Saccade Landing Point Prediction Based on Fine-Grained Learning Method

Area of Science:

• Neuroscience
• Computer Science
• Human-Computer Interaction

Background:

• Saccades are rapid eye movements crucial for visual attention.
• Predicting saccade landing points is vital for gaze-contingent systems to mitigate display-update latency.
• Current prediction methods struggle with the high speed and complexity of naturalistic viewing.

Purpose of the Study:

• To develop and evaluate a novel algorithm for early saccade landing point prediction.
• To improve the accuracy and reliability of eye-tracking-based gaze-contingent systems.
• To analyze factors influencing saccade landing point prediction errors.

Main Methods:

• Utilized Long Short-Term Memory (LSTM) networks for predictive modeling.
• Implemented a fine-grained loss function tailored for saccade prediction.
• Evaluated the algorithm on a large dataset of nearly 220,000 saccades from 75 participants during natural video viewing.
• Compared performance against state-of-the-art saccade landing point prediction algorithms.

Main Results:

• The proposed LSTM-based algorithm demonstrated superior performance compared to existing methods.
• Achieved up to a 50% reduction in prediction error.
• Identified key factors affecting prediction accuracy, including saccade duration, length, participant age, and intrinsic user characteristics.

Conclusions:

• Early prediction of saccade landing points is feasible with advanced deep learning techniques.
• The developed algorithm offers a significant improvement for real-time gaze-contingent applications.
• Further research into user-specific characteristics can enhance prediction accuracy.