Stop getting hit by self-driving cars with this one fashion trick that involves putting weird labels on all your clothing! Don’t be the last one to catch on to this new fashion trend.


In a hypothetical future where self-driving cars are increasingly common, they’ll have to do a really good job of automatically distinguishing between things that require sudden braking (e.g. a person in the roadway) and things that are OK to hit (e.g. a tumbling empty cardboard box).

The issue:

This is a hard problem. When a car gets data from its various cameras (and other sensors), it needs to figure out what exactly it is that it is seeing (Figure 1).


Fig. 1: This is probably a pedestrian in the roadway, but could it also be a billboard advertisement hundreds of feet away?

Although the specific “distant-billboard-or-close-pedestrian” question in Figure 1 can be answered just by using two cameras to estimate distance, there are situations where the problem must be resolved in a more complex fashion (Figure 2).


Fig. 2: Top: the image is interpreted correctly, and the car does NOT hit the pedestrian. Bottom: the car incorrectly believes that it sees a sunflower, and collides with it at full speed. Lest you think this is totally implausible, check out some specially-crafted adversarial examples (that can turn a panda into a banana) and a method of tricking lane-following algorithms into swerving the car into oncoming traffic.


We propose to place special “this is a human” symbols on articles of clothing that a human might wear (Figure 3).

When a car sees one of these unusual QR-code-like symbols, it will instantly say “ah, sunflowers do not wear specially-marked shoes, time to hit the brakes!”

To avoid this becoming a fashion disaster, these markings would not be apparently at normal human-visible wavelengths of light, but would only be detectable by special camera equipment.

Perhaps the markings could have fluorescent ink in them, and all cars could drive around with UV lights in the front.


Fig. 3: Left: this is what the shoe looks like to a human—the markings are invisible to the naked eye. Middle: the camera can see wavelengths of light beyond human ability, and can detect these special markings (shown here as yellow checkerboards). Right: the camera sees the checkerboard, and the object-classification algorithm realizes that this shoe is likely to be attached to a human.

One common objection to many self-driving-car-related issues is “couldn’t some criminal put these markers all over the city, to trick self-driving cars?”

The answer is yes, but it would be as equally illegal as it currently is to put mannequins on a winding road (which would also confuse human drivers).


This might be redundant with an infrared camera—in most locations, a human already is obviously distinguished from the background environment just by their warm-blooded glow in the infrared spectrum.

PROS: This will definitely make me a ton of money when it is licensed by major car manufacturers. Also, would someone please apply for and pay for a patent on my behalf? Thanks!

CONS: If one of these specially-marked shoes falls onto the roadway (perhaps by falling out of someone’s messenger bag while they’re biking), do we really want every car to come to a screeching halt at the sight of a single unattached shoe?