Robust Coding in a Global Subspace Model and its Collaboration with a Local Model for Visual Tracking