Scanning Fruit At Checkout Looks Clever, But Will It Actually Save Anything?

When Toshiba demonstrated its new grocery checkout scanner concept in early March at the RetailTech Japan tradeshow in Tokyo, the idea had a kind of geeky irresistibility: Instead of fumbling to find and type in product numbers for fruits and vegetables, checkers could simply hold each item up to an optical recognition system that would identify the item and ring up the correct price. Very clever—but not very practical for U.S. grocers.

Holding up each vegetable or piece of fruit in front of the scanner for one second? That's not a recipe for efficiency, even once the system actually works.

This video press release does a good job of capturing the tradeshow pitch: a camera inside the scanner that's specifically designed to isolate the item visually, plus logic to identify the item against a database. The demo sounds good. But like any good demo, it's tailored to a tradeshow booth.

What happens in a real store? Getting rid of individual stickers and abandoning RFID tags for produce sounds appealing. But will stores also get rid of plastic bags? If so, customers may have large quantities of loose produce rolling around in their baskets. If not, the checker will have to take out a piece of fruit out of the bag, scan it, put it back in the bag, weigh the whole bag and then send the bag on its way. Or can the scanner identify fruit inside a bag? (If it can do that reliably, wouldn't it have been part of the dog-and-pony show?)

Then there are more subtle issues. "In this demo, there are three kinds of apples: Fuji, Jonagold and Mutsu," Toshiba spokeman Keiichi Hasegawa says in the video. "The Fuji and Jonagold originally come from the same stock, so if you're not really familiar with apples, they might look the same. But this scanner can distinguish them, by recognizing subtle differences in pattern and coloration."

But that's not enough for a grocery chain's produce department. Can this scanner tell the difference between regular and organic tomatoes by visual inspection? (Can anyone?) Can it tell the difference between a small Large navel orange and a large Regular navel orange? Will it differentiate a Chiquita banana from a marked-down no-name version or locally grown produce from the same varieties that are shipped in? These are all essentially marketing issues, and they're not easy for machine logic to handle.

And in practice, any new checkout technology that fills this tiny a niche has to be really good at what it does. Hasegawa says the scanner "can recognize objects very fast, even if they're moving," thanks to technology that eliminates surroundings and other visually noisy images.

It's still not as fast as barcodes for packaged goods, though. That one-second hold in front of the camera means any packaged item with a printed barcode will go through more quickly using existing scanners.

Toshiba is busy building a produce database, so this is clearly aimed at that section of the store. But for a typical U.S. grocery chain, it's hard to see how this pays for itself. (Cutting out the human checker, as with Kroger's prototype scanning tunnel, seems like a better cost-saving approach for a new technology.)

Still, it's nice to see Toshiba coming up with this type of concept checkout. Like a concept car at an auto show, it can have some great ideas—or great technology—built into it. But it still doesn't have to be something you'd want to take for a spin, much less actually use every day.

And who knows? Cameras are getting cheaper. If Toshiba unties its product-recognition software from a fixed checkout system, and then rebuilds it as part of a smartphone-based self-checkout system, it might turn out to be a lot more useful—and cost effective—after all.