EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Robust Monocular Gaze Estimation Under Free Head Motion via Multi-Pose Calibration and State-Dependent Fusion

Jiarui Meng, Menghan Hu

IEEE Open Journal on Immersive Displays · 2026

DOI: 10.1109/ojid.2026.3738705 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Open Journal on Immersive Displays, 2026. Verfasst von Jiarui Meng und 1 weiteren Autor:innen. Die Arbeit nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Gaze estimation under unconstrained head motion remains a major challenge in eye-tracking research. Conventional single-pose calibration generalizes poorly across natural head rotations, often causing ocular feature degradation, gaze drift, and increased estimation noise at large head angles. We present a monocular RGB-based gaze estimation framework that enhances calibration coverage and temporal stability. A multi-pose calibration protocol collects eye-head correlation samples over a structured pose grid, extending the effective tracking region beyond single-pose calibration. A state-dependent dynamic fusion mechanism adaptively balances head-pose and ocular cues according to the current head-motion state. A decoupled dual-KF scheme independently tracks head pose and gaze state, suppressing trajectory jitter. Experiments on a private dataset show that the proposed framework reduces gaze RMSE by 24.3% with the full calibration protocol; trajectory analysis on free-viewing data (with head yaw up to 30 $^{\circ }$ ) further shows a 98.5% reduction in jump rate under a matched calibration budget. Component-wise analyses validate the effectiveness of adaptive feature screening, spatial correction, and temporal filtering. On the static MPIIFaceGaze benchmark, where temporal dynamic fusion is inapplicable, our method lowers the baseline error from 8.94 $^{\circ }$ to 4.86 $^{\circ }$ using HOG features and XGBoost; a static concatenation control at 8.69 $^{\circ }$ isolates this gain as arising from feature engineering and model choice. These results demonstrate a low-cost, hardware-agnostic solution for robust gaze estimation under dynamic head motion.

Zitieren

APA

Meng, J., & Hu, M. (2026). Robust Monocular Gaze Estimation Under Free Head Motion via Multi-Pose Calibration and State-Dependent Fusion. IEEE Open Journal on Immersive Displays. https://doi.org/10.1109/ojid.2026.3738705

Vancouver

Meng J, Hu M. Robust Monocular Gaze Estimation Under Free Head Motion via Multi-Pose Calibration and State-Dependent Fusion. IEEE Open Journal on Immersive Displays. 2026. doi:10.1109/ojid.2026.3738705

BibTeX
@article{Meng2026,
  author = {Meng, Jiarui and Hu, Menghan},
  title = {Robust Monocular Gaze Estimation Under Free Head Motion via Multi-Pose Calibration and State-Dependent Fusion},
  journal = {IEEE Open Journal on Immersive Displays},
  year = {2026},
  doi = {10.1109/ojid.2026.3738705}
}
Kostenlos starten →