Context-Adaptive Gait Biometrics for Real-Time Continuous Authentication

This paper proposes a continuous authentication system leveraging gait analysis, integrating real-time image processing and machine learning to address vulnerabilities in traditional one-time verification methods. By utilizing gait—a natural and spoof-resistant biometric—the framework enables non-intrusive, real-time identity validation. A Microsoft Kinect sensor captures 3D kinematic features (speed, stride length, joint angles), which are processed via machine learning models to authenticate users against pre-trained profiles. The system dynamically adapts to environmental variables (lighting, terrain) and mitigates gait variability (e.g., injuries) through confidence-based early stopping, optimizing computational efficiency. Privacy is preserved via encrypted data storage and processing. Evaluations demonstrate robust performance in controlled settings, with scalability for high-security applications such as access control. While achieving promising accuracy, challenges persist in heterogeneous environments, underscoring the need for multi-sensor integration. This work advances gait biometrics by balancing security and usability, offering a passive authentication solution that enhances both convenience and protection against impersonation threats.