GEOMETRIC MODEL-BASED 3D FACE TRACKING - 3D Face Processing: Modeling, Analysis and Synthesis - page 29

Chapter 4

GEOMETRIC MODEL-BASED 3D FACE
TRACKING

To achieve robust 3D non-rigid face tracking, facial motion model is needed.
In this chapter, we first review previous works on face tracking in Section 1.
Next, in Section 2, we describe the geometric MU-based facial motion track-
ing algorithm. Section 3 will describe applications of our geometric 3D face
tracking algorithm.

1. Previous Work

Analysis of human facial motion is the key component for many applications,
such as model-based very low bit rate video coding for visual telecommuni-
cation [Aizawa and Huang, 1995], audio/visual speech recognition [Stork and
Hennecke, 1996], expression recognition [Bartlett et al., 1999]. A large amount
of work has been done on facial motion tracking. Simple approaches only uti-
lize low-level image features. Although their computation complexity is low,
they are not robust enough. For example, Goto et al. [Goto et al., 2001] extract
edge information to find salient facial feature regions (e.g. eyes, lips, etc.). The
extracted low-level image features are compared with templates to estimate the
shapes of facial features. However, it is not robust enough to use low-level im-
age features alone. The error will be accumulated with the increase in number of
frames being tracked. High-level knowledge of facial deformation must be used
to handle error accumulation problem by imposing constraints on the possible
deformed facial shapes. It has been shown that robust tracking algorithm needs
to integrate low-level image information and high-level knowledge. Examples
of high-level constraints include: (1) parameterized geometric models such as
B-Spline curves [Chan, 1999], snake model [Kass et al., 1988], deformable
template [Yullie et al., 1992], and 3D parameterized model [DeCarlo, 1998];
(2) FACS-based models [Essa and Pentland, 1997, Tao and Huang, 1999]; and