It borders on reckless than Tesla isn't using LiDAR or a similar sensor.
Relying on images only requires you to perfectly identify every object in the scene in order to infer its bounding box. And that's an issue like we see here where you don't expect to see private jets near a Tesla and so it may not be modelled.
With LiDAR, Radar etc. at least they know "something" is there even if they don't know what is.