Real-time 3D face modeling with a commodity depth camera

Abstract

We propose a system that enables a person to quickly and easily generate a high-quality 3D model of their face using a low-cost depth camera such as the Kinect. The Kinect sensor provides depth images at video-rate, however, the depth images are often noisy and contain holes. Our system builds high-quality 3D face models by registering and integrating multiple depth images of a user's head. This method allows a user to generate a model of their face by simply moving their head in front of a fixed depth camera. Our method is built upon the KinectFusion system with modifications to enable segmented depth images to be robustly registered and integrated. Furthermore, we present a fast technique to segment a user's head from a depth image. Our system is able to run at a real-time rate on commodity hardware.