Send me the app please (iPhone 6S), I'll end you my opinion of its efficacy.

Information overload: the computer that's driving the self-driving car is already in great trouble telling the difference between a shadow and a pothole from a direct camera image.

Now you're asking the AI to interpret invisible objects from slight color variations in the shadows - which it already has troubles identifying correctly as shadows in the first place.

That leads to worse outcomes as the computer is putting one shaky interpretation on top of another, which makes the machine prone to imagining things that just aren't there, or missing what is obviously there. After all, we're still at the stage where the computer is not quite sure whether it's even on the road at all, or what "road" means.