3D Face Modeling and Head Pose Tracking with Comsumer RGB-D Cameras

Abstract

We propose a method to address the issue of face modeling and head pose estimation from a depth image. We calculate head poses by aligning face point cloud acquired by a consumer depth camera with a 3D face model. However, due to the uniqueness of each individual’s biometrics, the pre-defined face model may not completely fit the real face of each person, which leads to errors in the estimated head pose. In order to address this problem, we use a deformable model to characterize the biometric differences of the face. The deformable face model has been proven to be effective. The method is used for automatic head pose annotation for RGB face images.

Demo

• Raw 3D Face Scanning
• Non-Rigid 3D Model Fitting and Face Model Deformation
• Accurate Head Pose Tracking with A Comsumer RGB-D Camera