Data annotation method, device, system, equipment and storage medium for face recognition

Through the high-precision face recognition model, the problem of time-consuming and large errors in traditional manual labeling is solved, and efficient and accurate data labeling and model optimization are achieved.

CN114998954BActive Publication Date: 2025-08-22BEIJING MOMENTA TECH CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202210441189.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-04-25
Publication Date
2025-08-22
Estimated Expiration
2042-04-25

AI Technical Summary

Technical Problem

The traditional artificial ID classification method consumes huge time and errors in facial recognition data annotation, making it difficult to meet the needs of low precision.

Method used

The first face recognition model with high model accuracy is used to extract the feature values ​​of the face image to be marked, and similar face images are filtered from the registered face image, and automatically or semi-automatically annotated according to the similarity matching results, and identity annotated in combination with the recognition terminal.

Benefits of technology

This greatly reduces the cost of manual labeling, improves the accuracy and efficiency of data labeling, and optimizes the accuracy of the face recognition model through a closed-loop system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114998954B_ABST
    Figure CN114998954B_ABST
Patent Text Reader

Abstract

The present application discloses a data annotation method, apparatus, system, equipment and storage medium for face recognition, relating to the field of face recognition technology. The method comprises: obtaining a face image to be annotated, inputting the face image to be annotated into a pre-trained first face recognition model, extracting feature values ​​corresponding to the face image to be annotated, wherein the first face recognition model is a face recognition model whose model accuracy reaches a first preset value; based on the feature values ​​corresponding to the face image to be annotated, screening out similar face images corresponding to the face image to be annotated from a set of registered face images, wherein the set of registered face images includes registered face images containing user identity tags that are pre-entered into a system; annotating the face image to be annotated based on the similar face images to obtain an annotation result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of face recognition technology, and in particular to a data annotation method, apparatus, system, device and storage medium for face recognition. Background Art

[0002] Face recognition is based on human facial features. The face recognition model trained by neural networks further extracts the identity features contained in each face and compares them with known faces to identify the identity corresponding to each face.

[0003] In the facial recognition scenario of a fleet, the training of the facial recognition model requires the labeling of a large amount of real-time facial data. Specifically, the original data collection can be completed by taking facial images of the driver by vehicle, and the images with the same ID in the dataset are grouped together to achieve data labeling for the facial recognition model.

[0004] However, traditional labeling methods require labelers to classify all raw data by ID, which is extremely time-consuming and has certain errors in manual labeling, making it difficult for face recognition model training to meet low-precision requirements. Summary of the Invention

[0005] In view of this, the present application provides a data labeling method, device, system, equipment and storage medium for face recognition. The main purpose is to solve the problem that there are errors in the method of manual ID classification for face recognition data labeling in the existing technology, making it difficult for face recognition model training to meet low-precision requirements.

[0006] According to a first aspect of the present application, a data annotation method for face recognition is provided, comprising:

[0007] Obtaining a facial image to be annotated, and inputting the facial image to be annotated into a pre-trained first facial recognition model to extract feature values ​​corresponding to the facial image to be annotated, wherein the first facial recognition model is a facial recognition model whose model accuracy reaches a first preset value;

[0008] Based on the feature value corresponding to the facial image to be annotated, screening out similar facial images corresponding to the facial image to be annotated from a set of registered facial images, wherein the set of registered facial images includes registered facial images containing user identity tags that are pre-entered into the system;

[0009] The face image to be labeled is labeled according to the similar face image to obtain a labeling result.

[0010] Furthermore, the step of screening out similar facial images corresponding to the facial image to be labeled from a set of registered facial images based on the feature values ​​corresponding to the facial image to be labeled specifically includes:

[0011] Traversing the registered face images in the registered face image set, performing similarity matching between the feature values ​​corresponding to the face image to be annotated and the feature values ​​corresponding to the registered face images, and obtaining a similarity matching result;

[0012] According to the similarity rankings between the registered face image and the face image to be labeled in the similarity matching results, similar face images corresponding to the face image to be labeled are screened out from the registered face set.

[0013] Furthermore, the selecting, from the registered face set, similar face images corresponding to the face image to be labeled based on the similarity rankings between the registered face images and the face image to be labeled in the similarity matching results, includes:

[0014] According to the similarity rankings of the registered facial images and the facial images to be labeled in the similarity matching results, a registered facial image whose similarity ranking is before a preset value is selected as a similar facial image corresponding to the facial image to be labeled; or

[0015] According to the similarity rankings corresponding to the registered facial images and the facial images to be labeled in the similarity matching results, the registered facial image with the highest similarity ranking is selected as the similar facial image corresponding to the facial image to be labeled.

[0016] Furthermore, after selecting, based on the similarity rankings corresponding to the registered facial images and the facial images to be annotated in the similarity matching results, a registered facial image whose similarity ranking is before a preset value as a similar facial image corresponding to the facial image to be annotated, annotating the facial image to be annotated based on the similar facial images to obtain an annotation result, includes:

[0017] The similar face image corresponding to the face image to be labeled is output to the recognition terminal, so that the recognition terminal labels the user identity according to the similar face image and uses the labeled user identity label as the labeling result.

[0018] Furthermore, the method further comprises:

[0019] Key point detection and face alignment processing are performed on the face image to be labeled and the face images in the registered face image set.

[0020] According to a second aspect of the present application, a data annotation method for face recognition is provided, comprising:

[0021] Acquire a face image to be annotated through a data acquisition module, and send the face image to be annotated to a data annotation module;

[0022] The data labeling module receives the facial images to be labeled from the data acquisition module, labels the facial images to be labeled using the data labeling method of the first aspect mentioned above, and deploys the second face recognition model trained using the labeling results to the data acquisition module.

[0023] Furthermore, the data acquisition module is provided in a vehicle, and an image acquisition device is installed in the vehicle. The data acquisition module is used to obtain a facial image to be annotated, and the facial image to be annotated is sent to the data annotation module, specifically including:

[0024] The data acquisition module receives the face image to be annotated from the image acquisition device, and sends the face image to be annotated to the data annotation module through a scheduled task.

[0025] Furthermore, after deploying the second face recognition model trained using the annotation results to the data acquisition module, the method further includes:

[0026] The second face recognition model is loaded through the data acquisition module, and the face image from the image acquisition device is recognized using the second face recognition model.

[0027] Furthermore, sending the face image to be annotated to a data annotation module includes:

[0028] The data storage module receives and stores the face image to be annotated from the data acquisition module, and sends the face image to be annotated to the data annotation module.

[0029] Furthermore, the data storage module is set in the cloud or development end, and image storage space is reserved in the cloud or development end, and the image storage space meets the predefined capacity requirements. The sending of the face image to be annotated to the data annotation module includes:

[0030] The real-time capacity of the image storage space is detected by a data storage module, and when the real-time capacity reaches a preset value, the face image to be annotated is sent to the data annotation module.

[0031] Furthermore, the data annotation module is provided at the development end, and before the second face recognition model trained using the annotation results is deployed to the data acquisition module, the method further includes:

[0032] The model accuracy corresponding to the second face recognition model is tested, and if the model accuracy reaches a second preset value, it is determined that the training of the second face recognition model is completed.

[0033] According to a third aspect of the present application, a data annotation device for face recognition is provided, comprising:

[0034] an extraction unit, configured to obtain a facial image to be annotated, input the facial image to be annotated into a pre-trained first facial recognition model, and extract feature values ​​corresponding to the facial image to be annotated, wherein the first facial recognition model is a facial recognition model having a model accuracy reaching a first preset value;

[0035] a screening unit configured to screen out similar facial images corresponding to the facial image to be labeled from a set of registered facial images based on feature values ​​corresponding to the facial image to be labeled, the set of registered facial images including registered facial images containing user identity tags that have been pre-entered into the system;

[0036] The labeling unit is used to label the face image to be labeled according to the similar face image to obtain a labeling result.

[0037] Furthermore, the screening unit includes:

[0038] a matching subunit, configured to traverse the registered face images in the registered face image set, perform similarity matching between the feature values ​​corresponding to the face image to be annotated and the feature values ​​corresponding to the registered face images, and obtain a similarity matching result;

[0039] The screening subunit is used to screen out similar facial images corresponding to the facial image to be marked from the registered face set according to the similarity ranking corresponding to the registered facial image and the facial image to be marked in the similarity matching result.

[0040] Furthermore, the screening subunit is specifically configured to select, based on the similarity rankings of the registered facial images and the facial images to be labeled in the similarity matching results, a registered facial image whose similarity ranking is before a preset value as a similar facial image corresponding to the facial image to be labeled; or

[0041] According to the similarity rankings corresponding to the registered facial images and the facial images to be labeled in the similarity matching results, the registered facial image with the highest similarity ranking is selected as the similar facial image corresponding to the facial image to be labeled.

[0042] Furthermore, the labeling unit is specifically used to select the registered face image whose similarity ranking is before a preset value as the similar face image corresponding to the face image to be labeled according to the similarity ranking corresponding to the registered face image and the face image to be labeled in the similarity matching result, and then output the similar face image corresponding to the face image to be labeled to the recognition terminal, so that the recognition terminal labels the user identity according to the similar face image, and uses the labeled user identity label as the labeling result.

[0043] Furthermore, the device further comprises:

[0044] A processing unit is used to perform key point detection and face alignment processing on the face image to be labeled and the face images in the registered face image set.

[0045] According to the fourth aspect of the present application, a data labeling system for face recognition is provided, including: a data acquisition module, a data storage module, and a data labeling module, wherein the data acquisition module is connected to the data storage module, the data storage module is connected to the data labeling module, and the data labeling module is connected to the data acquisition module, and the data labeling module is the data labeling device of the third aspect mentioned above.

[0046] According to the fifth aspect of the present application, a data labeling device for face recognition is provided, including a storage medium, a processor, and a computer program stored on the storage medium and runnable on the processor, wherein the processor implements the above-mentioned data labeling method for face recognition when executing the program.

[0047] According to a sixth aspect of the present application, a storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned data labeling method for face recognition is implemented.

[0048] By means of the above technical solution, the present application provides a data annotation method, apparatus, system, device and storage medium for face recognition. The method inputs the face image to be annotated into a pre-trained first face recognition model, extracts the feature values ​​corresponding to the face image to be annotated, and based on the feature values ​​corresponding to the face image to be annotated, screens out similar face images corresponding to the face image to be annotated from the registered face image set, and then annotates the face image to be annotated based on the similar face images to obtain the annotation results. Compared with the data annotation method of manual ID classification for face recognition in the prior art, the embodiment of the present invention uses a first face recognition model with high model accuracy to assist in annotating the face image to be annotated, which greatly saves the cost of manual annotation, reduces the error of manual annotation, and controls the quality of data annotation.

[0049] Furthermore, the above-mentioned data labeling system for face recognition includes: a data acquisition module, a data storage module, and a data labeling module, and each module forms a closed loop. In this closed-loop system, the vehicle-deployed face recognition model can use the labeling results generated in the system to repeatedly learn and iterate, thereby continuously optimizing the model accuracy of the vehicle-deployed face recognition model through the closed loop. Since the training process of the face recognition model requires labeling a large number of samples, the closed-loop system uses a semi-automatic labeling process, which does not require manpower costs for data labeling, while ensuring data labeling efficiency and improving data labeling accuracy.

[0050] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0052] Figure 1 A schematic diagram of a process flow of a data annotation method for face recognition provided by an embodiment of the present application is shown;

[0053] Figure 2 A schematic diagram showing a flow chart of another data annotation method for face recognition provided by an embodiment of the present application;

[0054] Figure 3 A structural block diagram of a data annotation device for face recognition provided by an embodiment of the present application is shown;

[0055] Figure 4 A structural block diagram of a data annotation system for face recognition provided by an embodiment of the present application is shown;

[0056] Figure 5 A structural block diagram of another data annotation system for face recognition provided by an embodiment of the present application is shown;

[0057] Figure 6 A schematic diagram of a process for semi-automatically labeling facial images on a development end is shown in an embodiment of the present application;

[0058] Figure 7 A structural schematic diagram of a data annotation device for face recognition provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0059] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0060] This embodiment can be applied to facial recognition scenarios within fleets. In this scenario, cameras deployed in vehicles can capture driver facial images to collect raw data. Images with the same ID in the dataset are then grouped together to label the data for the facial recognition model. However, traditional labeling methods require human labelers to categorize all raw data by ID, which is time-consuming and prone to errors. This makes it difficult for facial recognition model training to meet low-precision requirements.

[0061] In order to solve the above problems, this application provides a data annotation method for face recognition, such as Figure 1 Shown, including:

[0062] 101. Obtain a facial image to be labeled, input the facial image to be labeled into a pre-trained first face recognition model, and extract feature values ​​corresponding to the facial image to be labeled.

[0063] 102. Based on the feature value corresponding to the face image to be labeled, select similar face images corresponding to the face image to be labeled from the registered face image set.

[0064] 103. Label the face image to be labeled according to similar face images to obtain a labeling result.

[0065] The above-mentioned first face recognition model is a pre-trained model with a large number of parameters and high model accuracy. It can usually accurately extract the feature values ​​in the face image. Here, a high-precision face recognition model can be used as a pre-trained high-precision model, which can assist in the automatic matching of the face recognition labeling process. The output of the high-precision face model can also be directly used as the labeling result, thereby saving the cost of manual labeling.

[0066] The collection, use and processing of facial information-related data involved in this application are in compliance with local laws, regulations and standards, and the data is obtained with the authorization of the owner.

[0067] In this embodiment, the registered face image set includes registered face images containing user identity tags that are pre-entered into the system. Specifically, in the process of screening out similar face images corresponding to the face image to be labeled from the registered face image set based on the feature values ​​corresponding to the face image to be labeled, the registered face images in the registered face image set can be traversed, and the feature values ​​corresponding to the face image to be labeled can be similarly matched with the feature values ​​corresponding to the registered face images to obtain similarity matching results, and based on the similarity ranking of the registered face images and the face images to be labeled in the similarity matching results, similar face images corresponding to the face image to be labeled can be screened out from the registered face set.

[0068] In the face recognition scenario within a fleet, the registered face image set can be the registered images of all drivers pre-entered into the system. The registered images have driver identity tags, which are equivalent to face images with predicted identity identification.

[0069] Since the similarity matching result can reflect the similarity between the registered face image and the face image to be annotated to a certain extent, the higher the similarity, the higher the probability that the face image to be annotated and the registered face image are the same face. Specifically, based on the similarity ranking of the registered face image and the face image to be annotated in the similarity matching result, the registered face image with the highest similarity ranking can be selected as the similar face image corresponding to the face image to be annotated. At this time, the data annotation process for face recognition is a fully automated process. Since the annotation results of the face recognition output by the model are limited by the model accuracy, they cannot reach 100% accuracy. On the other hand, considering that there may be a certain error in the similarity between the registered face image and the face image to be labeled, it is also possible to select the registered face image whose similarity ranking is before the preset value as the similar face image corresponding to the face image to be labeled based on the similarity ranking of the registered face image and the face image to be labeled in the similarity matching result, and further output the similar face image corresponding to the face image to be labeled to the recognition terminal, so that the recognition terminal labels the user identity according to the similar face image, and uses the labeled user identity label as the labeling result. At this time, the data labeling process used for face recognition is a semi-automatic process, and the labeling results output by the model are used as a reference for subsequent manual recognition. While improving the labeling efficiency, it can improve the accuracy of data labeling for face recognition.

[0070] In this embodiment, the feature value represents a set of vector values. Each face image can be described by a vector value, which includes various features of the face, such as the face contour and facial features.

[0071] In this embodiment, the feature values ​​of the registered facial images in the registered facial image set can also be extracted using the first facial recognition model, or other models can be used, which is not limited here. Since the feature values ​​are represented in vector form, specifically in the process of performing similarity matching based on the feature values ​​of the facial image to be annotated and the feature values ​​of the registered facial images in the registered facial image set, the cosine distance can be used to measure the similarity between the two feature values. The specific calculation formula is as follows:

[0072]

[0073] Where D is the pre-determined distance between the eigenvalues ​​of the face image to be annotated and the eigenvalues ​​of the registered face images in the registered face image set, is the normalized feature vector of the image to be labeled, It is the normalized feature vector of the registered face image in the registered face image set.

[0074] In this embodiment, the recognition terminal can be an intelligent communication terminal, which can be manually selected by the annotator to output the user identity tag. For example, five similar facial images are selected for the facial image to be annotated. These five similar facial images are facial images that the high-precision facial recognition model believes are closest to the facial image to be annotated, and the five similar facial images are further pushed to the recognition terminal. Here, the annotator at the recognition terminal can use the facial image to be annotated as a judgment basis for each similar facial image. If the annotator determines that the facial image to be identified belongs to any of the similar facial images, the identity identifier of the facial image to be identified can be obtained from the similar facial images. Of course, if the annotator determines that the facial image to be identified does not belong to any of the similar facial images, the identity identifier of the facial image to be identified cannot be obtained from the similar facial images.

[0075] Furthermore, to improve the accuracy of facial image annotation, key point detection and face alignment can be performed on the facial image to be annotated and the facial images in the registered facial image set before extracting the feature values ​​of the facial image to be annotated, and the facial image to be annotated can be standardized to ensure the accuracy of subsequent feature extraction. Specifically, five key points can be set for facial key point detection, including the center of the eyes, the tip of the nose, and the corners of the mouth. After detecting these five key points in the facial image, these five key points are aligned to a specified standard position. This standard position is equivalent to the key point position of the average face and can be obtained by average calculation or setting, thereby completing the standardization process.

[0076] Based on the above data annotation method for face recognition, this application provides another data annotation method for face recognition, such as Figure 2 Shown, including:

[0077] 201. Acquire a facial image to be annotated through a data acquisition module, and send the facial image to be annotated to a data annotation module;

[0078] 202. Receive the facial image to be annotated from the data acquisition module through the data annotation module, annotate the facial image to be annotated using the data annotation method, and deploy the second face recognition model trained using the annotation results to the data acquisition module.

[0079] Among them, the data labeling method can output the labeling results of the face image to be detected, which is equivalent to the identity label corresponding to the face image. Here, the second face recognition model is a model deployed in the vehicle for face image recognition. This model is a model trained in real time using labeled data, and the model accuracy is continuously improved during the training process. Compared with the manual ID classification labeling method, it can save the manual intervention cost of the entire labeling process, achieve semi-automatic labeling, and improve data labeling efficiency.

[0080] Furthermore, the data acquisition module is set in the vehicle, and an image acquisition device is installed in the vehicle, such as a camera, a camera, etc. The acquisition direction corresponding to the image acquisition device can be customized. For example, if it is necessary to capture the driver's image, the acquisition direction corresponding to the acquisition device is usually set to the area where the driver's face often appears.

[0081] Given the uncertainty surrounding facial images, the image acquisition device may not always capture facial images. To conserve transmission resources, the data acquisition module receives facial images to be annotated from the image acquisition device and sends them to the data annotation module via a scheduled task. For example, facial images captured by the image acquisition device may be transmitted every 10 minutes. Alternatively, facial images received by the data acquisition module may be transmitted to the image annotation module after a certain number of them have been received. The method for transmitting facial images is not limited here.

[0082] Optionally, since the data acquisition module can directly obtain facial images, the data acquisition module can also load the second facial recognition model after receiving the second facial recognition model trained using the annotation results, and use the second facial recognition model to perform recognition on the facial image from the image acquisition device.

[0083] In this embodiment, considering that the hardware in the vehicle is very small and the computational complexity is low, it is impossible to deploy a parameter model with a large amount of computation. The second face recognition model trained here can be trained in real time after semi-automatic annotation based on the collected face images. The computational complexity can be controlled in real time and can be better applied to the vehicle.

[0084] Optionally, due to the capacity limitations of the data acquisition module and the data annotation module, it is difficult to store a large number of facial images. Here, the data storage module can also receive and store the facial images to be annotated from the data acquisition module, and send the facial images to be annotated to the data annotation module.

[0085] In this embodiment, the data storage module is equivalent to a virtual storage space. The collected facial images are uploaded to the data storage module for storage, and then the facial images are transmitted to the data annotation module to facilitate the use of the data annotation module. Specifically, during the transmission process, it can be triggered according to the data requirements of the data annotation module, and a time interval trigger can also be set, which is not limited here.

[0086] Optionally, the data annotation module can be set up in the cloud or development end, and image storage space is reserved in the cloud or development end. The image storage space meets the predefined capacity requirements. The data annotation module can detect the real-time capacity in the image storage space, and when the real-time capacity reaches the preset value, the facial image to be annotated is sent to the data annotation module.

[0087] Optionally, the data labeling module is set at the development end. Taking into account the accuracy of face recognition, the data labeling module can also detect the model accuracy corresponding to the trained second face recognition model before deploying the second face recognition model trained using the labeling results to the data acquisition module. If the model accuracy reaches a second preset value, it is determined that the training of the second face recognition model is completed.

[0088] In this embodiment, since the accuracy of the model is limited by the amount of model calculations performed by the modules in the vehicle, the second preset value can be set based on actual needs.

[0089] The data labeling method for face recognition provided in the embodiment of the present application, since the training process of the face recognition model requires the labeling of a large number of samples, uses a face recognition model with high model accuracy to assist in labeling the face images to be labeled, and trains a second face recognition model, thereby greatly saving the cost of manual labeling. The second face recognition model trained using the labeling results is further deployed to the data acquisition module, so that the various modules in the labeling system form a closed loop, and use the labeling results generated in the system to repeatedly learn and iterate, thereby continuously optimizing the model accuracy of the vehicle-deployed face recognition model.

[0090] Accordingly, this application provides a data annotation device for face recognition, such as Figure 3 As shown, it includes: an extraction unit 31, a screening unit 32, and a labeling unit 33.

[0091] The extraction unit 31 may be configured to obtain a facial image to be annotated, input the facial image to be annotated into a pre-trained first facial recognition model, and extract feature values ​​corresponding to the facial image to be annotated, wherein the first facial recognition model is a facial recognition model having a model accuracy reaching a first preset value;

[0092] The screening unit 32 may be configured to screen out similar facial images corresponding to the facial image to be tagged from a set of registered facial images based on the feature values ​​corresponding to the facial image to be tagged, wherein the set of registered facial images includes registered facial images containing user identity tags that have been pre-entered into the system;

[0093] The labeling unit 33 may be used to label the face image to be labeled based on similar face images to obtain a labeling result.

[0094] In actual application scenarios, the screening unit 32 includes:

[0095] The matching subunit can be used to traverse the registered face images in the registered face image set, perform similarity matching between the feature values ​​corresponding to the face image to be annotated and the feature values ​​corresponding to the registered face images, and obtain a similarity matching result;

[0096] The screening subunit can be used to screen out similar face images corresponding to the face image to be labeled from the registered face set based on the similarity ranking between the registered face image and the face image to be labeled in the similarity matching result.

[0097] In actual application scenarios, the screening subunit can be specifically used to select, based on the similarity ranking between the registered face image and the face image to be annotated in the similarity matching result, the registered face image whose similarity ranking is before a preset value as the similar face image corresponding to the face image to be annotated; or

[0098] According to the similarity ranking between the registered face image and the face image to be labeled in the similarity matching result, the registered face image with the highest similarity ranking is selected as the similar face image corresponding to the face image to be labeled.

[0099] In actual application scenarios, the labeling unit can be specifically used to rank the similarities corresponding to the registered face image and the face image to be labeled in the similarity matching result, select the registered face image whose similarity ranking is before a preset value as the similar face image corresponding to the face image to be labeled, and then output the similar face image corresponding to the face image to be labeled to the recognition terminal, so that the recognition terminal labels the user identity according to the similar face image, and uses the labeled user identity label as the labeling result.

[0100] In actual application scenarios, the device also includes:

[0101] The processing unit can be used to perform key point detection and face alignment processing on the face images to be marked and the face images in the registered face image set.

[0102] Furthermore, based on the above-mentioned data annotation device for face recognition, the present application provides a data annotation system for face recognition, such as Figure 4 As shown, it includes: a data acquisition module 40, a data storage module 41, and a data annotation module 42. The modules in the data annotation system form a closed loop, wherein the data acquisition module 40 is connected to the data storage module 41, the data storage module 41 is connected to the data annotation module 42, and the data annotation module 42 is connected to the data acquisition module 40. Here, the data annotation module 42 is the above-mentioned data annotation device for face recognition.

[0103] The data acquisition module 40 can be used to obtain the facial image to be annotated from the image acquisition device and send the facial image to be annotated to the data storage module 41;

[0104] The data storage module 41 can be used to receive and store the facial images to be annotated sent by the data acquisition module 40, and send the facial images to be annotated to the data annotation module 42;

[0105] The data annotation module 42 can be used to receive the facial images to be annotated sent by the data storage module 41, annotate the facial images to be annotated using the pre-trained first facial recognition model, obtain the annotation results, and deploy the second facial recognition model trained using the annotation results to the data acquisition module 40.

[0106] It can be understood that the first face recognition model here is a face recognition model whose model accuracy reaches the first preset value. It mainly plays the role of auxiliary labeling. It is equivalent to a pre-trained high-precision model, which can assist in the automatic matching of face recognition labeling process. The second face recognition model is a model deployed in the vehicle for face image recognition. This model is a model trained in real time using labeled data, and the model accuracy is continuously improved during the training process. Compared with the manual ID classification labeling method, it can save the cost of manual intervention in the entire labeling process, achieve semi-automatic labeling, and improve data labeling efficiency.

[0107] The embodiment of the present application provides a data labeling system for face recognition, which includes: a data acquisition module, a data storage module, and a data labeling module, and each module forms a closed loop. In this closed-loop system, the face recognition model deployed on the vehicle can use the labeling results generated in the system to repeatedly learn and iterate, thereby continuously optimizing the model accuracy of the face recognition model deployed on the vehicle through the closed loop. Since the training process of the face recognition model requires labeling a large number of samples, the closed-loop system uses a semi-automated labeling process. By using a face recognition model with high model accuracy to assist in labeling the face images to be labeled, the cost of manual labeling is greatly saved, the error of manual labeling is reduced, and the quality of data labeling is controlled.

[0108] Optionally, the data acquisition module 40 is set in a vehicle, and an image acquisition device is installed in the vehicle. Here, the data acquisition module 40 can receive the facial images to be labeled sent by the image acquisition device, and transmit the facial images to be labeled to the data storage module 41 through a scheduled task.

[0109] Optionally, since the data acquisition module can directly obtain facial images, the data acquisition module 40 can also load the second facial recognition model after receiving the second facial recognition model trained using the annotation results, and use the second facial recognition model to perform recognition on the facial image from the image acquisition device.

[0110] Optionally, the data storage module 41 is set in the cloud or development end, and image storage space is reserved in the cloud or development end. The data storage module 41 can detect the real-time capacity in the image storage space. When the real-time capacity reaches a preset value, the facial image to be annotated is sent to the data annotation module 42.

[0111] Optionally, the data labeling module 42 is set at the development end, where the development end is usually a local server. Taking into account the accuracy of face recognition, the data labeling module 42 can also detect the model accuracy corresponding to the trained second face recognition model before deploying the second face recognition model trained using the labeling results to the data acquisition module. If the model accuracy reaches a second preset value, it is determined that the training of the second face recognition model is completed.

[0112] In another embodiment of the present application, a structural block diagram of another data annotation system for face recognition is provided, such as Figure 5As shown, a camera located in the vehicle first captures the driver's facial image and transmits it to the cloud. A scheduled task can then be used to stream the facial image stored in the cloud back to the development end, where the facial image is semi-automatically annotated. The annotated facial image is then used to train the facial recognition model. The trained facial recognition model is then deployed to the vehicle, forming a closed-loop iteration of the facial image annotation system to continuously improve the accuracy of the deployed model. In the field of smart car technology, facial recognition technology can be used in a car's driver monitoring system (DMS), occupant monitoring system (OMS), and other cabin monitoring systems. It is understood that the facial recognition data annotation method or system provided in the embodiments of this application can be used in the field of smart cars, as well as in other fields, such as smart homes and smart terminals, and this application does not limit this.

[0113] The specific process of semi-automatic annotation of facial images on the development side is as follows: Figure 6 As shown in the figure, semi-automatic annotation is an important part of the entire closed-loop process, which can save a lot of costs for manual intervention. Before the semi-automatic annotation begins, the registered face images are pre-processed to obtain standard face images, and then the corresponding registered face feature library is extracted through a high-precision face recognition model and saved in the system in advance. In this feature library, there is a set of feature values ​​corresponding to each registered face image. When semi-automatic annotation begins, the unlabeled face images will be pre-processed and output to a high-precision face recognition model to extract the feature values ​​of the corresponding image to be labeled. The feature values ​​of the image to be labeled are further matched with the registered face library for similarity, and the similarity ranking of the image to be labeled in the registered face images is obtained to find the images with the corresponding top N registered face IDs. These images are then displayed to the annotation engineer for selection to quickly complete the face image annotation process.

[0114] In this embodiment, the above-mentioned semi-automatic labeling process only needs to display the images of the first N IDs matched with the assistance of the high-precision model to the labeling personnel for selection, without the labeling personnel having to pay attention to the remaining large number of irrelevant registration IDs, which can greatly save the cost of manual labeling. At the same time, through the closed-loop semi-automatic data labeling method, the high-precision model assistance with large computing power can be used to reduce the human errors caused by manual labeling and control the quality of data labeling.

[0115] The present application also provides an electronic device, such as Figure 7As shown, the electronic device 700 includes a processor 701 and a memory 702. The memory 702 stores programs or instructions that can be run on the processor 701. When the program or instructions are executed by the processor 701, the various steps of the above-mentioned data labeling method embodiment for face recognition are implemented, and the same technical effect can be achieved. To avoid repetition, they are not repeated here.

[0116] It should be noted that the electronic devices in the embodiments of the present application include the above-mentioned mobile electronic devices and non-mobile electronic devices.

[0117] Memory 702 can be used to store software programs and various data. Memory 702 may primarily include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area may store an operating system, applications or instructions required for at least one function (such as a sound playback function or an image playback function), and the like. Furthermore, memory 702 may include volatile memory or non-volatile memory, or both. Non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DRRAM). The memory 702 in the embodiment of the present application includes but is not limited to these and any other suitable types of memory.

[0118] Processor 701 may include one or more processing units. Optionally, processor 701 integrates an application processor and a modem processor. The application processor primarily handles operations related to the operating system, user interface, and application programs, while the modem processor primarily processes wireless communication signals, such as a baseband processor. It is understood that the modem processor may not be integrated into processor 701.

[0119] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of the above-mentioned data labeling method embodiment for face recognition are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0120] An embodiment of the present application also provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned data labeling method embodiment for face recognition, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0121] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0122] An embodiment of the present application also provides a computer program product, which is stored in a storage medium and is executed by at least one processor to implement the various processes of the data labeling method embodiment for face recognition as described above, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0123] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, it should be noted that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0124] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. A data annotation method for face recognition, characterized in that: include: Obtaining a facial image to be annotated, and inputting the facial image to be annotated into a pre-trained first facial recognition model to extract feature values ​​corresponding to the facial image to be annotated, wherein the first facial recognition model is a facial recognition model whose model accuracy reaches a first preset value; According to the feature values ​​corresponding to the facial image to be labeled, similar facial images corresponding to the facial image to be labeled are screened out from a set of registered facial images, the set of registered facial images including registered facial images containing user identity tags that are pre-entered into the system, the registered facial images in the set of registered facial images are traversed, and the feature values ​​corresponding to the facial image to be labeled are similarly matched with the feature values ​​corresponding to the registered facial images to obtain a similarity matching result; according to the similarity ranking between the registered facial images and the facial images to be labeled in the similarity matching result, a registered facial image whose similarity ranking is before a preset value is selected as a similar facial image corresponding to the facial image to be labeled; or according to the similarity ranking between the registered facial images and the facial images to be labeled in the similarity result, a registered facial image with the highest similarity ranking is selected as a similar facial image corresponding to the facial image to be labeled; The similar facial image corresponding to the facial image to be labeled is output to the recognition terminal, so that the recognition terminal labels the user identity based on the similar facial image, and deploys the second facial recognition model trained using the labeling result to the data acquisition module, and the model accuracy corresponding to the trained model reaches a second preset value.

2. The method according to claim 1, characterized in that The method further comprises: Key point detection and face alignment processing are performed on the face image to be labeled and the face images in the registered face image set.

3. A data annotation method for face recognition, characterized in that: include: Acquire a face image to be annotated through a data acquisition module, and send the face image to be annotated to a data annotation module; The data annotation module receives the facial image to be annotated from the data acquisition module, annotates the facial image to be annotated according to the similar facial image using the data annotation method described in any one of claims 1-2, and deploys the second facial recognition model trained using the annotation result to the data acquisition module, and the similar facial image is obtained in the following manner: traversing the registered facial images in the registered facial image set, performing similarity matching on the feature values ​​corresponding to the facial image to be annotated and the feature values ​​corresponding to the registered facial images, and obtaining a similarity matching result; according to the similarity ranking of the registered facial images and the facial images to be annotated in the similarity matching result, selecting the registered facial image with a similarity ranking before a preset value as the similar facial image corresponding to the facial image to be annotated; or according to the similarity ranking of the registered facial images and the facial images to be annotated in the similarity result, selecting the registered facial image with the highest similarity ranking as the similar facial image corresponding to the facial image to be annotated.

4. The method according to claim 3, characterized in that The data acquisition module is provided in a vehicle, and an image acquisition device is installed in the vehicle. The data acquisition module is used to obtain a facial image to be annotated, and the facial image to be annotated is sent to the data annotation module, specifically including: The data acquisition module receives the face image to be annotated from the image acquisition device, and sends the face image to be annotated to the data annotation module through a scheduled task.

5. The method according to claim 4, characterized in that After deploying the second face recognition model trained using the annotation results to the data acquisition module, the method further includes: The second face recognition model is loaded through the data acquisition module, and the face image from the image acquisition device is recognized using the second face recognition model.

6. The method according to claim 3, characterized in that The step of sending the face image to be annotated to a data annotation module includes: The data storage module receives and stores the face image to be annotated from the data acquisition module, and sends the face image to be annotated to the data annotation module.

7. The method according to claim 6, characterized in that The data storage module is set in the cloud or development end, and the cloud or development end has reserved image storage space, and the image storage space meets the predefined capacity requirements. The facial image to be annotated is sent to the data annotation module, including: The real-time capacity of the image storage space is detected by a data storage module, and when the real-time capacity reaches a preset value, the face image to be annotated is sent to the data annotation module.

8. The method according to claim 4, characterized in that The data annotation module is provided at the development end. Before deploying the second face recognition model trained using the annotation results to the data acquisition module, the method further includes: The model accuracy corresponding to the second face recognition model is tested, and if the model accuracy reaches a second preset value, it is determined that the training of the second face recognition model is completed.

9. A data annotation device for face recognition, characterized in that: include: an extraction unit, configured to obtain a facial image to be annotated, input the facial image to be annotated into a pre-trained first facial recognition model, and extract feature values ​​corresponding to the facial image to be annotated, wherein the first facial recognition model is a facial recognition model having a model accuracy reaching a first preset value; a screening unit configured to screen out similar facial images corresponding to the facial image to be labeled from a set of registered facial images based on feature values ​​corresponding to the facial image to be labeled, the set of registered facial images including registered facial images containing user identity tags that have been pre-entered into the system; The screening unit includes: a matching subunit for traversing the registered facial images in the registered facial image set, performing similarity matching between the feature values ​​corresponding to the facial image to be labeled and the feature values ​​corresponding to the registered facial images, and obtaining a similarity matching result; a screening subunit for selecting, based on the similarity ranking between the registered facial images and the facial images to be labeled in the similarity matching result, a registered facial image with a similarity ranking before a preset value as a similar facial image corresponding to the facial image to be labeled; or selecting, based on the similarity ranking between the registered facial images and the facial images to be labeled in the similarity matching result, a registered facial image with the highest similarity ranking as a similar facial image corresponding to the facial image to be labeled; The labeling unit is used to output the similar facial image corresponding to the facial image to be labeled to the recognition terminal, so that the recognition terminal labels the user identity according to the similar facial image, and deploys the second facial recognition model trained using the labeling result to the data acquisition module, and the model accuracy corresponding to the trained model reaches a second preset value.

10. The device according to claim 9, characterized in that The device further comprises: A processing unit is used to perform key point detection and face alignment processing on the face image to be labeled and the face images in the registered face image set.

11. A data annotation system for face recognition, characterized in that: include: A data acquisition module, a data storage module, and a data annotation module, wherein the data acquisition module is connected to the data storage module, the data storage module is connected to the data annotation module, and the data annotation module is connected to the data acquisition module, and the data annotation module is the data annotation device described in any one of claims 9-10.

12. A data annotation device for face recognition, characterized in that: The method comprises a processor and a memory, wherein the memory stores a program or instruction running on the processor, and when the program or instruction is executed by the processor, the steps of the data labeling method for face recognition as described in any one of claims 1 to 8 are implemented.

13. A readable storage medium having a program or instruction stored thereon, characterized in that: When the program or instruction is executed by a processor, the steps of the data labeling method for face recognition as described in any one of claims 1 to 8 are implemented.

Citation Information

Patent Citations

  • Identity authentication method and apparatus

    CN108491805A

  • Active learning method and device suitable for face recognition model training, terminal and medium

    CN110765917A

  • Face recognition method and device, robot and storage medium

    CN112488053A