A cost function to determine the optimum filter and parameters for stabilising gaze data.

Area of Science:

• Human-Computer Interaction
• Biomedical Engineering
• Signal Processing

Background:

• Eye tracker software often uses filters to reduce noise in gaze data.
• Filtering enhances data precision but can introduce latency due to sliding window mechanisms.
• This latency affects the stabilization time of gaze data post-saccade.

Purpose of the Study:

• To examine the impact of various filters and parameter settings on accuracy, precision, and latency in eye-tracking data.
• To identify optimal parameters for filtering eye-tracking data using a cost function.
• To evaluate the trade-offs between filter-induced latency and data precision.

Main Methods:

• Investigated the effects of different filters and parameter settings (window length, removal threshold, etc.) on eye-tracking data quality.
• Utilized a cost function to determine optimal filter parameters.
• Analyzed noise characteristics using RMS/STD for both filtered and unfiltered data.

Main Results:

• Found that for Finite Impulse Response (FIR) filters, using standard deviation to remove 95% of samples in the sliding window achieves an optimal balance between filter-related latency and data precision.
• Confirmed that unfiltered data exhibits an RMS/STD ratio around 1, characteristic of white noise.
• Observed lower RMS/STD values for all tested filters, indicating effective noise reduction.

Conclusions:

• Optimal filter parameter selection is crucial for balancing eye-tracking data precision and latency.
• FIR filters, combined with a 95% sample removal based on standard deviation, offer a robust method for improving gaze data quality.
• Filtering significantly reduces noise, as evidenced by lower RMS/STD values compared to unfiltered data.