After the sudden economic fall, it’s essential to get business back to normal quickly but safely.
AI has been one of the many technologies used for tracking virus transmission and finding a cure, but it can also help to track physical distancing and reduce the related contagion risk. This has boosted the public debate around using AI and other technologies to help us adapt to the current situation. In this blog article, we discuss with our Lead AI Solution Architect Niko Vuokko about Silo.AI’s Physical distancing analyzer, a computer vision based tool for systematically monitoring and improving the contagion risk situation at malls, shops, restaurants and other semi-public spaces.
How does technology help businesses to restart safely after Covid-19?
Niko: This is a complex question as all the technical pieces exist, but we need public conversation and collaboration between governments, businesses and the general public to align various needs together. For the economy it is important to get business flowing and people moving sooner rather than later. At the same time, everyone needs to feel and be safe. Businesses need to be able to provide that guarantee: systematically monitor and improve their spaces to minimize the risk of contagion. To succeed, businesses need new tools, but using these technologies must be a value-driven choice: the solution must rather strengthen than undermine common European values and respect for privacy.
How would you describe the situation in Europe?
Niko: Europe has already taken into use a selection of, but not all, actions that Asian countries have deployed. Decision makers need to keep in mind that this won’t be the last epidemic. They need to somehow assure that we can reduce contagion risk in public spaces sustainably over a long period of time while bringing life back to normal. At the moment many countries in Europe require the use of masks when visiting stores, which obviously can’t go on forever. We need to push the responsibility of reducing contagion risk to businesses and the public sector which can work on it systematically. However, this will not happen without improved access to new data and tools.
What is Silo.AI’s Physical distancing analyzer about?
Niko: Our approach is about providing tools to commercial properties to systematically monitor and improve their success in reducing contagion risk. As said, businesses need tools to discover bottlenecks in their interior design, modify policies, and provide information to their customers on the safety status of their premises. To tackle these needs, we’ve built a tool to continuously assess the physical distancing status based on security camera feeds. This information is then converted into detailed analytics on the situation over e.g. a day and a week and into map-based drill-downs across the property. It can also be used for real-time awareness where the tool recommends pinpoint actions.
How can the Social distancing analyzer be put to use?
Niko: The tool can be deployed into properties with even very large numbers of cameras and works with all common camera installations. The tool is computing the distances between people while detecting related groups of people. It’s also assessing the use of masks in the premises and the ages of people present to provide risk group focused analyses. The tool is customizable including ready-made components, something we see as the right approach for a wide variation of spaces to which it may be deployed.
How is privacy taken into account in Silo.AI’s solution?
Niko: Key aspect of the solution design is about privacy. We choose not to sacrifice any of our freedoms to achieve the tiny bit of safety. We’ve been building a privacy threat model based on which we’ve built a number of features to stop the leaking of private information. We are also following the guidelines set by the EU in the GDPR legislation and its recent AI whitepaper. For example, the solution never stores identifiable image data and all processing and storage happens locally. We are also pushing computing close to the cameras so that while the feed may be relayed as such to the security center, we are removing identifiable information from it as soon as possible.
While camera surveillance is table stakes nowadays in many commercial properties, there’s always the threat of slippery slopes present when using data for new use cases. Different from regular camera surveillance, our solution does not require identifiable camera feed to be stored. In any case, we’re going to include also social science aspects to the design to ensure that we can realize our objectives and avoid unintended consequences.