In this paper, the authors examine the problem of tracking people in both bright and dark video sequences. In particular, this problem is treated as a background/foreground decomposition problem, where the static part corresponds to the background, and moving objects to the foreground. Having this into account, the problem is formulated as a rank minimization problem of the form X = L + S + E, where X is the captured scene, L is the low-rank part (background), S is the sparse part (foreground) and E is the corrupting uniform noise introduced in the capture process. Actually, low-rank and sparse structures are widely studied and some areas such as Robust Principal Component Analysis (RPCA) and Matrix Completion (MC) have emerged to solve this kind of problems. Here we compare the performance of three different methods in solving the RPCA optimization problem for background separation: augmented lagrange multiplier method, Bayesian markov dependency method, and bilateral random projections method. Furthermore, a preprocessing light normalization stage and a mathematical morphology based post-processing stage are proposed to obtain better results.