Text this: Dual‐view 3D human pose estimation without camera parameters for action recognition