Tim Hotze
Tim Hotze
Aug 22, 2017 · 1 min read

This is one of the reasons that I’m skeptical that systems based entirely on vision will be the way to go. If you had a system that synthesized say, cameras and radar, the camera might say “there’s a pedestrian up ahead” but the radar would say “there’s nothing for the next 50 meters except some object staying at a constant distance.” If the machine knew it was going at highway speeds, it’d go to follow that the object was also going at highway speeds — and so was an image on the back of the truck, and NOT a person.

Of course, any self-driving system will have to learn to deal with images of people: People on sidewalks (maybe OK as long as they’re travelling parallel to the vehicle and to the side), people on bicycles and motorcycles (which may be in parallel and in front of the vehicle), even people on bicycles standing, which might look a lot like a person.

)
    Tim Hotze

    Written by

    Tim Hotze