Object, Scene and Actions: Combining Multiple Features for Human Action Recognition