WO2022126464A1

WO2022126464A1 - Face recognition method, robot, and storage medium

Info

Publication number: WO2022126464A1
Application number: PCT/CN2020/137065
Authority: WO
Inventors: 曾钰胜; 程骏; 庞建新
Original assignee: 深圳市优必选科技股份有限公司
Priority date: 2020-12-17
Filing date: 2020-12-17
Publication date: 2022-06-23

Abstract

Disclosed in the present application are a face recognition method, a robot, and a storage medium. The face recognition method comprises: acquiring a face image to be recognized; performing face key point recognition on said face image by using a pre-trained face key point model, the face key point model being a lightweight neural network model; aligning said face image according to the face key point obtained by recognition to obtain an aligned face image; extracting a face feature in the aligned face image; and comparing the face feature with a registered face feature in a preset database to recognize a person in the face image to be recognized, the registered face feature being obtained by performing feature extraction on a registered front face image, the registered front face image being obtained by screening by a pre-trained face pose detection model. Embodiments of the present application achieve the effect of accurately completing face recognition under the condition of limited device end computing power.

Description

A face recognition method, robot and storage medium

technical field

The present application relates to the technical field of face recognition, and in particular, to a face recognition method, a robot and a storage medium.

Background technique

Face recognition is a biometric recognition technology based on human facial feature information. With the rapid development of science and technology, face recognition technology is increasingly Security equipment and robots, etc., based on this, the demand for technological innovation of face recognition technology is becoming more and more urgent.

Face key point recognition is an important part of face recognition algorithms and applications. The accuracy of key point recognition will directly affect the accuracy of face recognition. Existing to improve the accuracy of face key point recognition and the accuracy of subsequent face recognition often need to set up a relatively complex network model.

Due to the limited computing power on the robot side, complex models are difficult to use on the robot side. Therefore, there is an urgent need for a face recognition method that can be used on the robot side.

Application content

Based on this, it is necessary to propose a face recognition method, device, computer equipment and storage medium that can accurately complete face recognition under the condition of limited computing power on the device side.

In a first aspect, an embodiment of the present application provides a face recognition method, the method comprising:

Obtain the face image to be recognized;

Use a pre-trained face key point model to perform face key point recognition on the to-be-recognized face image, and the face key point model is a lightweight neural network model;

Align the to-be-recognized face images according to the identified face key points to obtain aligned face images;

extracting face features in the aligned face image;

Compare the facial features with the registered facial features in the preset database to identify the characters in the to-be-recognized facial images, and the registered facial features are characterized by performing a feature on the registered frontal image. Extracted and obtained, the registered frontal face image is obtained by screening through a pre-trained face pose detection model.

In a second aspect, an embodiment of the present application provides a computer device, including a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor is caused to perform the following steps:

Obtain the face image to be recognized;

extracting face features in the aligned face image;

Comparing the facial features with the registered facial features in the preset database to identify the characters in the to-be-recognized facial images, the registered facial features are characterized by characterizing the registered frontal face images. Extracted and obtained, the registered frontal face image is obtained by screening through a pre-trained face pose detection model.

In a third aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, causes the processor to perform the following steps:

Obtain the face image to be recognized;

extracting face features in the aligned face image;

Implementing the embodiments of the present application will have the following beneficial effects:

The embodiment of the present application uses a pre-trained face key point model to perform face key point recognition on the to-be-recognized face image, wherein the face key point model adopts a lightweight neural network model, which reduces computing power requirements, and The registered face features in the preset database are used for comparison to ensure the accuracy of recognition. The registered face features are accurate face features obtained through multiple models, which solves the problem of face obtained due to insufficient computing power on the device side. The problem that the key points are not accurate enough has achieved the effect of accurately completing face recognition under the condition of limited computing power on the device side.

Description of drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only These are some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without any creative effort.

in:

1 is an application environment diagram of a face recognition method in an embodiment of the present application;

2 is a flowchart of a face recognition method in an embodiment of the application;

3 is a flowchart of obtaining a face image to be recognized in a face recognition method according to an embodiment of the present application;

4 is a flowchart of a face recognition method in another embodiment of the present application;

5 is a flowchart of obtaining a registered face image in a face recognition method according to an embodiment of the present application;

6 is a structural block diagram of a face recognition device in an embodiment of the present application;

FIG. 7 is a structural block diagram of a robot in an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

FIG. 1 is an application environment diagram of a face recognition method in one embodiment. Referring to Fig. 1, the face recognition method is applied to a face recognition device. The face recognition device includes a terminal 110 and a server 120 . The terminal 110 and the server 120 are connected through a network. The terminal 110 may be a desktop terminal or a mobile terminal, and the mobile terminal may be at least one of a mobile phone, a tablet computer, a notebook computer, and a robot. The server 120 can be implemented by an independent server or a server cluster composed of multiple servers. The terminal 110 is used to obtain the face image to be recognized and upload it to the server 120, and the server 120 is used to receive the face image to be recognized, and use a pre-trained face key point model to perform face key point recognition on the to-be-recognized face image. , the face key point model is a lightweight neural network model; according to the identified face key points, the to-be-recognized face image is aligned to obtain an aligned face image; the aligned face image is extracted The facial features in the image; compare the facial features with the registered facial features in the preset database to identify the characters in the to-be-recognized facial images, and the registered facial features are obtained by comparing the registered facial features. The registered frontal face image is obtained by feature extraction, and the registered frontal face image is obtained by screening through a pre-trained face pose detection model.

In another embodiment, the above-mentioned text entity recognition method can be directly applied to the terminal 110, and the terminal 110 is used to obtain the face image to be recognized; a pre-trained face key point model is used to perform human analysis on the face image to be recognized. face key point recognition, the face key point model is a lightweight neural network model; according to the identified face key points, the to-be-recognized face image is aligned to obtain an aligned face image; Aligning the facial features in the face image; comparing the facial features with the registered facial features in the preset database to identify the person in the to-be-recognized facial image, the registered facial features It is obtained by performing feature extraction on the registered frontal face image, and the registered frontal face image is obtained by screening through a pre-trained face pose detection model.

As shown in FIG. 2, in one embodiment, a face recognition method is provided. The method can be applied to both a terminal and a server, and this embodiment is described by taking the application to a terminal as an example. The face recognition method specifically includes the following steps:

S110. Acquire a face image to be recognized.

In this embodiment, the terminal that executes the method is a robot, and the robot includes a camera. When the user needs to perform face recognition, the user can place the face in front of the camera of the robot, so that the robot can shoot the face through the camera, and the obtained image as the face image to be recognized.

Preferably, as shown in FIG. 3 , step S110 may specifically include the following steps:

S111. Acquire an image to be recognized.

S112. Detect and extract the face in the to-be-recognized image to obtain the to-be-recognized face image.

In this embodiment, the image captured by the robot is used as the image to be recognized. In order to ensure that there is a face in the image to be recognized during face recognition, it is also necessary to detect the face in the image to be recognized, so as to extract the image to be recognized. Only the image of the face part is included as the face image to be recognized, thereby improving the accuracy of face recognition and improving the efficiency of face recognition.

S120 uses a pre-trained face key point model to perform face key point recognition on the to-be-recognized face image, and the face key point model is a lightweight neural network model.

S130. Align the face images to be identified according to the identified face key points to obtain an aligned face image.

In this embodiment, after the face image to be recognized is obtained, the face key point recognition can be performed on the face image to be recognized, wherein the face key point recognition model can use a pre-trained face key point detection model, and the face key point detection model to be recognized After the face image is input to the face key point detection model, the face key point detection model can output the face key points of the face image to be recognized. After obtaining the face key points, the face image to be recognized can be aligned according to the identified face key points, thereby obtaining an aligned face image. The aligned face image is to convert the face image to be recognized based on the face key points. In order to obtain the frontal face image, the aligned face image can be used for the feature comparison of face recognition, so as to ensure the accuracy of face recognition. Among them, the face key point detection model can be specifically the neural network model of MobileNetV3_small. Specifically, MobileNetV3_small is a lightweight version of MobileNetV3 model, which has good accuracy under the condition of excellent speed performance, and the required computing power Lower, and the above MobileNetV3 combines the advantages of MobileNetV1 and MobileNetV2, and is a strong mobile-side feature extraction model. As an option, even if the computing power on the device is still limited, such as a small robot, the MobileNetV3_small model can be channel-cut to get MobileNetV3_smallx0.5, which has better performance even with a little loss of accuracy. face recognition effect.

S140, extracting face features in the aligned face image.

S150. Compare the face feature with the registered face feature in a preset database to identify the person in the to-be-recognized face image.

In this embodiment, after the aligned face image is obtained, the face feature in the aligned face image can be extracted, and then the face feature is compared with the registered face feature in the preset database, so as to identify The person in the face image to be recognized. Among them, the registered face features are obtained by feature extraction on the registered frontal face images, and the registered frontal face images are obtained by screening through a pre-trained face pose detection model. Specifically, before performing face recognition, to register the face of the user to be recognized, the same method as the above steps can be used to obtain the key point recognition of the registered user's face, and then extract the registered face features. The recognition accuracy also needs to be screened by a pre-trained face pose detection model to obtain a frontal face image, and the registered face features are obtained by feature extraction from the registered frontal face image. Each registered face feature includes the corresponding user. name. In this way, the registered face features of the frontal face images of all registered users are stored in the preset database, and when face recognition is performed, the person in the face image to be recognized can be easily identified by comparison.

As shown in FIG. 4 , in another embodiment, before step S110 of the face recognition method in the above embodiment, the method further includes:

S210. Acquire a registered face image.

In this embodiment, before performing face recognition, in order to register the face of the user to be recognized, it is first necessary to obtain a registered face image. The registered face image may be captured by a robot in real time, or uploaded by a user through a server. , which is used for the face images registered by the user in the preset database for comparison in subsequent face recognition.

Preferably, as shown in FIG. 5 , step S210 may specifically include the following steps:

S211. Acquire multiple face images to be registered.

S212. Detecting the faces in the multiple face images to be registered to obtain a registered face image, where the registered face image is to be registered with the highest facial feature value in the multiple face images to be registered face image.

In this embodiment, since the registered face image of the registered user is obtained, a plurality of face images to be registered can be obtained, and the best image among them can be selected as the registered face image. After obtaining multiple face images to be registered, firstly, the faces in the multiple face images to be registered can be detected to obtain the registered face images, so as to ensure that there are faces in the registered face images during registration. In addition, because There are multiple face images to be registered. After screening out the face images to be registered that do not have faces, the face image to be registered with the highest face feature value needs to be selected as the registered face image, that is, the registered face images are multiple. A face image to be registered with the highest face feature value among the face images to be registered, wherein the highest face feature value can be represented as the image size of the face in the multiple face images to be registered is the largest. Exemplarily, The image size of the face in the first face image to be registered is 100×100, and the image size of the face in the second face image to be registered is 10×10, then the face feature value in the first face image to be registered is the highest, The first face image to be registered is used as the registered face image.

S220, taking the registered face image as the input of the pre-trained face recognition model, the face recognition model includes a face key point detection model and a face gesture detection model, and the face key point detection model is the same as the The face gesture detection model is connected.

S230. Obtain the face pose output by the face pose detection model, and obtain the registered face key points output by the face key point detection model.

S240. When the face posture is a frontal face image, align the registered face images according to the registered face key points to obtain an aligned registered face.

S250, extracting the facial features in the aligned registered faces as registered facial features and storing them in a preset database.

In this embodiment, after obtaining the registered face image, the registered face image can be used as the input of the face recognition model to obtain the face posture and the key points of the registered face, wherein the face recognition model is pre-trained, The face recognition model includes a pre-trained face key point detection model and a pre-trained face pose detection model. The face key point detection model is connected with the face pose detection model. The face key point detection model is also used for Face key point extraction in face recognition, and face pose detection model can be simply constructed according to actual needs. Specifically, the face key point detection model is connected to the face pose detection model through the middle feature layer of the face key point detection model. Exemplarily, the middle feature layer in the face key point detection model is selected, and the middle feature layer is selected from the middle feature layer. After 2 convolutional layers with a stride of 2, it is connected to the face pose detection model to complete the decoupling effect on the extraction of face key points, which is used as the input of the face pose detection model, so as to realize the use of face key points to assist people. The prediction of face pose is a joint model obtained by connecting the intermediate feature layer of the face key point detection model with the face pose detection model, that is, the face recognition model has a good balance, and the prediction of face pose is very important for extracting faces. The dependency and calculation amount of key points are not high, and the problem of large errors in face pose prediction based on 3D standard modules is avoided.

It should be noted that the face recognition model is trained based on a preset loss function, and the preset loss function used by the face key point detection model can be the WingLoss loss function, specifically:

where C is a constant, ω is a positive number used to limit the range of the nonlinear part of the loss function to the interval [-ω,ω], and ε is the curvature that constrains the nonlinear region of the loss function, All of the above parameters can be selected according to actual requirements. For example, ω=10 and ε=2 are selected. Compared with the L1 loss function, the L2 loss function and the L1_smooth loss function, the WingLoss loss function has a higher response to subtle differences, and can achieve more refined positioning of the face key points during the training of the face key point detection model. The preset loss function used by the face pose detection model can be the L1_smooth loss function.

In this embodiment, the face key points include eye key points and other face key points, other face key points may be nose key points, mouth key points, and ear key points, and eye key points include left eye key points and The key point of the right eye, however, due to the difference in the size of the eyes, the diversification of the degree of eye opening and closing, and the complex situation of wearing glasses, the positioning of the key points of the eye is prone to large errors. The loss coefficient of the partial key points is greater than the loss coefficient of the other key points. Exemplarily, the loss coefficient of the eye key points is 3, and the loss coefficient of other face key points is 1, thereby increasing the weight of the eye key points. , which helps the eye key points to be positioned more accurately without affecting other key points.

Further, after the registered face image is input into the face recognition model, the face key point detection model in the face recognition model will output the registered face key points, while the face pose detection model in the face recognition model is different from the human face. The face key point detection model is connected. The face pose detection model will take the input registered face image and the face key point feature output from the middle feature layer of the face key point detection model as input, and output the face pose. The face pose realizes In order to realize the decoupling of the key points of the face through the auxiliary prediction of the key points of the face without affecting the accuracy of the key points of the face. After the face pose is obtained, it can be determined whether the registered face image is a frontal image according to the face pose. When the face pose is a frontal image, the registered face image can be aligned according to the key points of the registered face to obtain Align registered faces. Finally, the face features in the aligned registered faces are extracted as registered face features and stored in a preset database for use in subsequent face recognition.

In an alternative embodiment, acquiring multiple face images to be registered includes: acquiring multiple registered face images corresponding to the same registered user whose eyes are in different states. The registered face key points include registered eye key points and registered other face key points, and the acquiring the registered face key points output by the face key point detection model further includes: extracting the face key point detection model by using the face key point detection model. For the eye feature points in the multiple registered face pictures, the obtained eye feature points in different states are used as the registered eye key points of the same registered user. In this embodiment, in order to further reduce the error of the key points of the eyes, when acquiring multiple face images to be registered, multiple registered face images corresponding to the same registered user whose eyes are in different states are acquired. When the face key point detection model outputs the registered face key points, the face key point detection model can also be used to extract the eye feature points in multiple registered face pictures, and the obtained eye feature points in different states are used as the same. A registered user's registered eye key points, so as to improve the registration data of eye key points, and reduce the error in face recognition from the way of data source.

In the embodiment of the present application, the registered face image is used as the input of a pre-trained face recognition model, and the face recognition model includes a face key point detection model and a face gesture detection model. The face key point The detection model is connected with the face pose detection model, and the loss function weight and data source of the eye key points are added, which not only solves the problem of insufficient robustness of face pose prediction, but also solves the positioning of the eye key points. The problem of large errors is prone to occur, and the joint estimation of face key points and face pose is completed without losing the accuracy of face key points, and the effect of improving the accuracy of eye key point recognition is achieved.

As shown in FIG. 6 , in one embodiment, a face recognition device is provided. The face recognition device provided in this embodiment can execute the face recognition method provided by any embodiment of the present application, and has the corresponding method for executing the method. Functional modules and beneficial effects. The face recognition device includes an image acquisition module 100 , a key point extraction module 200 , a face alignment module 300 , a feature extraction module 400 and a face recognition module 500 .

Specifically, the image acquisition module 100 is used to acquire a face image to be recognized; the key point extraction module 200 is used to use a pre-trained face key point model to perform face key point recognition on the to-be-recognized face image, and the The face key point model is a lightweight neural network model; the face alignment module 300 is used to align the to-be-recognized face images according to the face key points to obtain an aligned face image; the feature extraction module 400 is used to extract out the facial features in the aligned face images; the face recognition module 500 is used to compare the facial features with the registered facial features in the preset database to identify the to-be-recognized face images characters in.

In one embodiment, the above-mentioned apparatus further includes a face registration module 600, and the face registration module 600 is used to obtain a registered face image; the registered face image is used as the input of the pre-trained face recognition model. The face recognition model includes a face key point detection model and a face gesture detection model, and the face key point detection model is connected with the face gesture detection model; obtain the face gesture output by the face gesture detection model , and obtain the registered face key points output by the face key point detection model; when the face pose is a frontal face image, align the registered face images according to the registered face key points to obtain Aligning the registered faces; extracting the facial features in the aligned registered faces as the registered face features and storing them in a preset database.

In one embodiment, the face key point detection model is connected to the face gesture detection model through an intermediate feature layer of the face key point detection model, and the face gesture detection model is used for registration according to the input The face image and the face key point feature determine whether the registered face image is a frontal face image, and the face key point feature is the output of the intermediate feature layer.

In one embodiment, the face registration module 600 is further configured to obtain a plurality of face images to be registered; the faces in the plurality of face images to be registered are detected to obtain a registered face image, and the registration The face image is the face image to be registered with the highest face feature value among the plurality of face images to be registered.

In one embodiment, the face registration module 600 is further configured to acquire multiple registered face pictures corresponding to the same registered user whose eyes are in different states; the registered face key points include registered eye key points and registered face key points. Other face key points, the face registration module 600 is also used to extract the eye feature points in the multiple registered face pictures by using the face key point detection model, and the obtained eye features in different states point as the registered eye key point of the same registered user.

In one embodiment, the face key points include eye key points and other face key points, and the face recognition model is obtained by training based on a preset loss function. The loss coefficient of the partial key point is larger than the loss coefficient of the other face key points.

In one embodiment, the image acquisition module 100 is also used to acquire an image to be recognized; the face in the image to be recognized is detected and extracted to obtain an image of the face to be recognized.

Figure 7 shows an internal structure diagram of the robot in one embodiment. As shown in Figure 7, the robot includes a processor, memory and network interface connected through a system bus. Wherein, the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the robot stores an operating system, and also stores a computer program. When the computer program is executed by the processor, the processor can implement the face recognition method. A computer program may also be stored in the internal memory, and when the computer program is executed by the processor, the processor may execute the age identification method. Those skilled in the art can understand that the structure shown in FIG. 7 is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the robot to which the solution of the present application is applied. More or fewer components are shown in the figures, either in combination or with different arrangements of components.

In one embodiment, a robot is proposed, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the following steps:

Obtain the face image to be recognized;

extracting face features in the aligned face image;

In one embodiment, before the acquiring the face image to be recognized includes:

Obtain a registered face image; take the registered face image as the input of a pre-trained face recognition model, the face recognition model includes a face key point detection model and a face gesture detection model, and the face key The point detection model is connected with the face gesture detection model; the face gesture output by the face gesture detection model is obtained, and the registered face key points output by the face key point detection model are obtained; When the posture is a frontal face image, the registered face image is aligned according to the key points of the registered face to obtain an aligned registered face; the facial features in the aligned registered faces are extracted as the registered face features and Stored in the preset database.

In one embodiment, the obtaining of the registered face images includes: obtaining a plurality of face images to be registered; detecting faces in the plurality of face images to be registered to obtain a registered face image, and the registration The face image is the face image to be registered with the highest face feature value among the plurality of face images to be registered.

In one embodiment, the acquiring multiple face images to be registered includes: acquiring multiple registered face images corresponding to the same registered user whose eyes are in different states; the registered face key points include the registered eye key and registering other face key points, the obtaining the registered face key points output by the face key point detection model further includes: using the face key point detection model to extract the Eye feature points, the obtained eye feature points in different states are used as the registered eye key points of the same registered user.

In one embodiment, the acquiring an image of a face to be recognized includes: acquiring an image to be recognized; and detecting and extracting a human face in the image to be recognized to obtain an image of the face to be recognized.

In one embodiment, a computer-readable storage medium is provided, which stores a computer program, and when the computer program is executed by a processor, causes the processor to perform the following steps:

Obtain the face image to be recognized;

extracting face features in the aligned face image;

Obtain a registered face image; take the registered face image as the input of a pre-trained face recognition model, the face recognition model includes a face key point detection model and a face gesture detection model, and the face key The point detection model is connected with the face gesture detection model; the face gesture output by the face gesture detection model is obtained, and the registered face key points output by the face key point detection model are obtained; when the face When the posture is a frontal face image, the registered face image is aligned according to the key points of the registered face to obtain an aligned registered face; the facial features in the aligned registered faces are extracted as the registered face features and Stored in the preset database.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium , when the program is executed, it may include the flow of the above-mentioned method embodiments. Wherein, any reference to memory, storage, database or other medium used in the various embodiments provided in this application may include non-volatile and/or volatile memory. Nonvolatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Road (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

The technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features It is considered to be the range described in this specification.

The above-mentioned embodiments only represent several embodiments of the present application, and the descriptions thereof are relatively specific and detailed, but should not be construed as a limitation on the scope of the patent of the present application. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which all belong to the protection scope of the present application. Therefore, the scope of protection of the patent of the present application shall be subject to the appended claims. Please enter the specific implementation content section.

Claims

A face recognition method, characterized in that the method comprises:

Obtain the face image to be recognized;

Use a pre-trained face key point model to perform face key point recognition on the to-be-recognized face image, and the face key point model is a lightweight neural network model;

Align the to-be-recognized face images according to the identified face key points to obtain aligned face images;

extracting face features in the aligned face image;

Comparing the facial features with the registered facial features in the preset database to identify the characters in the to-be-recognized facial images, the registered facial features are characterized by characterizing the registered frontal face images. Extracted and obtained, the registered frontal face image is obtained by screening through a pre-trained face pose detection model.
The method according to claim 1, wherein before the acquiring the face image to be recognized comprises:

Get the registered face image;

Taking the registered face image as the input of the pre-trained face recognition model, the face recognition model includes a face key point detection model and a face gesture detection model, and the face key point detection model is the same as the Face pose detection model connection;

obtaining the face pose output by the face pose detection model, and obtaining the registered face key points output by the face key point detection model;

When the face posture is a frontal face image, aligning the registered face image according to the registered face key points to obtain an aligned registered face;

The facial features in the aligned registered faces are extracted as registered facial features and stored in a preset database.
The method according to claim 2, wherein the face key point detection model is connected to the face gesture detection model through an intermediate feature layer of the face key point detection model, and the face gesture detection model The model is used to determine whether the registered face image is a frontal face image according to the input registered face image and face key point feature, and the face key point feature is the output of the intermediate feature layer.
The method according to claim 2, wherein the acquiring a registered face image comprises:

Obtain multiple face images to be registered;

Detecting the faces in the multiple face images to be registered to obtain a registered face image, the registered face image is the face to be registered with the highest facial feature value in the multiple face images to be registered image.
The method according to claim 4, wherein the acquiring a plurality of face images to be registered comprises:

Obtain multiple registered face pictures corresponding to the same registered user whose eyes are in different states;

The registered face key points include registered eye key points and registered other face key points, and the acquisition of the registered face key points output by the face key point detection model further includes:

The eye feature points in the multiple registered face pictures are extracted by using the face key point detection model, and the obtained eye feature points in different states are used as the registered eye key points of the same registered user.
The method according to claim 2, wherein the face key points include eye key points and other face key points, and the face key point detection model is obtained by training based on a preset loss function. The loss coefficient of the eye key point in the preset loss function is greater than the loss coefficient of the other face key points.
The method according to claim 1, wherein the acquiring the face image to be recognized comprises:

Get the image to be recognized;

Detecting and extracting the face in the to-be-recognized image to obtain the to-be-recognized face image.
A robot includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the following steps:

Obtain the face image to be recognized;

Use a pre-trained face key point model to perform face key point recognition on the to-be-recognized face image, and the face key point model is a lightweight neural network model;

Align the to-be-recognized face images according to the identified face key points to obtain aligned face images;

extracting face features in the aligned face image;

Comparing the facial features with the registered facial features in the preset database to identify the characters in the to-be-recognized facial images, the registered facial features are characterized by characterizing the registered frontal face images. Extracted and obtained, the registered frontal face image is obtained by screening through a pre-trained face pose detection model.
The robot according to claim 8, wherein before the acquiring the face image to be recognized comprises:

Get the registered face image;

Taking the registered face image as the input of the pre-trained face recognition model, the face recognition model includes a face key point detection model and a face gesture detection model, and the face key point detection model is the same as the Face pose detection model connection;

obtaining the face pose output by the face pose detection model, and obtaining the registered face key points output by the face key point detection model;

When the face posture is a frontal face image, aligning the registered face image according to the registered face key points to obtain an aligned registered face;

The facial features in the aligned registered faces are extracted as registered facial features and stored in a preset database.
The robot according to claim 9, wherein the face key point detection model is connected to the face gesture detection model through an intermediate feature layer of the face key point detection model, and the face gesture detection model The model is used to determine whether the registered face image is a frontal face image according to the input registered face image and face key point feature, and the face key point feature is the output of the intermediate feature layer.
The robot according to claim 9, wherein the acquiring a registered face image comprises:

Obtain multiple face images to be registered;

Detecting the faces in the multiple face images to be registered to obtain a registered face image, the registered face image is the face to be registered with the highest facial feature value in the multiple face images to be registered image.
The robot according to claim 11, wherein the acquiring a plurality of face images to be registered comprises:

Obtain multiple registered face pictures corresponding to the same registered user whose eyes are in different states;

The registered face key points include registered eye key points and registered other face key points, and the acquisition of the registered face key points output by the face key point detection model further includes:

The eye feature points in the multiple registered face pictures are extracted by using the face key point detection model, and the obtained eye feature points in different states are used as the registered eye key points of the same registered user.
The robot according to claim 9, wherein the face key points include eye key points and other face key points, and the face key point detection model is obtained by training based on a preset loss function. The loss coefficient of the eye key point in the preset loss function is greater than the loss coefficient of the other face key points.
The robot according to claim 8, wherein the acquiring the face image to be recognized comprises:

Get the image to be recognized;

Detecting and extracting the face in the to-be-recognized image to obtain the to-be-recognized face image.
A computer-readable storage medium storing a computer program, when executed by a processor, the computer program causes the processor to perform the following steps:

Obtain the face image to be recognized;

Use a pre-trained face key point model to perform face key point recognition on the to-be-recognized face image, and the face key point model is a lightweight neural network model;

Align the to-be-recognized face images according to the identified face key points to obtain aligned face images;

extracting face features in the aligned face image;

Comparing the facial features with the registered facial features in the preset database to identify the characters in the to-be-recognized facial images, the registered facial features are characterized by characterizing the registered frontal face images. Extracted and obtained, the registered frontal face image is obtained by screening through a pre-trained face pose detection model.
The storage medium according to claim 15, wherein before the acquiring the face image to be recognized comprises:

Get the registered face image;

Taking the registered face image as the input of the pre-trained face recognition model, the face recognition model includes a face key point detection model and a face gesture detection model, and the face key point detection model is the same as the Face pose detection model connection;

obtaining the face pose output by the face pose detection model, and obtaining the registered face key points output by the face key point detection model;

When the face posture is a frontal face image, aligning the registered face image according to the registered face key points to obtain an aligned registered face;

The facial features in the aligned registered faces are extracted as registered facial features and stored in a preset database.
The storage medium according to claim 16, wherein the face key point detection model is connected to the face gesture detection model through an intermediate feature layer of the face key point detection model, and the face gesture The detection model is used to determine whether the registered face image is a frontal face image according to the input registered face image and face key point feature, and the face key point feature is the output of the intermediate feature layer.
The storage medium according to claim 16, wherein the acquiring a registered face image comprises:

Obtain multiple face images to be registered;

Detecting the faces in the multiple face images to be registered to obtain a registered face image, the registered face image is the face to be registered with the highest facial feature value in the multiple face images to be registered image.
The storage medium according to claim 18, wherein the acquiring a plurality of face images to be registered comprises:

Obtain multiple registered face pictures corresponding to the same registered user whose eyes are in different states;

The registered face key points include registered eye key points and registered other face key points, and the acquisition of the registered face key points output by the face key point detection model further includes:

The eye feature points in the multiple registered face pictures are extracted by using the face key point detection model, and the obtained eye feature points in different states are used as the registered eye key points of the same registered user.
The storage medium according to claim 16, wherein the face key points include eye key points and other face key points, and the face key point detection model is obtained by training based on a preset loss function, In the preset loss function, the loss coefficient of the eye key point is greater than the loss coefficient of the other face key points.
The storage medium according to claim 15, wherein the acquiring the face image to be recognized comprises:

Get the image to be recognized;

Detecting and extracting the face in the to-be-recognized image to obtain the to-be-recognized face image.