We make better drivers and riders
Deep Dive: AI in road traffic safety
Cat Dow looks into the new cameras that police forces are using to detect drunk drivers and drug drivers – remotely
The ‘world’s first’ trials using AI- powered cameras to detect drug- and drink-impaired driving have been underway for almost a year now. The Devon and Cornwall police force became the pioneering constabulary to deploy the technology on a trial basis, following the success of a similar programme that detects offenders not wearing seatbelts and/or handling their mobiles. But should gaps in policing really be given to a ‘robo-cop’? And does this spell the beginning of the end for human officers?
The technology capitalises on pre-existing hardware; a fancy visual suite that combines long-and short-range cameras with a sophisticated software application that can detect if belts are correctly across a driver’s body, and if drivers are making hand-held calls or viewing their screens rather than the road ahead.
This original pilot was so successful, 10 more authorities, including Norfolk, Thames Valley and West Mercia, ran equally successful trials. Durham Constabulary reported that a week-long temporary installment of the system caught almost 1,000 drivers texting at the wheel, or not wearing seatbelts, and fined those flouting the law.
Geoff Collins, UK general manager of Acusensus explains, “By monitoring multiple attributes, for example, if a car is making progress in a controlled manner, or if it’s exhibiting behaviour that falls outside of what is expected, the system will flag it and relay the data to a police team that can intercept the car further down the road.”
Assessing impaired driving is trickier than photographing drivers breaking the law by not belting up. That snapped-in-a-second image holds a tangible fact, like the automated parameters set for speeding – with no room for reasonable doubt or argument (and thus, watertight prosecution). Identifying impaired driving is more complex, however.
Firstly, the algorithms that power the cloud-based analysis engine must be strict enough to catch offenders, but not so strict as to identify everyone veering momentarily from a narrowly prescribed boundary. Collins points out, “At present, we don’t quite know the number of variables we’ll need to consider to strike an appropriate balance between flagging every momentary lapse in concentration, and stopping too few cars.”
Then there’s timing. Since alcohol and drugs metabolise out of the body’s system over time –and affect every human body differently – these offences are much more time-sensitive, meaning the AI-based sensor system is only the first step in tackling this increasing road safety problem. A worrying 70% increase was recorded between 2014 and 2022, according to the Department for Transport’s figures.
“Op Limit” coordinates 45 constabularies over the festive period to clamp down on those drivers who’ll chance it – it’s a huge human operation, but in 2024, sadly, the number of drivers testing positive for drink and drug consumption doubled.
Nicholas Lyes, Policy and Standards Director at IAM RoadSmart, says, “Ultimately, we should be looking at tools that can capture those people that, frankly, do not care about other people's safety, their own safety, and have no perception of risk. Of course, the issue with technology is that [sometimes] it can be wrong.”
Lyes makes a salient point. Tech glitches are an obstacle for gaining public trust – and with a product in its infancy, Collins is right to address this, “Our impaired development is still in the oven, so to speak.” It is essential a machine doesn’t misread a medical condition, or even a basic yawn, and that it’s trained with enough diversity of faces, behaviours, or driving environments to eliminate prejudice. This isn’t just a technical problem, either. The idea of being judged – or even convicted – by an algorithm rubs a lot of people the wrong way.
As such, each slice of impaired driving footage flagged by the Acusensus suite is scrutinised by a human. Lyes says, “So long as there is a mechanism in place to review [what is captured], then you have a layer of technology that will assist you, not replace you.”
In 2023 in the UK alone, there were over 1,600 road deaths, and tens of thousands of serious injuries, so like many, Collins holds an urgent optimism, “The seatbelt and mobile phone programmes are going from strength to strength and I’m hoping we will soon add this [impaired driving] capability to it, too.”
So is AI the answer? In the short-term, it could prove more challenging. But in the longer term, with well-considered deployment, and awareness of its limitations, its capabilities could prove pivotal in reducing the number of avoidable fatalities on our roads.
The ‘world’s first’ trials using AI- powered cameras to detect drug- and drink-impaired driving have been underway for almost a year now. The Devon and Cornwall police force became the pioneering constabulary to deploy the technology on a trial basis, following the success of a similar programme that detects offenders not wearing seatbelts and/or handling their mobiles. But should gaps in policing really be given to a ‘robo-cop’? And does this spell the beginning of the end for human officers?
The technology capitalises on pre-existing hardware; a fancy visual suite that combines long-and short-range cameras with a sophisticated software application that can detect if belts are correctly across a driver’s body, and if drivers are making hand-held calls or viewing their screens rather than the road ahead.
This original pilot was so successful, 10 more authorities, including Norfolk, Thames Valley and West Mercia, ran equally successful trials. Durham Constabulary reported that a week-long temporary installment of the system caught almost 1,000 drivers texting at the wheel, or not wearing seatbelts, and fined those flouting the law.
Monitoring multiple patterns
To capture drivers potentially impaired by drink and/or drugs, those cameras are part of a more impressive suite of infrared and radar sensors. Acusensus – the equipment manufacturer – then configures the machine-learning algorithms of its proprietary software to identify and map patterns of driving behaviour, facial alertness in the driver and ocular awareness to give traffic police a heads-up on cars behaving unusually.Geoff Collins, UK general manager of Acusensus explains, “By monitoring multiple attributes, for example, if a car is making progress in a controlled manner, or if it’s exhibiting behaviour that falls outside of what is expected, the system will flag it and relay the data to a police team that can intercept the car further down the road.”
Assessing impaired driving is trickier than photographing drivers breaking the law by not belting up. That snapped-in-a-second image holds a tangible fact, like the automated parameters set for speeding – with no room for reasonable doubt or argument (and thus, watertight prosecution). Identifying impaired driving is more complex, however.
Firstly, the algorithms that power the cloud-based analysis engine must be strict enough to catch offenders, but not so strict as to identify everyone veering momentarily from a narrowly prescribed boundary. Collins points out, “At present, we don’t quite know the number of variables we’ll need to consider to strike an appropriate balance between flagging every momentary lapse in concentration, and stopping too few cars.”
Then there’s timing. Since alcohol and drugs metabolise out of the body’s system over time –and affect every human body differently – these offences are much more time-sensitive, meaning the AI-based sensor system is only the first step in tackling this increasing road safety problem. A worrying 70% increase was recorded between 2014 and 2022, according to the Department for Transport’s figures.
Why more tech also means more manpower
If AI is helping identify more offenders, that will in turn increase the demand for more manpower to verify the computer data with human blood and breath tests. Yet, police resources are finite. Chief Superintendent Marc Clothier, National Police Chiefs’ Council (NPCC) Operational Lead for Operation Limit (an annual Christmas campaign targeting drink- and drug-driving with hotspot- and intelligence-led checks), said, “In 2023, 19.6% of fatal collisions were assigned at least one drink or drugs related factor.”“Op Limit” coordinates 45 constabularies over the festive period to clamp down on those drivers who’ll chance it – it’s a huge human operation, but in 2024, sadly, the number of drivers testing positive for drink and drug consumption doubled.
Nicholas Lyes, Policy and Standards Director at IAM RoadSmart, says, “Ultimately, we should be looking at tools that can capture those people that, frankly, do not care about other people's safety, their own safety, and have no perception of risk. Of course, the issue with technology is that [sometimes] it can be wrong.”
Lyes makes a salient point. Tech glitches are an obstacle for gaining public trust – and with a product in its infancy, Collins is right to address this, “Our impaired development is still in the oven, so to speak.” It is essential a machine doesn’t misread a medical condition, or even a basic yawn, and that it’s trained with enough diversity of faces, behaviours, or driving environments to eliminate prejudice. This isn’t just a technical problem, either. The idea of being judged – or even convicted – by an algorithm rubs a lot of people the wrong way.
As such, each slice of impaired driving footage flagged by the Acusensus suite is scrutinised by a human. Lyes says, “So long as there is a mechanism in place to review [what is captured], then you have a layer of technology that will assist you, not replace you.”
Your car is watching, too
For their part, car makers have already integrated facial recognition, rest break suggestions and drowsiness alerts. For some time, the European Union has planned to bring in mandatory built-in breathalysers so the car is immobilised if the driver is over-the-limit. There has been talk of introducing the same requirements, particularly for repeat drink-drive offenders here in the UK, though nothing has yet been decided. Such steps could limit the collisions involving drink-driving, and would go some way to tackling some illegal drug use.In 2023 in the UK alone, there were over 1,600 road deaths, and tens of thousands of serious injuries, so like many, Collins holds an urgent optimism, “The seatbelt and mobile phone programmes are going from strength to strength and I’m hoping we will soon add this [impaired driving] capability to it, too.”
So is AI the answer? In the short-term, it could prove more challenging. But in the longer term, with well-considered deployment, and awareness of its limitations, its capabilities could prove pivotal in reducing the number of avoidable fatalities on our roads.
