CONF-LA: Real-Time Fixation-to-Line Assignment for Reading Gaze Data
Researchers propose CONF-LA (Confidence-score-based Online Fixation-to-Line Assignment), a low-latency method for assigning gaze fixations to lines of text in real-time during multi-line reading. The approach integrates knowledge of reading behavior and Gaussian line likelihoods to compute posterior-line scores, deferring assignments when uncertainty is high. Evaluated on open-source data, CONF-LA achieves stable performance in post hoc analysis, closes the online-offline gap by 1-2%, and has a mean per-fixation latency of 0.348 ms. It shows particular invariance to regressions, significantly improving ad hoc median accuracies. The work addresses challenges in remote and webcam-based eye tracking for real-time reading support.
Key facts
- CONF-LA stands for Confidence-score-based Online Fixation-to-Line Assignment
- Mean per-fixation latency is 0.348 ms
- Closes online-offline gap by 1-2%
- Evaluated on existing open-source data
- Invariant to regressions
- Improves ad hoc median accuracies
- Addresses noise factors and layout ambiguity in multi-line reading
- Deferred assignments when uncertainty is high
Entities
—