The Eye–Head Mover Spectrum: Modelling Individual and Population Head Movement Tendencies in Virtual Reality

Paper Title

The Eye–Head Mover Spectrum: Modelling Individual and Population Head Movement Tendencies in Virtual Reality

Publication Info

• Topic area: Individual differences in eye–head coordination during gaze shifts in VR.
• Keywords: Eye–head coordination, VR, gaze shifts, head movement, individual differences, foveated rendering, adaptive systems, user modeling, population distribution, functional PCA.

Background and Problem

• Problem / challenge: Existing research on eye–head coordination is limited by categorical definitions, reliance on thresholds, and small sample sizes. There is no continuous, scalable model to capture individual differences in head movement tendencies in VR.
• Significance: Understanding head movement tendencies is critical for designing adaptive VR systems, improving comfort, and enhancing interaction techniques.
• Motivation and related work: Prior studies in psychology and biomechanics identified head movers and non-head movers but relied on thresholds and small samples. HCI research has not systematically modeled head contribution in VR, leaving a gap in understanding individual differences and their implications for system design.

Solution

• Proposed approach: Introduction of the eye–head mover spectrum, a continuous model of head contribution to gaze shifts, using a soft-hinge parametric function to capture individual differences and population-level distributions.
• Novelty:
  1. Development of a user-specific, continuous model of head contribution to gaze shifts.
  2. Analysis of a large 360° video dataset (N=80) to map population-level distributions.
  3. Controlled user study (N=28) to test stability and context-dependence of individual strategies.
  4. Application of functional PCA to reveal dominant axes of variation in head movement tendencies.
• Procedure and key techniques:
  • Preprocessing of gaze and head data from a large VR dataset and a user study.
  • Fitting of three parametric models (linear, hinge, soft hinge) to individual data, with the soft hinge selected for its accuracy and interpretability.
  • Functional PCA to analyze population-level distributions and individual variability.
  • Comparison of head movement tendencies across tasks (abstract target selection vs. 360° video viewing).

Results

• Concrete findings:
  • The soft-hinge model explained 91.1% of variance in head contribution across participants in the dataset.
  • Head contribution varied from 5–15° at 20° target eccentricity to 30–45° at 50°, reflecting a wide spectrum of strategies.
  • In the user study, head contribution strategies were stable across tasks, with a significant correlation (r = 0.60, p < 0.001) between abstract and video tasks.
• Advantage over baselines:
  • The soft-hinge model outperformed linear and hinge models in explained variance (R²) and error (RMSE).
  • Avoided reliance on arbitrary thresholds and captured gradual transitions in head involvement.
• Experiments / evaluation:
  • Dataset analysis: D-SAV360 dataset (N=80) with 360° video free-viewing.
  • User study: Abstract target selection and 360° video tasks (N=28), using the Meta Quest Pro HMD.
  • Metrics: R², RMSE, functional PCA, and subjective feedback.
• Limitations and future work:
  • Limited to horizontal gaze shifts within ±50°; vertical and 3D coordination remain unexplored.
  • Tasks focused on free viewing and abstract selection; other VR contexts (e.g., locomotion, interaction) need investigation.
  • Demographic diversity and hardware ergonomics may influence results.
  • Future work should extend to larger datasets, dynamic contexts, and full 3D coordination.

Summary

This paper introduces the eye–head mover spectrum, a continuous model of head movement tendencies in VR, capturing individual differences in eye–head coordination. Using a soft-hinge parametric model, the authors analyzed a large 360° video dataset and conducted a controlled user study, revealing a spectrum of strategies from eye-leaning to head-leaning. The findings demonstrate that these strategies are stable within individuals but shift with task context. This work has direct implications for adaptive VR systems, including foveated rendering, viewport alignment, and inclusive design, and provides a foundation for future research on individual differences in VR interactions.