User Identification via Free Roaming Eye Tracking Data

TL;DR

This work tackles user identification from eye movements in real-world settings by introducing a dual roaming dataset (FR and TR) with 41 participants using a low-cost 200 Hz eye tracker. It develops a full identification pipeline around data collection, preprocessing, rich feature extraction (M3S2K) and an RBFN classifier with $k=32$ centers, leveraging both fixations and saccades with probabilistic fusion $P_{final}(i) = 0.5 \cdot P_{fix}(i) + 0.5 \cdot P_{sac}(i)$. The study reports non-lab accuracies up to $89.4\%$ (TR) and $87.3\%$ (FR), and shows competitive performance relative to BioEye RAN under lab conditions, including a downsampled case of $95.3\%$. Top features are duration-based (fixation/saccade duration), and end-fragment trajectories yield the strongest signals, highlighting the potential of inexpensive, real-world biometric identification. The findings motivate broader training across diverse tasks and larger cohorts to enhance robustness and generalization.

Abstract

We present a new dataset of "free roaming" (FR) and "targeted roaming" (TR): a pool of 41 participants is asked to walk around a university campus (FR) or is asked to find a particular room within a library (TR). Eye movements are recorded using a commodity wearable eye tracker (Pupil Labs Neon at 200Hz). On this dataset we investigate the accuracy of user identification using a previously known machine learning pipeline where a Radial Basis Function Network (RBFN) is used as classifier. Our highest accuracies are 87.3% for FR and 89.4% for TR. This should be compared to 95.3% which is the (corresponding) highest accuracy we are aware of (achieved in a laboratory setting using the "RAN" stimulus of the BioEye 2015 competition dataset). To the best of our knowledge, our results are the first that study user identification in a non laboratory setting; such settings are often more feasible than laboratory settings and may include further advantages. The minimum duration of each recording is 263s for FR and 154s for TR. Our best accuracies are obtained when restricting to 120s and 140s for FR and TR respectively, always cut from the end of the trajectories (both for the training and testing sessions). If we cut the same length from the beginning, then accuracies are 12.2% lower for FR and around 6.4% lower for TR. On the full trajectories accuracies are lower by 5% and 52% for FR and TR. We also investigate the impact of including higher order velocity derivatives (such as acceleration, jerk, or jounce).

Figures (8)

• Figure 1: User distribution based on the age and gender.
• Figure 2: Pupil labs Neon eye tracker.
• Figure 3: Boulevard at the University of Bremen.
• Figure 4: User distribution based on the length of their trajectories in various sessions.
• Figure 5: Average number of fixations across all users per VT value for the FR and the TR dataset.