Home / Uncategorized / Inside the convenience store without Amazon check-in, driven by surveillance

Inside the convenience store without Amazon check-in, driven by surveillance



For now, many have heard of Amazon's bolder attempt to shake up the retail world, the Go store, with no cash and no cash. Come in, take what you want and leave. I had the opportunity to do that recently, as well as choose the brain of one of its leading architects.

My intention was to try to steal something and see these complacent Amazon guys taking a nap. But it became clear when I entered that this was not going to be an option. I was never more than a foot or two from an Amazon public relations representative, and as Dilip Kumar, the vice president of project technology, convinced me, they had already provided such crude attacks on his system.

As you will have seen in the promotion video, enter the store (so far only accessible to Amazon employees) through a door that opens when you scan a QR code generated by the Amazon Go application on your phone. At this moment (well, actually, at the time you entered or perhaps before), your account is associated with your physical presence and the cameras begin to track each of your movements.

The many, many cameras

  1. OLYMPUS DIGITAL CAMERA

  2. OLYMPUS DIGITAL CAMERA

I wondered when the idea of ​​the Amazon ATM store was proposed for the first time how it would be achieved. Cameras in the ceiling, behind the showcases, on the pedestals? What type? Proximity and weight sensors, facial recognition? Where would all this be collected and processed?

Amazon's approach was not as complex as I expected, or rather not as expected. Mainly, the system is comprised of dozens and dozens of roof mounted camera units, covering and recovering every square inch of the store from multiple angles. I guess there are maybe a hundred in the store that I visited, which was the size of a common warehouse or a gas station.

These are common RGB cameras, made to measure with tables in the enclosure to do some basic works of artificial vision, presumably things like motion detection, basic identification of objects, etc.

They are complemented by separate depth sensing cameras (using a time-of-flight technique, or so I understood) from Kumar) that blend in the background like the rest, all matte black.

The images captured from these cameras are sent to a central processing unit (for lack of a better term, without knowing exactly what it is), which does the real job of quickly and accurately identifying different people in the store and objects that are collected or saved. Pick up something you add to your "virtual shopping cart" and you can store it in a bag or in a shopping bag as fast as you want. There is no need to hold it for the system to see it.

  1. OLYMPUS DIGITAL CAMERA

  2. OLYMPUS DIGITAL CAMERA

  3. OLYMPUS DIGITAL CAMERA

  4. OLYMPUS DIGITAL CAMERA

  5. OLYMPUS DIGITAL CAMERA

This is where the secret sauce is, Kumar told me, and I believe him . A problem as banal as it may seem to determine which person similarly dressed took a cup of yogurt almost identical, it is very difficult to hit the speed and precision necessary to base an entire business on it.

A student After all, with the resources available these days, you could probably design a version of this store in a few weeks that would work 80 percent of the time. But doing it right 99.9 percent of the time, without friction and instantly, is a challenge that requires a lot of work.

Remarkably, facial recognition is not used (I asked). Amazon perhaps felt from the beginning that this would make them reproach buyers aware of their privacy, although the idea that these people come to this store seems unlikely. Instead, the system uses other visual signals and monitors the continuity between the cameras: you never see a lens, so it's easy for the system to see how a buyer moves from one camera to another and establishes the connection.

In case there is a technical problem with a camera or somehow the lens is removed, the system does not decompose completely. It has been tested with missing cameras, although naturally it will not be long before a replacement is placed and the system recalibrates.

In addition to the cameras, there are weight sensors on the shelves, and the system is aware of the exact weight of each item, so do not try to grab two yogurts at once and shake the second, as I considered trying. You could do it Indiana Jones style, with an adequate amount of sand in a sack, but that's more effort than most store thieves are willing to turn off.

And, as Kumar pointed out to me, most people are not store robbers, and the system is designed for most people. Building a system that assumes bad intentions instead of simply detecting discrepancies is not always a good design option.

In fact, there is a human being in the circuit if the system is in a bind, but Kumar said that this was rare enough to just need to be considered. He also said that the difficulty of monitoring the store does not increase with square footage, although of course you will need more cameras and more processing power.

It has also been tested with serious crowds; we were there for a slow time in the middle of the afternoon, but shortly before it was lunchtime, they told me, when dozens and not just a handful of people could come and go without doing anything but show their phone to a sensor in the entry.

There may not be ATMs, but there are staff: storekeepers that replenish the inventory; an identification inspector (and sommelier, I'm sure) in the wine and beer section, and chefs in the back preparing fresh sandwiches and food kits. There is also someone in the entrance area to help people with the application, answer questions and receive returns.

The selection consisted mainly of lunches and snacks to take away, with the usual handful of household items that you take in the cellar. the way home. However, the prices were what would be expected in a supermarket instead of in a convenience store.

As for Amazon's expected gambitas that take advantage of their existing properties and hooks, few are found. The application is autonomous, and your purchases are tracked there instead of in your "main" Amazon account. The main members do not get lower prices. Whole Foods has a small section of its own, but there is no broader association (and there are no plans to convert any of those stores to Go, although I can not imagine why not).

In general, I am impressed with the perfection of the system, and I can see that these things work successfully here and there.

From the philosophical point of view, I am concerned, of course, a convenience store that you leave is a friendly mask in the face of a highly controversial application of technology: ubiquitous personal vigilance.

I think it's a bit of an exaggeration to replace a tester or a self-service box with a hundred cameras that record every little movement without blinking. What is there to gain? 20 or 30 seconds of your lap time? The lack of convenience has hardly been a complaint for this market, it is there in the name: "convenience store".

Like many companies are applying technology nowadays, this seems to me a lot of ingenuity and resources that are used to "solve" something that few people care about and even less consider it a problem. As a technical achievement it is remarkable, but so is a robotic dog.

The store works, that's all I can say. Where Amazon will get it out of here I could not say, and no one would respond meaningfully to my questions in this regard. Amazon Go will be open to the public starting this week, but it has not yet been seen if someone will find something more than a novelty.


Source link