Access control systems can be divided into: electronic password access control system, card access control system, fingerprint access control system, finger vein access control system, palmprint access control system, iris access control system, voice recognition access control system and face recognition access control system from the technical dimension.
Face Recognition Technology
Face Recognition (FR) is a kind of biometric technology that carries out identification based on the information of human face features. It is a series of related technologies that use a camera or a camera to capture an image or a video stream containing a human face, and automatically detect and track the face in the image, and then perform face recognition on the detected face, which is also commonly called portrait recognition and facial recognition.
The face, like other biometric features of the human body (fingerprints, iris, etc.) is inherent, and its uniqueness and good characteristics of not being easily copied provide the necessary prerequisites for identity identification. In comparison with other types of biometric identification face recognition has the following characteristics.
The user does not need to specially cooperate with the face acquisition equipment, and can obtain the face image almost unconsciously, so that the sampling method is not "mandatory".
The user does not need direct contact with the device to acquire face images.
Classifying, judging and recognizing multiple faces can be used in practical application scenarios
In addition, it also meets the visual characteristics: the characteristic of "recognizing people by their appearance", as well as the characteristics of simple operation, intuitive result and good concealment.
The research of face recognition system started in 1960s, and was improved after 1980s with the development of computer technology and optical imaging technology, while it really entered the primary application stage in late 1990s; in recent years, with the progress of artificial intelligence technology mainly deep learning, face recognition technology has been developed rapidly. "Face recognition system" integrates artificial intelligence, machine recognition, machine learning, model theory, expert system, video image processing and other professional technologies, and is a comprehensive and strong system engineering technology.
Face recognition system usually includes several processes: face image acquisition and detection, key point extraction, face regularization (image processing), face feature extraction and face recognition comparison.
Face image acquisition.
Different face images can be captured through the camera lens, such as static images, dynamic images, different positions, different expressions and other aspects can be well captured. When the user is within the capture range of the capture device, the capture device will automatically search and capture the user's face image.
Face detection is mainly used in practice for pre-processing of face recognition, i.e., to accurately calibrate the position and size of a face in an image.
Key point extraction (feature extraction).
The features that can be used in face recognition systems are usually classified as visual features, pixel statistical features, face image transformation coefficient features, face image algebraic features, etc. Face feature extraction is what is done for certain features of the face. Face feature extraction, also known as face characterization, is the process of feature modeling of a human face. The methods of face feature extraction are summarized into two main categories.
One is knowledge-based representation methods.
In addition to this representation mentioned above, another one is based on algebraic features or on statistical
Face regularization (preprocessing).
Image preprocessing for faces is the process of processing images based on face detection results and ultimately serving for feature extraction. The original image acquired by the system is often not directly usable due to various conditions and random interference, and it must be subjected to image preprocessing such as grayscale correction and noise filtering are the earliest stages of image processing.. For face images, the preprocessing process mainly includes light compensation for face images, grayscale transformation, histogram equalization, normalization, and geometric correction.
Face recognition comparison (matching and recognition).
By setting the threshold value, the feature data of the face image is extracted and searched, compared and matched with the feature templates stored in the database. When the threshold value exceeds the similarity, the matching result is obtained.Face recognition is to compare the features of the face to be recognized with the obtained face feature template, and to judge the identity information of the face according to the degree of similarity. It can be divided into 1:1, 1:N, and attribute recognition. Among them, 1:1 is to compare the eigenvalue vectors corresponding to 2 faces, and 1:N is to compare the eigenvalue vector of 1 face photo with the eigenvalue vector corresponding to another N faces, and output the face with the highest similarity or the top X similarity ranking.
The advantage of face recognition is its naturalness and the feature of not being detected by the measured individual.
By naturalness, we mean that the recognition method is the same as the biometric features used by humans (or even other organisms) for individual recognition. For example, face recognition is used by humans to distinguish and confirm identity by comparing faces, while iris recognition, voice recognition, body shape recognition, etc. have natural characteristics.
Undetectability is also important for a recognition method, which makes it less offensive and less likely to be spoofed because it is less likely to attract attention. Unlike fingerprint recognition or iris recognition, which require fingerprint capture using finger contact sensors or iris image capture using infrared light, these special capture methods can easily be detected and cause inconvenience.
Face recognition is considered to be one of the most difficult research topics in the field of biometric recognition and even in the field of artificial intelligence. The difficulties of face recognition are mainly brought about by the characteristics of human face as a biometric feature.
There is little difference between different individuals, all faces have similar structures, and even the structural appearance of face organs are similar. Such a feature is advantageous for localization using faces, but disadvantageous for using faces to distinguish human individuals. For example, the phenomenon of twins refers to the birth of two individuals in a single pregnancy in fetal animals. Twins can generally be divided into two categories: identical twins and dizygotic twins. In human societies, the average birth rate of twins worldwide is 1:89. For the human twin phenomenon, some twins have facial differences, and some twins are even extremely similar in terms of facial features, making it very challenging for face recognition systems to distinguish each individual almost biometrically.
The appearance of human faces is very unstable, and people can produce many expressions through changes in their faces, and the visual images of faces vary greatly in different observation angles. In addition, face recognition is also affected by lighting conditions (e.g., day and night, indoor and outdoor, etc.), many coverings of faces (e.g., masks, sunglasses, hair, beards, etc.), age, and many other factors.
Ease of aggressiveness.
With the development of digital photography, video synthesis technology, etc., it is becoming easier and easier to obtain face information or synthetic face information of a specified person. Even more, with the development of deep learning techniques of adversarial training, computers can synthesize biometric information such as the face of any person with high accuracy. Some of the anti-face recognition trained by generating adversarial networks have 99.5% success rate of identity spoofing, and even become the nemesis of many face recognition systems.
Scenario-based classification and grading model for products
Face recognition access control system is based on the innovation of advanced face recognition technology applied in the field of access control entrances and exits. Compared to other technical means such as key access control, IC card access control, fingerprint access control, iris access control, etc., face access control if has the following advantages.
Face recognition cardless entry, eliminating the extra work of carrying cards.
High-speed accuracy, convenience and speed, face recognition is usually completed within 1 second or even senseless natural passage.
Photo records can be retroactively verified.
At the same time face access control also has some risks.
There is a certain risk of misidentification for highly similar faces or twins, etc.
face recognition is vulnerable to attack by some technical means.
Face recognition involves risks such as privacy.
Overall face recognition access control is a major technical change in access control systems, bringing access control intelligent passage into a new AI era.
The key content of this white paper is the product application scenario of face access control, product technology form is divided into three categories and six levels, establishing a technical guide to the application and development of face access control products, which is also a major innovation of this white paper. This classification method combined with the access control application scenario and face recognition technology in-depth research demand refinement and technical analysis, welcome readers to put forward valuable comments so that the editors can modify and improve.
The application scenario of access control passage is divided into three categories.
Strong fit class.
Face and equipment distance within 0.5 meters, face angle within 15 degrees, face comparison mode for 1:1 person card comparison or <1000 people of small face library comparison. Such as family face door lock, office small face access control, face time and attendance, etc.
The distance between the face and the device is between 0.5m-1.5m, the angle of the face is within 30 degrees, and the capacity of the face comparison library is within 10,000 people in the medium-sized face library. Such as intelligent building floor face access control, community unit face access control, etc.
Natural passage class.
The distance between the face and the device is between 1 meter and 3 meters, the angle of the face is within 45 degrees, and the capacity of the face matching library is within 10,000 to 100,000 people. Such as park entrance/exit face passage, public transportation face gates, etc.
Level 1: Academic level. Face library capacity of about a hundred people, accuracy of 60% or less, for academic analysis of new technologies.
Level 2: Entertainment level. The capacity of the face library is around 500 people, and the accuracy is between 60-85%, which is used for entertainment games
Level 3: consumer level. Face database capacity within 1000 people, accuracy between 85-95%, used in the personal field or small and medium-sized enterprises.
Level 4: Commercial level. Face database capacity between 1,000-10,000, accuracy between 95-99%, used for medium-sized enterprise applications.
Level 5: Industry level. Face database capacity between 10,000 and 100,000 people, with accuracy between 99-99.999%, for super large-scale enterprises or mass population applications such as public security and transportation.
Level 6: Financial level. The face database capacity is above 100,000 people, the accuracy is above 99.999%, and the error rate is less than 1 in 100,000, which can be used in financial payment and other application fields.
Summary: By subdividing applications by three categories and six levels, the application scenarios and product technical features of face access control products can be better distinguished to provide guidance for subsequent product applications and product development.