A tracking algorithm based on online learning of discriminant dictionary has been proposed, which combines dictionary items with label information, thus making the discriminant dictionary both reconstructive and discriminant. In order to enhance the discriminant ability of the model, the classifier is embedded in the target representation model, with its candidate target to be determined according to the reconstruction error and discriminant classification score. The online dictionary learning algorithm is used to update both the dictionary and the classifier in the process of learning, so that the model can adapt to the dynamic changes of the target appearance and background environment. The experimental results show that the proposed method achieves satisfactory results in most of such tests concerning large occlusion, fast motion, strong light and attitude change.