A Computer Algorithm Can See When You’re Drunk

The temperature of your nose gives you away

Oh, alcohol. It can be so much fun, until it makes you do something stupid. But even without doing anything embarrassing, booze makes us blush. Now it turns out those rosy cheeks can be turned against you. Using infrared cameras, scientists have taught a computer algorithm to recognize drunk people by their facial flush.

After a glass of wine or a nice whiskey on the rocks, alcohol makes your blood vessels relax and widen. This expansion brings the blood vessels closer to your skin, making your face hot and red. The blush can deepen if you have a deficiency in the enzyme aldehyde dehydrogenase, which helps in the metabolism of alcohol. (That’s why people with Asian heritage can tend to “glow” when they’re drinking.)

To find out if they could train a computer to recognize drunk people, researchers in Greece got 41 study volunteers tipsy, and snapped thermal images of the participants’ faces before and after drinking four glasses of wine. Although different people reached different levels of intoxication from those four drinks, they all reached a blood alcohol content of at least 0.05. That’s below the legal driving limit of 0.08, but driving with a BAC of 0.05 would make you 38 percent more likely to get in a car crash compared to someone who’s completely sober.

Next the researchers had artificial neural networks compare the sober versus drunken faces, one pixel at a time. They found that the best indicator of a person’s state of intoxication is the temperature of the forehead and nose. Using that information, they built an algorithm that could determine who was sober and who was drunk with a 90 percent accuracy, even on faces that it hadn’t been trained on, and even if it didn’t know what the person looked like sober.

The researchers suggest the computer program could help police pinpoint hooligans who might start trouble, or it could be installed in cars and heavy machinery to prevent drunk driving.