Optimizing Computer Vision to Improve the Grocery Shopping Experience

Optimizing Computer Vision to Improve the Grocery Shopping Experience
Published on

For this smart cart tech provider, it's the computer vision algorithm that makes the difference

Supermarkets have long operated in a highly competitive environment. Slim profit margins, supply chain woes and competition amid inflation have forced businesses in this sector to seek an edge through customer experience (CX). Recently, analytics drawn from shopper behavior has powered merchandising, leading to more informed purchasing and display systems.

However, one area has remained steadfastly analog. The grocery shopping cart is an integral piece of the shopping experience, yet it remains untapped as a networked customer touchpoint. Some companies have introduced smart carts that offer self-checkout services by using AI-powered computer vision (CV) to identify the products that people place in the cart. This allows shoppers to skip the checkout lines, while providing valuable data insights to grocers.

Although widespread adoption of grocery smart carts has been slow, activity in this sector has been gaining momentum in recent months. One solution provider, Shopic, has offered a fresh angle on the smart cart trend, by developing a device that clips on to older grocery carts, which dramatically lowers the investments required to roll out the system.

Since there's less hardware and infrastructure, there's more pressure on device-based processing, to ensure that Shopic's computer vision tech is able to quickly and accurately "see" what products people are buying. Let's take a closer look at why this model might prove effective for supermarkets – and how Shopic's approach to AI makes it possible.

Perfecting Edge Algorithms

Smart carts are a great idea on paper. The cart centralizes a shopper's experience, giving them product suggestions, checkout abilities, and customized offers. However, in practice, the implementation isn't always perfect.

"Many of our competitors have sensors like scales, or more cameras from more angles," says Shopic's CTO and co-founder Eran Kravitz. "Unlike other solutions, we've set out a goal to achieve precision based on computer vision algorithms only."

Whereas other smart carts can be prohibitively expensive, taxing on network resources, and often require dedicated in-store storage areas, Shopic's clip-on is lighter, but that means it needs to perform better with less. "Having a separate, detachable, lightweight device, with no other hardware required, is part of our differentiation," Kravitz says. "Therefore we must put all of our algorithmic efforts into computer vision."

These efforts seem to be paying off. Shopic's website claims that the system can accurately identify 98.5% of all items that it has been trained on, and that it detects when an item is added or removed from the cart with 99.4% accuracy.

As with any electronic device, updates play an important role in ensuring performance for smart carts. Most importantly, Kravitz says, the company focuses on balancing the need for upgrades with operating costs. "It's a delicate balance: better hardware will increase the cost of the device; a newly added algorithm may slow down performance. We have a whole team responsible for device optimizations – constantly making improvements to save even just tenths of seconds wherever possible."

He notes that Shopic tracks development parameters carefully, ensuring each update takes performance and cost viability into account from the planning stage.

Flattening Pre-Rollout Training Curves

For computer vision systems to work well, they often require major investment in training. Shopic's decision to emphasize perfecting its CV algorithms has a significant effect on training times and investment. Staff can train the Shopic system to recognize a store item quickly, so the human resources and lag times required to roll out the system are likewise differentiators.

"We've put effort into research projects to shorten the amount of time required for data collection and training significantly," Kravitz explains. "This process entails both technological and operational challenges. We've come a long way since our first deployment in terms of time-to-market, due to the required training process."

Once Shopic has been deployed in a store, the need for the system to learn continues, as new products are introduced into the inventory mix. "Our system is based on learning algorithms," Kravitz asserts.

"That means that whenever the system fails to recognize something, it receives feedback, and then it can learn from it. This is true for new or changing items at the supermarket as well. If we fail to recognize an item for whatever reason, the system adds that event's data to its dataset, to learn from on the next training cycle. This allows an ongoing learning process without requiring the retailer to do manual labor."

Mitigating Exploitation Risks

Shopic's computer vision has another critical use case, aside from identifying grocery store items. It analyzes shopper behavior for telltale signs of shoplifting and checkout fraud, an issue that has beleaguered the industry and often compromises the adoption of self-service checkout tech.

"We obviously can't give away all of our secrets," Kravitz says, "but due to our high accuracy in recognizing events around the cart, we can also recognize suspicious incidents. We have some alerts that give the shopper a chance to correct the issue, and we also have some silent alerts to draw the attention of store employees."

This system improves upon existing self-checkout lanes, where shoplifting is a routine concern. With theft-deterring smart carts in place, supermarkets don't need to assign employees to monitor self-checkout lanes, reducing the resources they have to dedicate to customer experience.

Kravitz also points out that choosing to shop with a smart device builds trust in other ways. "Another thing to keep in mind is that, though not a must, in most deployments, the shopper is identified, as they log in with a loyalty number or something similar," he says. "Shoplifters usually like their anonymity very much – they wouldn't usually try to steal when identified personally."

The result is less shoplifting and fewer pilferage costs for supermarkets.

A Powerful Package

Shopic's decision to focus on computer vision, instead of revamping an analog shopping cart has caused it to pack its devices with significant value. AI-powered CV algorithms seamlessly blend into the shopper's experience and give stores the analytics they need to offer customized discounts, boosting profits.

Only the future will tell which way smart carts will evolve, but Shopic's CV-focused model holds a lot of promise.

Join our WhatsApp Channel to get the latest news, exclusives and videos on WhatsApp

                                                                                                       _____________                                             

Disclaimer: Analytics Insight does not provide financial advice or guidance. Also note that the cryptocurrencies mentioned/listed on the website could potentially be scams, i.e. designed to induce you to invest financial resources that may be lost forever and not be recoverable once investments are made. You are responsible for conducting your own research (DYOR) before making any investments. Read more here.

Related Stories

No stories found.
logo
Analytics Insight
www.analyticsinsight.net