Evaluating Eye Tracking Signal Quality with Real-time Gaze Interaction Simulation

TL;DR

The paper addresses evaluating eye-tracking signal quality in real-time gaze interaction using an offline dataset. It employs three classification approaches (IKF, IVT, IDT) and a Rank-1 fixation selection to define trigger-events, analyzing how buffer-period and dwell-time settings influence trigger-event detection and signal quality. Key findings show IKF provides the fastest trigger onset, IVT yields more accurate trigger-event identification, and Kalman-filter-based methods deliver better signal quality, though data-quality issues such as NaN handling and calibration offsets can cause substantial variability across participants. These insights guide algorithm selection for reliable, real-time gaze-based interfaces and underscore the need to account for data imperfections in practical deployments.

Abstract

We present a real-time gaze-based interaction simulation methodology using an offline dataset to evaluate the eye-tracking signal quality. This study employs three fundamental eye-movement classification algorithms to identify physiological fixations from the eye-tracking data. We introduce the Rank-1 fixation selection approach to identify the most stable fixation period nearest to a target, referred to as the trigger-event. Our evaluation explores how varying constraints impact the definition of trigger-events and evaluates the eye-tracking signal quality of defined trigger-events. Results show that while the dispersion threshold-based algorithm identifies trigger-events more accurately, the Kalman filter-based classification algorithm performs better in eye-tracking signal quality, as demonstrated through a user-centric quality assessment using user- and error-percentile tiers. Despite median user-level performance showing minor differences across algorithms, significant variability in signal quality across participants highlights the importance of algorithm selection to ensure system reliability.

Figures (2)

• Figure 1: Supplementary Fig. 1. Exemplar: Angular offset in trigger-events for noisy eye-tracking data. (A1-A2) Display horizontal and vertical channels of an eye-tracking signal from GazeBase RAN data containing NaNs, with five different targets represented in separate colors. (B1-B2) Illustrate the same signal as in (A1) and (A2) but with NaNs replaced (if there is any) using a forward-filling approach. (C1-C2) The defined trigger-events in both channels along with the targets. Trigger-events were defined using our simulation methodology with IKF, dwell time of 100 ms, and buffer-period of 1000 ms. The fourth row contains five subplots, each representing an individual target from the signal above with the respective defined trigger-event. Angular offsets are annotated above each subplot.
• Figure 2: Supplementary Fig. 2. Details are same as Figure \ref{['fig:1018']}