Model optimization. At the initial frame, the feature reliability is initialized with the same value of all features. After each frame, the reliability of the feature is calculated by equation (14). Using these features, the correlation filter is learned by equation (9).