Real-time 3D face localization and verification

Computer Vision;3D;Face Detection;Face Verification;Face Recognition;Head Pose Estimation;Depth Cameras

Meyer, Gregory Paul

关键词: Computer Vision;    3D;    Face Detection;    Face Verification;    Face Recognition;    Head Pose Estimation;    Depth Cameras;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/95341/MEYER-DISSERTATION-2016.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

We present a method for real-time 3D face localization and verification using a consumer-grade depth camera. Our approach consists of three parts: face detection, head pose estimation, and face verification. Face detection is performed using a standard detection framework which we significantly improve by leveraging depth information. To estimate the pose of the detected face, we developed a technique that uses a combination of the particle swarm optimization (PSO) and the iterative closest point (ICP) algorithm to accurately align a 3D face model to the measured depth data. With the face localized within the image, we can compare a database 3D face model to the depth image to verify the identity of the subject. We learn a similarity metric using a random decision forest to accurately authenticate the subject. We demonstrate state-of-the-art results for both face localization and face verification on standard datasets. Since the camera and our method operate at video rate, our system is capable of continuously authenticating a subject while he/she uses his/her device.