Fixation-based Self-calibration for Eye Tracking in VR Headsets

Overview

• This study proposes a novel self-calibration method for eye tracking in virtual reality (VR) headsets.
• The method assumes the user's viewpoint can move freely and the points of regard (PoRs) from different viewpoints are distributed within a small area on an object surface during visual fixation.
• The method first detects fixations from uncalibrated gaze directions using an extended 3D version of the I-VDT algorithm.
• The calibration parameters are then optimized by minimizing the dispersion of the PoRs.
• The proposed method can identify optimal calibration parameters without explicit user calibration, image processing, or marker-substitute objects.

Plain English Explanation

The study presents a new way to automatically calibrate eye tracking in virtual reality (VR) headsets. The key idea is that when a person is looking at an object, their gaze will be focused on a small area of that object, even as they move their head around. By analyzing this gaze pattern, the researchers can figure out how the person's eyes are oriented relative to the headset's cameras, without requiring the user to go through a tedious calibration process.

The method first detects when the user is fixating their gaze on an object, using an extended version of an existing algorithm. It then adjusts the calibration parameters to minimize the scatter or "dispersion" of the detected gaze points on the object surface. This allows it to determine the optimal alignment between the user's visual axis and the headset's optical axis, without any explicit user interaction.

In tests with 18 participants moving around in VR environments, this self-calibration approach achieved an average accuracy of 2.1 degrees, which is significantly better than the typical offset. The accuracy can be further improved by refining the fixation detection or optimization algorithm. This represents an important advance in making eye tracking more seamless and user-friendly for VR applications.

Technical Explanation

The proposed self-calibration method builds on prior work by leveraging the assumption that a user's points of regard (PoRs) will be clustered on object surfaces during visual fixation, even as the viewpoint changes.

First, the method detects fixations from the uncalibrated gaze direction data using an extension of the I-VDT (velocity and dispersion threshold identification) algorithm similar to this approach. This allows it to identify periods of stable gaze in a 3D virtual environment, rather than just on a 2D screen.

The calibration parameters are then optimized by minimizing the sum of a dispersion metric across the detected PoRs. This finds the optimal alignment between the user's visual axis and the headset's optical axis, without requiring explicit user calibration or additional sensors.

The method was evaluated using gaze data from 18 participants navigating two VR environments with significant occlusions similar to challenges in driver attention tracking. It achieved an average accuracy of 2.1 degrees, outperforming typical offsets, and the accuracy can be further improved by refining the fixation detection or optimization.

Unlike prior work that relied on head pose estimation or other external information, this self-calibration approach only uses the uncalibrated gaze data, making it a more practical and user-friendly solution for VR eye tracking.

Critical Analysis

The paper presents a compelling self-calibration approach that can significantly improve the accuracy of eye tracking in VR without burdening users. However, a few limitations and areas for further research are worth noting:

The method assumes the user's points of regard will be clustered on object surfaces during fixation. While this holds true in many scenarios, it may not apply in all VR experiences, particularly those with sparse or featureless environments. Additional research is needed to understand the robustness of the approach in a wider range of VR applications.

The study only evaluated the method with 18 participants. A larger-scale validation would help confirm the generalizability of the results. Additionally, testing the method across different VR hardware, user demographics, and real-world tasks could provide valuable insights.

The paper does not discuss how the method would handle dynamic scenes or rapid head movements, which can be common in VR. Extending the self-calibration to handle these more challenging scenarios could further enhance the practical applicability of the approach.

Overall, this research represents an important step forward in making eye tracking more seamless and accessible for VR users. With continued refinement and validation, the proposed self-calibration method could have significant implications for a wide range of VR applications, from gaming and entertainment to assistive technologies and human-computer interaction.

Conclusion

This study introduced a novel self-calibration method for improving the accuracy of eye tracking in virtual reality (VR) headsets. The key innovation is the ability to infer the optimal calibration parameters from the user's natural gaze patterns, without requiring explicit calibration or additional sensors.

The results demonstrate that this approach can achieve significantly better accuracy than typical offsets, with an average error of just 2.1 degrees. This represents an important advancement in making eye tracking more seamless and user-friendly for VR applications, which could have wide-ranging implications for gaming, human-computer interaction, assistive technologies, and beyond.

While the method shows promise, further research is needed to validate its robustness across a wider range of VR scenarios and user demographics. Nonetheless, this work represents an important step forward in the ongoing effort to enhance the capabilities and accessibility of VR technology.