3D Face Point Cloud Reconstruction and Recognition Using Depth Sensor

Area of Science:

• Computer Vision
• Artificial Intelligence
• Biometrics

Background:

• Facial recognition systems often rely on RGB data, limiting performance in low-light conditions.
• Current AI-driven facial recognition methods demand significant computational resources (CPU/GPU) and large datasets.
• Existing approaches struggle with varying illumination, posing challenges for real-world applications.

Purpose of the Study:

• To develop an effective 3D facial reconstruction and recognition method using depth sensors.
• To address the limitations of RGB-based facial recognition in low-light or dark environments.
• To create a computationally efficient facial recognition solution suitable for smart devices.

Main Methods:

• Acquiring multi-view 3D point clouds using a depth camera.
• Reconstructing 3D face models by stitching point clouds with the Iterative Closest Point (ICP) algorithm.
• Segmenting 3D models to isolate facial features and extracting geometric information (normal, curvature).
• Implementing a feature-based 3D facial similarity score for recognition.

Main Results:

• The proposed method successfully reconstructs denser 3D face models from depth data.
• Accurate 3D facial recognition is achieved even in dark environments.
• The system demonstrates effectiveness in real-world experiments, enabling correct person labeling.
• The method is suitable for integration into smart devices with depth sensing capabilities.

Conclusions:

• Depth sensor-based 3D facial recognition offers a robust alternative to RGB-based methods, especially in challenging lighting.
• The proposed feature-based similarity score effectively utilizes geometric information for accurate identification.
• This approach provides a computationally efficient and accurate solution for facial recognition on mobile platforms.