Blog

Amazon Wants To Kill the Barcode - Slashdot

Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way. Android 9000mah Battery Mobile Computer Scanner

Amazon Wants To Kill the Barcode - Slashdot

Amazon has a specific type of problem that it currently uses barcodes for, and there are enough problems with that application that they're trying to do better, this way or that way.

That doesn't mean all barcodes are on the way of the dodo. But journalism apparently is, right on the heels of slashdot editorship.

Barcode on cardboard. On a string. On a thing.

That'll be 1 million dollars, Jeff.

Yeah, I can see Amazon using this vision system to tell the difference between a string trimmer and a fishing pole, but what about items that differ only in subtle options? Can it tell the difference between a 4GB Pi and a 8GB Pi package just by the shape?

The article already mentioned problems in selecting similar products that differ only in color so they seem to have recognized some of the problems moving forward.

On the packaging of a product, the barcode should be on all sides.

On the packaging of a product, the barcode should be on all sides.

Because that causes other problems.

I bought such an item, a fairly big flat box which, indeed, had the bar code on all sides.

Guess what happened? As the till operative passed the box in front of the scanner it found three of the bar codes! It took much longer for them to cancel two of the supposed items from the till than it would have taken turning the box over to find a single barcode.

But thatâ(TM)s easily solved for a distributor like Amazon. When they come off the pallet, just affix a 3D barcode on every side of the box or an RFID tag. You can do it small enough that it wonâ(TM)t be intrusive to the packaging or cost all that much.

The logical approach would be to have a barcode detection model that can help locate the barcode on a product with relation to a positional frame of reference, then feed this in as input features to the barcode scanning robot. In parallel, Amazon is big enough that it can lobby for normalization of labelling regulations + barcode placement for the few products that cause it trouble.

Using AI to guess every object in their inventory sounds almost as dumb as killing barcodes.

"The robot has trouble finding the barcode. So we'll take pictures from various angles as it moves down the line..."

"Yeah, that'll make sure the barcode is seen."

"No... we're going to give the pictures to an AI to figure out what the product is."

If you make robots to turn objects around to recognize them, you can also program them to find the barcode to scan it.

Exactly. And 99% of the time the barcode is in the same location on each item. Find it once and remember forever.

Anyway, weren't RFIDs going to kill the barcode?

In 1996, the first patent for a batteryless RFID passive tag with limited interference was granted -- WP

Anyway, weren't RFIDs going to kill the barcode?

Anyway, weren't RFIDs going to kill the barcode?

Walmart is big into using RFID for store inventory -- Walmart RFID [google.com]

Walmart also uses bar codes at their cash registers. Last I heard, they wouldn't carry merchandise without bar codes, period.

Walmart also uses bar codes at their cash registers. Last I heard, they wouldn't carry merchandise without bar codes, period.

Walmart also uses bar codes at their cash registers. Last I heard, they wouldn't carry merchandise without bar codes, period.

I think the RFID tags are for detailed inventory tracking, so they don't have to physically scan everything, but simply walk by... Noting that not everything they carry has a chip yet. Barcodes still make sense for checkout scanning at this time.

For Walmart and really any other retailer, barcodes are the lowest common denominator. If 100% of products had an RFID they could theoretically change their barcode policy. But RFID on low margin items or food is likely not a solved problem in the industry.

RFID tags serve a different purpose for a retailer than UPC codes do, and will continue to do so. It's hard to image them ever replacing bar codes at checkout.

perhaps I have an overactive imagination.

Anyway, weren't RFIDs going to kill the barcode?

Anyway, weren't RFIDs going to kill the barcode?

Look Satan, we don't want your mark of the beast in our bologna packages.

Look Satan, we don't want your mark of the beast in our bologna packages.

Look Satan, we don't want your mark of the beast in our bologna packages.

It's the UPC that contains the MotB. The first two thin bars, the middle two thin bars and the last two thin bars each represent a '6'.

How about this for a wild idea? Have someone listing the item on Amazon tell Amazon where the barcode is.

Items don't magically appear in Amazon's database - someone had to enter that information into a database somewhere that Amazon synchronizes with to obtain their product information.

It's very easy for Amazon to demand that if you want Amazon to carry your product you tell Amazon where the barcode is. You can take photos of the item on various sides, then highlight the barcode. (Software can take a best g

But this tech could be a good step towards a robot that can sort, wash, dry and fold laundry.

I like the sci fi film where people pull on a new paper jumpsuit and put their old clothes in an incinerator. Saves from having to buy detergent and keeps microplastics out of the environment.

A further step is for the robot to wear the laundry for you so you can go off and do other things (Thanx Douglas Adams).

No no, this is better. The robot actually recognizes the *item*. Now, just to make it extra reliable, they should get everybody together and agree on some kind of standardized fiducial with encoded information to put on all packages that would provide detailed information, including for less capable systems.

Smart phones recognize them, even if they're oriented wrong.

Only if the orientation is close to a 90 degs multiple of the correct orientation otherwise forget it. Ditto if the lighting is to dark or bright or a shadow falls across the square.

Only if the orientation is close to a 90 degs multiple of the correct orientation otherwise forget it. Ditto if the lighting is to dark or bright or a shadow falls across the square.

Only if the orientation is close to a 90 degs multiple of the correct orientation otherwise forget it. Ditto if the lighting is to dark or bright or a shadow falls across the square.

Well, if they're using a robotic hand, they can "hand-wave" that problem away.

You're adding a ton of needless complexity to what could be a very simple system. QR codes just aren't a good fit.

You're adding a ton of needless complexity to what could be a very simple system. QR codes just aren't a good fit.

You're adding a ton of needless complexity to what could be a very simple system. QR codes just aren't a good fit.

I'm not the one adding the "needless complexity." Amazon is - FTFA:

Eventually, Amazon's AI experts and roboticists want to combine the technology with robots that identify items while picking them up and turning them around.

Eventually, Amazon's AI experts and roboticists want to combine the technology with robots that identify items while picking them up and turning them around.

Much easier to stick a QR code on opposite faces of the object, and a bunch of $30 FHD cameras feeding into a central image recognizer/QR code interpreter. No need for a robotic hand even.

Decent barcode scanners recognize barcodes in any orientation too. The difference is, you don't need a camera for a barcode.

QR codes were (and are) an alternative to barcodes where you need to encode more information. Since they do encode more information, they need more hardware and are potentially less reliable, so they didn't (and shouldn't have been expected to) replace barcodes. They get used for new things.

That "advantage" doesn't outweigh the many enormous drawbacks. Why are you so hot to defend QR codes anyway?

I suspect the answer is in the username.

I'm not "defending" QR codes. I'm saying that they have their uses - no need for a robotic hand to pick up and turn things around when a bunch of $30 cameras and a server with QR code software can do the job.

Same as RFID tags have their advantages and disadvantages.

And the GP and I both pointed out that a barcode does it even better. You don't need cameras or a server, and it works much more reliably.

"You can... include a lot more information" Which is irrelevant when you're in an environment with controlled connectivity. You only need a unique identifier.

"You can... include a lot more information" Which is irrelevant when you're in an environment with controlled connectivity. You only need a unique identifier.

"You can... include a lot more information" Which is irrelevant when you're in an environment with controlled connectivity. You only need a unique identifier.

Obviously you've never dealt with inventory control. It's VERY relevant when the same basic stock number can include multiple revisions (such as motherboards, or video cards - which bios version, for example). Same with cars, food, etc. "Yes, it's romaine lettuce from supplier x - but from which field (or have you forgotten the salmonella outbreaks where the individual field, and even the shift, turned out to be relevant)?

If you can item it down to the individual pallet, which product bar codes can't, yo

" you don't need a camera for a barcode"

A barcode reader is a camera, these days.

That was all in your imagination. There are many different kinds of 1d and 2d barcodes. One size does not fit all.

Using pictures of items in Amazon warehouses and training a computer model, ...

This "camera system" as an alternative is not good, Unless it's to Scan written standardized codes on the product's shell or packaging, because similar items can have the same picture. Example: A 16 Inch Macbook with 32 Gigabytes of RAM and another laptop in the exact same series, but the model configured with 64 Gigabytes of RAM soldered to the board instead of 32... the two products look exactly the same, but the SKUs are different; They would be impossible to reliably distinguish in a warehouse without being able to reliably read the codes on the box. Having a Unique Code that identifies the indistinguishable in outside appearance but very different products is the most reliable way to identify the two accurately.

I ordered a 2TB SATA SSD drive, I got a 1TB NVMe drive instead. The barcode sticker matched the product that I ordered; it was on the wrong box. So barcodes aren't perfect, but when two boxes are very close, that doesn't help either.

I had the opposite happen me once too. Bought a nice Sony 720p flat panel LCD during a Boxing Day sale back in the days when those things were still pretty pricy, got it home and while the box had 720p all over the outside of it inside was the much more expensive 1080p panel that wasn't on sale and cost nearly twice as much at 720p at regular price.

I was tempted to return it because of the extended warranty I'd bought, high priced item back then, but in the end I decided to risk it and saved about half the

Barcodes are used for many purposes, so manufacturers will still use barcodes regardless of what Amazon is doing. What "killing the barcode" means is that Amazon wants to use an alternate solution to the barcode for item recognition as part of its packing process. So, the only thing being killed is Amazon's use of the barcode for packing item. The barcode will still continue to exist in the future.

Amazon was using the barcode to double check that an item being packed was the right item. Their new solution is based on visual identification. Obviously, having either a robot or a human visually inspect an item is only possible if two potentially similar items are visually distinct. That's a problem since many items are either very similar (e.g., a 2x4 versus a 1x4) or the difference is entirely hidden inside the exterior (e.g., the iPhone with 128GB or 256GB).

Also, Amazon is claiming 99% recognition accuracy, which implies a 1% error rate. If an error requires a human to process the error, then that's a huge number of items that requires a human instead of a robot. I imagine that a practical error rate has to be quite a bit less than 1%.

I imagine that a practical error rate has to be quite a bit less than 1%.

I imagine that a practical error rate has to be quite a bit less than 1%.

To be useful, it only has to be less than the human error rate. Which they say it is.

Also, Amazon is claiming 99% recognition accuracy, which implies a 1% error rate. If an error requires a human to process the error, then that's a huge number of items that requires a human instead of a robot. I imagine that a practical error rate has to be quite a bit less than 1%.

Also, Amazon is claiming 99% recognition accuracy, which implies a 1% error rate. If an error requires a human to process the error, then that's a huge number of items that requires a human instead of a robot. I imagine that a practical error rate has to be quite a bit less than 1%.

It depends on how many human interventions are required with barcodes. If the robot can't find or read the barcode on 10% of the items, then a 1% error is still a big win.

> visually inspect an item is only possible if two potentially similar items are visually distinct. That's a problem since [...] the difference is entirely hidden inside the exterior (e.g., the iPhone with 128GB or 256GB).

Not to mention that there's always the danger of the robot holding it wrong.

This isn't the real reason they want to get rid of barcodes.

It's because they want to wrest any remaining control over their operations from a non-Amazon entity. In this case GS-1, the body that sets standards for barcodes internationally.

You mean your camera systems can't figure out how to identify a barcode, and do any necessary 3d transformations to make it straight?

Yes, what you describe can easily be done, and many places already do this without issue. The summary mentioned conveyor belts, and barcodes are typically only printed on one side of the box, so the camera cannot see it when the barcode is on the downside.

Someone at Amazon is bucking for VP, and this is their thesis project. It sounds like an over complicated solution to a very minor problem. I think they will likely learn this is too costly to maintain, as the model will need constant retraining as warehouse inventories keep changing and box-art is not always permanent for a product -- ever seen "New look, Same great product!" printed anywhere? I also think error rates will be too high to be independently useful, many items come in very generic outer packaging, and many items that share a common use also share a common look.

So, yeah, this looks like a waste of money.

Yes, what you describe can easily be done, and many places already do this without issue. The summary mentioned conveyor belts, and barcodes are typically only printed on one side of the box, so the camera cannot see it when the barcode is on the downside.

Yes, what you describe can easily be done, and many places already do this without issue. The summary mentioned conveyor belts, and barcodes are typically only printed on one side of the box, so the camera cannot see it when the barcode is on the downside.

So, make a transparent conveyor belt and call it a day

So, make a transparent conveyor belt and call it a day

So, make a transparent conveyor belt and call it a day

Not exactly, but yes this is the sort of solution that would actually work.

One way is a section of rollers, rather than belt, so that the rollers may have small gaps between them, and a high-ish speed camera (120fps sounds plenty fast) with sufficient light to capture continuous imagery of the bottom side. Then it's just a software problem to stitch the images together into a continuous stream and look for barcodes. Problems solved, no machine learning necessary, and they can use continue using the existing

Or you can just use your own QR codes. Easy to generate, can contain as little or as much info as you need, and smartphones already handle them. Maybe they can use the robotic arm to hold a smartphone? Stick the same QR code on opposite sides of the object and it will always be visible.

Sort of how we have printed labels on more than one side of packaging so it doesn't matter which side is facing the purchaser?

The problem is that Amazon accepts packages in loose retail packaging from suppliers. Packaging made for retail shelves require human handling to pick and place. Amazon needs to tell their suppliers to package items in standardized form factors that a robot can handle.

just consider the barcode as part of the item's image

Computers are still far better at reading bar codes, than they are at OCR. When I, as a human, try to read the numbers on some receipts, it's difficult to tell the difference between an 8 and a 0, when they do dumb stuff like putting a dot in the middle of the 0. OCR still isn't as accurate as human reading, so what we'd have is a larger number of scanning errors, if we were to switch from bar codes to OCR, for purposes such as store checkouts.

Instead of reading a plain barcode 100% perfect at all times and that can run in a meek Arduino, now they improve the method by introducing color images that need a programmed IA to associate them to the database of barcodes the rest of the world uses using a extremely powerful computer!

One somewhat overlooked aspect of the bar code is that it is legible to human beings. Not only are the codes themselves completely standardized and not that difficult to memorize, but products almost invariably have the numerals written below the bars that can be used as a backup for the average individual.

I can see other techniques supplementing bar codes on products, but not replacing them entirely.

Using image recognition to double-check that the barcode does indeed belong to a bannana and not a steak seems like a pretty good use of this technology, but I struggle to understand how it could be faster, more accurate, or cheaper than using an existing near-universal retail standard.

Why not take pictures of where the barcodes are if youâ(TM)re already training on this, then the robots can find it?

Side note, when weâ(TM)d use barcoding/ocr frameworks we never had issues with finding a barcode, it was a problem finding the right barcode.

Mail sorters not only can read bar codes off oddly shaped and oriented objects, they can read printed or even hand-written addresses off oddly-shaped objects and print barcodes back on those oddly shaped objects.

They're basically admitting that they're technologically inferior to the Post Office.

There may be more comments in this discussion. Without JavaScript enabled, you might want to turn on Classic Discussion System in your preferences instead.

Microsoft Acquires Startup Developing High-Speed Cables for Transmitting Data

Colgate's 9 Billion Toothpaste Tubes Defy Effort To Recycle Them

Amazon Wants To Kill the Barcode - Slashdot

Rugged Handheld Terminal Scanner 10 to the minus 6th power Movie = 1 Microfilm