A new driver aid claims to revolutionise road safety.

The artificial intelligence (AI) behavioural system claims to prevent more than 90% of crashes. It achieves this by predicting the actions of pedestrians and other road users, then alerting the driver to potential incidents.

Behaviour AI is available as a dash cam after-fit solution. However, Humanising Autonomy who make the system, are hoping it will be offered as standard fit within the next two to three years.

It links to advanced driver-assistance systems (ADAS), and the company is already in talks with several vehicle manufacturers.

The system has been installed on some London buses for more than a year. Trials have also begun with a van-based fleet operator. This is part of an innovation challenge between Transport for London (TfL) and 10 industry partners including Royal Mail, DPD, UPS and John Lewis Partnership.

Intelligent road safety

Co-founder and chief product officer Leslie Nooteboom said the system helps autonomous systems “to understand people”. It predicts what other road users will do and alerts the driver. It also provides the fleet operators with data about any near miss (as well as crashes). In turn, this data helps with driver training.

“It’s not just recognition; it predicts if they are going to cross the road or where they are moving to on the road,” he said. “Our data shows that the system can predict two seconds before the driver realises what is happening and send them an alert.”

Behaviour AI scans the road for vulnerable road users and predicts their next movement. Nooteboom claims it can detect objects up to 80 metres ahead at 40mph, twice the distance of the industry standard, with 99% accuracy. The system is frequently updated to take account of new forms of mobility, such as e-scooters.

“It is also continuously learning and not just from one driver, but an entire fleet,” says Nooteboom. “We believe it can prevent around 92% of accidents.”

Always learning

Development continues using the data that is continuously being fed into the central system. For example, initially it struggled to with scooter riders because they didn’t move their limbs. Now the operating has been developed to recognise the growing number of e-scooters.

“This is where we add intelligence to the system so it machine learns and applies that to the real world. It doesn’t just learn from the data we give it; it’s an interpretable AI approach. We can see how it’s making decisions, so we know how sure it is about making those decisions. Then we know what we need to do to make it even better, which we do with over-the-air updates.”

Statistical proof

The venture capitalist-based company has its own data to support its claims about accident reductions. Within the next six months it will have analysed copious amounts of real fleet data to provide robust proof. This includes using near-miss data for driver training purposes.

“We take the video of the near miss and extract information on the risk level for driver education, even where there was no accident,” said Nooteboom.

“We also provide the information for the insurance company so they can understand the lead up to the accident.”

The pricing structure has yet to be set, but simply preventing one crash would justify the investment, according to Nooteboom.

He added: “We believe we are first to market with this solution. It uses small chips so it can be used now. The competition is the predictive AI in the full autonomous vehicles.”

Already working

The company’s behavioural video analytics software was deployed last year to help TfGM (Transport for Greater Manchester) understand social distancing issues. It assessed how passenger behaviour is affected by the Covid-19 pandemic and temporary infrastructure changes need to be made.

By analysing CCTV footage, the system enabled TfGM to re-route public transport to hot spots where lots of people were waiting for a bus.

The data will also be used to better understand the capacity for walking and cycling, and how space is used at transport interchanges.