Revealed: what Britain’s Covid cameras saw
Cameras in Nottingham have been using AI to spot people breaking social-distancing rules. Are they a threat to people’s privacy? James O’Malley investigates
Cameras in a UK city have been using AI to spot people breaking social-distancing rules. Are they a threat to privacy? James O’Malley investigates.
When Britain locked down last spring, one of the many unknowns was whether people would comply with the new social distancing rules. Would we all keep our distance? Or would we grow tired of the restrictions and flout the rules as time wore on? The answer was vital, as compliance could have a big impact on the transmission and spread of Covid-19.
To find out what was happening on street corners, Innovate UK, the government’s innovation agency, awarded a grant to tech firm Vivacity Labs. The company was to use AI-powered cameras, which were installed on the streets of 16 British towns and cities.
The devices themselves are relatively simple cameras that are bolted on to existing infrastructure, such as lampposts and traffic lights. The cameras aren’t passively recording images like traditional CCTV, but they are analysing them with computer vision algorithms to identify different types of traffic. The cameras can differentiate between cars, pedestrians, bikes and other forms of traffic. And once traffic has been counted, they use regular mobile connectivity to beam data back to Vivacity’s servers.
To examine social distancing, the cameras watch where objects that are flagged as pedestrians and bikes move, and then determine whether they came within two metres of each other.
Almost a year on, after I submitted a freedom of information request to obtain the crucial data captured by these cameras, PC Pro can reveal the results of one city’s social distancing study, how the cameras are being used and why they raise important questions around privacy.
Counting rule-breakers
The Vivacity Labs’ cameras are mostly deployed by different local authorities, as well as two government agencies: Transport for London and the Canal & River Trust. However, after inquiring via the Freedom of Information Act, it appears that only Nottingham City Council took up Vivacity Labs’ offer to monitor social distancing, using around 100 cameras around the city.
The results offer a fascinating glimpse into how people’s behaviour changed during the pandemic, as rules were relaxed and tightened. Between June and the start of August, the cameras recorded around 20,000 instances per day of close passes (less than two metres) between pedestrians. But around a week into August, as “Eat Out to Help Out” began, and as venues such as bowling alleys and casinos were allowed to reopen, recorded infringements rocketed to around 55,000 per day.
This number only increased as schools reopened at the start of September and University of Nottingham students moved in later that month. The numbers then began to fall again as Tier 2 restrictions were advised and then imposed in October.
“It’s about helping [councils] to understand at a high level where are the problem areas and how those are evolving over time,” explained Vivacity Labs CEO Mark Nicholson. “High streets across the country already have lots of footfall monitors to look at how people are behaving and how many people are going past various shops. This is in many ways an extension of that.”
Nicholson said that the social distancing data collected by his company’s cameras could even be used to help determine when reopening may be safe, based on the interactions people are having. But since the data we obtained was first compiled by Vivacity Labs, Nottingham City Council has said that it hasn’t made use of any of the
collected data, has no plans to use it and will not be collecting this sort of data in the future.
Making traffic smarter
So what was the original purpose of the cameras? Vivacity Labs first conceived the cameras for monitoring road, not human, traffic. This was useful during the pandemic too.
“[Vivacity Labs’ cameras] give us a level of temporal and spatial granularity in terms of data… that opens up a huge array of possibilities,” said James Ashton, Nottingham City Council’s transport strategy manager. “To be able to control our network, to be able to manage it a lot better, to be able to respond to incidents, manage congestion, and to get people around the city a lot more efficiently.”
Ashton said the city was able to use the data to understand how changing lockdown rules impacted traffic flows. “We saw during the summer huge increases in cycling and walking activity, which has stayed pretty high actually, surprisingly even during the bad weather,” he said. “And obviously reductions in commuting traffic but we also saw higher than normal levels of vans and HGVs moving around the network.”
The data seen by PC Pro bears this out. According to the cameras, motor traffic in Nottingham fell dramatically at the start of lockdown in March, but by September had returned to almost pre-lockdown levels. Similarly, cycling in the city rapidly increased last March, but later in the year fell closer to the baseline.
Ashton also points to how the cameras will help with recovery once the pandemic is over, even if they just stick to traffic instead of specifically looking at social distancing compliance. “It’s going to be really difficult because, obviously, our public transport network was centred on the city centre,” he said. “But it’s now obvious to anyone using this data that people aren’t going to the city centre any more. The patterns are much more distributed. They’re going to out-of-town shopping centres, maybe by car, or going to local centres by walking or cycling.”
So, when it comes to the job of re-optimising the city’s transport infrastructure to reflect the new normal, the Vivacity Labs data will help inform the planners making the decisions. Ashton also hopes that Nottingham will be able to use the cameras to take advantage of a product that Vivacity Labs calls “smart junctions”, which uses real-time data to control traffic lights.
“You can put these sensors around a junction, and they can, in real-time, optimise the signal timings in a much more reactive and quicker way than we currently can,” he said, pointing out that because the cameras are connected, they can use data from much further afield. “If you can tell there’s a big chunk of traffic coming down the road, maybe five or ten minutes away, the junction will be prepared for it and set up for it.”
Perhaps unsurprisingly, the advent of AI cameras has also piqued the interest of privacy campaigners, although Vivacity Labs emphasises that its devices process data locally, and only transmit back the top-line, anonymised data, such as the counts for vehicle types, discarding any images collected.
The company is also aware of the potential privacy concerns caused by intelligent cameras, so much so that it prefers to describe them as “sensors” instead of “cameras”, on the basis that it isn’t images that are ultimately recorded.
That still leaves privacy questions to answer. What if, for example, the police were to ask Vivacity Labs for footage of an illegal gathering of more people than lockdown restrictions allow? Would it be able to provide that?
“We would be able to tell them absolutely nothing about those people,” said Nicholson. “We’d be able to simply say we saw five people at that period of time. There would be no video footage available. We have previously been asked by the police for this, but we say no every time because it’s physically not there. It’s simply not possible to use any of our systems for enforcement.”
So how much should we worry about AI cameras? In China, smart surveillance cameras have become ubiquitous as part of the country’s system of “social credit”, which in some cities can see citizens automatically identified by facial recognition and fined if they are detected, for example, crossing the road at the wrong time.
“Were such cameras to be used in a way that resulted in action being taken against members of the public then that would be a different matter, and potentially a serious concern,” said Phil Booth, a coordinator at medConfidential, which campaigns for confidentiality and consent in health and social care. “I’m not saying that this is the case in this particular instance or application – but putting ‘intelligence’ into CCTV cameras clearly provides them with extended capacities to do so, and provides opportunities for ‘feature creep’.”
While Booth isn’t particularly worried about Vivacity Labs’ deployment, he believes the important thing is to ensure all such systems are fully compliant with dataprotection rules and remain transparent. “I see no reason why legitimate, lawful purposes should not be permitted, but – especially with public surveillance ‘innovations’ such as these – that means dotting every ‘i’ and crossing every ‘t’,” he said.
The cameras process data locally and only transmit back the topline, anonymised data, discarding any images