DEEP: 3D Gaze Pointing in Virtual Reality Leveraging Eyelid Movement

Title of the Paper

DEEP: 3D Gaze Pointing in Virtual Reality Leveraging Eyelid Movement

Paper Information

• Subject Area: Gaze interaction and target selection techniques in virtual reality
• Keywords: Virtual reality, gaze interaction, target selection, eyelid movement, ambiguous target selection, depth adjustment

Research Background and Problems

• Problems and Challenges:

  • Although gaze-based interaction has gained attention due to its fast movement and hands-free operation, it faces three major challenges:
    1. Unintentional Activation: Natural gaze movements can lead to the "Midas touch" problem.
    2. Low Input Accuracy: Gaze jitter significantly affects selection precision, especially for small targets.
    3. Target Occlusion: Targets in 3D scenes may be partially or completely occluded, making selection difficult.
  • Existing methods have limited effectiveness in addressing these issues.

• Importance:

  • Solving the selection problem for occluded and dense targets in virtual reality (VR) environments has broad practical significance, including applications in CAD design, smart room interaction, and gaming.

• Research Motivation and Related Work:

  • Proposing a novel method that utilizes users' eyelid movements to adjust the visible depth of the selection target, overcoming the challenge of selecting occluded targets.
  • Improving upon existing approaches (e.g., gaze overshoot adjustment, dynamic layout restructuring, and statistical optimization selection methods) to make them suitable for gaze interaction.

Solution

• Method and Innovation:

  • Introducing a new technique called DEEP, which combines eyelid movement and probabilistic modeling to provide precise and occlusion-robust 3D gaze target selection.
  • Enhancement: For the first time, continuous eyelid movement is utilized to achieve visual depth adjustment, improving selection performance for dense targets.

• Key Design and Steps:

  1. Core Feature Design:
     • Angle of Eyelid Adjustment (AAE):
       • Users can perform continuous depth adjustments through blinking or widening their eyes, revealing occluded targets.
     • Probabilistic Input Prediction Model:
       • Combines target position and depth information to dynamically select the intended target.
  2. User Study Validation:
     • A series of user studies were conducted to test the naturalness, controllability of eyelid movements, and user behavior patterns during dwell-based selection tasks.
  3. Technical Implementation:
     • Provides visual feedback mechanisms, including a slider to display the current visual depth.
     • Accounts for gaze jitter by expanding the target area to improve selection robustness.

Research Outcomes

• Specific Results:

  • Experimental Validation:
    • Selection Performance: In all test scenarios (partial occlusion, full occlusion, dense target scenes), DEEP significantly outperformed baseline techniques (e.g., Naive Dwell), achieving an average target selection time of 2.5 seconds and reducing the error rate to 2.3%.
    • User Preference: The dynamic algorithm H-DEEP ranked highest in user satisfaction, successfully integrating the advantages of depth and position selection.
  • User Data Insights:
    • Controllability and Comfort: Experiments showed that users could easily perform natural and deliberate eyelid movements, maintaining good stability within non-extreme ranges.
    • Robustness: DEEP's dynamic depth adjustment mechanism improved the success rate of target selection.

• Advantages Over Existing Methods:

  • Demonstrated superior selection capability in fully occluded target scenarios.
  • Eliminated input uncertainty caused by jitter and dense regions in traditional techniques.

• Limitations and Future Directions:

  • Algorithm Improvement: The current dynamic weight allocation algorithm for depth and position selection can be further optimized; future work could explore machine learning or more complex contextual models.
  • Adaptability and Personalization: Algorithms need to be designed to accommodate individual differences, such as natural eye movement patterns.
  • Real-World Application Testing: Current tests were conducted in abstract VR environments; future work should validate the technology in complex real-world scenarios, incorporating background interference and diverse target forms.

This paper demonstrates how eyelid movement and probabilistic modeling can significantly improve gaze-based target selection techniques, providing a robust solution for virtual reality interaction.