That's the core problem. Cameras don't do distance well. Radar doesn't do size/shape well. And human eyeballs are generally bad at both. Only when multiple different sensors all see the same thing at the same time/space can we start extrapolating size and shape.