Embedded intelligent door lock face recognition system capable of correcting face features

Abstract

The invention relates to the technical field of intelligent door locks, in particular to an embedded intelligent door lock face recognition system capable of correcting face features, which comprises an image acquisition module, a face detection module, a face alignment module, a feature extraction module, a feature adaptive correction module and a recognition matching module, the image acquisition module is utilized to acquire original image data, the face detection module locates a face position based on an image acquired by the image acquisition module, then the face alignment module carries out posture correction according to a detected face area, and the subsequent feature extraction module extracts feature vectors for the aligned face, so that the face recognition accuracy is improved. And the feature adaptive correction module compares and corrects the extracted feature vectors with standard features, and finally, the recognition matching module judges whether the matching is successful by using the corrected features, so that a coherent face recognition process is integrally formed to cope with different face states, and the aim of correcting the face features is fulfilled. And the high-efficiency recognizability of face recognition in different states is met.

Description

Technical Field

[0001] The invention belongs to the technical field of smart door locks, and in particular relates to an embedded smart door lock face recognition system capable of correcting facial features. Background Art

[0002] The face recognition method of smart door locks is to capture the user's facial image through the camera on the door lock, and use the background system to compare the collected facial features. When the comparison is successful, the lock is unlocked. In order to complete this step, the face data must be entered in advance in the background of the smart door lock. The face data is usually collected by collecting information such as the front and left and right sides of the face to ensure the accuracy of recognition; For example, the patent application number is 201910262383.2, which discloses an intelligent access control system based on fingerprint recognition and face recognition, including a power module, a main control module, a WIFI module, a camera module, a fingerprint recognition module, a touch screen module, a doorbell, a collision switch module for detecting the switch status of the door, a door lock and a drive module; the system combines Internet of Things technology, biotechnology, mobile Internet technology and embedded system technology, etc., and designs an intelligent access control system based on the concept of Internet of Things with multi-network integration technology, and integrates the "Internet +" technology well into the access control system, thereby realizing intelligent control of the door and information transmission. Compared with the existing intelligent access control system, the system greatly increases the anti-theft security performance, and applies virtual passwords, fingerprint recognition, face recognition and other technologies to better protect our lives and property. At the same time, it combines telephone communication functions with touch screen touch and display functions for user convenience, thereby improving the intelligence of the access control system.

[0003] However, the above-mentioned smart access control system extracts and compares facial features during face recognition, but does not correct the facial features. If the user's facial features change, such as having different expressions, hairstyles or makeup, the face may not be recognized, and the efficient recognition of face recognition in different states cannot be met. Therefore, we need to propose an embedded smart door lock face recognition system that can correct facial features to solve the above-mentioned problems, so that it can adapt to and correct the changes in facial features according to the changes in facial features to meet the efficient recognition of face recognition in different states. Summary of the invention

[0004] In view of the above problems, the present invention provides an embedded intelligent door lock face recognition system capable of correcting facial features, comprising: an image acquisition module, the image acquisition module is used to acquire facial image data so as to provide original facial image data; A face detection module, which is used to accurately locate the face position in the collected face image data, so that the feature correction can be focused on the detected face area; A face alignment module, which is used to regularize faces in different postures and angles into a unified standard posture to reduce the impact of relative position changes in different postures. The input and output ends of the face detection module are respectively connected to the output end of the image acquisition module and the input end of the face alignment module; A feature extraction module is used to convert the face image into a feature vector form that can be quantified and convenient for computer processing. When different face features change over time or in the environment, the feature adaptive correction module is used to perform comparative analysis and correct the difference to achieve the purpose of accurate recognition; A feature adaptive correction module, which is used to dynamically adjust changes in facial features so that the embedded smart door lock can still accurately identify the face when there are various changes. The input and output ends of the feature extraction module are respectively connected to the output end of the face alignment module and the input end of the feature adaptive correction module; The recognition and matching module is used to compare the corrected features with the existing user features, and finally determine whether the user is a legitimate user, verify the effectiveness of the entire feature correction process, and complete the goal of face recognition. Only by accurately correcting the features can the matching success rate be improved. The output end of the feature adaptive correction module is connected to the input end of the recognition and matching module.

[0005] Furthermore, the image acquisition module collects image information through the camera on the embedded smart door lock. The camera contains an image sensor and an analog-to-digital converter. The image sensor inside the camera converts the light entering the lens into an electrical signal based on the photoelectric effect, and then further converts the electrical signal into a digital image signal through the internal analog-to-digital converter, providing a steady stream of image frames for the face detection module.

[0006] Furthermore, the process of the face detection module accurately locating the face position is as follows: A1. Receive the image information collected by the image acquisition module and perform grayscale processing on the image information. The grayscale processing formula is as follows: , where R, G and B are the pixel values ​​of the red channel, green channel and blue channel in the color image respectively; A2. Convert the grayscale processed image into an integral image. The integral image conversion formula is as follows: , where x and y are the coordinate positions of the integral graph, is the input image, i and j are the position coordinates of the pixels in the rectangular area covered by the input image from the upper left corner (0, 0) to (x, y), is the converted integral image; A3. Calculate the eigenvalue D of the face based on the integral image. The calculation formula is as follows: ,in, The coordinates of the upper left corner of the white rectangle The corresponding integral graph is, The coordinates of the lower right corner of the white rectangle The corresponding integral graph is, The coordinates of the upper left corner of the black rectangle The corresponding integral graph is, The coordinates of the lower right corner of the black rectangle The corresponding integral graph; A4. Input the extracted features into the classifier to determine whether it is a face region. If it is a face region, it is transmitted to the face alignment module for further operation. If it is not a face region, it returns to step A1.

[0007] Furthermore, in step A4, the formula of the classifier is as follows: , where T is the total number of iterations, is the weight obtained from the tth round of training, is the value of the sample in the feature D dimension, is the prediction result obtained from the tth round of training, that is, when When ,when When , H(x) is the classifier calculation value, sign is the symbol feature, and satisfies the following conditions: like , then H(x)=+1, indicating that the area corresponding to the sample x is determined to be a face area; like , then H(x)=-1, indicating that the area corresponding to the sample x is not a face area.

[0008] Furthermore, the process of the face alignment module to regularize faces in different postures and angles to a unified standard posture is as follows: B1. Normalize the face image. The normalization formula is as follows: , where r is a eigenvector; Z is the normalized eigenvector, is the first Quantity, is the number of elements in the vector; B2, convolution operation is performed on the preprocessed face image and the convolution kernel; B3. Determine the left eye coordinates based on the image features after convolution processing ( , ) and right eye coordinates ( , ), the slope k of the binocular line is calculated according to the left eye coordinates and the right eye coordinates, and the calculation formula is as follows: : B4. Calculate the yaw angle based on the calculated slope k , the calculation formula is as follows: ; B5. According to the calculated yaw angle , construct the affine transformation matrix A; B6. According to the affine transformation matrix A, the pixel coordinates of the face image are affine transformed to obtain the coordinates of rotating and translating the face to the standard posture. The coordinate affine transformation formula is as follows: , where x and y are the pixel coordinates of the face image, and are the transformed coordinates respectively.

[0009] Furthermore, in step B2, the convolution operation formula is as follows: , where m and p are the sizes of the convolution kernels, is the convolution kernel, j are the coordinates of the convolution kernel size, is the input image, For The convolution output value.

[0010] Furthermore, in step B5, the affine transformation matrix A constructed is in the following form: ,in, = , is the translation in the x direction, is the translation in the y direction.

[0011] Furthermore, the process of converting the face image into a feature vector form by the feature extraction module is as follows: C1. Use Gaussian kernels of different scales to perform convolution operations with face images to construct a scale space to simulate the appearance of face images at different resolutions. The convolution operation formula is as follows: ,in, is the face image after face alignment, is the Gaussian convolution scale, and are the coordinates of the pixels of the input face image, for Gaussian function at scale, is the scale space image; C2. In the constructed scale space, local extreme points are detected by comparing the size of each pixel with its neighboring pixels at the same scale and the neighboring pixels at the corresponding positions of the upper and lower adjacent scales; C3, with the extreme point as the center, calculate the gradient amplitude Q(x,y) and direction , the calculation formula is as follows: ; ,in, is the pixel value of the image after convolution, and and are the coordinates of the image pixels after convolution processing, For horizontal The pixel value of the coordinate shifted 1 unit position to the right, For horizontal The pixel value of the coordinate shifted 1 unit position to the left, For vertical direction The pixel value of the coordinate moving up by 1 unit position, For vertical direction The pixel value by which the coordinates are shifted down by 1 unit position.

[0012] Furthermore, the process of dynamic adjustment of the feature adaptive correction module is as follows: D1. Select two features extracted by the feature extraction module to calculate the distance. The calculation formula is as follows: ,in, and are two features extracted by the feature extraction module, o is the total number of features extracted by the feature extraction module, and d is the distance between the two features; D2. Set the distance threshold between two features and two weight coefficients and ; D3. Calculate the dynamic weight coefficient based on the calculated two feature distances d , the calculation formula is as follows: , And when When α= ; D4. According to the calculated dynamic weight coefficient Adjust the eigenvector. The eigenvector adjustment formula is as follows: ,in, is the corrected eigenvector, is the currently extracted facial feature vector, It is the standard feature vector of the user stored in the embedded smart door lock system.

[0013] Furthermore, the process of the identification and matching module comparing the corrected features with the existing user features is as follows: E1. Set the similarity threshold G between the modified features and the existing user features according to the face recognition requirements of the smart door lock; E2. Compare the similarity between the existing user features and the modified features Calculation, the calculation formula is as follows: , where M is the corrected feature vector, N is the existing user feature vector, is the modulus of the corrected eigenvector, is the modulus of the existing user feature vector; E3. Calculate the similarity value Comparing with the similarity threshold G, there are two situations: when , it is judged that the face recognition is successful and the door is opened; when , the face recognition is judged to be unsuccessful, the door opening operation is not triggered, and a prompt of recognition failure is given.

[0014] The beneficial effects of the present invention are: 1. The present invention uses the image acquisition module to obtain original image data through the cooperation of an image acquisition module, a face detection module, a face alignment module, a feature extraction module, a feature adaptive correction module and a recognition and matching module. The face detection module locates the face position based on the image acquired by the image acquisition module, and then the face alignment module performs posture correction according to the detected face area. The subsequent feature extraction module extracts feature vectors for the aligned face, and the feature adaptive correction module compares and corrects the extracted feature vectors with standard features. Finally, the recognition and matching module uses the corrected features to determine whether the match is successful. A coherent face recognition process is formed as a whole to cope with different face states, thereby achieving the purpose of correcting face features and meeting the efficient recognition of face recognition under different states.

[0015] 2. The present invention can obtain face image data in different postures by cooperating with the face alignment module of the face detection module, and perform affine coordinate transformation on the face pixel coordinates in different postures and angles, so that the face features extracted subsequently have better consistency and comparability, which is convenient for face recognition operation.

[0016] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A system block diagram according to an embodiment of the present invention is shown; Figure 2 A flowchart of accurately locating a face position by a face detection module according to an embodiment of the present invention is shown; Figure 3 A flowchart of how a face alignment module according to an embodiment of the present invention adjusts faces in different postures and angles to a unified standard posture is shown; Figure 4 A flowchart of a feature adaptive correction module for dynamic adjustment according to an embodiment of the present invention is shown; Figure 5 A flowchart of comparing the modified features with the existing user features according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] An embodiment of the present invention provides an embedded smart door lock face recognition system capable of correcting facial features, comprising an image acquisition module, a face detection module, a face alignment module, a feature extraction module, a feature adaptive correction module and a recognition and matching module, wherein an input end and an output end of the face detection module are respectively connected to an output end of the image acquisition module and an input end of the face alignment module, an input end and an output end of the feature extraction module are respectively connected to an output end of the face alignment module and an input end of the feature adaptive correction module, and an output end of the feature adaptive correction module is connected to an input end of the recognition and matching module.

[0021] The image acquisition module is used to acquire face image data so as to provide original face image data; The image acquisition module collects image information through the camera on the embedded smart door lock. The camera contains an image sensor and an analog-to-digital converter. The image sensor (such as a common CMOS sensor or CCD sensor) inside the camera converts the light entering the lens into an electrical signal based on the photoelectric effect, and then further converts the electrical signal into a digital image signal through the internal analog-to-digital converter (ADC). The digital image signal is transmitted frame by frame to the system memory in a certain format (such as a common YUV format or RGB format, etc.) through an interface (such as a USB interface, an SPI interface, etc.) connecting the camera to the main control board of the embedded smart door lock in a certain data transmission protocol (such as a batch transmission method specified in the USB protocol), thereby realizing continuous updating of image data and providing a continuous source of image frames for the subsequent face detection module; The face detection module is used to accurately locate the face position in the collected face image data, so that the feature correction can be focused on the detected face area to avoid erroneous processing of irrelevant areas such as the background; The process of the face detection module to accurately locate the face position is as follows: A1. Receive the image information collected by the image acquisition module and perform grayscale processing on the image information. The grayscale processing formula is as follows: , where R, G and B are the pixel values ​​of the red channel, green channel and blue channel in the color image respectively; A2. Convert the grayscale processed image into an integral image. The integral image conversion formula is as follows: , where x and y are the coordinate positions of the integral graph, is the input image, i and j are the position coordinates of the pixels in the rectangular area covered by the input image from the upper left corner (0, 0) to (x, y), is the converted integral image; A3. Calculate the eigenvalue D of the face based on the integral image. The calculation formula is as follows: ,in, The coordinates of the upper left corner of the white rectangle The corresponding integral graph is, The coordinates of the lower right corner of the white rectangle The corresponding integral graph is, The coordinates of the upper left corner of the black rectangle The corresponding integral graph is, The coordinates of the lower right corner of the black rectangle The corresponding integral graph; A4, input the extracted features into the classifier to determine whether it is a face region, if it is a face region, transmit it to the face alignment module for further operation, if it is not a face region, return to step A1; The formula of the classifier is as follows: , where T is the total number of iterations, is the weight obtained from the tth round of training, is the value of the sample in the feature D dimension, is the prediction result obtained from the tth round of training, that is, when When ,when When , H(x) is the classifier calculation value, sign is the symbol feature, and satisfies the following conditions: like , then H(x)=+1, indicating that the area corresponding to the sample x is determined to be a face area; like , then H(x)=-1, indicating that the area corresponding to the sample x is not a face area; The face alignment module is used to regularize faces in different postures and angles into a unified standard posture to reduce the impact of relative position changes in different postures and improve the accuracy of face recognition; The process of the face alignment module to regularize faces in different postures and angles to a unified standard posture is as follows: B1. Normalize the face image. The normalization formula is as follows: , where r is a eigenvector; Z is the normalized eigenvector, is the first Quantity, is the number of elements in the vector; B2. Perform convolution operation on the preprocessed face image and the convolution kernel. The convolution operation formula is as follows: , where m and p are the sizes of the convolution kernels, is the convolution kernel, j are the coordinates of the convolution kernel size, is the input image, For The convolution output value; B3. Determine the left eye coordinates based on the image features after convolution processing ( , ) and right eye coordinates ( , ), the slope k of the binocular line is calculated according to the left eye coordinates and the right eye coordinates, and the calculation formula is as follows: : B4. Calculate the yaw angle based on the calculated slope k , the calculation formula is as follows: ; B5. According to the calculated yaw angle , construct the affine transformation matrix A, the constructed affine transformation matrix A is in the following form: ,in, = , is the translation in the x direction, is the translation in the y direction; B6. According to the affine transformation matrix A, the pixel coordinates of the face image are affine transformed to obtain the coordinates of rotating and translating the face to the standard posture. The coordinate affine transformation formula is as follows: , where x and y are the pixel coordinates of the face image, and are the transformed coordinates respectively; By performing affine coordinate transformation on the facial pixel coordinates in different postures and angles, the facial features extracted subsequently have better consistency and comparability, which facilitates the operation of face recognition. The feature extraction module is used to convert the face image into a feature vector form that can be quantified and convenient for computer processing. When different face features change over time or in the environment, the feature adaptive correction module is used to perform comparative analysis and correct the difference to achieve the purpose of accurate recognition; The process of converting the face image into feature vector form by the feature extraction module is as follows: C1. Use Gaussian kernels of different scales to perform convolution operations with face images to construct a scale space to simulate the appearance of face images at different resolutions. The convolution operation formula is as follows: ,in, is the face image after face alignment, is the Gaussian convolution scale, and are the coordinates of the pixels of the input face image, for Gaussian function at scale, is the scale space image; C2. In the constructed scale space, local extreme points are detected by comparing the size of each pixel with its neighboring pixels at the same scale and the neighboring pixels at the corresponding positions of the upper and lower adjacent scales; C3, with the extreme point as the center, calculate the gradient amplitude Q(x,y) and direction , the calculation formula is as follows: ; ,in, is the pixel value of the image after convolution, and and are the coordinates of the image pixels after convolution processing, For horizontal The pixel value of the coordinate shifted 1 unit position to the right, For horizontal The pixel value of the coordinate shifted 1 unit position to the left, For vertical direction The pixel value of the coordinate moving up by 1 unit position, For vertical direction The pixel value of the coordinates moved down by 1 unit position; The feature adaptive correction module is used to dynamically adjust changes in facial features, so that the embedded smart door lock can still accurately identify various changes in the face (such as subtle facial changes, different expressions, and changes in facial light and shadow under different lighting), thereby ensuring the efficient recognition of face recognition.

[0022] The process of dynamic adjustment of the feature adaptive correction module is as follows: D1. Select two features extracted by the feature extraction module to calculate the distance. The calculation formula is as follows: ,in, and are two features extracted by the feature extraction module, o is the total number of features extracted by the feature extraction module, and d is the distance between the two features; D2. Set the distance threshold between two features and two weight coefficients and ; D3. Calculate the dynamic weight coefficient based on the calculated two feature distances d , the calculation formula is as follows: , And when season = ; D4. According to the calculated dynamic weight coefficient Adjust the eigenvector. The eigenvector adjustment formula is as follows: ,in, is the corrected eigenvector, is the currently extracted facial feature vector, is the standard feature vector of the user stored in the embedded smart door lock system; By comparing features and adjusting feature vectors, facial features can be adaptively corrected, so that the face recognition system can still accurately identify various reasonable changes in the face, thus ensuring its efficiency and accuracy.

[0023] The identification and matching module is used to compare the corrected features with the existing user features, and finally determine whether the user is a legitimate user, verify the effectiveness of the entire feature correction process, and complete the goal of face recognition. Only by accurately correcting the features can the matching success rate be improved, ensuring that the embedded smart door lock can correctly identify the door opener in different situations; The process of the identification and matching module comparing the corrected features with the existing user features is as follows: E1. Set the similarity threshold G between the modified features and the existing user features according to the face recognition requirements of the smart door lock; E2. Compare the similarity between the existing user features and the modified features Calculation, the calculation formula is as follows: , where M is the corrected feature vector, N is the existing user feature vector, is the modulus of the corrected eigenvector, is the modulus of the existing user feature vector; E3. Calculate the similarity value Comparing with the similarity threshold G, there are two situations: when , it is judged that the face recognition is successful and the door is opened; when , the face recognition is judged to be unsuccessful, the door opening operation is not triggered, and a prompt indicating recognition failure is given; Through the cooperation of image acquisition module, face detection module, face alignment module, feature extraction module, feature adaptive correction module and recognition and matching module, the image acquisition module is used to obtain original image data, the face detection module locates the face position based on the image obtained by the image acquisition module, and then the face alignment module performs posture correction according to the detected face area. The subsequent feature extraction module extracts feature vectors for the aligned face, the feature adaptive correction module compares and corrects the extracted feature vectors with standard features, and finally the recognition and matching module uses the corrected features to determine whether the match is successful. A coherent face recognition process is formed as a whole to cope with different face states, thereby achieving the purpose of correcting face features and meeting the efficient recognition of face recognition under different states.

[0024] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An embedded smart door lock face recognition system capable of correcting facial features, characterized in that: include: An image acquisition module, wherein the image acquisition module is used to acquire facial image data so as to provide original facial image data; A face detection module, which is used to accurately locate the face position in the collected face image data, so that the feature correction can be focused on the detected face area; A face alignment module, which is used to regularize faces in different postures and angles into a unified standard posture to reduce the impact of relative position changes in different postures. The input and output ends of the face detection module are respectively connected to the output end of the image acquisition module and the input end of the face alignment module; A feature extraction module is used to convert the face image into a feature vector form that can be quantified and convenient for computer processing. When different face features change over time or in the environment, the feature adaptive correction module is used to perform comparative analysis and correct the difference to achieve the purpose of accurate recognition; A feature adaptive correction module, which is used to dynamically adjust changes in facial features so that the embedded smart door lock can still accurately identify the face when there are various changes. The input and output ends of the feature extraction module are respectively connected to the output end of the face alignment module and the input end of the feature adaptive correction module; The recognition and matching module is used to compare the corrected features with the existing user features, and finally determine whether the user is a legitimate user, verify the effectiveness of the entire feature correction process, and complete the goal of face recognition. Only by accurately correcting the features can the matching success rate be improved. The output end of the feature adaptive correction module is connected to the input end of the recognition and matching module.

2. The embedded smart door lock face recognition system capable of correcting facial features according to claim 1, characterized in that: The image acquisition module collects image information through the camera on the embedded smart door lock. The camera contains an image sensor and an analog-to-digital converter. The image sensor inside the camera converts the light entering the lens into an electrical signal based on the photoelectric effect, and then further converts the electrical signal into a digital image signal through the internal analog-to-digital converter, providing a steady stream of image frames for the face detection module.

3. The embedded intelligent door lock face recognition system capable of correcting facial features according to claim 2, characterized in that: The process of the face detection module to accurately locate the face position is as follows: A1. Receive the image information collected by the image acquisition module and perform grayscale processing on the image information. The grayscale processing formula is as follows: , where R, G and B are the pixel values ​​of the red channel, green channel and blue channel in the color image respectively; A2. Convert the grayscale processed image into an integral image. The integral image conversion formula is as follows: , where x and y are the coordinate positions of the integral graph, is the input image, i and j are the position coordinates of the pixels in the rectangular area covered by the input image from the upper left corner (0, 0) to (x, y), is the integral image after transformation; A3. Calculate the eigenvalue D of the face based on the integral image. The calculation formula is as follows: ,in, The coordinates of the upper left corner of the white rectangle The corresponding integral graph is, The coordinates of the lower right corner of the white rectangle The corresponding integral graph is, The coordinates of the upper left corner of the black rectangle The corresponding integral graph is, The coordinates of the lower right corner of the black rectangle The corresponding integral graph; A4. Input the extracted features into the classifier to determine whether it is a face region. If it is a face region, it is transmitted to the face alignment module for further operation. If it is not a face region, it returns to step A1.

4. The embedded intelligent door lock face recognition system capable of correcting facial features according to claim 3 is characterized in that: In step A4, the classifier formula is as follows: , where T is the total number of iterations, is the weight obtained from the tth round of training, is the value of the sample in the feature D dimension, is the prediction result obtained from the tth round of training, that is, when When ,when When , H(x) is the classifier calculation value, sign is the symbol feature, and satisfies the following conditions: like , then H(x)=+1, indicating that the area corresponding to the sample x is determined to be a face area; like , then H(x)=-1, indicating that the area corresponding to the sample x is not a face area.

5. The embedded intelligent door lock face recognition system capable of correcting facial features according to claim 4 is characterized in that: The process of the face alignment module to regularize faces in different postures and angles to a unified standard posture is as follows: B1. Normalize the face image. The normalization formula is as follows: , where r is a eigenvector; Z is the normalized eigenvector, is the first Quantity, is the number of elements in the vector; B2, convolution operation is performed on the preprocessed face image and the convolution kernel; B3. Determine the left eye coordinates based on the image features after convolution processing ( , ) and right eye coordinates ( , ), the slope k of the binocular line is calculated according to the left eye coordinates and the right eye coordinates, and the calculation formula is as follows: ： B4. Calculate the yaw angle based on the calculated slope k , the calculation formula is as follows: ； B5. According to the calculated yaw angle , construct the affine transformation matrix A; B6. According to the affine transformation matrix A, the pixel coordinates of the face image are affine transformed to obtain the coordinates of rotating and translating the face to the standard posture. The coordinate affine transformation formula is as follows: , where x and y are the pixel coordinates of the face image, and are the transformed coordinates respectively.

6. The embedded intelligent door lock face recognition system capable of correcting facial features according to claim 5, characterized in that: In step B2, the convolution operation formula is as follows: , where m and p are the sizes of the convolution kernels, is the convolution kernel, j are the coordinates of the convolution kernel size, is the input image, For The convolution output value.

7. The embedded intelligent door lock face recognition system capable of correcting facial features according to claim 6, characterized in that: In step B5, the affine transformation matrix A constructed is in the following form: ,in, = , is the translation in the x direction, is the translation in the y direction.

8. The embedded intelligent door lock face recognition system capable of correcting facial features according to claim 7, characterized in that: The process of converting the face image into feature vector form by the feature extraction module is as follows: C1. Use Gaussian kernels of different scales to perform convolution operations with face images to construct a scale space to simulate the appearance of face images at different resolutions. The convolution operation formula is as follows: ,in, is the face image after face alignment, is the Gaussian convolution scale, and are the coordinates of the pixels of the input face image, for Gaussian function at scale, is the scale space image; C2. In the constructed scale space, local extreme points are detected by comparing the size of each pixel with its neighboring pixels at the same scale and the neighboring pixels at the corresponding positions of the upper and lower adjacent scales; C3, with the extreme point as the center, calculate the gradient amplitude Q(x,y) and direction , the calculation formula is as follows: ； ,in, is the pixel value of the image after convolution, and and are the coordinates of the image pixels after convolution processing, For horizontal The pixel value of the coordinate shifted 1 unit position to the right, For horizontal The pixel value of the coordinate shifted 1 unit position to the left, For vertical direction The pixel value of the coordinate moving up by 1 unit position, For vertical direction The pixel value by which the coordinates are shifted down by 1 unit position.

9. The embedded intelligent door lock face recognition system capable of correcting facial features according to claim 8, characterized in that: The process of dynamic adjustment of the feature adaptive correction module is as follows: D1. Select two features extracted by the feature extraction module to calculate the distance. The calculation formula is as follows: ,in, and are two features extracted by the feature extraction module, o is the total number of features extracted by the feature extraction module, and d is the distance between the two features; D2. Set the distance threshold between two features and two weight coefficients and ; D3. Calculate the dynamic weight coefficient based on the calculated two feature distances d , the calculation formula is as follows: ， And when season = ; D4. According to the calculated dynamic weight coefficient Adjust the eigenvector. The eigenvector adjustment formula is as follows: ,in, is the corrected eigenvector, is the currently extracted facial feature vector, It is the standard feature vector of the user stored in the embedded smart door lock system.

10. The embedded intelligent door lock face recognition system capable of correcting facial features according to claim 9, characterized in that: The process of the identification and matching module comparing the corrected features with the existing user features is as follows: E1. Set the similarity threshold G between the modified features and the existing user features according to the face recognition requirements of the smart door lock; E2. Compare the similarity between the existing user features and the modified features Calculation, the calculation formula is as follows: , where M is the corrected feature vector, N is the existing user feature vector, is the modulus of the corrected eigenvector, is the modulus of the existing user feature vector; E3. Calculate the similarity value Comparing with the similarity threshold G, there are two situations: when , it is judged that the face recognition is successful and the door is opened; when , the face recognition is judged to be unsuccessful, the door opening operation is not triggered, and a prompt of recognition failure is given.

Images