Robust foreground detection using block-based RPCA

Abstract Robust foreground detection is difficult in real-life applications due to complex factors such as illumination change and background interference. A robust foreground detection approach using block-based RPCA is proposed in this paper. The input matrix is segmented into different blocks according to the initial segmentation. Trajectories of moving objects are utilized for weights calculation. Foreground detection is described as a block-based RPCA which is solved using IALM. Final processing is used to improve the detection results. Both qualitative and quantitative experiments between our detector and some commonly used detectors indicate the precision and robustness of the proposed foreground detection approach.