Eye Movement Feature-Guided Signal De-Drifting in Electrooculography Systems
Lianming Hu, Xiaotong Zhang, Kamal Youcef-Toumi · 2025
Read Eye Movement Feature-Guided Signal De-Drifting in Electrooculography Systems by Lianming Hu, Xiaotong Zhang, Kamal Youcef-Toumi (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-07416.txt.
Introduction
Eye-tracking signals drift over time, and that drift corrupts gaze estimates in human-robot teamwork. Lianming Hu, Xiaotong Zhang, and Kamal Youcef-Toumi present a de-drifting method in a 2025 research paper. Their method works on electrooculography, which measures eye movement through voltages at the skin. It uses recognized eye-movement features to rebuild the signal baseline.
The approach corrects drift adaptively while keeping the waveform's shape intact. In simulation, the average error fell by more than a third. In real-world recordings, it fell by about a quarter, despite extra unpredictable noise. The method beat conventional de-drifting techniques in both settings. Cleaner eye signals could make gaze control steadier for assistive robots, hands-free interfaces, and performance support tools used by workers.
Catalog record
| CREATOR | Lianming Hu, Xiaotong Zhang, Kamal Youcef-Toumi |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-07416
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Eye Movement Feature-Guided Signal De-Drifting in Electrooculography Systems Lianming Hu, Xiaotong Zhang, Kamal Youcef-Toumi Abstract Electrooculography (EOG) is widely used for gaze tracking in Human-Robot Collaboration (HRC). However, baseline drift caused by low-frequency noise significantly impacts the accuracy of EOG signals, creating challenges for further sensor fusion. This paper presents an Eye Movement Feature-Guided De-drift (FGD) method for mitigating drift artifacts in EOG signals. The proposed approach leverages active eye-movement feature recognition to reconstruct the feature-extracted EOG baseline and adaptively correct signal drift while preserving the morphological integrity of the EOG waveform. The FGD is evaluated using both simulation data and real-world data, achieving a significant reduction in mean error. The average error is reduced to 0.896° in simulation, representing a 36.29% decrease, and to 1.033° in real-world data, corresponding to a 26.53% reduction. Despite additional and unpredictable noise in real-world data, the proposed method consistently outperforms conventional de-drifting techniques, demonstrating its effectiveness in practical applications such as enhancing human performance augmentation. I INTRODUCTION Human-robot collaboration (HRC) has become a trending solution in both industry and daily life scenarios. In some cases, it has also proven to be a better solution than traditional robotic systems, as it combines the strengths of both humans and robots: humans exhibit superior flexibility and adaptability, while robots excel in accuracy, cost-effectiveness, and efficiency [ 1 ] [ 2 ] . However, to ensure an efficient and productive workflow in HRC, robust and effective communication between humans and robots must be established. As naturally efficient collaborators in HRC, humans have various ways to convey information, either explicitly or implicitly [ 3 ] . Among all modalities of information transfer, the gaze is an implicit yet rich source of data that enables robots to infer human states, revealing both physical and cognitive activities, including attention, intention, and relevance [ 4 , 5 , 6 ] . This information is particularly valuable in enhancing task execution, planning, and prediction in HRC. Currently, there are various ways of tracking the human gaze. A common method for gaze tracking nowadays is a camera-based oculography system that tracks certain parts of the eye’s movement relative to the reference point of the eye to predict the gaze [ 7 ] . However, certain limitations are also found in the camera-based eye gaze tracking method. For example, a stable environment with an ideal level of illumination and a high-resolution camera is required to ensure the necessary eye image or video can be well captured. Also, this method cannot work if the user turns to the other direction where the camera fails to capture the eye region [ 8 ] . All these limitations make camera-based eye-tracking expensive and cumbersome for a more dynamic environment such as HRC. Electrooculography (EOG) is another gaze-tracking technology that utilizes sensors to detect changes in corneal-retinal potential (CRP) caused by human eye movement. It relies on tiny skin-surface electrodes as the only input, making it a simpler setup for many user-controlled devices while offering greater flexibility, as users can move freely. Additionally, since EOG does not require a camera, it avoids high computational resource demands, large power supply requirements, and privacy concerns. As a result, it has the potential to become a more practical and accessible gaze tracking technique in the HRC field. Current EOG solutions still have limitations, and many challenges remain active topics of research today. One major challenge for EOG to be used in a long-term timeframe is the drift of the signal. The drift will slowly add error to the baseline of the EOG signal, causing a shifting of the user’s absolute gaze angle prediction from the true value. Much work has been done to address different solutions in mitigating the drift effect, including high pass filtering [ 9 ] , polynomial fitting [ 10 ] , signal differencing [ 11 ] , wavelet decomposition [ 12 ] , and techniques based on baseline component recognition [ 13 ] . These techniques are common methods for achieving de-drifting. However, each has its limitations, affecting practicality in different circumstances. For example, while high-pass filtering is a simple and effective solution for both offline and real-time de-drifting, it fails to preserve the original morphology of the EOG signal, distorting the waveform and altering the amplitude of key features such as saccades. In contrast, more complex systems, such as baseline component recognition, can successfully de-drift while preserving signal morphology but require prior knowledge of each subject’s cue-target information during signal acquisition. This necessitates a fixed trial protocol, which must be predefined and known by the system before data processing, significantly limiting its practicality in free-movement EOG use cases. In this work, a novel Feature-Guided De-drifting (FGD) method was presented to minimize the drift effect. The FGD method extracts the drift trend from the raw EOG signal by first identifying saccades, one of the most critical features of eye movement. It then applies an adaptive window to accurately exclude saccades from the EOG signal, reconstruct the remaining signal as a new baseline, and finally use 1D multilevel wavelet decomposition to approximate the drift trend for de-drifting. Notably, while both FGD and the baseline component recognition system by N. Barbara et al. [ 13 ] perform de-drifting through feature extraction on the EOG baseline, FGD does not require prior knowledge from a fixed trial protocol. This makes it well-suited for most HRC scenarios, where the user’s gaze angle and gaze transitions are not predetermined but freely chosen. Additionally, as a feature-guided approach, the proposed method has lower sampling frequency requirements, making it more suitable for cost-effective systems with greater practicality. In both simulation and real data testing, this method has successfully removed the drift without distorting the morphology of the EOG signal. More specifically, the key contributions of this paper are as follows: • A novel de-drifting approach for EOG signals in free-movement use cases, leveraging active feature extraction to enhance de-drifting while preserving the morphology of the original signal. • An adaptive feature extraction method that dynamically aligns with the actual saccade time window while reconstructing the baseline, improving the de-drifting process’s precision, robustness, and signal stability. • A seamless transition from simulation to real-world implementation, demonstrating the proposed method’s effectiveness in preserving EOG signal morphology while achieving accurate de-drifting. II RELATED WORK To the best of our knowledge, our work is very unique. In the following sub-sections, some works related to ours are summarized. II-A Human Eye Gaze Tracking The primary patterns of human eye movement can be categorized into three types: fixation, blink, and saccade [ 12 ] . Various technologies have been developed to track eye movements and predict gaze direction. For instance, contact-based tracking methods require physical contact with the user such as contact lenses, electrodes, head-mounted devices, etc. In contrast, contactless eye trackers have also been developed to minimize intrusiveness by tracking eye movement without physical contact. These systems are mostly vision-based, with camera-based eye tracking being the most common approach [ 7 ] . II-A 1 Camera-Based Eye Tracking Camera-based eye tracking requires the camera to monitor the features of the human eye and perform computer vision-related analyses of the figure or video captured. For example, pupil-corneal reflection techniques use a