A robust tracking approach based on residual unscented particle filter (RUPF) and discriminative sparse representation is proposed to track an object accurately and efficiently. The Unscented Kalman filter is used to bring the target's observation into its proposal distribution. Then the Markov chain Monte Carlo (MCMC) improves the sampling result. The accuracy of filtering is enhanced and problems such as particle degeneration and dilution can be restrained with our RUPF. Observation likelihood is modeled based on the discriminative sparse representation

which improves the ability to extract the target from the background. The L1-regularized least squares problem in sparse representation is solved using the alternating direction method of multipliers (ADMM). Both quantitative and qualitative experiments are conducted on several challenging image sequences and the comparisons with other state-of-the-art trackers demonstrate that our tracker is more accurate than some common trackers and owns less computational complexity than traditional sparse-based trackers. With the development of the hardware

a quickly operated UKF guarantees a faster speed of our tracker with a high tracking accuracy.