⎝⎛im体育娱乐⎞⎠

National Service Hotline: 021-67756215 67756203

⎝⎛im体育娱乐⎞⎠> Technical Support
Face recognition technology and its application

Three modes of face recognition

Three application modes of face recognition: The 1: 1 authentication mode is essentially a process in which a computer quickly compares the current face with a portrait database and determines whether it matches. It can be simply understood as proof that you are you. 1: N is to find and match the face data of the current user in the massive portrait database. M: N is the process of facial recognition of everyone in the scene by computer and comparison with the portrait database. M: N, as a dynamic face comparison, has a very high usage rate and can be fully applied to a variety of scenarios, such as public security, welcome, robot (300024) applications, etc.

The main business application scenarios of face recognition include security, access control, new retail, and integration of witnesses and other fields: in the security field, with the deepening of the concept of intelligent security, there is a need for intelligent identification and analysis of front-end cameras and back-end processing; intelligent access control systems It is one of the earliest application carriers of face recognition in the commercial field. Among the advantages of face recognition access control system over other biometric access control systems are: natural, non-mandatory, non-contact and intelligent data output; intelligent welcome system Based on dynamic capture and non-cooperative face recognition technology, linked with access control, and combined with the background management system can achieve fast and accurate identification of guest identity, visitor invitation, visitor registration, visitor data statistics, visitor query and other functions; face recognition is also AmazonGo The key technology represented by the new retail application scenario; in addition, the ID card combination product combines ID card identification and face recognition technology, and the market demand is strong.

The current profit model of face recognition is mainly to B. At present, the main profit models of face recognition companies include enterprise-level technical services and software and hardware sales. Domestic deep learning companies represented by Shangtang Technology and Megishi Technology, in the B2B2C business model, cooperate with leading companies in various industries to jointly promote the application and realization of face recognition in various industries. They output their technical capabilities to the B-side, and use the model of sharing, charging according to the license, and charging based on the number of uses of the technology to bind the business growth of the B-side serving C-side customers, and use the industry resources of the B-side to open up the market.

Face recognition has been commercialized in the core business of several listed companies. Face recognition technology has been successfully applied to multiple industries represented by the security industry, and has entered the core business and core product systems of multiple listed companies. Not only has it greatly improved the operational efficiency of the original business of each company, but it has also created multiple incremental markets with broad prospects. 智慧城市 等大型项目,通过项目形式盈利。 Taking the security industry as an example, there are three main business models: 1. Providing security vendors with video structured, face control, face search, vehicle recognition, crowd analysis, and other software and hardware integrated forms; providing customized GPUs, smart cameras, and other integrations; Charge solutions based on the number of surveillance video processed; 2. Integrate face recognition into identity recognition products, gates, counter products, etc., and make profits through industry channels through productization; and 3. cooperate with traditional security vendors Cooperate to win bids for large-scale projects such as safe cities and smart cities through their channels, and make profits through project forms.

First, the three modes of face recognition

1.1 1: 1 mode for face recognition

The 1: 1 authentication mode is essentially a process in which a computer quickly compares the current face and portrait database and finds out whether or not they match. It can be simply understood as proof that you are you. 1: 1 as a static comparison, the potential commercial value in the core of pan-finance and information security is huge. For example, in the process of airport security check, the process of matching the appearance of the cardholder with the ID information is a typical 1: 1 scenario. However, the human eye recognition rate is only about 95%, and it will be affected by the external environment. Therefore, airport security personnel will ensure the accuracy of the recognition by changing shifts. The emergence of face recognition technology solves the disadvantages of manual recognition, and can be fully applied to the examination of the identity of test candidates, hotel check-in, train station ticket integration, mobile payment and other scenarios that require real-name system.

1.2 1: N mode for face recognition

1: N is to find and match the face data of the current user in the massive portrait database. 1: N has the characteristics of dynamic comparison and non-cooperation. Dynamic comparison refers to the process of obtaining face data and further comparison through interception of dynamic video streams. Non-cooperation is non-mandatory and efficient in the recognition process. Performance, the recognition object can complete the recognition work without going to a specific location. Because of these two characteristics, the 1: N authentication mode can quickly land in scenarios such as public safety management and VIP customer face recognition, but its difficulty is much higher than static 1: 1, because the machine is facing overexposure, backlighting, side Face, distance, and more.

1.3 M: N mode for face recognition

M: N is the process of facial recognition of everyone in the scene by computer and comparison with the portrait database. M: N, as a dynamic face comparison, has a very high usage rate and can be fully applied to a variety of scenarios, such as public security, welcome, robot applications, etc. However, the M: N mode still has a lot of disadvantages, because it must rely on a large amount of face databases to run, and because of the large recognition base and insufficient device resolution, the M: N mode will generate a high error rate. Thus affecting the recognition results.

Although the maturity of face recognition technology can replace part of the labor force, it still cannot be used as the sole verification method. It needs to be combined with manual recognition to make accurate judgments. For example, under the interference of the external environment, face recognition technology will generate erroneous data, and manual assistance is required to complete the recognition and confirmation process; or in enterprise applications, places with high confidentiality requirements can use face recognition Two-factor authentication with credit card to further ensure security.

Second, the main commercial application scenarios of face recognition

2.1 Security industry

The security industry has broad application space. The scale of China's security industry has increased from 230 billion yuan at the end of the "Eleventh Five-Year Plan" to 500 billion yuan at the end of the "Twelfth Five-Year Plan", a growth rate of up to 18%. The growth rate of the domestic security market is much higher than that of the world, and it is expected to reach 65.4 million yuan in 2017. Among them, video surveillance, as an important link in the construction of a safe city, will be one of the most valuable application scenarios for face recognition.


Intelligent video surveillance is imminent. With the rapid increase in the number of cameras and the continuous advancement of the camera network process, the generation of massive video data has far exceeded the management scope of monitors, and the intelligentization of video surveillance is becoming more and more urgent. 人工智能 引入视频监控中,使其具备自动整理与分类的功能,将数据结构化处理,并使处理后的结构性数据能大规模被用于检索,分析与统计中,最终通过针对性深度挖掘使其成为有意义的情报数据。 Intelligent management is to introduce artificial intelligence into video surveillance, make it have the functions of automatic organization and classification, structure the data, and enable the processed structured data to be used in retrieval, analysis and statistics on a large scale. Finally, it becomes meaningful intelligence data through targeted deep mining.


Driven by deep learning technology, face recognition technology can simultaneously have the ability to identify person attributes and identities. Driven by deep learning, face recognition can realize real-time detection under any face occlusion and perspective, which overcomes several problems in face detection at one time: side faces, semi-occlusions, and blurred faces, greatly improving each Face detection effect in a real situation. At the same time, it can identify gender, age, expression and various facial physiological characteristics, not only can accurately identify the gender and age of the person in the photo, but also provide expression, face value (beauty index), wear glasses, make-up makeup, apply lipstick, wear There are more than 40 attributes such as hat, hair color, and beard style, with an average accuracy rate of more than 90%, and the average error of age prediction is less than 3 years old.

The field of video surveillance can realize human behavior and vehicle identification in multiple scenarios. Pedestrian detection algorithms based on deep learning can accurately find the position of pedestrians in various situations of occlusion, and can further analyze the posture and movement of pedestrians, which can be applied to traffic monitoring, assisted driving, and unmanned driving. It can detect a variety of vehicles from different angles in driving scenes, traffic monitoring scenes, and bayonet scenes, and provide physical characteristics such as license plate number, car brand, model, and color.

The field of video surveillance can realize human behavior and vehicle identification in multiple scenarios. Pedestrian detection algorithms based on deep learning can accurately find the position of pedestrians in various situations of occlusion, and can further analyze the posture and movement of pedestrians, which can be applied to traffic monitoring, assisted driving, and unmanned driving. It can detect a variety of vehicles from different angles in driving scenes, traffic monitoring scenes, and bayonet scenes, and provide physical characteristics such as license plate number, car brand, model, and color.

The intelligent transformation of cameras is another market opportunity in the security field. The accuracy of the camera will affect the accuracy of the final face recognition. Previously, a large number of cameras had three fatal shortcomings: 1. It is easy to lose distance information when three-dimensional to two-dimensional; 2. Cannot complete full-view monitoring and is greatly affected by light sources; It is not possible to combine depth shooting with full-angle shooting. At present, the technology companies represented by Green Eyes are focusing on developing three types of cameras to solve these problems:

1.Depth Video

It generates a Depth Video based on the principle of visual overlap of predator eyes, using structured light, and then analyzes human behavior based on the data. Even if there is occlusion, it can still accurately determine everyone's movement trajectory and behavior according to the overhead radar chart.

2.Light field camera

The compound eyes of the insect can capture the light source to the greatest extent and transmit it to the nerves to form a response quickly. At the same time, it can complete the 360-degree observation of the external environment without dead angles. The camera uses the lens or sensor array to receive the largest light field according to this imaging principle, and distributes the light field and calculates the result.

3.Eye camera

The principle of the human eye camera is actually similar to that of mammalian eyeballs. The human eye contains a macula, which concentrates 75% of the effective pixels of the human eye, and the remaining 25% is distributed at a regular angle of 160 degrees. According to this feature, the human eye camera works together with a detection system with a wide field of view angle but insufficient resolution and a macular system with a small field of view angle but high resolution to complete the shooting task.

2.2 access control

The access control system, also known as the access control system, is an intelligent system that controls access to the entrance. It can be summarized in general terms: the management of people's access rights-that is, who is managed and when and which doors can be accessed at any time. The initial form of the access control system is a mechanical door lock, but with the external environment, such as the situation, location, level of authority, subdivision of work processes, and increased human flow, the management of keys has become more difficult. The emergence of electronic card locks (magnetic cards and radio frequency cards), electronic password locks, etc. has improved people's ability to manage entrances and exits to a certain extent, and channel management has entered the electronic era. However, with the continuous application of these two types of electronic locks, their own advantages and disadvantages gradually appear.

Intelligent access control system is one of the earliest application carriers of face recognition in the commercial field. Among them, the advantages of face recognition access control system over other biometric access control systems are:

1.Nature

Naturalness refers to the manner in which face data is collected and compared using a camera device, and includes both voice recognition and body shape recognition, while fingerprint recognition and iris recognition do not have the feature of naturalness.

2.Not mandatory

Non-mandatory means that the identified object can complete the recognition task without active cooperation during the recognition process. For example, face recognition uses visible light to obtain face image information for comparison. Different from fingerprint recognition or iris recognition, it is necessary to use an electronic pressure sensor to collect fingerprints, or use infrared rays to collect iris images. And the face recognition access control system is more secure and efficient.

3. Non-contact

Compared with other biometric technologies, face recognition is non-contact, users do not need to contact the device directly, and can support high-concurrency processing. At the same time, they can satisfy the sorting, judgment and recognition of multiple faces in actual application scenarios. .

4.Intelligent data output

Face recognition access control can collect facial data for everyone in the scene, help managers to establish easy to retrieve, contrast management files, and provide the necessary basis for improving management decisions.

2.3 Smart Welcome

Enterprise intelligent welcome system. The intelligent welcome system is based on dynamic capture and non-cooperative face recognition technology, in conjunction with the access control, and combined with the background management system to achieve fast and accurate identification of guest identity, guest invitations, visitor registration, visitor data statistics, visitor query and other functions, so that Corporate offices are more intelligent.

Show welcome sign-in system. The sign-in form of the exhibition has evolved from the original paper identity card to various forms such as bar codes, magnetic cards, IC (smart) cards, mobile phone QR codes, etc. However, these methods have only realized paperless management of meeting sign-in, although to a certain extent Improved efficiency. However, there are still many hidden dangers, such as the loss of information security and impersonation. The show welcome sign-in system is also based on face recognition technology. The organizer only needs to collect the facial images of the participants (collection or search online) and enter them into the system portrait database. Visitors only need to pass within the camera range, and the camera equipment can quickly capture and identify, and further allocate seats to complete the sign-in task. At the same time, the system can also be linked with other devices, such as label printers, guest face search, etc.

2.4 New retail represented by AmazonGo

Face recognition technology is the key to the new retail application scenario represented by AmazonGo. When a consumer enters the store, the camera device first performs face recognition on it and sends it back to the Amazon user portrait database for comparison. Each product in the mall is equipped with a gravity sensor and sensor, which captures consumer behavior and gestures through a camera device, and accurately recognizes the information of the product it picks up. At the same time, it collects images after the consumer leaves the shelf and transmits them to the information center . The computer uses the comparison result and the gravity sensing information transmitted by each product to jointly determine whether the consumer is shopping or not, without delay. The microphone in the store can determine the location of the consumer based on the ambient sound, thereby assisting the camera device to locate and track the person. When leaving the store, the scanner can scan and record the goods purchased by the consumer, perform secondary confirmation at the same time, and complete the deduction task in its consumer account.

2.5 Witnesses

The essence of face recognition and witness combination products is a new application that combines identity card identification and face recognition technology. It retains the part of "machine-readable" documents in traditional offline audits, and uses face recognition technology to hold them. The cardholder performs on-site face collection and cross-checks with the image information in the ID card to complete the verification process. According to the structure of the system framework, intelligent authentication will be divided into network comparison mode and terminal comparison mode. At the same time, according to the different application scenarios, the device form is divided into a witness and a gate machine and a certification and a machine.

Profit model of face recognition

Third, the profit model of face recognition

At present, face recognition technology companies mainly rely on enterprise-level technical services and software and hardware sales as a profit model. Domestic deep learning technology companies represented by Shangtang Technology and Megishi Technology mainly cooperate with leading companies in various industries based on the B2B2C business model to jointly promote the application and realization of face recognition in various industries. Generally, this type of company outputs technical capabilities to the B side, and uses the model of sharing, charging by license, and charging by the number of uses of the technology to bind the business growth of the B side to serve the C side customers, thereby opening up the market with the help of B's industry resources.

4.Face recognition has been commercialized in the core business of many listed companies

In 2014, the DeepID deep learning model developed by the Computer Vision Research Group led by Professor Tang Xiaoou of the Chinese University of Hong Kong achieved a recognition rate of 99.15% on the LFW (Labeled Faces in the Wild) database for the first time. The recognition rate on the LFW is 97.52%). It has been three years since the major breakthrough in face recognition technology.

In 2017, after three years of technology dividend digestion and diffusion, face recognition has been successfully applied to multiple industries represented by the security industry, and has entered the core business and core product systems of multiple listed companies. Not only has it greatly improved the operational efficiency of the original business of each company, but it has also created multiple incremental markets with broad prospects. Taking the security industry as an example, there are three main business models:

First, to provide security vendors with integrated software and hardware forms such as video structuring, face deployment, face search, vehicle recognition, and crowd analysis. Provide integrated solutions such as custom GPUs and smart cameras, and charge according to the number of monitored video channels. ;

Second, integrate the face recognition function into identity recognition products, gates, counter products, etc., and make products profitable through industry channels;

Third, cooperate with traditional security vendors to jointly win large-scale projects such as safe cities and smart cities through their channels, and make profits through project forms.

We believe that in the context of artificial intelligence rising to the national strategic level, face recognition will greatly expand.

七乐彩 天天电玩城 欢乐拼三张 手机电玩平台