A high-precision face recognition method based on binocular vision

Abstract

The application relates to the technical field of face recognition, and particularly discloses a high-precision face recognition method based on binocular vision, which comprises the following steps: simultaneously collecting a left-view image and a right-view image of a face by using a binocular camera, pre-processing the images in the collected face image pair, and extracting face features from the pre-processed left-view image and right-view image; finding the corresponding relationship of the face feature points in the left-view image and the right-view image through a stereo matching algorithm, accurately estimating the posture of the face, calculating the parallax of the face feature points in the left-view image and the right-view image, obtaining the rotation and translation information of the face, thereby better coping with face images with different postures and reducing the influence of posture changes on the recognition result; since the image information of two visual angles is simultaneously acquired, the influence of light changes can be reduced to some extent by comparing and fusing the information of the two images, and the accuracy of face recognition is improved.

Description

Technical Field

[0001] This invention belongs to the field of face recognition technology, specifically relating to a high-precision face recognition method based on binocular vision. Background Technology

[0002] Facial recognition technology is widely used in access control systems. It extracts a set of vectors that can represent facial features, namely eigenfaces, by performing principal component analysis and other methods on facial images. The facial image to be recognized is then projected into the eigenface space, and its similarity to known facial feature vectors is calculated for recognition.

[0003] For example, patent application number 202410308502.4 discloses a face recognition method, including: acquiring an image to be recognized, wherein the image to be recognized contains a face to be recognized; performing pose estimation on the face to be recognized in the image to be recognized to determine the pose type of the face to be recognized, wherein the pose type is used to characterize the orientation of the face to be recognized; in response to the pose type being a preset type, reconstructing the image to be recognized using three-dimensional reconstruction technology to obtain a reconstructed image, wherein the preset type is used to characterize the orientation of the face to be recognized as a side orientation; performing face recognition on the face to be recognized in the reconstructed image to generate a recognition result. This invention solves the technical problem of poor accuracy in face recognition.

[0004] However, the aforementioned face recognition methods typically use a monocular camera to capture face images. Excessive or insufficient light, as well as uneven lighting distribution, can affect the quality of face images, making it difficult to accurately extract facial features. For example, in backlit environments, shadows may appear on the face, leading to the loss of some features and thus affecting the accuracy of face recognition. Therefore, we need to propose a high-precision face recognition method based on binocular vision to address these problems. This method should utilize a binocular camera to simultaneously acquire image information from two perspectives, mitigating the impact of lighting changes to some extent by comparing and fusing information from the two images, thereby improving the accuracy of face recognition. Summary of the Invention

[0005] The purpose of this invention is to provide a high-precision face recognition method based on binocular vision, which can simultaneously acquire image information from two perspectives using a binocular camera. To a certain extent, it can reduce the impact of lighting changes by comparing and fusing the information from the two images, thereby improving the accuracy of face recognition and solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A high-precision face recognition method based on binocular vision includes the following steps:

[0008] S1. Use a binocular camera to simultaneously capture left and right view images of a face, ensuring that a pair of face images with a certain parallax are acquired at the same time.

[0009] S2. Preprocess the images in the acquired face image pairs to improve image quality;

[0010] S3. Extract facial features from the preprocessed left and right view images, and at the same time use the characteristics of binocular vision to calculate the depth information of the face and extract depth-related features.

[0011] S4. Use a stereo matching algorithm to find the correspondence between facial feature points in the left and right view images and calculate the disparity to obtain the three-dimensional coordinate information of the face.

[0012] S5. Based on the three-dimensional coordinate information, perform pose estimation and correction on the face, unify the faces with different poses to the standard pose, and perform accurate feature matching.

[0013] S6. Match the extracted facial features to be identified with the known facial features in the database, calculate the similarity, and determine whether they are the same person based on the similarity threshold. If the similarity is greater than the threshold, the identification is successful; otherwise, the identification fails.

[0014] Preferably, the process for the binocular camera to acquire left and right view images of a face is as follows:

[0015] S11. Calibrate the binocular camera to determine its internal and external parameters. The internal parameters include focal length and principal point coordinates, while the external parameters include rotation and translation relationships.

[0016] S12. Fix the binocular camera at a position that can capture the left and right view images, so that the optical axes of the binocular camera are nearly parallel and the brightness and angle of the two cameras are consistent.

[0017] S13. Based on actual needs and camera performance, set the acquisition parameters of the binocular camera, including resolution, frame rate and exposure time.

[0018] S14. Use a synchronous triggering mechanism to trigger and control the two cameras to ensure that the two cameras capture images at the same time. When the cameras are triggered, the binocular cameras simultaneously capture the left and right view images of the face and transmit the images to the storage device for storage.

[0019] Preferably, in step S14, the synchronization triggering mechanism uses a synchronization signal generator to send a precise synchronization signal to the binocular cameras, so that the two cameras start image acquisition at the same time, ensuring that the acquired left view image and right view image are synchronized in time. A precise time synchronization algorithm is written to assist the synchronization signal generator in synchronization. Before image acquisition, the internal clocks of the two cameras are calibrated to keep the time references of the two cameras consistent. The parallax range of the centering image of the face image acquired by the binocular cameras is between 5 and 20 pixels.

[0020] Preferably, the preprocessing procedure for the images in the face image pair is as follows:

[0021] S21. Convert the left and right view images to grayscale to obtain grayscale images. The formula for grayscale conversion is:

[0022] ,in, R represents the pixel value of the grayscale image, and G and B represent the red, green, and blue color channel component values ​​of the pixel in the left or right view image, respectively. 、 and These are the weighting coefficients for the red, green, and blue color channels of pixels in the left or right view image, respectively.

[0023] S22. Perform mean filtering on the grayscale image to obtain the pre-filtered image. The mean filtering formula is:

[0024] ,in, These are the pixel values ​​of the image after preliminary filtering. represents the pixel value of the grayscale image to be filtered, and N is the total number of pixels in the neighborhood window;

[0025] S23. Perform median filtering on the initially filtered image to remove noise. The median filtering formula is as follows:

[0026] ,in, The coordinates after preliminary filtering are Image pixel values, Center coordinates The filtered pixel values;

[0027] S24. Normalize the filtered pixels to adjust the pixel range of the image to a fixed interval. The normalization formula is:

[0028] ,in, These are the normalized pixel values. These are the pixel values ​​of the original image after filtering. This represents the minimum pixel value of the filtered image. This represents the maximum pixel value of the filtered image.

[0029] Preferably, in step S3, the facial feature extraction process is as follows:

[0030] S31. Construct the Gaussian difference pyramid. The formula for constructing the Gaussian difference pyramid is:

[0031] ,in, For the function of the difference of Gaussian pyramid, These are the image coordinate values. The standard deviation is used to control the degree of Gaussian blur.

[0032] S32. Apply Gaussian blur to the image at different scales using the difference of Gaussian pyramid function, and calculate the difference between adjacent scale images to detect feature points at different scales.

[0033] S33. By performing scale space extremum point detection in the Gaussian difference pyramid, points with extrema in both scale space and image space are found as key points.

[0034] S34. Calculate the orientation of the key point and determine the main orientation of the key point based on the gradient orientation of the pixels around the key point.

[0035] S35. Taking the key point as the center, take... The feature region of a certain size is divided into multiple sub-regions, and the gradient magnitude and direction of the image within each sub-region are calculated.

[0036] S36. Based on the gradient magnitude and direction, construct the gradient direction histogram of the pixels in the sub-region, and then connect each histogram to form a feature descriptor.

[0037] S37. Based on the baseline distance and focal length of the binocular camera, the depth of corresponding points in the left and right view images is obtained by calculating the disparity of corresponding points in the left and right view images.

[0038] Preferably, in step S37, the depth calculation formula is:

[0039] ， in, Here, Q represents the depth value, and Q represents the baseline distance between the two cameras. The focal length of the binocular camera. This represents the difference in the x-coordinate of the same object in the left and right views.

[0040] Preferably, in step S4, the correspondence between facial feature points in the left and right view images is established by calculating the similarity between feature points extracted from the left and right views, then finding the point in the right view that best matches each feature point in the left view, thereby establishing the correspondence between the feature points of the left and right views, and then calculating the disparity and three-dimensional coordinates. The similarity calculation formula is as follows:

[0041] ,in, The similarity between feature points in the left and right views. For the dimension of the feature descriptor, For the vector of feature descriptors of the left view, For the vector of feature descriptors of the right view, This represents the number of feature descriptors.

[0042] Preferably, in step S5, pose estimation and correction are performed by minimizing the reprojection error to solve for the camera's pose parameters. The formula for calculating the reprojection error is:

[0043] ,in, E For reprojection error, For three-dimensional point coordinates, These are the coordinates of the projection of the 3D point onto the image. R For rotation matrix, T It is a translation vector. For projection function, This represents the number of three-dimensional points.

[0044] Preferably, in step S6, the method for matching the facial features to be identified with known facial features in the database is to first calculate the distance between the facial features to be identified and the known facial features, and then use a measurement model based on the distance to predict the difference between the facial features to be identified and the known facial features to confirm whether they match. The distance calculation formula is as follows:

[0045] ,in, The distance between the facial features to be identified and the known facial features. For the facial features to be identified, The facial features are known.

[0046] Preferably, the measurement model is implemented using the following formula:

[0047] Where L is the difference between the facial features to be identified and the known facial features, M is the boundary threshold, and Y represents the label of whether they match. When Y=1, it means that the facial features match, and when Y=0, it means that the facial features do not match. Indicates taking 0 and The maximum value in, when When it is less than M, When the facial features to be identified are matched with known facial features, When M is greater than or equal to M, The facial features to be identified do not match the known facial features.

[0048] The high-precision face recognition method based on binocular vision proposed in this invention has the following advantages compared with existing technologies:

[0049] 1. This invention acquires three-dimensional information of a face by simulating the human eyes' perception of depth using binocular vision. Compared to monocular vision, it adds depth dimension information, which helps to more accurately describe facial features and improve recognition accuracy. By accurately estimating the face's pose and calculating the disparity of facial feature points in the left and right view images, the rotation and translation information of the face is obtained, thus better handling face images with different poses and reducing the impact of pose changes on recognition results. Since image information from two perspectives is acquired simultaneously, the impact of illumination changes can be mitigated to some extent by comparing and fusing the information from the two images, thereby improving the accuracy of face recognition.

[0050] 2. This invention acquires face image pairs with a certain parallax and can dynamically adjust the exposure time through an automatic exposure control algorithm under different lighting conditions to ensure that the acquired face images have good contrast and brightness. Attached Figure Description

[0051] Figure 1 A flowchart according to an embodiment of the present invention is shown;

[0052] Figure 2 A flowchart illustrating the process of acquiring left and right view images of a human face using a binocular camera according to an embodiment of the present invention is shown.

[0053] Figure 3 A flowchart illustrating the preprocessing of images in a face image pair according to an embodiment of the present invention is shown;

[0054] Figure 4 A flowchart of the facial feature extraction process according to an embodiment of the present invention is shown. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] This invention provides, for example Figure 1-4 The high-precision face recognition method based on binocular vision shown includes the following steps:

[0057] S1. Simultaneously capture left and right view images of a face using a binocular camera to ensure that a face image pair with a certain parallax is acquired at the same time. The face image pair consists of a left view image and a right view image.

[0058] like Figure 2 As shown, the process of the binocular camera acquiring left and right view images of a face is as follows:

[0059] S11. Calibrate the binocular camera to determine its internal and external parameters. The internal parameters include focal length and principal point coordinates, while the external parameters include rotation and translation relationships.

[0060] S12. Fix the binocular camera at a position that can capture the left and right view images, so that the optical axes of the binocular camera are nearly parallel and the brightness and angle of the two cameras are consistent.

[0061] S13. Based on actual needs and camera performance, set the acquisition parameters of the binocular camera, including resolution, frame rate, and exposure time. Higher resolution can provide more detailed facial image information, but it will also increase the amount of data and processing time. The frame rate is determined according to the actual application scenario. Generally speaking, for facial recognition systems with high real-time requirements, the frame rate should be high enough to ensure image continuity. The exposure time setting should take into account the lighting conditions to avoid overexposure or underexposure. Under different lighting conditions, the exposure time can be dynamically adjusted through an automatic exposure control algorithm to ensure that the acquired facial image has good contrast and brightness.

[0062] S14. Use a synchronous triggering mechanism to trigger and control the two cameras to ensure that the two cameras capture images at the same time. When the cameras are triggered, the binocular cameras simultaneously capture the left and right view images of the face and transmit the images to the storage device for storage.

[0063] The synchronization triggering mechanism uses a synchronization signal generator to send a precise synchronization signal to the binocular cameras, enabling both cameras to start image acquisition at the same time. This greatly reduces the acquisition time difference and ensures that the acquired left and right view images are synchronized in time. A precise time synchronization algorithm is written to assist the synchronization signal generator in synchronization. Before image acquisition, the internal clocks of the two cameras are calibrated to ensure that the time bases of the two cameras are consistent. The parallax range of the centering image of the face acquired by the binocular cameras is between 5 and 20 pixels. For example, when the baseline distance is 60mm and the focal length is 8mm, the parallax range is 5-20 pixels.

[0064] S2. Preprocess the images in the acquired face image pairs to improve image quality;

[0065] like Figure 3 As shown, the preprocessing process for the images in the face image pair is as follows:

[0066] S21. Convert the left and right view images to grayscale to obtain grayscale images. The formula for grayscale conversion is:

[0067] ,in, R represents the pixel value of the grayscale image, and G and B represent the red, green, and blue color channel component values ​​of the pixel in the left or right view image, respectively. , and These are the weighting coefficients for the red, green, and blue color channels of pixels in the left or right view image, respectively. Commonly used weighting coefficients include... =0.299、 =0.587 and =0.114;

[0068] S22. Perform mean filtering on the grayscale image to obtain the pre-filtered image. The mean filtering formula is:

[0069] ,in, These are the pixel values ​​of the image after preliminary filtering. represents the pixel value of the grayscale image to be filtered, and N is the total number of pixels in the neighborhood window;

[0070] S23. Perform median filtering on the initially filtered image to remove noise. The median filtering formula is as follows:

[0071] ,in, The coordinates after preliminary filtering are Image pixel values, Center coordinates The filtered pixel values;

[0072] S24. Normalize the filtered pixels to adjust the pixel range of the image to a fixed interval. The normalization formula is:

[0073] ,in, These are the normalized pixel values. These are the pixel values ​​of the original image after filtering. This represents the minimum pixel value of the filtered image. This represents the maximum pixel value of the filtered image.

[0074] S3. Extract facial features from the preprocessed left and right view images, and at the same time use the characteristics of binocular vision to calculate the depth information of the face and extract depth-related features.

[0075] like Figure 4 As shown, the facial feature extraction process is as follows:

[0076] S31. Construct the Gaussian difference pyramid. The formula for constructing the Gaussian difference pyramid is:

[0077] ,in, For the function of the difference of Gaussian pyramid, These are the image coordinate values. This is the standard deviation used to control the degree of Gaussian blur, such as the initial standard deviation when the number of layers in the difference of Gaussians pyramid is 4. It is 1.6;

[0078] S32. Apply Gaussian blur to the image at different scales using the difference of Gaussian pyramid function, and calculate the difference between adjacent scale images to detect feature points at different scales.

[0079] S33. By performing scale space extremum point detection in the Gaussian difference pyramid, points with extrema in both scale space and image space are found as key points.

[0080] S34. Calculate the orientation of the key point and determine the main orientation of the key point based on the gradient orientation of the pixels around the key point.

[0081] S35. Taking the key point as the center, take... The feature region of a certain size is divided into multiple sub-regions, and the gradient magnitude and direction of the image within each sub-region are calculated.

[0082] S36. Based on the gradient magnitude and direction, construct the gradient direction histogram of the pixels in the sub-region, and then connect each histogram to form a feature descriptor.

[0083] S37. Based on the baseline distance and focal length of the binocular cameras, the depth of corresponding points in the left and right view images is obtained by calculating the disparity of corresponding points in the left and right view images. The depth calculation formula is as follows:

[0084] ， in, Here, Q represents the depth value, and Q represents the baseline distance between the two cameras. The focal length of the binocular camera. The difference in the x-coordinate of the same object in the left and right views;

[0085] S4. Use a stereo matching algorithm to find the correspondence between facial feature points in the left and right view images and calculate the disparity to obtain the three-dimensional coordinate information of the face.

[0086] The correspondence between facial feature points in the left and right view images is established by calculating the similarity between feature points extracted from the left and right views, then finding the point in the right view that best matches each feature point in the left view, thus establishing the correspondence between the feature points of the left and right views. Disparity and 3D coordinates are then calculated. The similarity calculation formula is as follows:

[0087] ,in, The similarity between feature points in the left and right views. For the dimension of the feature descriptor, For the vector of feature descriptors of the left view, For the vector of feature descriptors of the right view, The number of feature descriptors;

[0088] S5. Based on the three-dimensional coordinate information, perform pose estimation and correction on the face, unify the faces with different poses to the standard pose, and perform accurate feature matching.

[0089] Pose estimation and correction are performed by minimizing the reprojection error to solve for the camera's pose parameters. The formula for calculating the reprojection error is:

[0090] ,in, E For reprojection error, For three-dimensional point coordinates, Let R be the coordinates of the projection of the 3D point onto the image, R be the rotation matrix, and T be the translation vector. For projection function, The number of three-dimensional points;

[0091] S6. Match the extracted facial features to be identified with the known facial features in the database, calculate the similarity, and determine whether they are the same person based on the similarity threshold (the similarity threshold is determined by statistically analyzing the false recognition rate of the 95% confidence interval). If the similarity is greater than the threshold, the identification is successful; otherwise, the identification fails.

[0092] The method for matching the facial features to be identified with known facial features in the database is to first calculate the distance between the facial features to be identified and the known facial features, and then use a metric model based on the distance to predict the difference between the facial features to be identified and the known facial features, thus confirming whether they match. The distance calculation formula is as follows:

[0093] ,in, The distance between the facial features to be identified and the known facial features. For the facial features to be identified, Given known facial features;

[0094] The measurement model is implemented through the following formula:

[0095] Where L is the difference between the facial features to be identified and the known facial features, M is the boundary threshold, and Y represents the label of whether they match. When Y=1, it means that the facial features match, and when Y=0, it means that the facial features do not match. Indicates taking 0 and The maximum value in, when When it is less than M, When the facial features to be identified are matched with known facial features, When M is greater than or equal to M, The facial features to be identified do not match the known facial features;

[0096] By employing binocular vision to simulate the depth perception of human eyes, three-dimensional information of the face can be obtained. Compared to monocular vision, this adds depth dimension information, which helps to more accurately describe facial features and improve recognition accuracy. By accurately estimating the facial pose and calculating the disparity of facial feature points in the left and right view images, the rotation and translation information of the face can be obtained. This allows for better handling of facial images with different poses and reduces the impact of pose changes on recognition results. Since image information from two perspectives is acquired simultaneously, the impact of illumination changes can be mitigated to some extent by comparing and fusing the information from the two images, thus improving the accuracy of face recognition.

[0097] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A high-precision face recognition method based on binocular vision, characterized in that: Includes the following steps: S1. Using a binocular camera, simultaneously capture left and right view images of a face to ensure that a face image pair with parallax is acquired at the same time. A synchronous triggering mechanism is used to trigger and control the two cameras to ensure that the two cameras acquire images at the same time. When the camera is triggered, the binocular camera simultaneously captures left and right view images of the face. The synchronous triggering mechanism uses a synchronous signal generator to send a precise synchronous signal to the binocular camera, so that the two cameras start image acquisition at the same time, ensuring that the acquired left and right view images are synchronized in time. The parallax range of the face image pair acquired by the binocular camera is between 5 and 20 pixels. S2. Preprocess the images in the acquired face image pairs to improve image quality; S3. Extract facial features from the preprocessed left and right view images, and at the same time use the characteristics of binocular vision to calculate the depth information of the face and extract depth-related features. S4. Use a stereo matching algorithm to find the correspondence between facial feature points in the left and right view images and calculate the disparity to obtain the three-dimensional coordinate information of the face. S5. Based on the three-dimensional coordinate information, the pose of the face is estimated and corrected, and the faces with different poses are unified to the standard pose for accurate feature matching. The pose estimation and correction is to solve the camera pose parameters by minimizing the reprojection error. S6. Match the extracted facial features to be identified with the known facial features in the database, calculate the similarity, and determine whether they are the same person based on the similarity threshold. If the similarity is greater than the threshold, the identification is successful; otherwise, the identification fails. The process of acquiring left and right view images of a person's face using the binocular camera is as follows: S11. Calibrate the binocular camera to determine its internal and external parameters. The internal parameters include focal length and principal point coordinates, while the external parameters include rotation and translation relationships. S12. Fix the binocular camera at a position that can capture the left and right view images, so that the optical axes of the binocular camera are nearly parallel and the brightness and angle of the two cameras are consistent. S13. Based on actual needs and camera performance, set the acquisition parameters of the binocular camera, including resolution, frame rate and exposure time. S14. Use a synchronous triggering mechanism to trigger and control the two cameras to ensure that the two cameras capture images at the same time. When the cameras are triggered, the binocular cameras simultaneously capture the left and right view images of the face and transmit the images to the storage device for storage. The process for preprocessing images in face image pairs is as follows: S21. Convert the left and right view images to grayscale to obtain grayscale images. The formula for grayscale conversion is: ,in, R represents the pixel value of the grayscale image, and G and B represent the red, green, and blue color channel component values ​​of the pixel in the left or right view image, respectively. , and These are the weighting coefficients for the red, green, and blue color channels of pixels in the left or right view image, respectively. S22. Perform mean filtering on the grayscale image to obtain the pre-filtered image. The mean filtering formula is: ,in, These are the pixel values ​​of the image after preliminary filtering. represents the pixel value of the grayscale image to be filtered, and N is the total number of pixels in the neighborhood window; S23. Perform median filtering on the initially filtered image to remove noise. The median filtering formula is as follows: ,in, The coordinates after preliminary filtering are Image pixel values, Center coordinates The filtered pixel values; S24. Normalize the filtered pixels to adjust the pixel range of the image to a fixed interval. The normalization formula is: ,in, These are the normalized pixel values. These are the pixel values ​​of the original image after filtering. This represents the minimum pixel value of the filtered image. This represents the maximum pixel value of the filtered image.

2. The high-precision face recognition method based on binocular vision according to claim 1, characterized in that: In step S14, a precise time synchronization algorithm is written to assist the synchronization signal generator in synchronizing. Before image acquisition, the internal clocks of the two cameras are calibrated to ensure that the time references of the two cameras are consistent.

3. The high-precision face recognition method based on binocular vision according to claim 1, characterized in that: In step S3, the facial feature extraction process is as follows: S31. Construct the Gaussian difference pyramid. The formula for constructing the Gaussian difference pyramid is: ,in, For the function of the difference of Gaussian pyramid, These are the image coordinate values. The standard deviation is used to control the degree of Gaussian blur. S32. Apply Gaussian blur to the image at different scales using the difference of Gaussian pyramid function, and calculate the difference between adjacent scale images to detect feature points at different scales. S33. By performing scale space extremum point detection in the Gaussian difference pyramid, points with extrema in both scale space and image space are found as key points. S34. Calculate the orientation of the key point and determine the main orientation of the key point based on the gradient orientation of the pixels around the key point. S35. Taking the key point as the center, take a feature region of size m*m around the key point, divide the feature region into multiple sub-regions, and calculate the gradient magnitude and direction of the image in each sub-region. S36. Based on the gradient magnitude and direction, construct the gradient direction histogram of the pixels in the sub-region, and then connect each histogram to form a feature descriptor. S37. Based on the baseline distance and focal length of the binocular camera, the depth of corresponding points in the left and right view images is obtained by calculating the disparity of corresponding points in the left and right view images.

4. The high-precision face recognition method based on binocular vision according to claim 3, characterized in that: In step S37, the depth calculation formula is: ， in, Here, Q represents the depth value, and Q represents the baseline distance between the two cameras. The focal length of the binocular camera. This represents the difference in the x-coordinate of the same object in the left and right views.

5. The high-precision face recognition method based on binocular vision according to claim 1, characterized in that: In step S4, the correspondence between facial feature points in the left and right view images is established by calculating the similarity between feature points extracted from the left and right views, then finding the point in the right view that best matches each feature point in the left view, thus establishing the correspondence between the feature points of the left and right views, and subsequently calculating disparity and 3D coordinates. The similarity calculation formula is as follows: ,in, The similarity between feature points in the left and right views. For the vector of feature descriptors of the left view, For the vector of feature descriptors of the right view, This represents the number of feature descriptors.

6. The high-precision face recognition method based on binocular vision according to claim 1, characterized in that: In step S5, pose estimation and correction are achieved by minimizing the reprojection error to solve for the camera's pose parameters. The formula for calculating the reprojection error is as follows: ,in, E For reprojection error, For three-dimensional point coordinates, These are the coordinates of the projection of the 3D point onto the image. R For rotation matrix, T It is a translation vector. For projection function, This represents the number of three-dimensional points.

7. The high-precision face recognition method based on binocular vision according to claim 1, characterized in that: In step S6, the method for matching the facial features to be identified with known facial features in the database is to first calculate the distance between the facial features to be identified and the known facial features, and then use a measurement model based on the distance to predict the difference between the facial features to be identified and the known facial features to confirm whether they match. The distance calculation formula is as follows: ,in, The distance between the facial features to be identified and the known facial features. For the facial features to be identified, The facial features are known.

Images