[论文解读] Cybersickness Detection through Head Movement Patterns: A Promising Approach

Masoud Salehi, Nikoo Javadpour+3 more|arXiv (Cornell University)|Feb 5, 2024

Robotics and Automated Systems被引用 7

一句话总结

本论文将头部运动模式作为一种连续的、非侵入式的头晕感检测标志在VR中进行研究，并报告了有希望的预测性能。

ABSTRACT

Despite the widespread adoption of Virtual Reality (VR) technology, cybersickness remains a barrier for some users. This research investigates head movement patterns as a novel physiological marker for cybersickness detection. Unlike traditional markers, head movements provide a continuous, non-invasive measure that can be easily captured through the sensors embedded in all commercial VR headsets. We used a publicly available dataset from a VR experiment involving 75 participants and analyzed head movements across six axes. An extensive feature extraction process was then performed on the head movement dataset and its derivatives, including velocity, acceleration, and jerk. Three categories of features were extracted, encompassing statistical, temporal, and spectral features. Subsequently, we employed the Recursive Feature Elimination method to select the most important and effective features. In a series of experiments, we trained a variety of machine learning algorithms. The results demonstrate a 76% accuracy and 83% precision in predicting cybersickness in the subjects based on the head movements. This study contribution to the cybersickness literature lies in offering a preliminary analysis of a new source of data and providing insight into the relationship of head movements and cybersickness.

研究动机与目标

• 通过一种新颖的生理标志物（头部运动）在VR中推动对 cybersickness 的检测。
• 使用公开的VR数据集，在六个轴上提取丰富的头部运动特征。
• 比较基于头部运动特征的机器学习模型来预测 cybersickness。

提出的方法

• 从公开的VR数据集中提取六个轴向的头部运动数据。
• 计算导数（速度、加速度、跃度）并导出统计、时域和频域特征。
• 应用递归特征消除以选择相关特征。
• 训练一系列机器学习分类器并在 cybersickness 预测上评估性能。

实验结果

研究问题

• RQ1 头部运动模式是否能作为VR中 cybersickness 的可靠指标？
• RQ2 头部运动的哪些特征（统计、时域、频域）最能预测 cybersickness？
• RQ3 在使用头部运动特征进行 cybersickness 检测时，不同机器学习模型的有效性如何？

主要发现

• 基于头部运动的特征可以以 76% 的准确度和 83% 的精确度预测 cybersickness。
• 三类特征集（统计、时域、频域）捕捉了与 cybersickness 相关的有意义信息。
• 递归特征消除有助于识别对分类最具信息量的特征。
• 该研究初步提供了头部运动与 cybersickness 之间关联的证据，并为未来的研究提出了新的数据来源。