Game Development Reference
In-Depth Information
Figure 3. For the creation of the eigenfeature database, several images of
the studied features are segmented, then normalized and finally analyzed
using Principal Component techniques. Diagram courtesy of the Instituto
de Matemática e Estatística at the Universidade de São Paulo.
∂
y
u
=
∂
x
v
=
are the pixel displacements between two images. Each
point on the image has one equation with two unknowns,
u
and
v
, which implies
that motion cannot be directly computed. There exist different methods that try
to solve (1) iteratively.
A complete bibliographical compilation of different optical flow methods can be
found in Wiskott (2001).
where
and
∂
t
∂
t
Principal component analysis — Eigen-decomposition
Optical flow methods are extensively used in shape recognition, but they do not
perform well in the presence of noise. If we want to identify a more general class
of objects, it is convenient to take into account the probabilistic nature of the
object appearance and, thus, to work with the class distribution in a parametric
and compact way.
The Karhunen-Loève Transform meets the requirements needed to do so. Its
base functions are the eigenvectors of the covariance matrix of the class being
modeled:
Search WWH ::
Custom Search