CN116229556A - Face recognition method and device, embedded device, and computer-readable storage medium - Google Patents

Abstract

The invention discloses a face recognition method, a face recognition device, an embedded device and a nonvolatile computer readable storage medium. The face recognition method comprises the following steps: detecting face information in the acquired image to generate a face image; detecting the quality of the face image based on a preset face screening model; and calculating the similarity between the face feature vector of the face image with the quality reaching the preset condition and the preset face feature vector based on the reconfigurable calculating unit so as to carry out face recognition. According to the face recognition method, the face recognition device, the embedded equipment and the nonvolatile computer readable storage medium, some face images with poor quality can be removed, so that the number of face images needing face recognition is reduced, and the recognition efficiency is improved. And the reconfigurable computing unit can reduce the resource occupancy rate and time consumption of the embedded device in the face recognition process so as to meet the operation requirements of the embedded device on other works.

Description

Face recognition method and device, embedded device, and computer-readable storage medium

technical field

The present application relates to the technical field of face recognition, and more specifically, to a face recognition method, a face recognition device, an embedded device, and a non-volatile computer-readable storage medium.

Background technique

In recent years, with the rapid improvement of computer performance and the continuous improvement of deep learning methods, major breakthroughs have been made in the fields of pattern recognition and artificial intelligence. People have achieved excellent results in many pattern recognition tasks through deep learning methods, and face recognition is no exception. However, in the process of face recognition, it is necessary to extract face features first and search in the face feature database, but the data volume of face features is too large, which will lead to low face recognition efficiency.

Contents of the invention

Embodiments of the present application provide a face recognition method, a face recognition device, an embedded device, and a non-volatile computer-readable storage medium.

The face recognition method of the embodiment of the present application includes detecting the face information in the collected image to generate a face image; based on a preset face screening model, detecting the quality of the face image; and based on a reconfigurable computing unit and calculating the similarity between the face feature vector of the face image whose quality reaches the preset condition and the preset face feature vector, so as to perform face recognition.

The face recognition device in the embodiment of the present application includes a generation module, a first detection module and a recognition module. The generation module is used to detect face information in the captured image to generate a face image. The first detection module is used to detect the quality of the face image based on a preset face screening model. The identification module is used to calculate the similarity between the face feature vector of the face image whose quality reaches the preset condition and the preset face feature vector based on the reconfigurable computing unit, so as to perform face recognition.

The embedded device in the embodiments of the present application includes a processor. The processor is used to detect face information in the collected image to generate a face image; based on a preset face screening model, detect the quality of the face image; and based on a reconfigurable computing unit, the calculation quality reaches the preset The similarity between the face feature vector of the conditional face image and the preset face feature vector is used for face recognition.

The non-transitory computer-readable storage medium of the embodiment of the present application contains a computer program, and when the computer program is executed by one or more processors, the processor is made to perform the following face recognition method: detecting Face information to generate a face image; based on a preset face screening model, detect the quality of the face image; and based on a reconfigurable computing unit, calculate the face features of the face image whose quality reaches the preset condition The similarity between the vector and the preset face feature vector for face recognition.

In the face recognition method, face recognition device, embedded device, and non-volatile computer-readable storage medium of the embodiments of the present application, before performing face recognition on a face image, a face screening model based on a preset , to detect the quality of the face image, only when the quality of the face image meets the preset conditions, the face recognition will be performed based on the reconfigurable computing unit, that is, before the face recognition is performed on the face image, some quality Poor face images, thereby reducing the number of face images that need face recognition to improve recognition efficiency, and the reconfigurable computing unit can reduce the workload of the processor, thereby reducing the face recognition process of embedded devices The resource occupancy rate and time consumption used in the system can meet the running requirements of embedded devices in other tasks.

Additional aspects and advantages of the embodiments of the application will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the embodiments of the application.

Description of drawings

The above and/or additional aspects and advantages of the present application will become apparent and understandable from the description of the embodiments in conjunction with the following drawings, wherein:

FIG. 1 is a schematic flow diagram of a face recognition method in some embodiments of the present application;

Fig. 2 is a schematic diagram of a face recognition device in some embodiments of the present application;

3 is a schematic plan view of an embedded device in some embodiments of the present application;

FIG. 4 is a schematic flow diagram of a face recognition method in some embodiments of the present application;

FIG. 5 is a schematic flow diagram of a face recognition method in some embodiments of the present application;

FIG. 6 is a schematic flow diagram of a face recognition method in some embodiments of the present application;

FIG. 7 is a schematic diagram of a scene of a face recognition method in some embodiments of the present application;

FIG. 8 is a schematic flow diagram of a face recognition method in some embodiments of the present application;

FIG. 9 is a schematic flowchart of a face recognition method in some embodiments of the present application;

FIG. 10 is a schematic flow diagram of a face recognition method in some embodiments of the present application;

FIG. 11 is a schematic flow diagram of a face recognition method in some embodiments of the present application;

Fig. 12 is a schematic flowchart of a face recognition method in some embodiments of the present application;

FIG. 13 is a schematic flowchart of a face recognition method in some embodiments of the present application;

Fig. 14 is a schematic diagram of a connection state between a non-volatile computer-readable storage medium and a processor in some embodiments of the present application.

Detailed ways

Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary, are only for explaining the embodiments of the present application, and should not be construed as limiting the embodiments of the present application.

Referring to FIG. 1 , an embodiment of the present application provides a face recognition method. The face recognition method comprises steps:

01: Detect face information in the captured image to generate a face image;

03: Based on the preset face screening model, detect the quality of the face image; and

05: Based on the reconfigurable computing unit, calculate the similarity between the face feature vector of the face image whose quality meets the preset conditions and the preset face feature vector for face recognition.

Referring to FIG. 2 , the embodiment of the present application provides a face recognition device 10 . The face recognition device 10 includes a generation module 11 , a first detection module 12 and a recognition module 13 . The face recognition method in the embodiments of the present application can be applied to the face recognition device 10 . Wherein, the generation module 11 , the first detection module 12 and the identification module 13 are used to execute step 01 , step 03 and step 05 respectively. That is, the generation module 11 is used to detect the face information in the captured image to generate a face image. The first detection module 12 is used for detecting the quality of the face image based on the preset face screening model. The recognition module 13 is used to calculate the similarity between the face feature vector of the face image whose quality reaches the preset condition and the preset face feature vector based on the reconfigurable computing unit, so as to perform face recognition.

Referring to FIG. 3 , the embodiment of the present application also provides an embedded device 100 . The face recognition method in the embodiments of the present application can be applied to the embedded device 100 . The embedded device 100 includes a processor 20 . The processor 20 is used to execute step 01, step 03 and step 05. That is, the processor 20 is used to detect the face information in the collected image to generate a face image; based on the preset face screening model, detect the quality of the face image; and based on the reconfigurable computing unit, the calculation quality reaches the preset The similarity between the face feature vector of the conditional face image and the preset face feature vector is used for face recognition.

Wherein, the embedded device 100 further includes a casing 30 and a camera 40 . The housing 30 can be used to install functional modules such as a display device, an imaging device, a power supply device, and a communication device of the embedded device 100, so that the housing 30 provides protection against dust, drop, and water for the functional modules. Camera 40 can be used to capture images. The embedded device 100 may be a mobile phone, a digital camera, a smart watch, a head-mounted display device, a game console, a robot, and the like. As shown in FIG. 3 , the embodiment of the present application is described by taking the embedded device 100 as a mobile phone as an example. It can be understood that the specific form of the embedded device 100 is not limited to the mobile phone.

Specifically, after the image is collected by the camera 40, the processor 20 may detect face information in the collected image to generate a face image. The face information may include the position of the face in the image and the positions of key points of the face (such as the positions of eyes, mouth, nose, and ears). Wherein, the processor 20 may use a face detection algorithm to obtain face information in the image. The face detection algorithm can be FaceBoxes face detection network, retinaface face detection network. In this way, the processor 20 can generate the face image in the image according to the obtained face information.

Next, the processor 20 may screen the face according to a preset face image to detect the quality of the face image. Among them, the face screening model can use a 16-layer convolutional neural network, which can include a convolutional layer, an activation layer, a normalization network (BN) layer, and a loss function layer. Its loss function layer is the loss function of robust regression (Huber regression loss function).

The preset face screening model is a pre-trained screening model, and the training samples in the face screening model may include high-quality face samples and low-quality face samples.

More specifically, high-quality face samples are samples with stable and easy-to-recognize facial feature points, and low-quality face samples are samples with difficult-to-recognize face feature points. Therefore, when the processor 20 detects the quality of the human face image by the preset human face screening model, the human face image can be input to the preset human face screening model, and the processor 20 can obtain the quality of the human face image. The quality is good or bad.

For example, when the feature points of the face image are difficult to identify, the preset face screening model will output a low quality face image. For another example, when the feature points of the face image are easy to identify, the preset face screening model will output a higher quality face image.

In some implementations, the preset face screening model can also be set to correspond to different scores for different quality training samples, that is, the preset face screening model can be used to identify the number of feature points according to the number of identified feature points Images corresponding to different numbers of feature points are scored. It can be understood that the more feature points are recognized, the higher the score of the corresponding face image is.

In this way, when the quality of the face image meets the preset condition, the processor 20 will perform face recognition based on the face image.

In one embodiment, the preset conditions can be divided into high quality and low quality, so that when the quality of the face image is high quality, the processor 20 will perform face recognition based on the face image.

In another embodiment, the preset condition may be a specific score value, such as 80 points, 90 points, etc. of a percentile system. Then when the processor 20 detects the quality score of the human face image by the preset human face screening model, it can be compared with the preset condition to determine whether the quality score of the human face image is greater than or equal to the score of the preset condition , the processor 20 will perform recognition based on the face image.

It can be understood that only when the quality reaches the preset condition, the processor 20 will perform face recognition based on the face image. That is, face recognition will not be performed on low-quality face images, thereby filtering out low-quality face images, thereby reducing the amount of recognition for face recognition, and performing face recognition only based on high-quality face images , can also improve the recognition rate.

More specifically, when the processor 20 performs face recognition based on a face image, the processor 20 may use a reconfigurable computing unit to extract a face feature vector of a face image whose quality meets a preset condition.

Next, the processor 20 may input the preset face feature vectors in the preset face information database to the reconfigurable computing unit for calculation by the reconfigurable computing unit. Wherein, the reconfigurable computing unit is a (Reconfigurable Computing Unit, RCU) device. Calculating the similarity between the feature vector of the human face to be tested and the pre-stored feature vector by the RCU device can reduce the workload of the processor 20, thereby reducing the resource usage and time consumption used by the embedded device 100 in the face recognition process, thereby Satisfy the running requirements of the embedded device 100 on other tasks.

In the face recognition method, face recognition device 10, and embedded device 100 of the embodiment of the present application, before performing face recognition on a face image, the quality of the face image will be detected based on a preset face screening model. , only when the quality of the face image reaches the preset condition, the face recognition will be performed based on the reconfigurable computing unit, that is, before the face recognition is performed on the face image, some face images with poor quality will be eliminated, so that Reduce the number of face images that need face recognition to improve recognition efficiency, and the reconfigurable computing unit can reduce the workload of the processor 20, thereby reducing the resource occupation used by the embedded device 100 in the face recognition process rate and time consumption, so as to meet the running requirements of the embedded device 100 on other tasks.

Please refer to Fig. 2, Fig. 3 and Fig. 4, in some embodiments, step 01: detect the face information in the collected image, to generate a face image, including steps:

011: Based on the preset face detection model, detect the position of the face in the collected image;

012: Generate face images based on face positions.

In some implementations, the generation module 11 is used to execute step 011 and step 012. That is, the generation module 11 is used for detecting the position of the face in the captured image based on the preset face detection model; and generating the face image according to the position of the face.

In some implementations, the processor 20 is configured to execute step 011 and step 012. That is, the processor 20 is configured to detect the position of the face in the captured image based on the preset face detection model; generate a face image according to the position of the face.

Specifically, the face information includes the face position of the target face, that is, the face position in the image collected by the camera 40 . When the processor 20 detects the face information in the collected image to generate the face image, it may be: the processor 20 detects the position of the face in the collected image based on a preset face detection model. According to the above, the preset face detection model can be FaceBoxes face detection network, retinaface face detection network, etc.

After the processor 20 obtains the image captured by the camera 40, it can first obtain the position of the face in the captured image according to the preset face detection model, so as to generate the face image according to the position of the face.

In this way, the image data that does not belong to the face part in the image collected by the camera 40 can be eliminated, thereby reducing the amount of data to be detected when performing quality inspection on the face image, so as to improve recognition efficiency.

Please refer to Fig. 2, Fig. 3 and Fig. 5, the face recognition method of the embodiment of the present application also includes steps:

07: Based on the preset face detection model, detect the position of the key points of the face in the face image; and

09: Align the face images according to the positions of the faces and key points of the faces, so that the pose of the target face of the aligned face images is adjusted to a preset pose.

Partial Step 03: Detect the quality of the face image, including steps:

031: Detect the quality of the aligned face images.

In some implementations, the face recognition device 10 further includes a second detection module 14 and an alignment module 15 . The second detection module 14 is used to execute step 07. The alignment module 15 is used to execute step 09 . The second detection module 14 is used to execute step 031 . That is, the second detection module 14 is configured to detect the position of key points of the face in the face image based on the preset face detection model. The alignment module 15 is used for aligning the face images according to the positions of the faces and key points of the faces, so that the pose of the target face in the aligned face images is adjusted to a preset pose. The first detection module 12 is used to detect the quality of the aligned face images.

In some embodiments, the processor 20 is configured to execute step 07 , step 09 and step 031 . That is, the processor 20 is used to detect the position of the key point of the face in the face image based on the preset face detection model; align the face images according to the position of the face and the position of the key point of the face, so that the aligned The pose of the target face of the face image is adjusted to a preset pose; and the quality of the aligned face image is detected.

Specifically, the face information also includes key points of the face. After the face image is generated, the processor 20 can also detect the positions of key points of the face in the face image according to a preset face detection model. Wherein, the positions of the key points of the human face may be positions such as eyes, nose, mouth, and ears.

In this way, after the processor 20 obtains the position of the human face and the positions of the key points of the human face, the face images can be aligned according to the position of the human face and the positions of the key points of the human face, so that the target human face of the aligned human face image The posture is adjusted to the default posture. Wherein, the preset posture may be a front view of a face, a side view of a face, and the like. Preferably, in order to ensure that the feature points in the face image are easy to extract, and considering that the face image is used for face recognition, the preset pose is a front view of the face.

Specifically, the processor 20 may perform affine transformation on the face image according to the position of the face and the positions of the key points of the face, so that the angle of the face faces the front, that is, the face image is a front view of the face.

Further, when the processor 20 detects the quality of the face image based on the preset face screening model, it detects the quality of the aligned face image.

In this way, it can be ensured that when the processor 20 detects the quality of the human face image based on the preset human face screening model, the feature points of the obtained human face image are relatively obvious, thereby ensuring that the detection of the quality of the human face image has high accuracy .

Please refer to Fig. 2, Fig. 3 and Fig. 6, in some embodiments, part of step 011: detecting the position of the key points of the face in the collected image also includes the steps of:

0111: According to the size of the captured image, generate multiple candidate frames of preset sizes, and output the score of each candidate frame;

0113: Determine that the candidate frame whose score is greater than the preset score is a face frame;

0115: Obtain the coincidence degree between any two face frames with overlapping parts, and determine that among the face frames whose coincidence degree is greater than the preset coincidence degree, the face frame with the highest score is the target face frame; and

0117: Output the position of the target face frame as the face position.

In some implementations, the second detection module 14 is used to execute step 0111 , step 0113 , step 0115 and step 0117 . The second detection module 14 is used to generate a plurality of candidate frames of preset size according to the size of the captured image, and output the score of each candidate frame; determine that the candidate frame whose score is greater than the preset score is a human face frame; obtain any two existing The degree of overlap between the face frames of the overlapping parts, determine the face frame whose degree of overlap is greater than the preset degree of overlap, the face frame with the highest score is the target face frame; and output the position of the target face frame as the face Location.

In some embodiments, the processor 20 is configured to execute step 0111 , step 0113 , step 0115 and step 0117 . That is, the processor 20 is used to generate a plurality of candidate frames of a preset size according to the size of the captured image, and output the score of each candidate frame; determine that the candidate frame with a score greater than the preset score is a face frame; obtain any two existing The degree of overlap between the face frames of the overlapping parts, determine the face frame whose degree of overlap is greater than the preset degree of overlap, the face frame with the highest score is the target face frame; and output the position of the target face frame as the face Location.

Specifically, when the processor 20 detects the position of the face in the image based on the preset face detection model, the processor 20 can generate a plurality of candidate frames of preset sizes according to the size of the captured image, and output each candidate frame box rating. in. The preset size is proportional to the size of the captured image, that is, the larger the size of the captured image, the larger the size of the candidate frame of the preset size.

As shown in FIG. 7( a ), if the size of the captured image P1 is 1600*900, the size S1 of the candidate frame is 200*150. As shown in FIG. 7( b ), if the size of the captured image P2 is 1920*1080, the size of the candidate frame S2 is 240*200.

When the processor 20 scores the multiple candidate frames, the multiple candidate frames can be scored based on a pre-trained face detection model. For example, the pre-trained face detection model contains scores corresponding to training samples of different qualities. When the candidate frames are input into the pre-trained face detection model, the scores of each candidate frame will be obtained. For example, the more features of a face in a proposal box, the higher the score.

Next, the processor 20 may determine a candidate frame whose score is greater than a preset score as a human face frame according to the score of each candidate frame. Wherein, the preset score may be an artificially set score, such as 80 points, 90 points and so on. It can be understood that, the candidate frame whose score is greater than the preset score has more facial features in the corresponding image in the candidate frame.

Further, the processor 20 may acquire the degree of coincidence between any two human face frames with overlapping parts, so as to determine the target human face frame according to the degree of coincidence and the scoring of the candidate frames.

More specifically, the more overlapping parts of any two face frames, the greater the overlapping degree. When the coincidence degree is greater than the preset coincidence degree, the processor 20 considers the face frame with the highest score among the two face frames as the target face frame. It can be understood that when the coincidence degree is greater than the preset coincidence degree, it means that the two face frames share more areas.

The target face frame is the face frame with the highest score among all the face frames determined by the processor 20 whose coincidence degree is greater than a preset coincidence degree.

In this way, the processor 20 can output the position of the target face frame as the face position. It can be understood that the position of the target face frame is the most accurate face position in the collected image, so as to ensure that the quality of the face image is detected more accurately.

Please refer to Fig. 2, Fig. 3 and Fig. 8, in some embodiments, step 05: Based on the reconfigurable computing unit, calculate the face feature vector and the preset face feature vector of the face image whose quality reaches the preset condition The similarity for face recognition also includes steps:

051: Based on the preset feature extraction model, extract the features of the face image to generate a face feature vector; and

053: Establish a face information database based on the face feature vector.

In some implementations, the recognition module 13 is used to execute step 051 and step 053 . That is, the recognition module 13 is used to extract features of the face image based on a preset feature extraction model to generate a face feature vector; and establish a face information database according to the face feature vector.

In some implementations, the processor 20 is configured to execute step 051 and step 053 . That is, the processor 20 is configured to extract features of the face image based on a preset feature extraction model to generate a face feature vector; and establish a face information database according to the face feature vector.

Specifically, when the processor 20 performs face recognition based on a face image, it can also extract features of the face image based on a preset feature extraction model to generate a face feature vector, and establish a face feature vector based on the face feature vector. face information database.

Among them, the preset feature extraction model is a pre-trained extraction model, which can use a 72-layer convolutional neural network, including convolutional layers, pooling layers, activation layers, fully connected layers, and loss layers. The loss function used is the weighted sum of softmax-loss and center-loss. Softmax-loss is used to improve the degree of intra-class aggregation of samples in the feature space; center-loss is used to increase the inter-class distance of samples in the feature space.

In the face recognition method of the embodiment of the present application, when face recognition is performed, face information capable of face recognition must first be stored in the face information database. Therefore, before face recognition, users need to enter their own face information. That is, a process of offline recording of human face images is required.

The specific process is shown in FIG. 9 : first, the camera 40 of the embedded device 100 collects images for detection by the processor 20 . Based on the preset human face detection model, the processor 20 detects human face information (human face position and human face key point position) to align the human face images. Next, the processor 20 detects the quality of the human face based on the preset human face screening model. When the quality of the face image reaches the preset condition, the processor 20 will extract the features of the face image based on the predetermined feature extraction model to generate a face feature vector and establish a face information database. And when the quality of the face image does not meet the preset condition, the camera 40 will prompt the user to recapture the image, so as to repeat the above steps until the quality of the face image reaches the preset condition.

It can be understood that the face images stored in the face information database are all face images of satisfactory quality. In this way, when performing face recognition, it can be ensured that the quality of face images that can pass face recognition is better, so as to ensure the accuracy of face recognition. Accuracy.

Please refer to Fig. 2, Fig. 3 and Fig. 10, in some embodiments, step 05: Based on the reconfigurable computing unit, calculate the face feature vector and the preset face feature vector of the face image whose quality reaches the preset condition The similarity for face recognition also includes steps:

055: Based on the preset feature extraction model, extract the features of the face image to generate the face feature vector to be tested;

057: Input the pre-stored feature vector in the preset face information database to the reconfigurable computing unit to calculate the similarity between the tested face feature vector and the pre-stored feature vector, and determine the target feature vector according to the similarity; and

059: When the similarity between the target feature vector and the face feature vector to be tested is greater than a preset threshold, determine that the face authentication is successful.

In some implementations, the recognition module 13 is used to execute step 055, step 057 and step 059. That is, the identification module 13 is used to extract the features of the face image based on the preset feature extraction model to generate the face feature vector to be tested; input the pre-stored feature vector in the preset face information database to the reconfigurable computing unit , to calculate the similarity between the face feature vector to be tested and the pre-stored feature vector, and determine the target feature vector according to the similarity; and when the similarity between the target feature vector and the face feature vector to be tested is greater than a preset threshold, determine Face authentication is successful.

In some embodiments, the processor 20 is configured to execute step 055 , step 057 and step 059 . That is, the processor 20 is used to extract the features of the face image based on the preset feature extraction model to generate the face feature vector to be tested; input the pre-stored feature vector in the preset face information database to the reconfigurable computing unit , to calculate the similarity between the face feature vector to be tested and the pre-stored feature vector, and determine the target feature vector according to the similarity; and when the similarity between the target feature vector and the face feature vector to be tested is greater than a preset threshold, determine Face authentication is successful.

Specifically, when the processor 20 performs face recognition based on a face image, the processor 20 may extract features of the face image based on a preset feature extraction model, so as to generate a face feature vector to be tested.

Next, the processor 20 may input the pre-stored feature vectors in the preset face information database to the reconfigurable computing unit for calculation by the reconfigurable computing unit. Wherein, the reconfigurable computing unit is a (Reconfigurable Computing Unit, RCU) device. Calculating the similarity between the feature vector of the human face to be tested and the pre-stored feature vector by the RCU device can reduce the workload of the processor 20, thereby reducing the resource usage and time consumption used by the embedded device 100 in the face recognition process, thereby Satisfy the running requirements of the embedded device 100 on other tasks.

More specifically, after the RCU device calculates the similarity between the feature vector of the face to be tested and the pre-stored feature vector, the processor 20 can determine the target feature vector according to the similarity. When the similarity between the target feature vector and the face feature vector to be tested is greater than a preset threshold, the processor 20 may determine that the face authentication is successful. It can be understood that the target feature vector is a feature vector or a plurality of vectors among all the pre-stored feature vectors whose similarity with the feature vector of the face to be tested is greater than a preset threshold.

Wherein, the preset threshold may be any numerical value set artificially, such as 90%, 95%, 98% and so on. When the similarity between the target feature vector and the feature vector of the face to be tested is greater than a preset threshold, it indicates that the face image can be recognized and successfully unlocked.

Like this, processor 20 just can carry out the calculation of similarity according to RCU device to be measured face feature vector and prestored feature vector, avoid the work of processor 20, thereby reduce the resource occupation of face recognition work to embedded device 100, with The time consumption of face recognition is reduced, and the running requirements of the embedded device 100 on other tasks can be met.

Please refer to Fig. 2, Fig. 3 and Fig. 11, in some embodiments, step 057: input the pre-stored feature vector in the preset face information database to the reconfigurable computing unit to calculate the face feature vector and The similarity of the pre-stored feature vectors, and determining the target feature vector according to the similarity, also includes steps:

0571: According to the memory capacity of the reconfigurable computing unit, divide the pre-stored feature vectors in the face information database into multiple feature sets, so that the memory occupied by each feature set is less than the memory capacity;

0573: Input each feature set into the reconfigurable computing unit in turn to calculate the similarity between the face feature vector to be tested and the pre-stored feature vector of each feature set;

0575: Sorting according to the corresponding similarity of each pre-stored feature vector, to determine the pre-stored feature vector of the preset sort as the target feature vector.

In some implementations, the identification module 13 is used to perform step 0571 , step 0573 and step 0575 . That is, the recognition module 13 is used to divide the prestored feature vectors in the face information database into multiple feature sets according to the memory capacity of the reconfigurable computing unit, so that the memory occupied by each feature set is less than the memory capacity; feature set into the reconfigurable computing unit to calculate the similarity between the face feature vector to be tested and the pre-stored feature vector of each feature set; sort according to the similarity of each pre-stored feature vector to determine the preset sorting The pre-stored feature vector is the target feature vector.

In some embodiments, the processor 20 is configured to perform step 0571 , step 0573 and step 0575 . That is, the processor 20 is used to divide the prestored feature vectors in the face information database into multiple feature sets according to the memory capacity of the reconfigurable computing unit, so that the memory occupied by each feature set is less than the memory capacity; feature set into the reconfigurable computing unit to calculate the similarity between the face feature vector to be tested and the pre-stored feature vector of each feature set; sort according to the similarity of each pre-stored feature vector to determine the preset sorting The pre-stored feature vector is the target feature vector.

Specifically, when the reconfigurable computing unit (RCU device) calculates the similarity between the face feature vector to be tested and the pre-stored feature vector, it can first divide the pre-stored feature vector in the face information database according to the memory capacity of the RCU device. for multiple feature sets.

More specifically, taking N pre-stored eigenvectors as an example, the processor 20 can be divided into M parts (that is, divided into M feature sets) according to the memory capacity of the RCU device, and the number of pre-stored eigenvectors in each part Then it is (N+M-1)/M. In this way, it can be guaranteed that the memory capacity of the RCU device can be divided into M parts, and the difference in the number of pre-stored feature vectors in each part will not be large, and each feature set The occupied memory is less than the memory capacity.

Next, the processor 20 may sequentially input each feature set into the RCU device to calculate the similarity between the feature vector of the face to be tested and the pre-stored feature vector of each feature set. Among them, due to the small memory capacity of the RCU device, during the calculation process, the RCU device can only calculate the similarity between the pre-stored feature vector in one feature set and the face feature vector to be tested. Therefore, each feature set is input to the RCU in turn. In the device, it can be ensured that the RCU device can calculate the similarity between the pre-stored feature vectors in all feature sets and the face feature vector to be tested.

Finally, the processor 20 may obtain the similarity corresponding to each pre-stored feature vector, so as to sort the similarity of the pre-stored feature vectors, and use the preset sorted pre-stored feature vector as the target feature vector. Wherein, the processor 20 may perform sorting in a descending manner according to the similarity corresponding to each pre-stored feature vector. And the preset ordering can be 5, 10, 15, etc. For example, when the preset sorting is 5, the first five pre-stored feature vectors are the target feature vectors, that is, the five pre-stored feature vectors with the highest similarity are the target feature vectors.

In some embodiments, after the first feature set is input into the RCU device, the RCU device can calculate the similarity between each pre-stored feature vector in the first feature set and the face feature vector to be tested, so as to obtain K The similarity between the feature vector of the face to be tested and the pre-stored feature vector of each feature set. Among them, K is the similarity between the feature vector of the face to be tested and the pre-stored feature vector of each feature set. After sorting from large to small, the first K similarities, that is, the K similarities are K indivual.

Then, the processor inputs the second feature set to the RCU device, and the RCU can calculate the similarity between each pre-stored feature vector in the second feature set and the face feature vector to be tested, so as to obtain K1 similarities again, and with The K similarities obtained from the first feature set are compared and updated to obtain updated K2 similarities. Wherein, K1 and K2 are equal to K. It can be understood that the K2 similarities are the K2 similarities with larger similarities after the similarity calculation between the first feature set and the second feature set.

By analogy, when the processor inputs a feature set to the RCU device, it will calculate the similarity with the feature vector of the face to be tested, and compare and update it with the K similarities obtained from the previous feature set until the input is complete. All the feature sets, so as to obtain the larger K similarities among the similarities between the pre-stored feature vectors in all feature sets and the face feature vectors to be tested. That is, the pre-stored feature vectors corresponding to the K similarities are target feature vectors.

Please refer to Fig. 2, Fig. 3 and Fig. 12, in some embodiments, step 057: calculating the similarity between the feature vector of the human face to be tested and the pre-stored feature vector, also includes the steps of:

0577: Calculate the Euclidean distance between the feature vector of the face to be tested and the pre-stored feature vector; and

0579: Determine the similarity between the feature vector of the face to be tested and the pre-stored feature vector according to the Euclidean distance.

In some implementations, the identification module 13 is used to perform step 0577 and step 0579. That is, the identification module 13 is used to calculate the Euclidean distance between the feature vector of the human face to be tested and the pre-stored feature vector; and determine the similarity between the feature vector of the human face to be tested and the pre-stored feature vector according to the Euclidean distance.

In some embodiments, the processor 20 is configured to perform step 0577 and step 0579. That is, the processor 20 is used to calculate the Euclidean distance between the feature vector of the human face to be tested and the pre-stored feature vector; and determine the similarity between the feature vector of the human face to be tested and the pre-stored feature vector according to the Euclidean distance.

Specifically, when the RCU device calculates the similarity between the face feature vector to be tested and the pre-stored feature vector, it can also first calculate the Euclidean distance between the face feature vector to be tested and the pre-stored feature vector, and then determine the face feature to be tested according to the Euclidean distance The similarity between the vector and the pre-stored feature vector.

The calculation formulas of the Euclidean distance are shown in the following formula (1), formula (2) and formula (3):

Wherein, formula (1) is a calculation formula of two-dimensional space, formula (2) is a calculation formula of three-dimensional space, and formula (3) is a calculation formula of N-dimensional space, specifically, the face feature vector to be tested and the pre-stored feature vector The specific number is calculated by selecting the corresponding formula. ρ, d(x, y) are the specific values of the Euclidean distance, (x ₁ , y ₁ ) is the coordinate of the feature vector of the face to be tested, and (x ₂ , y ₁ ) is the coordinate of the pre-stored feature vector.

More specifically, the smaller the Euclidean distance, the greater the similarity between the feature vector of the face to be tested and the pre-stored feature vector. In this way, after the Euclidean distance between the feature vector of the face to be tested and the pre-stored feature vector is calculated, one or more of the pre-stored feature vectors with a greater similarity to the feature vector of the face to be tested can be determined according to the Euclidean distance. Stored feature vectors.

In one embodiment, the RCU device may sequentially calculate the Euclidean distance between the feature vectors of the face to be tested and the pre-stored feature vectors in each feature set, and store only K feature vectors of the face to be tested with smaller Euclidean distances. Next, the O test face vectors in each feature set are compared in turn to obtain O test face feature vectors with smaller Euclidean distances in all feature sets. Wherein, K and O are any positive integers greater than 0, and K and O may be equal.

In the face recognition method of the embodiment of the present application, when the user uses the embedded device 100 to perform face recognition, as shown in FIG. Face detection to get face images. Then align the face images according to the positions of the faces and key points of the faces, so that the face images are screened according to the preset face screening model, that is, quality inspection. When the quality reaches the preset condition, face feature extraction is performed on the face image; when the quality does not reach the preset condition, the embedded device 100 will end the face recognition and remind the user of the face recognition fail. After extracting the features of the face image according to the preset feature extraction model to generate the face feature vector to be tested, the processor 20 can calculate the feature vector to be tested and the pre-stored feature vector in the face information database according to the RCU device. The similarity of the eigenvectors, so as to obtain the target eigenvectors that are relatively similar to the pre-stored eigenvectors, that is, return the face similarity ID. And when the similarity between the target feature vector and the face feature vector to be tested meets a preset threshold, it is determined that the face authentication is successful.

Referring to FIG. 14 , the embodiment of the present application further provides a non-volatile computer-readable storage medium 300 containing a computer program 301 . When the computer program 301 is executed by the one or more processors 20, the one or more processors 20 are made to execute the face recognition method in any one of the above-mentioned embodiments.

For example, when the computer program 301 is executed by one or more processors 20, the processors 20 are made to perform the following face recognition methods:

01: Detect face information in the captured image to generate a face image;

03: Based on the preset face screening model, detect the quality of the face image; and

05: Based on the reconfigurable computing unit, calculate the similarity between the face feature vector of the face image whose quality meets the preset conditions and the preset face feature vector for face recognition.

For another example, when the computer program 301 is executed by one or more processors 20, the processors 20 are made to perform the following face recognition method:

07: Based on the preset face detection model, detect the position of the key points of the face in the face image; and

09: Align the face images according to the positions of the faces and key points of the faces, so that the pose of the target face of the aligned face images is adjusted to a preset pose.

In the description of this specification, descriptions with reference to the terms "certain embodiments", "in one example", "exemplarily" and the like mean that specific features, structures, materials or characteristics described in conjunction with the embodiments or examples are included in the present application In at least one embodiment or example of . In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the described specific features, structures, materials or characteristics may be combined in any suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples and features of different embodiments or examples described in this specification without conflicting with each other.

Any process or method descriptions in flowcharts or otherwise described herein may be understood to represent modules, segments or portions of code comprising one or more executable instructions for implementing specific logical functions or steps of the process , and the scope of preferred embodiments of the present application includes additional implementations in which functions may be performed out of the order shown or discussed, including in substantially simultaneous fashion or in reverse order depending on the functions involved, which shall It should be understood by those skilled in the art to which the embodiments of the present application belong.

Although the implementation of the present application has been shown and described above, it can be understood that the above-mentioned implementation is exemplary and should not be construed as limiting the application, and those skilled in the art can make the above-mentioned The embodiments are subject to changes, modifications, substitutions and variations.

Claims (18)

1. A face recognition method, comprising:

detecting face information in the acquired image to generate a face image;

detecting the quality of the face image based on a preset face screening model; and

And based on the reconfigurable computing unit, computing the similarity between the face feature vector of the face image with the quality reaching the preset condition and the preset face feature vector so as to conduct face recognition.

2. The face recognition method according to claim 1, wherein the face information includes a face position of a target face, and the detecting acquires the face information in the image to generate the face image includes:

detecting the face position in the acquired image based on a preset face detection model;

and generating the face image according to the face position.

3. The face recognition method according to claim 2, wherein the face information further includes a face key point, the face recognition method further comprising:

Detecting the positions of the face key points in the face image based on a preset face detection model;

aligning the face image according to the face position and the position of the face key point so that the posture of the target face of the aligned face image is adjusted to be a preset posture;

the detecting the quality of the face image includes:

and detecting the quality of the face image after alignment.

4. The face recognition method according to claim 2, wherein the detecting the face position in the captured image includes:

generating a plurality of candidate frames with preset sizes according to the sizes of the acquired images, and outputting the scores of the candidate frames;

determining the candidate frames with scores greater than a preset score as face frames;

acquiring the coincidence degree between any two face frames with coincidence parts, and determining the face frame with the highest score as a target face frame in the face frames with the coincidence degree larger than the preset coincidence degree;

and outputting the position of the target face frame as the face position.

5. The face recognition method according to claim 1, wherein the calculating, based on the reconfigurable calculating unit, the similarity between the face feature vector of the face image having the quality reaching the preset condition and the preset face feature vector to perform face recognition includes:

Extracting the features of the face image based on a preset feature extraction model to generate a face feature vector;

and establishing a face information database according to the face feature vector.

6. The face recognition method according to claim 1, wherein the calculating, based on the reconfigurable calculating unit, the similarity between the face feature vector of the face image having the quality reaching the preset condition and the preset face feature vector to perform face recognition includes:

extracting the features of the face image based on a preset feature extraction model to generate a face feature vector to be detected;

inputting a pre-stored feature vector in a preset face information database to the reconfigurable computing unit so as to calculate the similarity between the face feature vector to be detected and the pre-stored feature vector, and determining a target feature vector according to the similarity;

and under the condition that the similarity between the target feature vector and the face feature vector to be detected is larger than a preset threshold value, determining that the face authentication is successful.

7. The face recognition method according to claim 6, wherein inputting the pre-stored feature vector in the pre-set face information database to the reconfigurable computing unit to calculate the similarity between the face feature vector to be detected and the pre-stored feature vector, and determining the target feature vector according to the similarity, comprises:

Dividing the pre-stored feature vectors in the face information database into a plurality of feature sets according to the memory capacity of the reconfigurable computing unit, so that the memory occupied by each feature set is smaller than the memory capacity;

sequentially inputting each feature set into the reconfigurable computing unit to compute the similarity of the face feature vector to be tested and the pre-stored feature vector of each feature set;

and sorting according to the corresponding similarity of each pre-stored feature vector to determine the pre-stored feature vector with preset sorting as the target feature vector.

8. The method of claim 6, wherein said calculating the similarity between the feature vector of the face to be detected and the pre-stored feature vector comprises:

calculating Euclidean distance between the face feature vector to be detected and the pre-stored feature vector; and

And determining the similarity of the face feature vector to be detected and the pre-stored feature vector according to the Euclidean distance.

9. A face recognition device, comprising:

the generation module is used for detecting face information in the acquired image to generate a face image;

The first detection module is used for detecting the quality of the face image based on a preset face screening model; and

The recognition module is used for calculating the similarity between the face feature vector of the face image with the quality reaching the preset condition and the preset face feature vector based on the reconfigurable calculation unit so as to carry out face recognition.

10. An embedded device, comprising a processor for detecting face information in an acquired image to generate a face image; detecting the quality of the face image based on a preset face screening model; and calculating the similarity between the face feature vector of the face image with the quality reaching the preset condition and the preset face feature vector based on the reconfigurable calculating unit so as to carry out face recognition.

11. The embedded device of claim 10, wherein the face information includes a face position of a target face, and the processor is configured to detect the face position in the face image based on a preset face detection model; and generating the face image according to the face position.

12. The embedded device of claim 11, wherein the face information further comprises a face key point, and the processor is configured to detect a location of the face key point in the face image based on a preset face detection model; aligning the face image according to the face position and the position of the face key point so that the posture of the target face of the aligned face image is adjusted to be a preset posture; and detecting the quality of the face image after alignment.

13. The embedded device of claim 11, wherein the processor is configured to generate a plurality of candidate frames of a preset size according to the size of the captured image, and output a score for each of the candidate frames; determining the candidate frames with scores greater than a preset score as face frames; acquiring the coincidence degree between any two face frames with coincidence parts, and determining the face frame with the highest score as a target face frame in the face frames with the coincidence degree larger than the preset coincidence degree; and outputting the position of the target face frame as the face position.

14. The embedded device of claim 10, wherein the processor is configured to extract features of the face image based on a preset feature extraction model to generate an acquired face feature vector; and establishing a face information database according to the collected face feature vectors.

15. The embedded device of claim 10, wherein the processor is configured to extract features of the face image based on a preset feature extraction model to generate a face feature vector to be detected; inputting a pre-stored feature vector in a preset face information database to the reconfigurable computing unit so as to calculate the similarity between the face feature vector to be detected and the pre-stored feature vector, and determining a target feature vector according to the similarity; and under the condition that the similarity between the target feature vector and the face feature vector to be detected is larger than a preset threshold value, determining that the face authentication is successful.

16. The embedded device of claim 15, wherein the processor is configured to divide the pre-stored feature vectors in the face information database into a plurality of feature sets according to a memory capacity of the reconfigurable computing unit, such that memory occupied by each of the feature sets is smaller than the memory capacity; sequentially inputting each feature set into the reconfigurable computing unit to compute the similarity of the face feature vector to be tested and the pre-stored feature vector of each feature set; and sorting according to the corresponding similarity of each pre-stored feature vector to determine the pre-stored feature vector with preset sorting as the target feature vector.

17. The embedded device of claim 15, wherein the processor is further configured to calculate a euclidean distance between the face feature vector to be measured and the pre-stored feature vector; and determining the similarity of the face feature vector to be detected and the pre-stored feature vector according to the Euclidean distance.

18. A non-transitory computer readable storage medium, storing a computer program which, when executed by one or more processors, performs the face recognition method of any one of claims 1-8.

Images