Correlation filter tracker based on sparse regularization

To lower the boundary effects, we propose a multi-scale ℓ1 regularized correlation filter tracker (MSL1CFT), which leverages the different regularization parameters to penalize each correlation filter coefficient in the learning process. Our method can learn the correlation filter model on a significantly larger set of negative training samples, without worsening the positive samples. We further present a fast solver to our model utilizing the Alternating Direction Method of Multipliers (ADMM) technique. The extensive empirical evaluations on two benchmark datasets: OTB2013 and VOT2015 demonstrate that our method outperforms the state-of-the-art approaches in tracking accuracy and robustness.