Text this: Video based object representation and classification using multiple covariance matrices.