Automatic background subtraction in a sparse representation framework
Ciprian David, Vasile Gui · 2013
An automatic sparse representations based approach for background subtraction is proposed in this paper. The background model is composed of a dictionary and a set of average decomposition coefficients. For this purpose a set of training frames is used to obtain a K-SVD dictionary. The same training set is used to compute decomposition coefficients by orthogonal matching pursuit. The final background-foreground segmentation is obtained by a two step thresholding operation. First, a locally thresholded image is obtained by using a previously estimated threshold. The final binary map is computed by a k-means estimated global threshold. Our approach is compared both quantitatively and qualitatively with state-of-the-art approaches.