Automotive

AI system determines if drivers are ready to take control

Existing systems are already able to determine if the driver is getting tired, but the Fraunhofer system can actually tell what they're up to – such as if they're talking on their smartphone
M. Zentsch/Fraunhofer IOSB
Existing systems are already able to determine if the driver is getting tired, but the Fraunhofer system can actually tell what they're up to – such as if they're talking on their smartphone
M. Zentsch/Fraunhofer IOSB

Although self-driving cars definitely are coming, the first ones will likely still need to periodically have the driver take manual control of the vehicle. A new system could check if the driver is currently able to do so, by monitoring what they're doing.

Presently being developed by a team at Germany's Fraunhofer Institute for Optronics, System Technologies and Image Exploitation, the technology is designed to work with cameras located inside the vehicle's cabin. These could be regular visible-light cameras, infrared night-vision cameras, or depth-sensing 3D cameras.

Live video from those devices is analyzed in real time by an onboard computer. It utilizes artificial-intelligence-based algorithms to establish the body poses of the driver and other passengers, and then matches those poses up to known activities such as engaging in face-to-face conversation, sleeping, or watching the road. The system can also recognize objects that may be a source of distraction, such as smartphones.

In a situation where the driver needs to take control – such as if the vehicle is approaching a road construction zone – the system alerts them, plus it checks that they're ready to do so. If it's determined that they aren't, the system calculates how long it will take them to return their attention to the road, based on their current activity. It then holds off on relinquishing control, waiting until they're ready.

None of the video footage is recorded, nor is it transmitted outside of the vehicle, so privacy issues shouldn't be a concern.

Additionally, once the technology is developed further, it might also be used to recognize hand gestures made by the driver to the car. If they were to say "Park over there" and point in a certain direction, for instance, the system could tell where they were pointing. It could additionally ensure that the driver and all the passengers had fastened their seatbelts, before allowing the car to proceed on its trip.

"The technology is ready for pilot production," says project manager Michael Voit. "We are already in contact with companies who want to use our technology."

Source: Fraunhofer

  • Facebook
  • Twitter
  • Flipboard
  • LinkedIn
5 comments
guzmanchinky
Isn't that kind of like what Cadillac Supercruise already does with the little camera that watches your face?
Ralf Biernacki
"privacy issues shouldn't be a concern" Yeah, I believe it. It's another one of these technologies, like Siri or GPS tracking, that start out oh so helpful and convenient, and when they are realized they somehow magically transform into increasingly invasive surveillance. The biometric devices that the Taliban now use to identify collaborators are a case in point.
Daishi
All the smart people in 2015: "We should have level 5 autonomy by 2016". All the smart people in 2021: "We should have the details worked out for the AI that monitors the driver for level 2 autonomy by 2030"
Pmeon
That would be the single most annoying feature to have in a car. Stricter driving tests followed by an IQ test would be far more effective at reducing accidents and moving road hazards.
S Redford
Human nature will always find a way round any system. The real question should be whether there is any sense in having fully autonomous systems whether monitored by a human or not. As the Donald Rumsfeld quote goes, “it’s the unknown unknowns - the ones we don't know we don't know” that are difficult to deal with! Despite the ‘I’ in AI meaning ‘Intelligence’, there is nothing truly intelligent about autonomous driver systems – they simply respond to learned scenarios. If a scenario isn’t known (wasn’t learned), an ‘AI’ system can’t react properly – although it may be solvable, look what appears to be happening with Tesla assistant mode when it finds the route blocked by a static emergency vehicle with flashing lights. Only real intelligence can respond to the unknown unknowns. If you don’t want to take responsibility and drive, then take a taxi (with driver), bus or train!