Abstract
Human eye activity monitoring is essential across diverse fields, including psychology, neuroscience, and human-computer interaction (HCI). While electrooculogram (EOG) is the current gold standard, its reliance on contact electrodes limits long-term usability. This paper proposes a novel, non-contact method called continuous wave Doppler oculogram (cDOG). By utilizing a continuous wave Doppler radar sensor (cDRS), the system remotely measures sub-millimeter displacements of the eye surface during rotation and blinking. Experimental results demonstrate a high correlation between cDOG and EOG signals, achieving an eye movement detection accuracy of 92.35% and a timing precision within 280.5 ms. Crucially, the system remains effective even when eyes are closed, offering a robust alternative for sleep and fatigue monitoring.
Introduction
Monitoring the "windows to the soul" has traditionally required a trade-off between accuracy and comfort. The electrooculogram (EOG) measures the standing electrical potential between the cornea and retina, but requires "wet" electrodes to be adhered to the face, which can cause skin irritation and discomfort. Conversely, video-based eye tracking is non-invasive but raises privacy concerns and fails when eyes are closed, such as during REM sleep or micro-sleep episodes in drivers.
This study explores whether low-cost, millimeter-wave radar can bridge this gap. By treating the human eye as a moving target, we can detect the phase shifts in radar echoes caused by eyeball and eyelid movements. This approach, termed cDOG, provides a high-accuracy, non-contact monitoring solution that works regardless of ambient light levels or whether the subject's eyes are open or shut.
Methodology and System Design
The proposed system utilizes a 24 GHz K-Band Doppler sensor module (NJR4262) paired with a high-precision I/Q mixer and an STM32 microcontroller. The core principle relies on the fact that while the eyeball is nearly spherical, its rotation causes slight changes in the vertical distance between the radar and the eye surface. These sub-millimeter displacements modulate the phase of the radar echo.
To extract meaningful data, the system employs arctangent demodulation to calculate micro-movements. The signal processing pipeline includes DC offset removal to eliminate environmental clutter, phase unwrapping for linearity, and a Discrete Wavelet Transform (DWT) for denoising and removing low-frequency baseline drift. Validation is performed by comparing the radar-derived cDOG signal against simultaneous EOG recordings using Wavelet Coherence analysis.
Experimental Setup
The feasibility of cDOG was tested with 10 healthy volunteers (5 male, 5 female). Subjects performed a series of standardized eye movements, including right-left saccades, up-down saccades, blinks, and fixations. These movements were recorded under both eyes-open and eyes-closed conditions. To assess the system's robustness, data was collected at varying distances (10 cm, 30 cm, 50 cm) and angles (0°, 45°, and 90° relative to the face).
Results and Discussion
The experimental results confirm a strong correlation (coefficient > 0.7) between cDOG and EOG signals, particularly in the low-frequency range (< 1 Hz) where primary eye movement components reside. At a distance of 10 cm with the face directly facing the radar, the system achieved a detection accuracy of 92.35%.
Performance remained high at 30 cm (90.86% accuracy) but degraded at 50 cm (65.12%), especially for closed-eye movements. Angular sensitivity tests showed that the system is effective up to a 45° angle (75.27% accuracy), though performance drops at 90° due to signal blockage by facial features like the nose bridge. Notably, the ability to maintain high accuracy with closed eyes suggests significant potential for clinical applications in sleep medicine.
Conclusion
This study demonstrates that continuous wave Doppler radar is a feasible and accurate method for non-contact eye activity monitoring. While the current prototype is sensitive to head motion artifacts and has higher power consumption (161 mW) than traditional EOG, these are engineering challenges for future iterations. Subsequent research will focus on multi-antenna arrays to isolate eye movements from head motion and enable two-dimensional gaze tracking, further expanding the utility of cDOG in HCI and medical diagnostics.
Summary updated