Researchers are developing facial recognition tech that can spot drunk driving from just a 3-second video clip.
The goal is for it to detect if a driver is intoxicated, angry, or fatigued—and thus a potential hazard on the road. The technology can also estimate a person’s blood alcohol level.
Australian researchers have spent roughly two years developing the project with AI.
“The algorithm currently can detect five expressions: whether a person is happy, sad, angry or showing disgust; whether they are tired or not tired or fatigued or not fatigued; and their blood alcohol level as well,” Dr Zulqarnain Gilani from Perth’s Edith Cowan University said in a conversation with First Up.
Gilani noted that testing involved simulator videos of drivers in three intoxication levels, each with distinct blood alcohol concentrations.
He said the current technology achieves 93% accuracy.
He said the current algorithm underwent a proof-of-concept test on a small cohort of 65 people. Next steps involve gathering more diverse data for real-world implementation.
When asked how the tech detects mood, Gilani explained it all boils down to psychology.
“Psychology literature tells us that humans display different, either expressions or psychological states, and their faces show that.”
“For example, they say that if somebody is drunk, they blink really fast. And the time for which they close their eyes slows down, so they close it for more time.”
“Whereas if someone is tired, their eyes are droopy. Now, the interesting thing is that if somebody is very fatigued and someone is intoxicated, they show almost the same sort of behaviour.”
Researchers identified two practical scenarios for real-world implementation. First, equip roadside cameras with the tech to detect impaired drivers and flag them somehow.
The other is to embed the tech in personal cars. Gilani noted that with modern electric ignitions, a driver-facing camera could detect impairment and prevent the engine from starting.
“We are actively working with different collaborators and partners and also applying for different funding so that we can collect more data and make this thing practical,” Gilani said.