New camera-based system can detect alcohol impairment in drivers by checking their faces

[midian182]

In brief: Glassy eyes, drooping eyelids, a slack jaw: these are all signs that someone might have had one drink too many. It's often obvious when someone is drunk just by looking at their face, and interior vehicle cameras could eventually use these tell-tale signs to help prevent drink-driving incidents.

Researchers at Edith Cowan University in Australia are developing a new technology that uses camera footage to detect whether a driver is alcohol impaired.

In a paper that was published earlier this year, the team describes how they devised an in-vehicle machine learning system that harnesses standard commercial RGB cameras to predict critical levels of blood alcohol concentration.

The researchers tested the system using 60 volunteers and an indoor driving simulator. Each person drove at different levels of inebriation: sober, low, and severe.

By analyzing facial characteristics such as features, gaze direction, and head position, the machine learning system was able to identify even low levels of alcohol impairment 75% of the time.

The system is an improvement over current driver-impairment identification methods that rely on factors such as pedal usage, steering patterns, and vehicle speeds, meaning they only work once someone has been driving for an extended time. At that point, it might be too late to prevent an accident.

With the camera-based technology, a car's computer could identify if a driver was drunk as soon as they got in, at which point the system would prevent the vehicle from starting.

The system can also use 3D and infrared footage of a driver's face, along with rearview camera videos that can show their posture. Steering interactions, event logs, and screen recordings of driving behavior can also be incorporated.

"Our system has the capability to identify intoxication levels at the beginning of a drive, allowing for the potential prevention of impaired drivers from being on the road," said Ensiyeh Keshtkaran, a doctoral student at Edith Cowan University, Australia, who contributed to the project.

Edith Cowan University Senior Lecturer Dr. Syed Zulqarnain Gilani said the next step was to define the image resolution needed to use the algorithm. If low-resolution videos are proven sufficient, this technology could even be employed by surveillance cameras installed on the roadside, allowing authorities to better identify drunk drivers.

NHTSA statistics show that about 37 people in the US die in drunk-driving crashes every day. In 2022, 13,524 people died in alcohol-impaired driving traffic deaths.

In December, the NHTSA said it had taken the first step in making anti-drunk driving technology a requirement in vehicles.

Permalink to story:

New camera-based system can detect alcohol impairment in drivers by checking their faces

 

[trparky]

Yes, but sleepiness can be seen as drunk. What? Are you not going to be allowed to drive your car if you only had three hours of sleep?

 

[NumberNine]

Car manufacturers collecting even more biometric data...

What could possibly go wrong? Looks like we'll find out.

 

[Endymio]

Yes, but sleepiness can be seen as drunk. What? Are you not going to be allowed to drive your car if you only had three hours of sleep?

If sleepiness is affecting your behavior to the point this system can detect it, then yes, you probably shouldn't be driving. In any case, I imagine the algorithm can fairly reliably differentiate between the two states.

 

[PurpaFur]

Seems like a great idea on paper, but who in their right minds wants a camera watching you the whole time you drive, ready and waiting to shut down your car at a moments notice? How easy would it be for either a government entity or a malicious actor to use that camera data to disable vehicles driven by selected individuals?

I'll always be on the side of less restrictions/monitoring.

 

[brucek]

No, your car should not refuse to start, or provide "evidence" against you, based on a technology that is "75% accurate."

Warning a driver that others are likely to perceive them as impaired could be useful, if we didn't know that these signals/logs would also end up being sent to insurance companies, car warranty providers ("your broken power window switch is probably from you mishandling it while looking drunk so we don't have to honor our warranty obligations"), and anyone else with an agenda ("Your honor, my husband should not have any custody rights because this software says he looked drunk on timestamp X").

 

[brucek]

Yes, but sleepiness can be seen as drunk. What? Are you not going to be allowed to drive your car if you only had three hours of sleep?

Hopefully not by this unreliable mechanism, but if I was on the jury I'd have no trouble holding someone civilly accountable for knowingly driving in that condition. Studies (as well as any adult's own life experience) have demonstrated that sleep deprivation is just as debilitating as alcohol. Do yourself a favor, if you must move on three hours of sleep, call an Uber.

Edit: there's a reason truck drivers, fighter pilots, etc. have mandatory rest periods and are not allowed to operate when they have not been met.

 

[Endymio]

No, your car should not refuse to start, or provide "evidence" against you, based on a technology that is "75% accurate."

Don't confuse Type I (false positive) with Type II (false negative) errors. This system will almost certainly have a higher Type II error rate than Type I. Also, that quoted 75% value is for low levels of inebriation; one can only assume that higher levels -- the ones which would likely wind up in court -- would be much more definitively pinpointed.

 

[Tantor]

If sleepiness is affecting your behavior to the point this system can detect it, then yes, you probably shouldn't be driving. In any case, I imagine the algorithm can fairly reliably differentiate between the two states.

That's not true. Driving when tired is not the same as when inebriated.

 

[Tantor]

Hopefully not by this unreliable mechanism, but if I was on the jury I'd have no trouble holding someone civilly accountable for knowingly driving in that condition. Studies (as well as any adult's own life experience) have demonstrated that sleep deprivation is just as debilitating as alcohol. Do yourself a favor, if you must move on three hours of sleep, call an Uber.

Edit: there's a reason truck drivers, fighter pilots, etc. have mandatory rest periods and are not allowed to operate when they have not been met.

Sleep deprivation and inebriation are NOT the same.

How exactly do you judge someone's sleep status? There is no way to attribute an integer value to it. I know people who ALWAYS look sleepy and tired.

Regarding sleep deprivation, I went for years with only 4 hours/night. I did a lot of driving without any problems. The primary danger is the possibility of actually falling asleep. But that's an entirely different issue, because now you're comparing a sleep state vs an inebriated state.

 

[Endymio]

That's not true. Driving when tired is not the same as when inebriated.

Not legally, no. But:

"... a 2018 survey by the AAA Foundation for Traffic Safety indicates that the percentage of crashes [found] that there were more than 6,000 crashes per year involving a fatigued driver that resulted in at least one death and that more than one in five fatal car crashes involved a drowsy driver. ... That means drowsy driving ranks nearly as high as drunk driving on the list of reasons for crashes...."

 

[human7]

Seems like a great idea on paper, but who in their right minds wants a camera watching you the whole time you drive, ready and waiting to shut down your car at a moments notice? How easy would it be for either a government entity or a malicious actor to use that camera data to disable vehicles driven by selected individuals?

I'll always be on the side of less restrictions/monitoring.

Car manufacturers collecting even more biometric data...

What could possibly go wrong? Looks like we'll find out.

For better or for worse, Congress has required that car manufactures install in new vehicles passive technology to detect and prevent driving while intoxicated as early as 2026, so the manufacturers don't really have a choice. The law does not mandate any form of technology, that's up to the NHTSA. There will certainly be court cases about it at some point.

 

[Nobina]

This is a violation of privacy and freedom. I don't embrace drunk driving, if you do it, you should have your licence revoked. But **** having your car be your prison or a vehicle for government to get up in your business.

A car should be stupid, it should do what you want it to do, not the other way around.

 

[AxelAminoff]

I have lazy eye and on two separate occasions, driving different VAG cars, the cars have notified me to take a break after less than 1 hour driving. Both cars are from 2017 - an Audi A3 and a Skoda something. So there is already some monitoring going on. I miss my 1991 Saab 900i, which lacked any "helpful" crap.

 

[brucek]

This is a violation of privacy and freedom. I don't embrace drunk driving, if you do it, you should have your licence revoked. But **** having your car be your prison or a vehicle for government to get up in your business.

A car should be stupid, it should do what you want it to do, not the other way around.

That's ideologically where I start from. Of course being killed by a drunk driver is also a violation of my privacy and freedom. And since this issue of serious accidents caused by drunk driving is now decades old, with lots of hard data available, I'm willing to accept that experience has shown that prosecution after-the-fact has not and will not adequately solve the problem as much as I'd like it to.

For now I remain opposed based on practical matters such as reliability, cost to install, cost to repair, etc. And it may never get past that. But if we ever did get a magic technology bullet where cars could reliably stop themselves from being driven drunk, I think that's what they'd have to do. Not necessarily from legislation saying they have to, which I'd probably oppose ideologically, but because auto makers would rightfully be sued for negligence after each death if it was possible to prove they had an easy, effective way to prevent it and they did not. That last part does fit within my ideology.

 

[human7]

That's ideologically where I start from. Of course being killed by a drunk driver is also a violation of my privacy and freedom. And since this issue of serious accidents caused by drunk driving is now decades old, with lots of hard data available, I'm willing to accept that experience has shown that prosecution after-the-fact has not and will not adequately solve the problem as much as I'd like it to.

For now I remain opposed based on practical matters such as reliability, cost to install, cost to repair, etc. And it may never get past that. But if we ever did get a magic technology bullet where cars could reliably stop themselves from being driven drunk, I think that's what they'd have to do. Not necessarily from legislation saying they have to, which I'd probably oppose ideologically, but because auto makers would rightfully be sued for negligence after each death if it was possible to prove they had an easy, effective way to prevent it and they did not. That last part does fit within my ideology.

On your last point, would it be fair to say that Congress is negligent if they had a provably easy, effective way to prevent the ~13,000 deaths per year caused by drunk driving? I don't like this provision in the law (the Infrastructure Bill which mandated this technology) very much, but it's also hard to deny the necessity of it if we are to save lives. I do believe that due process and other civil rights and liberties can be respected while implementing this technology (not that it necessarily will be, knowing how much firms and the government love to scoop up data). Private action of lawsuits for negligence, from my point of view, is a very messy way to get to safety, especially when you look at the body count required to even file the suits. Of course at this point, there's already a high body count.

 

[brucek]

On your last point, would it be fair to say that Congress is negligent if they had a provably easy, effective way to prevent the ~13,000 deaths per year caused by drunk driving?

Letting a marketplace of ideas and products sort itself out is indeed messy and error prone and far from perfect, it's just a lot better than going with the first idea a bunch of random congressional subcommittee staff members came up with in say the two weeks they had to work on a given topic before moving on for the next several decades.

I'm OK with government using a liability framework that has evolved over decades (centuries?) to adjudicate between disputing parties. I'm a lot less OK with government taking over the role of one of those parties itself.