A Data-Driven Framework for Intention Prediction via Eye Movement With Applications to Assistive Systems

Area of Science:

• Biomedical Engineering
• Computer Science
• Human-Computer Interaction

Background:

• Accurate human intention prediction is crucial for developing effective assistive devices.
• Eye movement tracking offers a non-invasive method for inferring user intentions.
• Existing gaze-based systems lack the accuracy needed for reliable assistive technology.

Purpose of the Study:

• To develop a data-driven framework for accurate human intention prediction using eye movement patterns.
• To enhance the performance of assistive devices for individuals with motor or communication limitations.

Main Methods:

• Utilized deep learning to analyze spatial and temporal eye movement patterns.
• Employed gaze point clustering to identify regions of interest (ROIs).
• Applied hidden Markov models (HMMs) to determine ROI transition sequences and transfer learning for object recognition.

Main Results:

• Achieved an average classification accuracy of 97.42% in predicting intended daily-life activities.
• Demonstrated significantly higher accuracy compared to existing gaze-based intention prediction studies.
• Successfully predicted user intentions both after task completion and during early task stages.

Conclusions:

• The proposed framework offers a highly accurate and reliable method for gaze-based human intention prediction.
• This advancement holds significant potential for improving the functionality and user experience of assistive technologies.
• The data-driven approach integrating deep learning with eye-tracking provides a robust solution for intention recognition.