Visual tracking using structural local DCT sparse appearance model with occlusion detection