Slides and Notes for SIGGRAPH Course 30: Performance-Driven Facial Animation

Section: Markerless Face Capture and Automatic Model Construction

Part II: Automatic Model Construction

Li Zhang  lizhangATcsDOTcolumbiaDOTedu

Latest Version:

Slides click here: zhang-slides.ppt

Complementary to the first part on "markerless face catpure" taught by Chris, this part of the section is focused on constructing realistic 3D face models. The constructed models can be used for facial animation directly, or they can be used in the model-based face tracking algorithms.

Specifically, the key technical problem we will be discussing is the fundamental computer vision problem of 3D reconstruction from 2D images. In particular, we will focus on the methods that will capture highly realistic 3D face models, and more importantly, the way how faces change from one pose to another. 

To construct the face models, two important procedures are involved: reconstructing the shapes and computing the motion between shapes. For shape reconstruction, we will discuss triangulation based methods as well as non triangulation based ones. The former includes laser scanners, structured light sensors, and stereo systems; the latter includes defocus sensors and time-of-flight sensors. For motion estimation, we will discuss marker based approach and template fitting approach.

Laser Scanners:

Structured Light Sensors:

Stereo Systems:

Time-of-Flight Sensors:

Defocus Sensors:

Marker-based Motion Estimation:

Template-Fitting for Motion Estimation:

Commercial Systems: