Identity authentication method and system based on posture estimation
Through the identity authentication method based on posture estimation, deep learning and posture analysis technology are used to solve the accuracy and privacy issues in existing biometric technologies, and high-precision and anti-spoofing identity authentication is achieved.
Patent Information
- Application Number
- CN202510203077.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-24
- Publication Date
- 2025-06-10
AI Technical Summary
Existing biometric technologies have accuracy problems, such as fingerprint wear, facial recognition is affected by light and angle, voiceprint recognition is affected by time changes and noise, and privacy issues and risk of spoofing attacks.
The identity authentication method based on posture estimation is adopted, and the user image is captured in real time through the camera, and feature extraction and pose analysis is used to use deep learning models. The connection and pose information between key points is determined by combining non-maximum suppression, greedy algorithms and dynamic programming, and the pose feature vector is generated for identity authentication.
It realizes high-precision identity authentication, has the ability to resist spoof and forgery attacks, and does not require physical contact, is easy to use, and is suitable for a variety of environments.
Smart Images
Figure CN120124034A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of biometric technologies, and particularly to an identity authentication method and system based on body pose estimation. Background Art
[0002] Existing identity authentication methods include fingerprint recognition, face recognition, voiceprint recognition, and iris recognition, which have achieved remarkable results in ensuring secure and convenient identity verification. However, these technologies also have certain limitations.
[0003] For fingerprint recognition, when fingerprints are worn, damaged, or stained, it may affect its accuracy; in addition, the phenomenon of fingerprint forgery or replication cannot be ignored. For face recognition, face recognition systems are vulnerable to changes in light, angle, and expression, and may be deceived by masks or cosmetics. For voiceprint recognition, the voice may change over time due to factors such as disease or aging; at the same time, background noise will affect its accuracy. For iris recognition, iris recognition systems require special equipment, with relatively low convenience, and iris diseases or injuries may lead to a decline in recognition accuracy.
[0004] In addition, existing biometric technologies also have the following general deficiencies: privacy issues, biometric data is highly sensitive, and its collection and storage processes may raise concerns about privacy rights; spoofing attacks, attackers with certain skills may use various technical means to fraudulently attack biometric systems, such as forging fingerprints or making masks; acceptability, some people may not accept the application of biometric technologies due to concerns about privacy or other reasons. Summary of the Invention
[0005] The objective of the present invention is to provide an identity authentication method and system based on body pose estimation, which has the ability to resist spoofing and forgery attacks, high accuracy, and designs a user-friendly interface to improve the usability of the system.
[0006] The present invention adopts the following technical solution: an identity authentication method based on body pose estimation, comprising the following steps:
[0007] S1. Real-time capture the original image of the identity authentication user through a camera, preprocess the image, and use a deep learning model for feature extraction to obtain a feature map and an affinity field.
[0008] S2. Based on the feature map, use non-maximum suppression to obtain the candidate positions of the key points.
[0009] S3. Based on the affinity field, use a parser and a greedy algorithm to traverse the candidate positions of the key points, and use dynamic programming to determine the connections and pose information between the key points.
[0010] S4. Convert the pose information into a vector representation to obtain a pose feature vector, and store it in the database.
[0011] S5. Use the info function to calculate the distance ratio between the body features in the original image of step S1 and the pose feature vector of step S4 respectively, so as to obtain a similarity score. Compare the similarity score with a similarity threshold to achieve identity authentication.
[0012] Further, in step S1, obtaining the feature map and the affinity field includes the following:
[0013] Perform image scaling, image padding, and data formatting on the original image to obtain a preprocessed image.
[0014] Convert the preprocessed image into a tensor in the PyTorch framework. Check whether there is an NVIDIA GPU that supports CUDA. If so, use CUDA to move the tensor to the GPU, and input the tensor into the deep learning model. After forward propagation processing, obtain the feature map and the affinity field.
[0015] Further, image padding includes filling the scaled image with a set pixel value, then converting the image type to floating point type, and performing normalization processing to convert the pixel value range from [0, 255] to [-0.5, 0.5].
[0016] Further, in step S2, obtaining the candidate positions of the key points includes the following:
[0017] S201. Smooth the feature map using Gaussian filtering, and filter the smoothed feature map using the gaussian_filter function to obtain a first feature map.
[0018] S202. Process the first feature map using non-maximum suppression. Compare the heat map value of each pixel point in the feature map with the heat map values of the pixel points in its 3x3 neighborhood, and retain the pixel point with the largest local heat map value. If the heat map value of this point is greater than the set threshold, it is the candidate position of the key point.
[0019] Further, in step S3, use the parser and the greedy algorithm to traverse the candidate positions of the key points, sort the affinity integrals between the candidate points through the greedy algorithm, select the effective connection combination with the highest score, and exclude the key points with repeated allocation to determine the connections and pose information between the key points.
[0020] Further, in step S5, the info function is used to calculate the ratios of the distance from the nose to the left ear to the distance from the left ear to the left eye, and the distance from the nose to the neck to the distance from the neck to the left shoulder in the original image of step S1 and the pose feature vector of step S4, respectively. The alpha function is used to calculate the Euclidean distances of the two distance ratios in the original image of step S1 and the pose feature vector of step S4, respectively, so as to obtain a similarity score. The similarity score is compared with a similarity threshold. When the similarity score < the similarity threshold, it is determined as a legitimate user and unlocking is triggered; when the similarity score ≥ the similarity threshold, authentication is rejected and the locked state is maintained to implement identity authentication.
[0021] Further, the present invention also proposes an identity authentication system based on body pose estimation, including:
[0022] A feature extraction module, configured to capture the original image of the identity authentication user in real time through a camera, preprocess the image, and perform feature extraction using a deep learning model to obtain a feature map and an affinity field.
[0023] A pose feature vector acquisition module, configured to, based on the feature map, use non-maximum suppression to obtain candidate positions of key points; based on the affinity field, use a parser and a greedy algorithm to traverse the candidate positions of key points, and use dynamic programming to determine the connections and pose information between key points; convert the pose information into a vector representation to obtain a pose feature vector, and store it in a database.
[0024] An identity authentication module, configured to use the info function to calculate the distance ratios between body features in the original image of the feature extraction module and the pose feature vector of the pose feature vector acquisition module, respectively, so as to obtain a similarity score, and compare the similarity score with a similarity threshold to implement identity authentication.
[0025] Further, the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the identity authentication method based on body pose estimation are implemented.
[0026] Further, the present invention also proposes a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is run by a processor, the identity authentication method based on body pose estimation is executed.
[0027] Compared with the prior art, the present invention adopts the above technical solutions and has the following technical effects:
[0028] The present invention has a high level of security and is highly robust against fraud and forgery attacks. In terms of social engineering attacks, such as phishing or shoulder surfing, etc., it has a high resistance.
[0029] The present invention is non-invasive, does not require physical contact, and is easy to use. It does not require complex registration or authentication processes and can achieve remote authentication through the network or mobile devices.
[0030] The present invention is applicable to a variety of environments, including but not limited to indoor and outdoor scenarios.
[0031] The continuous authentication function of the present invention can continuously monitor the user's posture to prevent fraud or unauthorized access. The crowd recognition ability helps to identify specific individuals in a crowd and provides a higher level of privacy protection. Description of the Drawings
[0032] Figure 1 is the overall implementation flowchart of the present invention.
[0033] Figure 2 is the operation interface diagram in the embodiment of the present invention.
[0034] Figure 3 is the result diagram of authentication failure in the embodiment of the present invention.
[0035] Figure 4 is the result diagram of successful authentication in the embodiment of the present invention. Detailed Embodiment
[0036] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention and should not be used to limit the protection scope of the present invention.
[0037] To achieve the above object, the present invention proposes an identity authentication method based on body pose estimation, as Figure 1 shown, the specific steps are as follows:
[0038] S1. Real-time capture the original image of the identity authentication user through a camera, preprocess the image, and use a deep learning model for feature extraction to obtain a feature map and an affinity field; the specific content is:
[0039] Perform image scaling, image padding, and data formatting (ensure that the image data is continuous in memory) on the original image to obtain the preprocessed image.
[0040] Among them, the image is scaled according to a given scale factor. The image is padded by filling the scaled image with a set pixel value, then converting the image type to floating point and normalizing it so that the pixel value range is converted from [0, 255] to [-0.5, 0.5].
[0041] Each frame of the video stream is read in real time through the OpenCV library, and the preprocessed image is converted into a tensor in the PyTorch framework. It is checked whether there is an NVIDIA GPU that supports CUDA. If so, the tensor is moved to the GPU using CUDA, and the tensor is input into the deep learning model. After forward propagation processing, a feature map and an affinity field are obtained, which can more accurately locate the connections between key points. The use of CUDA enables real-time or near-real-time pose estimation, which is particularly important for video stream processing.
[0042] S2. Based on the feature map, non-maximum suppression is used to obtain the candidate positions of the key points; the specific content is as follows:
[0043] S201. The feature map is smoothed using Gaussian filtering to reduce noise, and the smoothed feature map is filtered using the gaussian_filter function to obtain the first feature map.
[0044] S202. The first feature map is processed using non-maximum suppression. The heatmap value of each pixel point in the feature map is compared with the heatmap values of the pixel points in its 3x3 neighborhood for local maximum. The pixel point with the largest local heatmap value is retained. If the heatmap value of this point is greater than the set threshold, it is the candidate position of the key point to identify different parts of the human body, such as the head, shoulders, wrists, etc.
[0045] S3. Based on the affinity field, the candidate positions of the key points are traversed using a parser and a greedy algorithm, and dynamic programming is used to determine the connections and pose information between the key points; the specific content is as follows:
[0046] The candidate positions of the key points are traversed using a parser and a greedy algorithm. The affinity integrals between the candidate points are sorted using the greedy algorithm, the valid connection combination with the highest score is selected, and the key points with repeated assignments are excluded to determine the connections and pose information between the key points. Such as from the shoulder to the elbow, from the elbow to the wrist, etc.
[0047] S4. The pose information is converted into a vector representation to obtain a pose feature vector, which is stored in the database.
[0048] S5. Use the info function to calculate the ratios of the distance from the nose to the left ear to the distance from the left ear to the left eye, and the distance from the nose to the neck to the distance from the neck to the left shoulder in the original image of step S1 and the pose feature vectors (stored_info1, stored_info2) of step S4 respectively. Use the alpha function to calculate the Euclidean distances of the two distance ratios in the original image of step S1 and the pose feature vectors of step S4 respectively, so as to obtain the similarity score. Compare the similarity score with the similarity threshold. When the similarity score < similarity threshold, it is determined as a legitimate user. When the similarity score ≥ similarity threshold, the authentication is rejected. The verification process information is recorded in the CSV log file in real time to achieve identity authentication.
[0049] In this embodiment, the value range of the similarity threshold is 0.1 - 0.5. 0.1 is more strict and 0.5 is more loose, which can balance security and user experience.
[0050] If the pose feature vector of user A is [0.8, 1.2], that of user B is [0.79, 1.18], and the threshold is 0.1, then the similarity score = (0.8 - 0.79)^2 + (1.2 - 1.18)^2 = 0.0001 + 0.0004 = 0.0005 < 0.1, so the verification passes.
[0051] If the pose feature vector of user A is [0.8, 1.2], and the feature vector of user C is [1.5, 0.6], and the threshold is 0.1, then the similarity score = (0.8 - 1.5)^2 + (1.2 - 0.6)^2 = 0.49 + 0.36 = 0.85 > 0.1, so the verification fails.
[0052] As Figure 2 shown, it is the home page of the identity authentication system based on body posture detection. There are mainly 5 buttons on the home page, namely "Register", "Stop Recognition", "Lock", "Sensitivity", and "Confirm". Click the left mouse button on the button, and it will jump to the corresponding function execution page.
[0053] Click the "Register" button, and the human body features in front of the screen can be stored in the database. Users can add the key point information of new users to the identity authentication system based on body posture detection through the registration process for future identity verification; click the "Stop Recognition" button, and the computer will turn off the camera and stop the recognition of body posture estimation; click the "Lock" button, and the interface will be locked again; click the "Exit" button, and the operation will be exited. The range of sensitivity is 0.1 - 0.5, and the smaller the number, the stricter the unlocking condition. After adjustment, click "OK" to start recognition according to this sensitivity.
[0054] The verification result will be fed back to the user interface. If the user is determined to be legitimate, the information "The lock has been opened" will be displayed on the interface. If the authentication is rejected, the information "The lock has been closed" will be displayed on the interface.
[0055] Continuous verification: The identity authentication system based on body posture detection will continuously execute the above steps in each frame of the video stream to verify the user's identity in real time. The user's posture information is used for identity authentication to ensure high accuracy and reliability.
[0056] If the system detects that there is no human body in front of the camera, the following operations will be performed:
[0057] 1. Stop identity authentication: The system will stop the identity authentication process.
[0058] 2. Enter the waiting state: The system will enter the waiting state, waiting for a human body to appear in front of the camera.
[0059] Collect the user's body posture data through the camera or other sensors, including information such as human body bone key points, joint angles, body contours, etc. Extract representative features from the collected body posture data to form the user's body posture feature code. When the user performs identity authentication, the current body posture feature code will be extracted and matched with the registered body posture feature codes in the database. If the match is unsuccessful, the lock will not be opened, and the interface will be as Figure 3 shown. If the match is successful, it will show that the lock has been opened, and the interface will be as Figure 4 shown.
[0060] Figure 3 、 4 In, purple line: the line connecting the right eye and the right ear. Dark purple line: the line connecting the right eye and the tip of the nose. Dark pink line: the line connecting the tip of the nose and the left eye. Pink line: the line connecting the left eye and the left ear. Blue line: the line connecting the tip of the nose and the chest. Yellow line; the line connecting the right shoulder and the right elbow. Red line: the line connecting the right shoulder and the chest. Orange line: the line connecting the chest and the left shoulder.
[0061] The embodiment of the present invention also proposes an identity authentication system based on body posture estimation, including a feature extraction module, a posture feature vector acquisition module, an identity authentication module, and a computer program that can run on a processor. It should be noted that each module in the above system corresponds to the specific steps of the method provided by the embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. For the technical details not described in detail in this embodiment, reference can be made to the method provided by the embodiment of the present invention.
[0062] An embodiment of the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. It should be noted that when the processor executes the computer program, it corresponds to the specific steps of the method provided by the embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method. For the technical details not described in detail in this embodiment, reference may be made to the method provided by the embodiment of the present invention.
[0063] An embodiment of the present invention further provides a computer-readable storage medium, which stores a computer program. It should be noted that when the computer program is run by a processor, it corresponds to the specific steps of the method provided by the embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method. For the technical details not described in detail in this embodiment, reference may be made to the method provided by the embodiment of the present invention.
[0064] The above are only the preferred embodiments of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.
Claims
1. An identity authentication method based on posture estimation, characterized in that: include: S1. Capture the original image of the identity authentication user in real time through the camera, pre-process the image, and use the deep learning model to extract features to obtain the feature map and affinity field; S2, based on the feature map, use non-maximum suppression to obtain the candidate positions of the key points; S3, based on the affinity field, use the parser and greedy algorithm to traverse the candidate positions of the key points, and use dynamic programming to determine the connection and posture information between the key points; S4, converting the posture information into a vector representation, obtaining a posture feature vector, and storing it in a database; S5. Use the info function to calculate the distance ratio between the body features in the original image of step S1 and the posture feature vector of step S4, and then obtain a similarity score, compare the similarity score with the similarity threshold, and realize identity authentication.
2. The identity authentication method based on posture estimation according to claim 1, characterized in that: In step S1, the feature map and affinity field are obtained including the following contents: Perform image scaling, image filling and data formatting on the original image to obtain a preprocessed image; The preprocessed image is converted into a tensor in the PyTorch framework. Check whether the device has a CUDA-supported NVIDIA GPU. If so, use CUDA to move the tensor to the GPU and input the tensor into the deep learning model. After forward propagation processing, the feature map and affinity field are obtained.
3. The identity authentication method based on posture estimation according to claim 2, characterized in that: Image filling includes filling the scaled image with the set pixel value, converting the image type to a floating point type, and performing normalization so that the pixel value range is converted from [0, 255] to [-0.5, 0.5].
4. The identity authentication method based on posture estimation according to claim 1, characterized in that: In step S2, the candidate positions of key points are obtained including the following: S201, using Gaussian filtering to smooth the feature map, and using the gaussian_filter function to filter the smoothed feature map to obtain a first feature map; S202. Process the first feature map using non-maximum suppression, compare the heat map values of each pixel in the feature map with those of the pixels in its 3x3 neighborhood for the local maximum value, retain the pixel with the largest local heat map value, and if the heat map value of the point is greater than the set threshold, it is a candidate position for the key point.
5. The identity authentication method based on posture estimation according to claim 1, characterized in that: In step S3, the candidate positions of the key points are traversed using the parser and the greedy algorithm, the affinity scores between the candidate points are sorted by the greedy algorithm, the valid connection combination with the highest score is selected, and the duplicated key points are excluded to determine the connection and posture information between the key points.
6. The identity authentication method based on posture estimation according to claim 1, characterized in that: In step S5, the info function is used to calculate the ratio of the distance from the nose to the left ear to the distance from the left ear to the left eye, and the ratio of the distance from the nose to the neck to the distance from the neck to the left shoulder in the original image of step S1 and the posture feature vector of step S4, and the alpha function is used to calculate the Euclidean distance of the two distance ratios in the original image of step S1 and the posture feature vector of step S4, and then the similarity score is obtained. The similarity score is compared with the similarity threshold. When the similarity score is less than the similarity threshold, it is determined to be a legitimate user and unlocking is triggered; when the similarity score is greater than or equal to the similarity threshold, the authentication is rejected and the locked state is maintained to achieve identity authentication.
7. A system for the identity authentication method based on posture estimation according to claim 1, characterized in that: include: The feature extraction module is used to capture the original image of the identity authentication user in real time through the camera, pre-process the image, and use the deep learning model to extract features to obtain the feature map and affinity field; The posture feature vector acquisition module is used to obtain the candidate positions of key points based on the feature map using non-maximum suppression; based on the affinity field, the candidate positions of key points are traversed using the parser and greedy algorithm, and the connection and posture information between key points are determined using dynamic programming; the posture information is converted into a vector representation to obtain the posture feature vector and store it in the database; The identity authentication module is used to use the info function to calculate the distance ratio between the original image of the feature extraction module and the posture feature vector of the posture feature vector acquisition module, and then obtain the similarity score, compare the similarity score with the similarity threshold, and realize identity authentication.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the identity authentication method based on body posture estimation described in any one of claims 1 to 6 are implemented.
9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the identity authentication method based on posture estimation according to any one of claims 1 to 6 is executed.