Since launching its commercial offering last year, Veesion said it now detects more than 100,000 ‘suspicious gestures’ per month in more than 2,000 stores worldwide.
French start-up Veesion is trying to tackle the threat of shoplifting through gesture recognition technology.
The issue of shrinkage in shop stocks is a constant problem for various retail stores. A recent survey referred to it as a $100bn problem worldwide. Nearly 75pc of survey respondents said external theft has become a higher priority for them over the past five years.
To tackle this issue, French start-up Veesion said it has developed an AI platform that can detect theft in real time, alerting staff so they can prevent items from being stolen.
It accomplishes this by detecting gestures that indicate a person is trying to steal an item, such as by grabbing an item and sticking it in their pocket.
Veesion senior account executive Sean Ward told SiliconRepublic.com that it took two years to develop the AI to its current level, with thousands of hours of footage being used to train the model.
“It can be open-source videos, where anyone can find them,” Ward said. “Then obviously some of our own ones, which we can tweak a little bit more specifically, so that it can recognise different gestures that we want [the AI] to recognise.”
Veesion has grown rapidly since it launched its commercial offering last year, reaching more than 2,000 stores in more than 20 countries. Ward said around 20 to 30 stores in Ireland currently use this software, including brands such as Spar, SuperValu and “independent stores”.
Learning from stores
Ward said the AI is designed to be integrated with a store’s existing CCTV system. Once the AI recognises a gesture that is related to theft, the store staff get a video notification that shows the potential theft.
“They can then determine whether they investigate further or they just leave it, it’s totally up to them,” Ward said.
He added that giving an average success rate for the technology is difficult, as it can be different in every store that uses it. However, he said Veesion believes it helps to reduce theft by roughly 50pc to 70pc.
Veesion said its technology also incorporates “active learning”, which means it can improve its performance based on stored data from the stores it operates in.
Ward said the data is stored in each store independently, for legal reasons such as GDPR and also because “we just don’t have the space to have all this different video footage to keep on file”.
“There’s the main [AI] model, which has the set gestures that they look for,” Ward said. “And then in terms of each store, it will actually learn the store’s algorithm, as we say, so that it can adapt to the store’s policies.
Ward added that new features are coming “in the next couple of years”.
“We have to then retrain the AI on the newer features and make sure that integrates correctly,” Ward said.
Some stores have been very supportive of the technology. In March, a Nisa store in London claimed it was able to cut shoplifting losses by 90pc through Veesion’s software.
The company looks set to expand further as it raised €10m in March to improve its AI technology and grow in Europe and the US.
AI is being used by various companies and government organisations for the purpose of biometric identification, such as facial recognition technology.
Biometric identification is broad and can include simple technology such as fingerprint detection or more controversial systems such as emotion analysis.
The potential misuse of these systems has raised concerns among regulators. For example, a group of MEPs and an NGO coalition recently discussed the threat biometric surveillance technology poses to democracy and called for it to be banned in the upcoming AI Act.
The UK’s Information Commissioner’s Office has also warned against the use of emotion analysis technology and is developing guidance on the wider use of biometric tech such as facial, fingerprint and voice recognition systems.
Despite the potential for regulation around biometric tech, Ward is confident that gesture recognition such as Veesion’s AI will still “get the green light”.
“I think it’s fair to say that when it comes to theft, everyone moves in similar ways,” Ward said. “They’re going to be taking your product, they’re going to be concealing it in the bag or on their person.”
It will be more difficult, he believes, for technology that veers more into facial or emotional recognition, “Where you’re screenshotting someone’s face, you’re working out their gender, age. Even some cases it might be, not heart rate, but their general feeling. So it could be do they look nervous? Are they content? Are they calm? That’s where I think it’ll be trickier.”
10 things you need to know direct to your inbox every weekday. Sign up for the Daily Brief, Silicon Republic’s digest of essential sci-tech news.