Comparing the classification performances of handcrafted 3D features against learned representations viadeep networks for 3D point clouds