Social distancing and '1.5 metres' are concepts we have been confronted with every day in the past year. Although it is of vital importance to follow the rules, it can be difficult at times to get a feel for what 1.5 metres looks like. This is why Maarten Sukel, a PhD student at the Amsterdam Business School (ABS) and AI lead at the municipality of Amsterdam, and his team developed the 1.5 metre monitor: a camera that shows how closely people observe the required distance. 'It's quite a simple technique that works on the basis of machine learning.'
Maintaining distance is tricky, especially when this has been communicated via the same channels for almost a year, namely through stickers or people in the street carrying signs. 'We've become immune to it,' says Sukel. 'Amsterdam's council and public library came up with the idea to look for new ways of communicating. The monitor relies on the existing YOLO (You Only Look Once) model, which is already used by the council and others in the battle against litter. The model has been trained with over 300,000 photos of people’s faces, enabling it to recognise them more or less without fail. By linking this to a webcam and LED screen, we can launch an interactive campaign.'
The system was first tested at the entrance to the library, but Maarten and his team wanted to reach far more people. Sukel comments: 'We especially want to be visible in places where there are enforcement issues such as language barriers. We then thought up the idea to introduce the 1.5 metres monitor in a busier spot, leading to a plan to take Black Friday and park a lorry with a 3x5-metre LED screen right on Dam Square as well as in various other busy places.'
Sukel is being supervised by Dr Stevan Rudinac and Prof. Marcel Worring (FNWI).. The project is being developed further by ABS interns, who want to know how we can expand the monitor's functionality without infringing privacy. 'Right now, the system doesn’t store any data whatsoever,' Sukel explains. 'It's actually a kind of digital mirror that hides the faces of the public using coloured smiley overlays. A green smiley means you're keeping enough distance, an orange smiley indicates you're in the danger zone and a red smiley shows you're too close to other people. Another challenge was working with 2D images. We specifically chose to implement this project with cheap, widely available technology and not with, let's say, a more expensive 3D camera. As a result, we have to constantly calibrate the system by estimating a 1.5 metres on the basis of such things as the number of paving stones in an image.'
Sukel says that the project illustrates the user-friendliness of AI in public spaces. Most of all, it is a nice demonstration of the many different ways you can use image recognition. He adds: 'We hope this project helps us to contribute something to society. There are still some limitations, including the system's inability to recognise whether people belong to the same household and can therefore be closer together. We don't really mind such limitations because we see that, if anything, the 1.5 metres monitor leads to interesting conversations. Children, for example, are curious and test the system with great enthusiasm. In this way, we also get their parents to talk about the importance of social distancing.'
Sukel hopes that many other people will take the system on board. One instance of a project is the application of the AI system to Amsterdam's Marineterrein, where 'smart' lighting helps athletes to work out in a corona-proof environment. As Sukel observes: 'Nothing stops companies from using the system themselves and applying it to lobbies and meeting rooms, just to give a few examples. The more the monitor is used, the more people will think about how far they need to be away from others.'
Amsterdam City Council pursues an open and transparent policy towards AI. All information about the system can be found in the council's register of algorithms. Do you wish to take a deeper dive into the technological aspects of the 1.5 metres monitor or try it out yourself? That’s certainly possible as the project's source code is public domain and you can surf to https://github.com/Amsterdam-AI-Team/1.5-meter-monitor for a demo of how the system work.