Improved Action Recognition by Combining Multiple 2D Views in the Bag-of-Words Model