Video foreground detection via sparse representation

In this article we propose a novel background model method based on sparse representation theory. In this framework the image is divided into N×N non-overlapping blocks, each block of a new input image has a sparse representation in the space spanned by background model. The sparse representation is achieved by solving an ℓ₁-regularized least squares problem which is computed by the preconditioned conjugate gradients (PCG) algorithm. Then the block with large error vector is taken as the foreground region. The experimental results show that proposed algorithm produces more accurate and stable results.