What you need to know about facial recognition technology

Zaur Abutalimov

Director of product service cloud video surveillance and video analytics for business Ivideon.

Elena Glazkova

Marketer Ivideon.

For the state, face recognition - an important part of the security system and an impressive budget item. For journalists - either a panacea or an instrument of a world conspiracy. For business - a tool or product. Whose side nor accept, the basic questions still remain. Answers to these users routinely search the Internet (an average of 28 704 a query on the subject of face recognition in a month), but the show is not always. To correct the situation.

What is face recognition

Separate flies from cutlets. Users increasingly faced with face recognition in their own smartphonesWhere biometric identification is used to unlock the device and gain access to only the owner could data. During the recognition process necessarily involved 3D-camera, it was impossible to deceive a photo gadget.

Yet there is identification of individuals in real time and real conditions, in which case it is inseparably associated with video surveillance systems, where people literally "snatching away" of the shoot cameras video stream.

Imagine a high-quality modern video camera, placed just above the average height of a man in a well-lit place. Before her every day it takes about the same amount about the same people. Move they are not very fast.

The captured video can be stored in the cloud archive. The camera connects analytical module: a complex combination of algorithms (artificial Intelligence, Neural networks, that's all) plus the user interface. Module "grabs" the face of the video stream, determines the gender and age, and puts data into the database.

Gradually the image becomes larger. The system memorizes all detected faces automatically and stores them in a file, and the user with a tolerance specifies additional information: name, position, status, and other marks ( «VIP-guest" or "thief"). You can upload a photo of the desired person, and the module will find the archive in the detection of all of that person.

Once a person with a mark again passes in front of the camera, the system detects it as an important event and sends a push-notification to interested users.

Detection of face recognition in the context - this is the situation when the algorithm basically realized that the person in front of him, instead of an apple or a mermaid with Starbucks cups. The computing power required him first to do so, and only then it can be compared with the database or the person to remember.

If you've read the previous few paragraphs to the end, congratulations, now you know how the face detection in an ideal situation. Description suitable for any system, from those used in the Moscow metro, for small business solutions.

The main thing to understand: the ideal situation in real life to create difficult, especially when it comes to the whole city, not the office or shop. For example, in the subway a lot of people, all different, they go fast. Cameras need a lot, they are worth the money, place them need competent experts.

Is it possible to trick the algorithm face recognition

Despite the case of a miss, the accuracy of machine recognition has often superior to the one with which people define the face. China will soon beChina to build giant facial recognition database to identify any citizen within seconds system capable of Find a specific person among the 1.3 billion other inhabitants for 3 seconds with an accuracy of 90%.

Yet clearly this question is difficult to answer because the only perfect face recognition algorithm does not exist. Big glasses glued a beard, a cap, high-speed movement, the special make-up (for example, painted on the face grille"Black Swan", seals, circles and sticks. How to escape from the face recognition systems using makeup) - all of this is able to confuse algorithm. Especially in combination, because it is enough to recognizeHow to cheat detection system whether 70% open face. Now imagine that you must use the above shifts in the real city. That does not sound so simple, right?

Is it possible to recognize individuals online

the Internet - a place of paradox: the people here can simultaneously worry about, does not determine whether each a second camera on the streets of their personality, and genuinely want to "recognize other people's faces for photos online. " Consider this line of face recognition separately.

facial recognition software - it is either above analysis module (CCTV Camera + software + cloud storage), or soft, similar to the well-known (slightly scandalous) service FindFace. Today download recognition program entities "for free and without registration," in most cases, of course, impossible.

The dream of a user who enters a query, obviously, is as follows: Go to the site, upload photos human filmed surreptitiously on the subway, the program recognizes the face and gives a link to the profile in the social network. Aha, I caught! Or is it: to download programs to your computer, connect to her webcam and her cat raspoznaosh muzzle. Success - now you will receive a notification every time a cat steals sausage.

The reality is cruel. The first site that offers similar, refuses to work, and the other - requires the Python programming skills. More or less like a dream app called SearchFaceWhich recently restartSearchface restarted with authorization through "VKontakte". But the social network closed this function called FindClone. You upload your photos, and the algorithm tries to identify the same person in the database of social network "VKontakte". References to the application has not issued the profile, only the pictures - and it does not matter who they were loaded. If the user has long been active in social networks, photo issue created eerie "biographical" effect, but if not, the recognized image can laugh.

Actually, the example SearchFace clearly answers the question "How to use social networks face recognition?". More accurate to cformulirovat it this way: "As social networks are used to recognize faces?" The answer is simple: the database. An infinite number of unique combinations of numbers (this is for the algorithms Facebook"VKontakte" and the other person look at the photo) forms the basis for the training of neural networks, which are the basis of a decision face recognition.

Solutions are all different, and the neural network is also different, and the details and technical specifications, customers and suppliers of services, as a rule, are not disclosed. In particular, gender and age recognition module is able to determine due to the fact that it can learn from the information contained in the "Classmates", "VKontakte", Instagram and Facebook.

As programmed face recognition

You should never respond to questions, and developers for developers, if you are not a developer. Therefore, we turned for help to a specialist.

Dmitry Soshnikov

Member of the Russian Association for Artificial Intelligence and senior expert on the development of AI systems and machine learning Microsoft.

Face detection (as well as other related operations) - it is a typical problem. Therefore, many companies provide complete services in the form of a cloud API (programming intermediaries between applications) for high-quality solutions to these problems. In addition to IT-giants like Microsoft and the Google, face recognition are also involved in the specialized companies, including Russian. Their products are developing rapidly and provide an even more interesting features such as the identification of persons and silhouettes in the crowd.

Himself from the ground to train a neural network is much more complex. Need a large and high-quality set of input data, that is, hundreds of thousands (or better even more!) Of photos of people. In addition, it will need substantial computing resources and knowledge in the field of AI and machine learning. Big companies have all these tools, so solve the problem much better.

There is also an intermediate solution - already used to train the neural network, for example, OpenFace. This option is likely to be to work a little worse than ready cloud service, however, will allow to have full control over the system. This will require a certain level of understanding of the work neural networks and neural network frameworks and, apparently, some knowledge of Python, which has gained popularity as the main programming language among the Data Science professionals.

Indeed, it is convenient to carry out various experiments to visualize data and produce effective matrix calculations thanks to the excellent NumPy package. This is not the best language for commercial development because it does not contain any effective means to create more security software systems, however, alternatives to him in the field of training deep neural networks yet not.

How does facial recognition in business

Demand for face recognition in fintehe, retail and other types of business directly related to the increased availability of technology. The mechanics is simple: all enterprises and in all organizations there are surveillance cameras, which are used as tools for data collection and subsequent analysts. In the world of the surveillance system is removed in the past month terabytes of video in Full HD, that is, information processing is stored is really a lot.

Required software for data analysis can be "stitched" to the device manufacturer. Cameras with video analytics "on board" are usually quite expensive.

Alternative - analytics in the cloud, that is, remote data center, which is connected to any inexpensive camera. It is much cheaper, plus provides flexibility - you can tailor solutions for specific business.

recognition technology popularity persons in different fields of activity increases. For example, the Savings Bank - one of the leaders in terms of the announcement of various high-profile projects face recognition, and argueHe recognizes you from a thousand ATM determine the client's eyes with him in this regard may perhaps that "Tinkoff". In 2017, Sberbank acquiredSavings invested in facial recognition technology 25.07% of the company VisionLabs, creating software for facial recognition. For the 2018 financial institution has managed to test the face recognition in the Moscow subway, and even catchThanks to the system of recognition of Sberbank of persons caught 42 criminals 42 criminal testHe recognizes you from a thousand ATM determine the client's eyes ATMs with the identification of persons that attackers can not withdraw money from other people's cards, as well as to announce the collection of biometric data (voice audio, video face) customers. In April this year, Sberbank got control of the developer of voice recognition systems and people - "Speech Technology Center" (MDG).

Another thing is that the preview, test, pilot and buying decisions - not to actually implement. That right now it is really used in the Savings Bank (and if used) is safe to say may actually only German Gref.

With retailers all transparency. In fact, there are three problems that face detection solves.

Firstly, theft. The stores are operating fraudsters, With often the same people in the same network. Face recognition makes it possible to determine the "drifting thieves" and others, previously violated the order. As soon as one listed in the base of the offender will go to the store, the protection will be notified in the messenger or other convenient way.

Second, the difficulty of working with our regular customers. Data on purchases and birthdays to personalize offers for VIP-clients and fans of the brand, is simply not enough. Face detection can be integrated with CRM - ie software in which managers are entered all the information on all the transactions of the organization. In cases with thieves and VIP face recognition works about the same: a person entered in the black or white list, and when it reappears, the system will beep person with access. Gender and age are automatically detected, and additional information to add the responsible officer.

Third, the identification of persons in reteyle used for targeted advertising. For example, in some stores X5 Retail Group establishedX5 include computer vision camera to recognize facial expressions and customer age. By analyzing this data, the system displays on the screen on the trading floor products that can please the man. More vivid illustration - Case Lolli & Pops, large pastry shop in the United States. Face recognition system determinesYour future in-store loyalty program will be fed by facial recognition regular customers and sends their smartphones notice with products that can please them (taking into account individual preferences and even allergies to foods).

Another striking example of the use of technology in reteyle - shops without merchants and banks. For example, Alibaba Tao CafeAmazon Go vs Alibaba Tao Cafe: Staffless Shop Showdown - a cafe and supermarket, located in Hangzhou. It sells drinks, snacks, food, toys, backpacks and the like. Tao Cafe is only open to users of the site Taobao.

When buying drinks camera system with support for Face Recognition automatically identifies the client associated with their account in the online store and process the payment. Buyers go through the room, equipped with several sensors, which identify both the client and products. Scanning works even if people put the purchase in your pocket or bag.

As the developing face recognition technology

video surveillance systems with the identification of individuals is really taking over the world. In Moscow, the number of cameras in 2019 to reachHigh technology and security: How many cameras will appear in this year 174 thousand. This does not mean that all of these default device can recognize the personality: the most commonly reportedRecognition System criminals are wanted by the camcorder will work in Moscow in 2019 about 160 thousands of cameras with this feature. Nevertheless, at the end of 2018 of the Moscow City Hall announced its intentionThe authorities in Moscow in 2019 are going to replace your camcorder and run the facial recognition system replace all the surveillance devices and form a completely innovative system in the next year.

The paradox is that the 160 thousand - it's not so much. China - particularly when compared with other leading search engines queries on the subject of face recognition. There, at the end of 2017 wasIn Your Face: China's all-seeing state more than 170 million video surveillance cameras and for the next three years plannedChina's 'Big Brother' surveillance technology is not nearly as all-seeing as the government wants you to think connect to the network has about 400 million.

Proper and correct use of face recognition works primarily to enhance the safety and comfort. People usually penetrates quickly confidence in the technologies that eliminate them from the queue to a football match (smiling chamber - passed), to prevent theft and hooliganism, or help less to spend on purchases (loyalty program). All this, of course, requires a certain regulation - specifically for this protection laws are made personal data.

In the future, probably, the scope of face recognition in video surveillance systems will be regulated similarly to the current practice of working with the identification of individuals on the Internet. Seeking privacy people just do not load in excess Network - Partial failure of service SearchFace proves that this strategy is effective.

Of course, one can not indefinitely limit itself in walking in the streets, where the cameras are mounted on each intersection, but the possibility to remain anonymous is formed, if it is requested by the society.