INTRODUCTIONpropose a simple and fast background

subtraction scheme without background model updating, and yet it is tolerable to changes in room lighting for indoor surveillance using Independent Component Analysis (ICA).

ICA-BASED BACKGROUND SUBTRACTIONA. Basic ICA ModelThe observed mixture signals X can be expressed

as X = AS where A is an unknown mixing matrix, and

S represents the latent source signals.

The ICA model describes how the observed mixture signals X are generated by a process that uses the mixing matrix A to mix the latent source signals S.

ICA-BASED BACKGROUND SUBTRACTIONA. Basic ICA ModelThe source signals are assumed to be mutually

statistically independent. Based on the assumption, the ICA solution is obtained in an unsupervised learning process that finds a de-mixing matrix W. The matrix W is used to transform the observed mixture signals X to yield the independent signals, i.e., WX = Y. The independent signals Y are used as the estimates of the latent source signals S. The components of Y, called independent components, are required to be as mutually independent as possible.

ICA-BASED BACKGROUND SUBTRACTIONB. Proposed ICA Model In order to separate foreground objects from the

background in a scene image, we need at least two sample images to construct the mixture signals in the ICA model.

Let the sample images be of size m X n. Each sample image is organized as a row vector of K dimensions, where K = m ∙ n. Denote by Xb = [xb1 , xb2 , ∙ ∙ ∙ , xbK]the reference background image containing no foreground objects, and Xf = [xf1 , xf2 , ∙ ∙ ∙ , xfK]the foreground image containing an arbitrary foreground object in the stationary background.

ICA-BASED BACKGROUND SUBTRACTIONB. Proposed ICA Model In the training stage of the proposed background

subtraction scheme, the ICA model is given by Y = W∙XT

EXPERIMENTAL RESULTSSingle Reference Background

EXPERIMENTAL RESULTSSingle Reference Background

EXPERIMENTAL RESULTSMultiple Reference Background

EXPERIMENTAL RESULTSMultiple Reference Background

EXPERIMENTAL RESULTSEffect of Varying Foreground Objects