Today's Amazon Go has burst social media at home and abroad.
Amazon Go stunning place, that is, you can go straight into the supermarket, select the goods immediately after the departure, do not have long rows in the front of the checkout, waiting for checkout.
And even inexplicable there is a pleasure from the store "robbery" & rdquo;
We say that mobile payment to subvert the existing commodity trading, then this "robbery" Amazon Go even "transaction sense" are eliminated.
In short, people are attracted to this novel new shopping paradigm that seems to imply some kind of black technology. Not only ordinary people, and even Stanford professor Li Feifei, O'reilly, founder of the famous science and technology publishing company Tim O'Reilly are specifically to promote the point of praise.
Li Feifei: computer vision is very significant and wonderful application!
Tim O'Reilly: The future of retailing, which I have been expecting for many years.
Amazon Go highlights
Users enter the store, through the gates of the time to open the phone allowed to identify, when the phone in the system starts, and ready to update the list of items.The stunning part is,The moment the user takes or returns the item, the system in the phone automatically updates the list. Finally, the user leaves the item with satisfaction, the phone automatically charge.
This is what in the end there is black technology? On the official release of the video point of view, there are three technical highlights:Machine vision, depth learning algorithms and sensor fusion.
But only a few key words is not enough, Lei Feng network then to the Amazon employees in an interview, but the other said Amazon Go also belong to confidential items, the company asked not to disclose any details.
On Quora, however, a user named Brian Roemmele dug up two copies of Amazon's patent file a few years ago. The system depicted was very similar to the current Amazon Go. The two documents are named: "Detecting item interaction and movement" and "Transitioning items from the materials handling facility". The two documents are listed below. 2013 and 2014 applications (interested friends can answer the keyword in the public number "patent", download 2 copies of patent documents).
The two patent documents described the very complex application scenarios, including: warehouses, shipping, retail market, so many people had to see this document, all think this is Amazon for their own storage system tailored Customized, but today all the truth.
What are the two patents?
"Detecting item interaction and movement" This patent's summary is:
The user picks up or drops an item from a shelf and the goods management system can detect the action and update the list in the user's mobile device.
"Transition of items from the material handling facility" This patent summary is:
The item is recognized, and the item is automatically associated with the user when the user is picking up the item. When a user enters or passes through a "Transition Area", the picked items are automatically transferred to the user without any input from the user.
In general, the identified object is associated with the user, can automatically update the list, and in the "transfer zone" for automated transactions ("transfer zone" in the Amazon Go refers to the store's exit).
We can from the following figure, glimpse of the basic prototype of the Amazon Go.
Figure 208, the object is the camera, about 10 (some are blocked by the body). 208 cameras were placed on the ceiling (as shown on the left and right sides of each have a relatively large camera), shelves and shelves on both sides of the internal. 208 is used to capture the image of the user and the surrounding environment, and 208 within the shelf is used to determine the position of the item or the movement of the user's hand (the position of the user & apos; Entering and leaving shelves).
From the video, we can see the vaguely placed in the shelves of the internal camera.
In addition, the patent document also added: Here the camera can be of various types, can be RGB camera, or depth-sensing camera. In addition to the camera, you can also have other input devices, such as pressure sensors, infrared sensors, volume displacement sensors, screens and so on. The pressure sensor can detect when items are removed and entered, and infrared sensors can be used to differentiate between the user and the hands and objects.
We do not know exactly how far Amazon Go has been using computer vision technology. For example, when the user out of the store, they need to identify the user, the file repeatedly mentioned can be achieved using face recognition technology, but will continue to add additional information can be detected by mobile phones and other mobile devices.
Quora friends Anurag Ranjan is a machine vision and depth of field of quasi-PhD, he said:
Amazon Go uses the camera in the store to identify the user, may take a few photos for the user, relying on a number of standard depth learning network to form a feature map, such as Resnet ,, GoogLeNet or VGG such. In short, when the user stepped into the store the moment, Amazon Go to know what the user looks like.
But according to the patent and the existing experience, in the whole process of using computer vision to track the user, the price is too high. We can generally determine is that the rows of shelves on the camera should be the user "Motion recognition, And to identify whether the object was taken (here may also be the sensor to achieve).
However, the whole process of the difficulty is how to bind the items with the user??
Cloud Advanced Technology Engineer Zhou Xiang told Lei Feng, Amazon Go specific use which technology, at present we can only guess, he mentioned the possible implementation mechanism is:
Through the phone to do the precise positioning of people, and then the specific location of the object can probably determine, once people near the items to leave, while the nearest person's ID will be associated with the item ID.
Shelf certainly uses a pressure sensor, once the items left, it will clearly know what items in the place to leave, once the items back, the pressure sensor will also be sensed.
Depth learning may be learning people's shopping habits, may use computer vision to do a human action recognition, and then according to the items and the person's ID to associate, that is, the second confirmation. While the analysis of a person's shopping preferences, speculation that he would buy something to do some pre-sentence in advance.
Zhou Xiang proposed,Amazon Go may not have used face recognition technology, "Foreign privacy of human face is relatively high, without permission to collect other people's face will be prosecuted". Lei Feng network (public number: Lei Feng network) is also concerned that in the article of Li Feifei tweets, there are users of this conflict, that if the face collection, violation of their privacy.
Amazon Go has a lot of amazing people, but not everyone welcomed, in addition to the privacy issues mentioned above, there are many people worry that if the widespread popularity of this technology (we know that Amazon is more like 2B Business), a large number of supermarket cashiers will lose their jobs, after all, Amazon promised in the video in early 2017, on-line in Seattle, the first Amazon Go.
But this technology how many "real operation", there are still many doubts, for example:
If several people crowded together to buy the same kind of thing (such as shopping malls to grab the goods, which is often happen), then the goods and the person's ID association can maintain accurate it?
If the user put the goods back to the location is wrong, it will be charged?
If someone makes fun of their own empty juice bottles to replace the store's fresh juice, then the Amazon Go out of recognition?
"Therefore, people here require a very quality shopping for the job," Zhou Xiang Lei Feng said. We have a lot of doubts about Amazon Go, perhaps a few months will be able to see the outcome in Seattle.