Facial Motion Capture with Sparse 2D-to-3D Active Appearance Models

Student thesis: Master thesis (including HD thesis)

  • Esben Plenge
This thesis presents a framework for markerless 3D facial motion capture. In particular it considers, if emotions can be conveyed by 3D facial motion capture from sparse geometry. A method dubbed 2D-to-3D Active Appearance Models has been developed for retrieving sparse 3D geometry from 2D images. The established motion capture framework is modular. In its current implementation it includes a tool, developed in Matlab, for acquiring 3D facial geometry data for training a statistical model, a method for synthesizing 3D facial geometry from 2D images has been written in C++, and a 3D animation front end, that visualizes the captured face data, also written in C++. The AAM-API, an open source C++ implementation of Active Appearance Models, has been used in this work. A thorough description of the theory behind the framework is given in the thesis, and its implementation is discussed in detail. The emotion conveyance capability of the framework implementation is tested, and it is shown, that the emotions embedded in the captured and animated facial expressions can be interpreted correctly. Keywords: Facial Motion Capture, Stereopis, Facial Animation, 3D Geomety Synthesis, Active Appearance Models, Human-Computer Interfaces.
Publication date2008
Number of pages96
Publishing institutionAalborg University Copenhagen
ID: 14979083