Digital Signal Processing Reference
In-Depth Information
12.4 Background Modeling
Background modeling refers to the process of creating, and subsequently maintain-
ing, a model of the appearance of the background in the field of view of a cam-
era. Background modeling is often used in identifying moving objects from a static
camera. One of the commonly used methods to extract moving objects is back-
ground subtraction method. Background subtraction involves background modeling,
background initialization, background maintenance and foreground detection. Back-
ground modeling is the crucial step in foreground detection. In real time applications,
there are few problems that influence background modeling [
17
]. Some of the major
problems in background modeling can be listed as:
Illumination changes:
A sudden or gradual change in illumination alters the back-
ground. Consequently, there is a deviation of the background pixels from the back-
ground model. This results in increase in the number of falsely detected foreground
regions and in the worst case the whole image may appear as foreground.
Moving objects:
The background may contain moving objects e.g. waving trees.
A foreground object that becomes motionless cannot be distinguished from a back-
ground object that moves and then becomes motionless (sleeping person) and these
objects should not be considered as a part of the foreground [
18
]. During initializa-
tion, many background modeling algorithms require a scene with no moving objects
in the model as the moving object occludes the actual background.
Shadows:
A shadow occurs when an object partially or totally occludes direct
light from a source of illumination [
19
]. Shadows may be classified into two major
classes: self and cast shadows. A self shadow occurs in the portion of an object which
is not illuminated by direct light and cast shadow is the area projected by the object
in the direction of direct light. Cast shadows may be classified as foreground due to
change of illumination in the shadow region. Background modeling system must be
insensitive to shadows, but it should also be able to detect the moving objects.
12.4.1 Background Modeling Using Principal Component Analysis
Principal component analysis, or PCA, is a linear dimensionality reduction technique
that is widely used in lossy data compression, feature extraction, and data visualiza-
tion [
20
]. Automated learning of low-dimensional linear models from training data
has become a standard paradigm in computer vision. Principal Component Analysis
(PCA) in particular is a popular technique for parameterizing shape, appearance,
and motion. These learned PCA representations have also proven useful for solving
problems such as face and object recognition, tracking, detection, and background
modeling.
PCA is also known as the Karhunen-Loeve transform. PCA can be defined as the
orthogonal projection of the data onto a lower dimensional linear space, known as
the principal subspace, such that the variance of the projected data is maximized.
