Retailers such as Target and Rite Aid have reported huge losses due to theft. Computer vision and image recognition tools can help identify and track products.
Nvidia introduced AI workflows to help retailers tackle theft and fraud.
Released on Thursday, Nvidia Retail AI Workflows include workflows for retail loss prevention, multicamera tracking and retail store analytics.
Retail Loss Prevention AI Workflow comes with pre-trained models that recognize hundreds of products usually lost to theft, the AI hardware and software vendor said.
Multi-Camera Tracking AI Workflow enables developers to create systems that track objects across multiple cameras throughout retail stores. It tracks objects and store associates and keeps a unique ID for each object. The vendor said the objects are tracked through visual embeddings or vector representations of images, not personal biometric information.
Retail Store Analytics AI Workflow uses computer vision to track the customer journey around the store, including the products customers carry, according to Nvidia.
The workflows are built on Nvidia Metropolis Microservices, a suite that enables developers to build vision AI applications.
A big problem for retailers
The new Nvidia tools are in response to the retail industry's shrinkage problem -- the loss of goods due to theft, damage and misplacement.
Shrinkage affects all retailers, including giants like Walmart, Rite Aid and Target. Last October, Rite Aid's chief revenue officer reported during an earnings call that the company is considering putting all its products behind showcases due to the rising theft problem. Target said it lost $400 million in profits the previous fiscal year due to shoplifting.
When you look at retail, right now, things like image recognition are very promising because they have multiple use cases.
Robert HetuAnalyst, Gartner
Theft and shrinkage are huge problems for brick-and-mortar retailers these days, but the technology to address the problem is also advancing quickly, Gartner analyst Robert Hetu said.
"When you look at retail, right now, things like image recognition are very promising because they have multiple use cases," Hetu said.
Knowing that someone is stealing something is part of solving the problem, he said. However, tools such as Nvidia's that can recognize the products based on an image or see when it's not in the right place also help reduce theft.
"This is an example where, if you can apply computer vision to faster checkout, less shrink, better stock, better customer satisfaction, tracking and monitoring customer behavior … it is hugely beneficial," Hetu said. "Being able to solve the shrinkage and theft problem is important and has lots of implications from a cost perspective, both retailer and consumer."
Both retailers and vendors like Nvidia are aware of this problem. They're trying to find the balance for using computer vision tools to address the theft and shrinkage problem, while not violating consumers' privacy.
"They're trying to walk that line and say, 'We're using these tools to help make operation more profitable,'" Hetu said. "It isn't necessarily driven around identification of each individual consumer or trying to recognize them."
Another part of Nvidia's retail workflow systems that might benefit retailers is that one of the workflows includes trained models.
"That's a huge benefit for the retailer because, if they can just plug in models that have the ability to recognize all these products, that has really huge implications," Hetu said.
However, retailers not only need to have technologies such as these, but they also need to deploy them correctly, which requires traditional visual observation, Hetu added.
"You can't necessarily accuse somebody of theft while they're standing in your store, so you still [need to] have the operational needs," he said.
