As the COVID-19 pandemic took hold of the UK in spring 2020, the streets emptied. Shops shut down, and anyone who ventured outside had to obey the ‘two-metre’ rule. In London, the authorities realised that they needed a way to monitor the changes in street activity in real time, to gain a better understanding of how lockdown was affecting city life, and what interventions were needed to allow the city’s nine million people to keep socially distanced.
Working with the Greater London Authority (GLA) and Transport for London (TfL), a team at The Alan Turing Institute launched Project Odysseus. This was a rapid repurposing of an existing project that had been monitoring air pollution around London. The team modified its algorithms, feeding them data from London’s traffic sensors and cameras to estimate pedestrian and vehicle numbers. The authorities used this information to understand changes in activity across London, pinpointing where more social distancing measures were needed.
“There was an urgency and a passion among our team to contribute to the fight against the virus,” says Theo Damoulas, Turing AI Fellow and leader of the project. “We wanted to help in whatever way we could.”
How did it start?
The roots of Project Odysseus stretch back to 2015, when the original London air pollution project was launched by the University of Warwick and GLA. Since 2017, the project has been part of the Turing’s data-centric engineering programme (of which Damoulas is the Deputy Programme Director), which seeks to apply data science techniques to engineering problems and is funded by the Lloyd’s Register Foundation.
During the pollution project, the researchers developed computer models and algorithms for combining data from a multitude of sensors across London, to estimate and forecast the city’s air quality. The project is still ongoing, and its ultimate goal is to produce nightly, short-term forecasts that Londoners can access through an app, giving them insights into when and where to travel in order to limit their pollution exposure.
In March 2020, as the UK went into its first lockdown, the data-centric engineering team asked GLA what it could do to support London’s response to the pandemic. GLA flagged up a lack of data about the city’s ‘busyness’ levels, and so the idea was born to adapt the air pollution tools and datasets to monitor London activity.
“The Turing was really valuable in bringing together specialist teams from different universities,” says Paul Hodgson, Senior Manager for City Data in GLA’s City Intelligence Unit. “It would have been impossible for GLA to uncover those teams in such a short timescale without the Turing’s help.”
Within a couple weeks, researchers from the Turing, Warwick, the University of Cambridge and UCL were working full-time on Project Odysseus, using their previous work as a springboard.
“Monitoring traffic was something we were already doing,” says Damoulas. “We had been using the images from TfL’s ‘JamCams’ [a network of over 900 CCTV cameras that monitor the city’s traffic] to detect different vehicle types, for information about emissions. We had also developed a cloud-based, computational infrastructure for combining and processing the sensor data, and interfaces for displaying the data. This all laid the groundwork for Odysseus.”
The majority of the team’s work has focused on analysing the stream of data that flows from TfL’s JamCams – some 30 gigabytes per day. These cameras are positioned at traffic intersections, but they also capture the pedestrians on road crossings and pavements. The team adapted its vehicle detection algorithms to detect pedestrians, providing authorities with near real-time estimates of pedestrian and vehicle density. Extra information about traffic density also came from the 11,000+ ‘inductive loop’ devices on London’s roads, which detect vehicles that pass over them.
In addition, the team crafted algorithms that use the JamCam footage to calculate the distances between pedestrians – a tricky problem of perspective because the cameras aren’t all at the same height and angle. This allowed the researchers to estimate the average social distancing in any given area.
A key aspect of the project, described by Damoulas as the “backbone”, was the systems and servers on which the data was stored and processed. Working with the Turing’s Research Engineering team, the researchers honed the infrastructure that they’d developed for the air pollution project, so that the data could be processed quicker and more securely. They also created a piece of software (an ‘Application Programming Interface’, API) that the authorities could use to analyse the data, as well as a prototype dashboard (below) for data visualisation.
The API was used by TfL to monitor pedestrian density during the pandemic, enabling the authorities to quickly identify where social distancing interventions were required. On Brixton high street, for example, the tool demonstrated that there was too much crowding on pavements, particularly near bus stops. As a result, TfL extended the pavement and moved a bus stop to create more space.
The tool similarly identified overcrowding at Borough High Street, Camden High Street and near London Bridge station. Temporary barriers were installed at these locations to extend the pavement, allowing pedestrians to maintain a safe distance. TfL says that it implemented over 700 such interventions at the height of the pandemic’s first wave, and that the Turing’s tool provided key data for those decisions.
Move the slider to see Borough High Street in London before and after the pavement was extended during the pandemic using temporary barriers (images are from a Transport for London JamCam). The ‘heatmaps’ overlaid by the Odysseus team show average pedestrian density between 28 March and 2 May 2020 (before image) and 2 May 2020 and 19 April 2021 (after image), adjusted to account for the different sampling periods. Yellow corresponds to high pedestrian density – the increased density in the extended pavement areas (red boxes) shows that pedestrians were using these areas to socially distance
Throughout this work, the team has been careful to consider the ethical implications of monitoring CCTV footage. “The JamCam footage is very low resolution, so there is no risk of being able to identify anyone,” says Damoulas. “From the start of the project, we have been committed to developing algorithms that are incapable of performing facial recognition or tracking people. We have a rolling review with the Turing’s Ethics Advisory Group to make sure that our data and algorithms are always preserving the privacy of the general public.”
What does the future hold?
An exciting future direction for this work, with research currently underway, is route-planning algorithms, which could use the data from London’s cameras to find the quietest pedestrian journeys. The researchers say that these algorithms could be used in an app that allows Londoners to avoid the busiest streets – something that will be useful even post-pandemic.
Another possible application of Project Odysseus is monitoring changes in economic activity as London recovers from the pandemic. Although the JamCams’ positioning at traffic intersections makes them less useful for tracking commercial activity, the London boroughs have thousands of CCTV cameras in their 600 high streets and town centres. Damoulas says that these are higher resolution than the JamCams, so the team would need to make sure that its algorithms and data processes continue to protect people’s anonymity. The algorithms could then be run on the footage, enabling authorities to record footfall and find out how it compares to pre-pandemic levels.
“This tool would help us to understand how our high streets are evolving,” says Hodgson. “High streets have social, community and cultural uses, in addition to retail. Bricks and mortar retail is decreasing, and so high streets are needing to adapt. The most successful high streets have a broad range of uses, such as libraries, offices and arts events. Recording footfall will give us insights into how different high streets are recovering, so we can focus our efforts.
“This collaboration with the Turing has created a lasting legacy that we intend to build on,” he adds.
Other city authorities have expressed an interest in the team’s work, and the researchers plan to make their tools and code freely available, to maximise the amount of people who can use and benefit from the research.
Looking further ahead, Project Odysseus represents a step towards a ‘digital twin’ of the city of London – a computer model that would be a bits-and-bytes representation of the real thing. The idea is that if researchers collect enough data about a city, they will gain a good enough understanding of its workings to be able to build a model to simulate it. Digital twins, says Damoulas, will allow authorities to understand the effects of interventions and policies before they introduce them.
“Let’s say you wanted to create a new Ultra Low Emission Zone [where drivers pay a charge if their vehicle doesn’t meet certain emissions standards],” he says. “A digital twin could show you the best place to put it, based on the effects it would have on air pollution, traffic and economic activity.”
As with Project Odysseus, the driving aim of this research is to improve the lives and livelihoods of city dwellers. As we recover from the devastating effects of the pandemic, this is looking more important than ever.
Top image: gemphoto / Shutterstock