Reconstructing Volume and Motion from Real Fluid Phenomena with Minimal Number of Camera Views