Uncovering and Addressing Blink-Related Challenges in Using Eye Tracking for Interactive Systems

Title of the Paper

Uncovering and Addressing Blink-Related Challenges in Using Eye Tracking for Interactive Systems

Paper Information

• Research Domain: Human-Computer Interaction and Eye Tracking Technology
• Keywords: Human-Computer Interaction, Eye Tracking, Blink, Data Loss, Data Interpolation, Interactive Systems, Data Processing, Data Quality, Machine Learning, Data Standardization

Research Background and Issues

• Research Questions:
  • Current eye tracking data suffers from data loss caused by blinking, which significantly impacts system performance.
  • The methods for handling missing data lack standardization, making it difficult to reproduce and compare results across studies.
  • The impact of blinking on eye tracking data extends beyond data loss, as artifacts within the surrounding time window may also cause issues.
• Significance:
  • Eye tracking technology is widely applied in interactive systems, including direct control, motion prediction, and gaze gestures. Improving data quality is directly tied to system performance and user experience.
• Motivation and Related Work:
  • Common methods for handling missing data include ignoring missing data, interpolating, or removing samples containing blink-related data. However, these approaches fail to fully account for the impact of artifacts on data accuracy.
  • Recent studies highlight the limitations of use-case-specific methods in terms of generalizability and reproducibility, underscoring the need for standardized data processing methods.

Proposed Solution

• Core Methods:
  • Conduct a systematic literature review and large-scale dataset analysis to understand the impact of existing processing methods on data quality.
  • Propose improved data processing workflows and provide specific guidelines.
• Innovations:
  • Introduce standardized blink detection and missing data processing workflows.
  • Identify the time range of artifacts occurring before and after blinks and quantify their specific impact on system performance.
  • Validate the error rates of different data interpolation methods using publicly available datasets to ensure broad applicability.
• Implementation Steps:
  1. Literature Review: Review recent scientific literature related to blinking and missing data processing, using the PRISMA method for study selection.
  2. Dataset Analysis: Select 11 open-source eye tracking datasets, unify preprocessing formats, convert data to visual angle degrees, and detect the time range of blink-related artifacts.
  3. Method Validation: Evaluate the effectiveness of different interpolation methods using artificially generated blink data and compare baseline methods with improved workflows.
  4. Guideline Development: Based on the results, establish standards for data preprocessing and recommend optimal blink interpolation methods.

Research Findings

• Key Discoveries:
  • Among 60 reviewed papers, none reported how blink data was handled, and inconsistent reporting hindered reproducibility.
  • Blink-related artifacts affect data within 70 milliseconds before and 118 milliseconds after a blink, necessitating the exclusion of data within these time windows during processing.
  • Significant differences in blink frequency and duration were observed across different eye tracking devices (e.g., 30Hz vs. 1000Hz).
  • Linear interpolation and cubic spline interpolation methods demonstrated the best performance in terms of average error rates for data recovery.
• Advantages Over Existing Solutions:
  • Provides clear criteria for algorithm selection and detailed guidance on window size and interpolation methods for research.
  • Resolves inconsistencies in blink data processing in current studies, enhancing data processing standardization and laying a foundation for future research.
• Experimental and Evaluation Results:
  • Experimental results show that proper handling of blink data and artifacts can significantly improve the practicality of eye tracking in interactive systems.
  • Excessively large window sizes lead to substantial reductions in usable data; interpolation methods minimize data wastage and enhance system responsiveness.
• Limitations and Future Directions:
  • Results are based on existing public datasets; future work should validate applicability to non-public datasets or more complex interactive scenarios.
  • Current interpolation methods can be further optimized by exploring more nonlinear approaches.
  • Development of new methods for real-time processing is needed to enhance performance and user experience in online interactive systems.