Vision-Based Human Action Recognition: A Sparse Representation Perspective