Search papers, labs, and topics across Lattice.
This paper introduces GazeFS, a novel approach for online target-centered gaze-trajectory forecasting and stabilization that leverages gaze-head history to predict the next target-center direction and a short-horizon Search/Focus estimate without requiring target information during inference. The method was validated through extensive experiments involving 7,960 acquisition episodes from 30 participants, demonstrating significant reductions in bias and dispersion of gaze trajectories. Key results indicate that GazeFS achieves a reduction in target error and maintains a high level of accuracy, suggesting its effectiveness in enhancing gaze interaction systems.
GazeFS reduces gaze trajectory errors by up to 0.400 degrees while maintaining temporal smoothness, revolutionizing target-centered gaze interaction.
Target-centered gaze interaction requires more than suppressing frame-to-frame fluctuations: target acquisition produces task-aligned changes in gaze-head dynamics, while a gaze trace may retain a persistent target-relative residual direction. We formulate gaze correction as online target-centered gaze-trajectory forecasting and stabilization and introduce GazeFS, which maps a variable-length gaze-head history to the next target-center direction and a short-horizon Search/Focus estimate without target information at inference. Across 7,960 acquisition episodes from 30 participants, Search-Focus differences remain stable under quality control, onset exclusion, and duration matching. History windows improve phase decoding over the current endpoint, but explicit task progress remains a strong control. Under the 30-participant, five-fold grouped out-of-fold protocol across three seeds, the reductions relative to raw hold in Focus episode bias, within-episode dispersion, and P90 target error are 0.182 degrees, 0.257 degrees, and 0.400 degrees, with participant-bootstrap 95% confidence intervals excluding zero. Endpoint-free replay from empty history preserves the Focus advantage and yields raw-network phase balanced accuracy/AUPRC of 0.925/0.993; coordinate controls further show that recent history contributes beyond explicit progress metadata. GazeFS therefore improves Focus target centering and empirical residual contraction while leaving temporal smoothness as a separate objective.
