Abstract:
This study presents a novel approach for fast and robust passive live face detection
designed for enhanced access control systems. The proposed solution integrates stereo
vision technology with advanced deep learning models to deliver a high-performance
authentication mechanism. Utilizing the OAK-D Lite stereo vision camera, the system
captures both 2D and 3D facial data, enabling the extraction of depth information to
distinguish between live human faces and spoofing attempts such as photos, videos, or
masks. A custom deep learning model is developed, specifically optimized to recognize
subtle depth variations in real time. The study emphasizes the model’s capability to operate
without active user engagement, such as blinking or head movements, providing a seamless
and secure authentication process. To train the model, a comprehensive dataset of stereo
facial images was collected, followed by meticulous data preprocessing and augmentation.
The training process included the use of state-of-the-art deep learning techniques,
leveraging stereo-view depth information to improve the model's robustness against
spoofing attacks. The performance of the model was evaluated based on several metrics,
including accuracy, precision, recall, and F1 score, showcasing a significant improvement
in spoof detection accuracy when compared to traditional 2D facial recognition systems.
The results indicate that the integration of stereo vision and deep learning dramatically
reduces the false acceptance rate in face authentication systems while maintaining real
time processing capabilities crucial for access control environments. By addressing key
limitations in current facial recognition technology, this research contributes a highly
secure and efficient face authentication method, paving the way for its application in
industries requiring rigorous access control such as banking, healthcare, and corporate
security.
