Polarization-resolved Imaging Reduces Gaze Error by 10, 16% and Improves Eye Tracking Precision

Understanding where a person is looking is fundamental to human interaction, and accurate eye tracking technologies are increasingly important for a range of applications, from virtual reality to assistive devices. Mantas Žurauskas, Tom Bu, and Sanaz Alali, along with colleagues at various institutions, now demonstrate a significant improvement in eye tracking accuracy by incorporating the polarization of light. Their research reveals that analysing the polarization state of reflected near-infrared light, in addition to its intensity, unveils trackable features on the eye’s surface, particularly the sclera and cornea, that are otherwise invisible. Across a large study involving 346 participants, machine learning models trained on data from this polarization-enabled eye tracking system reduce gaze error by up to 16% compared to conventional methods, even when faced with challenges like eyelid movement, varying eye position, and changes in pupil size, positioning this technology as a promising advancement for future wearable devices.

Polarization Imaging for Robust Eye Tracking

This research details the development of a new eye tracking system based on polarization imaging, offering several advanced features. The system utilizes polarization-sensitive imaging to track the eye, potentially overcoming limitations of traditional infrared systems. It captures comprehensive polarization information, providing a detailed understanding of how light interacts with the eye. Integration of metasurfaces further enhances the system’s capabilities, potentially improving image quality and compactness. A key innovation is the system’s ability to capture characteristics of the sclera using polarization, opening possibilities for biometric identification and authentication.

This technology could also potentially detect conditions like keratoconus and monitor the progression of myopia. Furthermore, the system can estimate the eye’s accommodation state, or focus, using polarization information, and potentially measure ocular hydration levels. The use of metasurfaces and advanced optics aims for a compact and efficient design. In essence, this research presents a sophisticated eye tracking system that leverages the power of polarization imaging and advanced optics to achieve high accuracy, robustness, and a range of novel capabilities, with the potential to significantly advance fields from virtual reality to medical diagnostics and biometric security.

Polarization Imaging Enhances Robust Eye Tracking

The study pioneered a polarization-resolved near-infrared imaging system to enhance eye tracking accuracy and robustness. Scientists engineered a system pairing a polarization-filter-array camera with a linearly polarized near-infrared illuminator, revealing trackable features on the sclera and cornea absent in standard intensity images. The camera records raw intensities at multiple linear orientations, allowing reconstruction of full-resolution images for each polarization. The system computes total intensity, the degree of linear polarization, and the angle of linear polarization, masking pixels with low intensity to suppress artifacts.

Specifically, calculations use established formulas, with adjustments to ensure numerical stability. For visualization, total intensity is normalized, and a composite image is created where hue encodes the angle of polarization, saturation scales with the degree of polarization, and value represents intensity. Experiments were conducted with a binocular setup, employing two cameras per eye, mounted to capture distinct perspectives and evaluate performance across varying distances. Participants were stabilized with a chinrest, and eye positions were adjusted to sample a broad range of positions. Interpupillary distance was measured for each participant and recorded. Gaze targets were presented on a monitor, and sequences were captured under various conditions, including different backgrounds and simulated slippage, to assess sensitivity to changes in position. All participants completed a series of visual target sequences used for both training and evaluation, with occasional adjustments to target positions to mitigate occlusions.

Polarization Improves Gaze Estimation Accuracy Significantly

Scientists have developed a new polarization-enabled eye tracking system that significantly improves gaze estimation accuracy and robustness. This work demonstrates how measuring the polarization of light reflected from the eye, in addition to its intensity, reveals trackable features on the sclera and cornea, features largely invisible in standard intensity-only images. Experiments with a cohort of 346 participants show that machine learning models trained on data from this system reduce the median gaze error by 16% compared to equivalent models trained on standard intensity images. The research team employed a polarization-filter-array camera paired with a linearly polarized near-infrared illuminator to capture detailed polarization information from ocular tissues.

This approach allows for the creation of richer, more textured images of the eye’s surface, providing more reliable tracking cues even when traditional methods are compromised by eyelid occlusion, changes in eye position, or variations in pupil size. Data analysis focused on the 95th percentile of per-frame absolute gaze error to emphasize the reliability of the system under challenging conditions, a metric strongly correlated with user experience. Results demonstrate that the system consistently achieves lower error values across the participant group, signifying a substantial gain in the tail performance of the eye tracking system, meaning fewer large errors and a more consistent user experience. The team trained and validated convolutional neural networks on data collected from the 346 participants, splitting the data into training and validation sets to ensure robust performance evaluation. The study confirms that this system provides a simple, robust sensing modality with potential for integration into future wearable devices, offering a new direction for eye tracking technology in extended reality and AI-powered glasses.

Polarization Improves Robust, Accurate Eye Tracking

This research demonstrates a significant advancement in eye-tracking technology through the development of polarization-enabled eye tracking. By measuring the polarization of reflected near-infrared light, alongside its intensity, scientists have created a system capable of identifying trackable features on the sclera and cornea that are not visible using traditional methods. Testing across a large cohort of participants, the team achieved a 16% reduction in gaze error compared to existing intensity-based systems, even under challenging conditions such as eyelid occlusion, varying eye relief, and changes in pupil size. The core achievement lies in harnessing light-tissue polarization effects to improve the accuracy and robustness of eye tracking, particularly in compact wearable devices.

The system reveals previously unseen collagen structures within the eye, which may correlate with various eye conditions, offering potential for new health monitoring applications if validated through long-term studies. While the current work focuses on demonstrating the technology’s capabilities, the researchers acknowledge the need for further investigation with larger groups over extended periods to fully understand the potential of tracking these structural changes. This work establishes a practical pathway toward reliable, low-burden eye-based input for human-computer interaction by tying algorithmic gains to the fundamental physics of light and tissue interaction.