Adaptive pixel-block based background subtraction using low-rank and block-sparse matrix decomposition

We present three stages of a novel backgrounds subtraction method in this paper: a new pixel-block based randomized arrangement is utilized to preprocess all the frame images, so that low-rank property of background and sparsity of foreground can be separated more easily; different foreground regions have different sparsity, we use a set of adaptive parameters for subtracting foregrounds according to the variances of frame pixels; finally, background model is built via an improved low-rank and block-sparse matrix decomposition based on the proposed adaptive pixel-block background subtraction. All these key measurements guarantee the considerable performance in background subtraction, which are also demonstrated in our experimental results.