Graphics Reference
In-Depth Information
correspondence across 3D model sequences provides a set of motion trajectories (vertex flow)
of 3D face scans. The vertex flow can be depicted on the adapted generic model (tracking
model) through the estimation of the displacement vector from the tracked points of the current
frame to the corresponding points of the first frame with a neutral expression. The vertex flow is
described by the facial motion vector U
u n ], where n is the number of vertices
of the adapted generic model. They used the Hidden Markov Model to model and train facial
dynamics.
=
[ u 1 ,
u 2 ,...,
Expression Transferring
Also known as expression cloning or performance capture when facial animation uses the
performance of an actor to animate virtual models. The steps discussed earlier, namely,
template fitting and tracking, allow expression transferring from real-time acquired 3D data
to a virtual model or puppetry. Several papers were published to transfer facial animation to
templates, puppetry or personalized models, yong Noh and Neumann (2001), Sumner and
Popovic (2004), Vlasic et al. (2005), Pyun et al. (2003), Zhang et al. (2004), Weise et al.
(2009), Weise et al. (2011), etc.
1.5 Summary and Conclusions
Creating 3D face models that look and deform realistically in an important issue is many
applications such as person-specific facial animation, 3D-based face recognition, and 3D-
based expression recognition. This chapter is a survey of successful state-of-the-art techniques
that sometimes led to commercial systems. These techniques are within a static/dynamic
(moving) face modeling-guided taxonomy. Each of the presented techniques is based on one
of the following well-known concepts: (i) depth from triangulation, (ii) shape from shading,
and (iii) depth from ToF. Obviously, other approaches exist in the literature but we limited our
survey to those based on the aforementioned concepts. In this section, we will put forward,
a comparative study of the mentioned approaches according to the intrinsic and extrinsic
factors. The intrinsic factors are related to the sensor, such as its cost, its spatial (in the
case of static modeling) or spatio-temporal resolutions (in the case of dynamic modeling),
its measurement accuracy, and its intrusiveness/need user cooperation. The extrinsic factors
include variations due to illumination changes, motion speed of the observed face, and details
in the face (wrinkles, scars, etc.). Figure 1.17 gives an evaluation of approaches according to
these criteria.
Laser-stripe scanning is intended for static faces due to the processing time required to
project the laser stripe on the whole face. The sensor is expensive and needs user cooperation
to perform face acquisition (a distance less than 1.5 m is required). Commercial systems
such as the Minolta Non-contact 3D Digitizer VIVID-910 3
produced texture and depth
images of the same resolution 640
480. The system accurately measures the 3D object
with a depth-accuracy of around 0.1 mm. It takes 2.5 s for the fine mode and 0.5 s for the fast
mode to produce a scan, thus no motion during the scan is tolerated. Laser-based techniques
×
3 http://www.konicaminolta.com/instruments/download/catalog/3d/pdf/vivid910 e9.pdf
 
Search WWH ::




Custom Search