Behavioral Biometrics in VR: Changing Sensor Signal Modalities.

Area of Science:

• Computer Science
• Biometrics
• Virtual Reality

Background:

• Virtual reality (VR) and metaverse development necessitates advanced user verification.
• Biometric authentication offers enhanced security over traditional passwords.
• Existing methods using controller data with CNNs face challenges with non-continuous Euler angles.

Purpose of the Study:

• To address limitations in VR biometric authentication using CNNs.
• To propose a novel modality transformation for dynamic signal generation.
• To improve the accuracy and robustness of user verification in VR environments.

Main Methods:

• Generated acceleration and angular velocity signals from trajectory and orientation data.
• Employed quaternion algebra for modeling dynamic signals.
• Trained various CNN architectures (Vanilla, attention-enhanced, Multi-Input) on original and transformed data.

Main Results:

• The modality transformation approach significantly improved F1-score accuracy across datasets.
• Accuracy increased from 0.80 to 0.82 (Comos), 0.77 to 0.82 (Quest), and 0.83 to 0.92 (Vive).
• The proposed method demonstrated superior performance compared to using raw trajectory data.

Conclusions:

• The novel modality transformation effectively enhances VR biometric authentication.
• Quaternion algebra-based signal generation improves CNN performance for user verification.
• This approach offers a more robust and accurate solution for securing metaverse interactions.