Switched-randomized robust PCA for foreground and background separation in video surveillance

Abstract

In this paper we propose a new robust principal component analysis method to separate the background and foreground scenes in video surveillance. Our approach uses a random projection method called Bilateral Random Projections (BRP) in conjunction with a switching between random projection matrices and a singular value estimation technique to separate the background and moving objects. The proposed approach called switched randomized robust principal component analysis (SR-RPCA) switches among different random projection matrices and chooses the best one in order to obtain a lower distortion. To demonstrate the effectiveness of our approach, we conducted experiments on two real-time datasets and experimental results are reported.