conventional blink-related features showed a significant drop in correlation from 0.59 (at 30 fps) to 0.28 (at 3 fps).
• The proposed method demonstrated superior performance and stability at low frame rates.

Conclusions:

• Eyelid variability features provide a more accurate and stable method for drowsiness estimation from low-frame-rate videos.
• The proposed approach offers a computationally efficient solution for real-time drowsiness monitoring.
• This method has significant potential for applications requiring low-cost, low-frame-rate video analysis.