Access control system based on face recognition

Abstract

The invention discloses an access control system based on face recognition, relates to the technical field of biological recognition, and solves the problem that feature parameters are difficult to extract according to face images and iris images. Secondly, face comparison detection is difficult to carry out according to the extracted feature parameters of the face image; then, it is difficult to analyze the iris image of the incomer according to the face comparison result, and it is difficult to analyze the abnormal behavior according to the monitoring image; and finally, the risk coefficient is difficult to comprehensively analyze, and the risk level of the access control system is difficult to assess. By analyzing the face image and the iris image, the identity verification accuracy of the access control system is improved; abnormal behaviors of strangers in a monitoring image are analyzed by using a deep learning model, and a risk coefficient is comprehensively analyzed according to an identification failure rate and a potential abnormal coefficient, so that the risk level of the access control system is evaluated.

Description

Technical Field

[0001] The invention belongs to the technical field of biometric identification, and in particular relates to an access control system based on face recognition. Background Art

[0002] With the rapid development of computer technology and the Internet, traditional access control verification methods can no longer meet people's needs. "Cardless" and "Security" have become research hotspots in the field of information security. The access control system is an effective measure to solve the entrances and exits of some important departments and achieve security management. It is mainly used in various confidential departments, such as banks, hotels, armories, offices and other places; the access control system is a system for controlling entrances and exits, aiming to improve security and management efficiency.

[0003] The existing technology has the following problems: first, it is difficult to extract feature parameters based on face images and iris images; second, it is difficult to perform face comparison detection based on the feature parameters of the extracted face images; then, it is difficult to analyze the iris image of the entrant based on the face comparison results, and it is difficult to analyze abnormal behavior based on the monitoring images; finally, it is difficult to comprehensively analyze the risk coefficient and it is difficult to evaluate the risk level of the access control system. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an access control system based on face recognition, which is used to solve the following technical problems:

[0005] First, it is difficult to extract feature parameters based on facial images and iris images; second, it is difficult to perform facial comparison detection based on the extracted feature parameters of facial images; then, it is difficult to analyze the iris image of the entrant based on the facial comparison results, and it is difficult to analyze abnormal behavior based on the surveillance images; finally, it is difficult to comprehensively analyze the risk factor and to evaluate the risk level of the access control system.

[0006] To solve the above problems, the first aspect of the present invention provides an access control system based on face recognition, comprising the following modules:

[0007] Image acquisition module: collects face images, iris images and surveillance images of entrants;

[0008] Database module: stores the facial images and iris templates of authorized personnel;

[0009] Image processing module: pre-processing based on the collected face images, iris images and surveillance images, including: denoising and image enhancement;

[0010] Feature extraction module: extracts feature parameters of face images and iris images based on the collected and preprocessed face images and iris images;

[0011] Image analysis module: perform face comparison detection based on the feature parameters of the extracted face image; analyze the iris image of the entrant based on the face comparison result; analyze the abnormal behavior of strangers based on the collected and processed surveillance images; analyze and calculate the recognition failure rate and potential abnormality coefficient;

[0012] Comprehensive analysis module: Comprehensively analyze and calculate the risk coefficient, and evaluate the risk level of the access control system based on the calculation results.

[0013] As a further solution of the present invention: extracting characteristic parameters of the face image and the iris image according to the collected and pre-processed face image and iris image, including the following steps:

[0014] Based on the collected and pre-processed face image, the face position is located from the face image and the face area is cropped by using a face detection algorithm; the feature points of the face area are extracted by using a feature point detection algorithm;

[0015] Extracting feature parameters based on the feature points of the detected facial area of ​​the human face, including: feature parameters of geometric shapes, color feature parameters and texture feature parameters; the feature parameters of the geometric shapes include: the X coordinate and Y coordinate of the center point of the eye, the X coordinate and Y coordinate of the center point of the nose, and the X coordinate and Y coordinate of the center point of the mouth; the color feature parameters include: the RGB value of the skin color of the human face; the texture feature parameters include: the length and depth of the wrinkles on the human face, and the area and number of the spots;

[0016] Iris features are extracted from the processed iris image, including shape features, texture features and color features, and an iris template is generated based on the extracted features.

[0017] As a further solution of the present invention: performing face comparison detection according to the feature parameters of the extracted face image includes the following steps:

[0018] Extract features from the facial images of authorized personnel in the database, and perform normalization and cleaning based on the extracted feature parameters of authorized personnel and entrants;

[0019] According to the extracted and processed feature parameters, the feature parameters of the facial area of ​​the entrant are compared with the feature parameters of the facial image of the authorized person in the database;

[0020] By analyzing the formula:

[0021] Get the geometric shape feature parameter similarity J, color feature parameter similarity Y and texture feature parameter similarity W; where (X 1k , Y 1k) represents the coordinate point of the geometric shape feature parameter of the authorized person; (X 2k , Y 2k ) represents the coordinate point of the geometric shape feature parameter of the entrant, when k=1, it represents the coordinate of the center point of the eye, when k=2, it represents the coordinate of the center point of the nose, and when k=3, it represents the coordinate of the center point of the mouth; R1, G1 and B1 represent the RGB values ​​of the facial skin color of the authorized person, and R2, G2 and B2 represent the RGB values ​​of the facial skin color of the entrant; M L 、M S 、M A and M Q They represent the wrinkle length, wrinkle depth, spot area and number of spots of the entrant, respectively. L 、N S 、N A and N Q Respectively indicate the length of wrinkles, the depth of wrinkles, the area of ​​spots and the number of spots of the authorized personnel;

[0022] By analyzing the formula: R = (J*ε1+Y*ε2+W*ε3)×100%

[0023] Obtain the facial similarity R of the human face; wherein J represents the similarity of geometric shape feature parameters, Y represents the similarity of color feature parameters, W represents the similarity of texture feature parameters, and ε1, ε2 and ε3 represent the weight coefficients of the similarity of geometric shape feature parameters, color feature parameters and texture feature parameters respectively;

[0024] Compare the facial similarity with the threshold, analyze and determine whether the face of the entrant matches the facial image of the authorized person in the database, and obtain the face comparison result; the comparison result includes: face comparison result match success and match failure;

[0025] When the face comparison result is a successful match, the entrant is an authorized person and the access control is opened; when the face comparison result is a failed match, the face image of the entrant is re-collected and the face comparison is performed again to determine whether the face comparison result is a successful match; when the re-collected face image of the entrant matches successfully, the access control is opened, otherwise the iris image of the entrant is analyzed to determine whether the entrant is an authorized person.

[0026] As a further solution of the present invention: analyzing the iris image of the entrant according to the face comparison result, comprising the following steps:

[0027] When the re-collected face image of the entrant fails to match, the iris image of the entrant is analyzed; an iris template is generated based on the iris features extracted from the iris image;

[0028] The iris features include: shape features, texture features and color features; wherein the feature values ​​of the shape features include: the inner and outer circle radii and eccentricity of the iris; the feature values ​​of the texture features include: decomposing the iris image into LL channel, HL channel and LH channel by Haar wavelet transform, and calculating the mean and variance of the energy of the decomposed LL channel, HL channel and LH channel to obtain the feature values ​​of the texture features; the feature values ​​of the color features include: the mean and standard deviation of the RGB values ​​of the iris area;

[0029] The extracted eigenvalues ​​are combined into a eigenvector, which is a multidimensional array, in which each element corresponds to an eigenvalue, and the eigenvalue of the i-th dimension represents the eigenvalue corresponding to the i-th feature, and the constructed eigenvector is used as an iris template;

[0030] Compare the iris template of the entrant with the iris template of the authorized person stored in the database;

[0031] By analyzing the formula:

[0032]

[0033] Get the iris template similarity H; where, A represents the dot product of the iris template of the entrant and the iris template of the authorized person. i represents the eigenvalue of the iris template of the entrant in the i-th dimension, B i represents the feature value of the iris template of the authorized person in the i-th dimension, n represents the number of dimensions of the iris template, and the number of dimensions represents the number of features;

[0034] The iris template similarity is compared with the threshold. When the iris template similarity exceeds the threshold, it means the verification is successful; otherwise, it means the verification fails. When the verification fails, the entrant is marked as a stranger.

[0035] As a further solution of the present invention: analyzing the abnormal behavior of strangers based on the collected and processed surveillance images includes the following steps:

[0036] Based on the collected and processed surveillance images, abnormal behaviors of strangers in the surveillance images are annotated, and the annotated abnormal behaviors are input into the deep learning model for training; the abnormal behaviors include: wandering, following others into the surveillance area, staying in the surveillance area for a long time, and covering the face;

[0037] According to the trained deep learning model, the access control system based on face recognition collects and processes surveillance images in real time, inputs them into the trained deep learning model to analyze the abnormal behavior of strangers and outputs the analysis results and confidence levels.

[0038] As a further solution of the present invention: analyzing and calculating the recognition failure rate and the potential abnormality coefficient, the following steps are included:

[0039] By analyzing the formula:

[0040] Get the recognition failure rate F; where S represents the total number of recognitions, S f Indicates the total number of recognition failures;

[0041] By analyzing the formula:

[0042] Get the potential abnormal coefficient Q; where C j represents the confidence of the deep learning model output of the i-th abnormal behavior, δ j represents the impact weight of the i-th abnormal behavior, and n represents the total number of abnormal behaviors.

[0043] As a further solution of the present invention: comprehensively analyzing and calculating the risk coefficient, and evaluating the risk level of the access control system according to the calculation result, including the following steps:

[0044] By analyzing the formula: β=(F*W1+Q*W2)×100%

[0045] The risk coefficient β is obtained; wherein F represents the recognition failure rate, Q represents the potential abnormality coefficient, and W1 and W2 represent the weights of the recognition failure rate and the potential abnormality coefficient respectively; a threshold is set according to the risk coefficient to evaluate the risk level of the access control system;

[0046] When β≤20%, the risk level is low risk;

[0047] When 20%<β<50%, the risk level is moderate;

[0048] When β ≥ 50%, the risk level is high.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The present invention realizes dual biometric identification by analyzing face images and iris images, greatly improving the accuracy of identity authentication of the access control system; face and iris, as unique biometric features of the human body, effectively prevent illegal intrusion; through multiple biometric identification, the user's passage efficiency is improved, which helps to reduce the user's waiting time;

[0051] The present invention realizes automatic monitoring and analysis by using a deep learning model to analyze the abnormal behavior of strangers in surveillance images, thus reducing the possibility of manual intervention and misjudgment.

[0052] The present invention comprehensively analyzes the risk factor by utilizing the recognition failure rate and the potential abnormality coefficient, and evaluates the risk level of the access control system, which helps to understand the overall risk status of the access control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only 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.

[0054] Figure 1 It is a schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION

[0055] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than 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.

[0056] See also Figure 1 The first embodiment of the present invention provides a door access control system based on face recognition, comprising the following modules:

[0057] Image acquisition module: collects face images, iris images and surveillance images of entrants;

[0058] Database module: stores the facial images and iris templates of authorized personnel;

[0059] Image processing module: pre-processing based on the collected face images, iris images and surveillance images, including: denoising and image enhancement;

[0060] Feature extraction module: extracts feature parameters of face images and iris images based on the collected and preprocessed face images and iris images;

[0061] Image analysis module: perform face comparison detection based on the feature parameters of the extracted face image; analyze the iris image of the entrant based on the face comparison result; analyze the abnormal behavior of strangers based on the collected and processed surveillance images; analyze and calculate the recognition failure rate and potential abnormality coefficient;

[0062] Comprehensive analysis module: Comprehensively analyze and calculate the risk coefficient, and evaluate the risk level of the access control system based on the calculation results.

[0063] Specifically, the facial image and iris template data of the authorized person are imported into the database module to ensure that the data is accurate. When someone enters the access control area, the image acquisition module is started to collect the facial image through the camera. The iris acquisition device is used to collect the iris image of the entrant. At the same time, the image of the monitoring area is collected by the monitoring camera for subsequent abnormal behavior analysis. The collected facial image, iris image and monitoring image are denoised to remove noise and interference in the image. The image clarity and contrast are improved by image enhancement technology. The feature extraction algorithm in the image processing module is used to extract feature parameters from the preprocessed facial image, including: the position and shape of the key points of the facial contour, eyes and nose. Similarly, feature parameters are extracted from the preprocessed iris image, including iris shape, texture and color. The extracted facial feature parameters are compared with the facial image of the authorized person in the database to determine whether the entrant is an authorized person. If the facial comparison fails, the iris image of the entrant is further analyzed and compared with the iris template in the database to ensure the authenticity and accuracy of the identity. The abnormal behavior of strangers is detected by using the behavior analysis algorithm in the monitoring image and image processing module. The recognition failure rate and potential abnormality coefficient are calculated based on the results of face comparison and iris analysis; the risk coefficient of the access control system is calculated based on the face comparison results, iris analysis results, abnormal behavior analysis results, recognition failure rate and potential abnormality coefficient, and the risk level of the access control system is evaluated based on the value of the risk coefficient.

[0064] In one embodiment of the present invention, extracting characteristic parameters of the face image and the iris image according to the collected and pre-processed face image and iris image includes the following steps:

[0065] Based on the collected and pre-processed face image, the face position is located from the face image and the face area is cropped by using a face detection algorithm; the feature points of the face area are extracted by using a feature point detection algorithm;

[0066] Extracting feature parameters based on the feature points of the detected facial area of ​​the human face, including: feature parameters of geometric shapes, color feature parameters and texture feature parameters; the feature parameters of the geometric shapes include: the X coordinate and Y coordinate of the center point of the eye, the X coordinate and Y coordinate of the center point of the nose, and the X coordinate and Y coordinate of the center point of the mouth; the color feature parameters include: the RGB value of the skin color of the human face; the texture feature parameters include: the length and depth of the wrinkles on the human face, and the area and number of the spots;

[0067] Iris features are extracted from the processed iris image, including shape features, texture features and color features, and an iris template is generated based on the extracted features.

[0068] Specifically, a face detection algorithm is used to locate the position of a face from a preprocessed face image; based on the positioning result, a face region is cut out for subsequent feature point detection and feature parameter extraction. Feature points are extracted from the cut face region using a feature point detection algorithm. Feature points include key points of key parts of eyes, nose, and mouth. Based on the detected feature points, geometric shape feature parameters are extracted by calculating the X-coordinates and Y-coordinates of the center points of eyes, nose, and mouth. The RGB value of the skin color of the face is extracted as a color feature parameter. Image processing techniques including edge detection and morphological operations are used to analyze wrinkles, and a spot detection algorithm is used to perform spot analysis. The wrinkle and spot texture feature parameters of the face are extracted, including wrinkle length and depth, spot area and number. An iris positioning algorithm is used to locate the iris region, and the iris region is cut out according to the positioning result. The shape parameters of the outer circle and inner circle of the iris are analyzed. The iris texture is extracted using a texture analysis method, and the mean and standard deviation of the color feature RGB value of the iris region are calculated. The extracted iris features include shape features, texture features, and color features, and an iris template is generated.

[0069] In one embodiment of the present invention, face comparison detection is performed based on the extracted feature parameters of the face image, including the following steps:

[0070] Extract features from the facial images of authorized personnel in the database, and perform normalization and cleaning based on the extracted feature parameters of authorized personnel and entrants;

[0071] According to the extracted and processed feature parameters, the feature parameters of the facial area of ​​the entrant are compared with the feature parameters of the facial image of the authorized person in the database;

[0072] By analyzing the formula:

[0073] Get the geometric shape feature parameter similarity J, color feature parameter similarity Y and texture feature parameter similarity W; where (X 1k , Y 1k ) represents the coordinate point of the geometric shape feature parameter of the authorized person; (X 2k , Y 2k ) represents the coordinate point of the geometric shape feature parameter of the entrant, when k=1, it represents the coordinate of the center point of the eye, when k=2, it represents the coordinate of the center point of the nose, and when k=3, it represents the coordinate of the center point of the mouth; R1, G1 and B1 represent the RGB values ​​of the facial skin color of the authorized person, and R2, G2 and B2 represent the RGB values ​​of the facial skin color of the entrant; M L 、M S 、M A and M QThey represent the wrinkle length, wrinkle depth, spot area and number of spots of the entrant, respectively. L 、N S 、N A and N Q Respectively indicate the length of wrinkles, the depth of wrinkles, the area of ​​spots and the number of spots of the authorized personnel;

[0074] By analyzing the formula: R = (J*ε1+Y*ε2+W*ε3)×100%

[0075] Obtain the facial similarity R of the human face; wherein J represents the similarity of geometric shape feature parameters, Y represents the similarity of color feature parameters, W represents the similarity of texture feature parameters, and ε1, ε2 and ε3 represent the weight coefficients of the similarity of geometric shape feature parameters, color feature parameters and texture feature parameters respectively;

[0076] Compare the facial similarity with the threshold, analyze and determine whether the face of the entrant matches the facial image of the authorized person in the database, and obtain the face comparison result; the comparison result includes: face comparison result match success and match failure;

[0077] When the face comparison result is a successful match, the entrant is an authorized person and the access control is opened; when the face comparison result is a failed match, the face image of the entrant is re-collected and the face comparison is performed again to determine whether the face comparison result is a successful match; when the re-collected face image of the entrant matches successfully, the access control is opened, otherwise the iris image of the entrant is analyzed to determine whether the entrant is an authorized person.

[0078] Specifically, a face feature extraction algorithm is used to extract geometric shape feature parameters, color feature parameters and texture feature parameters from the authorized person's face image and the entrant's face image stored in the database. The extracted feature parameters are normalized to ensure that different feature parameters have the same scale when compared. Invalid or abnormal feature parameters are cleaned to improve the accuracy of the comparison. The Euclidean distance between the coordinate points of the entrant and the authorized person is calculated based on the coordinates of the center points of the eyes, the center points of the nose and the center points of the mouth to obtain the geometric shape feature parameter similarity. The color feature parameter similarity is obtained by calculating the difference in RGB values ​​between the entrant and the authorized person. For wrinkle length, depth, and spot area and quantity features, the texture feature parameter similarity is calculated using the Euclidean distance. By combining the geometric shape feature parameter similarity, color feature parameter similarity and texture feature parameter similarity and their respective weight coefficients, the facial similarity of the face is calculated. Among them, the weight coefficient of the geometric shape feature parameter similarity is 0.5, the weight coefficient of the color feature parameter similarity is 0.3, and the weight coefficient of the texture feature parameter similarity is 0.2. This formula comprehensively considers the contribution of multiple feature parameters and improves the accuracy of the comparison. By setting the comparison threshold to 50%, it is determined whether the face of the entrant matches the authorized person in the database, where the value of the weight coefficient and the comparison threshold are dynamically adjusted according to historical data and actual conditions. The calculated face similarity is compared with the set comparison threshold. When the face similarity is less than or equal to the threshold, the face comparison result is considered to be a successful match; otherwise, it is considered to be a match failure; when the face comparison result is a successful match, the entrant is considered to be an authorized person, and the system controls the door to open. When the face comparison result is a match failure, the face image of the entrant is re-collected and the face comparison is performed again. If the re-collected image still fails to match, the iris recognition process is turned to analyze and judge the iris image of the entrant.

[0079] In one embodiment of the present invention, analyzing the iris image of the entrant according to the face comparison result includes the following steps:

[0080] When the re-collected face image of the entrant fails to match, the iris image of the entrant is analyzed; an iris template is generated based on the iris features extracted from the iris image;

[0081] The iris features include: shape features, texture features and color features; wherein the feature values ​​of the shape features include: the inner and outer circle radii and eccentricity of the iris; the feature values ​​of the texture features include: decomposing the iris image into LL channel, HL channel and LH channel by Haar wavelet transform, and calculating the mean and variance of the energy of the decomposed LL channel, HL channel and LH channel to obtain the feature values ​​of the texture features; the feature values ​​of the color features include: the mean and standard deviation of the RGB values ​​of the iris area;

[0082] The extracted eigenvalues ​​are combined into a eigenvector, which is a multidimensional array, in which each element corresponds to an eigenvalue, and the eigenvalue of the i-th dimension represents the eigenvalue corresponding to the i-th feature, and the constructed eigenvector is used as an iris template;

[0083] Compare the iris template of the entrant with the iris template of the authorized person stored in the database;

[0084] By analyzing the formula:

[0085]

[0086] Get the iris template similarity H; where, A represents the dot product of the iris template of the entrant and the iris template of the authorized person. i represents the eigenvalue of the iris template of the entrant in the i-th dimension, B i represents the feature value of the iris template of the authorized person in the i-th dimension, n represents the number of dimensions of the iris template, and the number of dimensions represents the number of features;

[0087] The iris template similarity is compared with the threshold. When the iris template similarity exceeds the threshold, it means the verification is successful; otherwise, it means the verification fails. When the verification fails, the entrant is marked as a stranger.

[0088] Specifically, when the re-collected face image of the entrant does not match the face image of the authorized person stored in the database, the iris recognition process is triggered. The inner and outer circle radii of the iris are calculated by edge detection and circle fitting algorithm, and the eccentricity is obtained by the ratio of the inner and outer circle radii of the iris or the eccentricity based on ellipse fitting. The iris image is decomposed into three channels of LL low frequency-low frequency, HL high frequency-low frequency and LH low frequency-high frequency using Haar wavelet transform. The mean and variance of the energy of each channel are calculated, and these values ​​will be used as the eigenvalues ​​of the texture feature. The RGB values ​​are extracted in the iris area, and the RGB value mean and standard deviation are calculated as the eigenvalues ​​of the color feature. The eigenvalues ​​of the extracted shape features, texture features and color features are combined into a feature vector. The feature vector is a multidimensional array, each element corresponds to an eigenvalue, and the number of dimensions of the array is equal to the number of features. The constructed feature vector is stored as an iris template for subsequent comparison. The iris template is a compact, easy to store and compare feature vector, which represents the uniqueness of the iris of the entrant. The similarity is obtained by taking the dot product of the iris template of the entrant and the iris template of the authorized person, divided by the product of their respective module lengths and adding the average of the sum of the Euclidean distances of the iris templates. The calculated iris template similarity is compared with the threshold, which is 50%. The threshold is adjusted dynamically according to the actual situation and historical data. When the iris template similarity exceeds the threshold, the verification is considered successful, that is, the entrant is an authorized person. When the iris template similarity does not exceed the threshold, the verification is considered failed, that is, the entrant is a stranger. When the iris verification is successful, the system controls the door to open and allows the entrant to enter.

[0089] In one embodiment of the present invention, analyzing abnormal behavior of strangers based on collected and processed surveillance images includes the following steps:

[0090] Based on the collected and processed surveillance images, abnormal behaviors of strangers in the surveillance images are annotated, and the annotated abnormal behaviors are input into the deep learning model for training; the abnormal behaviors include: wandering, following others into the surveillance area, staying in the surveillance area for a long time, and covering the face;

[0091] According to the trained deep learning model, the access control system based on face recognition collects and processes surveillance images in real time, inputs them into the trained deep learning model to analyze the abnormal behavior of strangers and outputs the analysis results and confidence levels.

[0092] Specifically, first, high-quality surveillance cameras are deployed to ensure that clear and complete surveillance images can be captured. The collected surveillance images are preprocessed, including denoising, contrast enhancement, and resolution adjustment, to improve image quality and analysis accuracy. Abnormal behaviors of strangers are manually marked from the collected and processed surveillance images. Abnormal behaviors include but are not limited to: wandering, following others into the surveillance area, staying in the surveillance area for a long time, and covering the face; wandering refers to walking around in the surveillance area repeatedly without obvious purpose; following others into the surveillance area refers to following a person or group into the surveillance area; staying in the surveillance area for a long time refers to staying in the same position for more than a preset time threshold; covering the face refers to deliberately covering the face with an object to avoid being recognized. The labeled abnormal behavior images are used as samples to construct a data set, and the constructed data set is input into the deep learning model for model training. During the training process, the model will learn how to distinguish between abnormal and normal behaviors. By continuously adjusting the parameters and structure of the model, the recognition accuracy and generalization ability of the model are improved. The trained model is evaluated using the validation set to check the recognition performance of the model. Based on the evaluation results, the model is optimized and adjusted to improve its recognition accuracy and robustness. The access control system based on face recognition collects images of the monitoring area in real time. The collected images are preprocessed, including denoising and contrast enhancement, to improve image quality. The processed monitoring images are input into the trained deep learning model for behavior analysis. The model outputs the analysis results and confidence, that is, it determines whether the stranger in the image has abnormal behavior and gives the corresponding confidence value.

[0093] In one embodiment of the present invention, analyzing and calculating the recognition failure rate and the potential abnormality coefficient includes the following steps:

[0094] By analyzing the formula:

[0095] Get the recognition failure rate F; where S represents the total number of recognitions, S f Indicates the total number of recognition failures;

[0096] By analyzing the formula:

[0097] Get the potential abnormal coefficient Q; where C j represents the confidence of the deep learning model output of the i-th abnormal behavior, δ j represents the impact weight of the i-th abnormal behavior, and n represents the total number of abnormal behaviors.

[0098] Specifically, first, collect data on all recognition attempts over a period of time, including the total number of recognitions and information on whether each recognition is successful. The total number of recognitions can be counted by recording each time the system starts the recognition function. The total number of recognition failures can be counted by judging whether the recognition results are consistent with the actual situation. When the recognition results are inconsistent with the actual situation, it is considered a recognition failure. Calculate the recognition failure rate according to the formula. Deploy surveillance cameras or sensors and other equipment to capture abnormal behaviors in the monitoring area in real time. Label and classify the captured abnormal behaviors, including but not limited to wandering, following others into the monitoring area, staying in the monitoring area for a long time, and covering the face. Use the trained deep learning model to process the labeled abnormal behavior data. The model outputs the confidence of each abnormal behavior, indicating the probability that the behavior is identified as an abnormal behavior. According to factors such as the nature, severity, and possible consequences of the abnormal behavior, an impact weight is assigned to each abnormal behavior. The impact weight of the first abnormal behavior wandering is 0.3, the impact weight of the second abnormal behavior following others into the monitoring area is 0.3, the impact weight of the third abnormal behavior staying in the monitoring area for a long time is 0.2, and the impact weight of the fourth abnormal behavior covering the face is 0.2. The value of the impact weight can be adjusted dynamically according to the actual situation. The potential abnormality coefficient is calculated according to the formula based on the confidence level and impact weight of the abnormal behavior.

[0099] In one embodiment of the present invention, the risk coefficient is calculated by comprehensive analysis, and the risk level of the access control system is evaluated according to the calculation result, including the following steps:

[0100] By analyzing the formula: β=(F*W1+Q*W2)×100%

[0101] The risk coefficient β is obtained; wherein F represents the recognition failure rate, Q represents the potential abnormality coefficient, and W1 and W2 represent the weights of the recognition failure rate and the potential abnormality coefficient respectively; a threshold is set according to the risk coefficient to evaluate the risk level of the access control system;

[0102] When β≤20%, the risk level is low risk;

[0103] When 20%<β<50%, the risk level is moderate;

[0104] When β ≥ 50%, the risk level is high.

[0105] Specifically, the risk coefficient is obtained by multiplying the recognition failure rate and the potential abnormality coefficient by the corresponding weights; among which, the weight of the recognition failure rate is 0.4 and the weight of the potential abnormality coefficient is 0.6. The size of the weight should be dynamically allocated according to the degree of influence of the two on the risk level of the access control system and the actual situation. According to the security requirements and risk tolerance of the access control system, the threshold of the risk coefficient is dynamically set. In order to accurately evaluate the risk level of the access control system. The calculated risk coefficient is compared with the set threshold to obtain the risk level of the access control system, including: low risk, medium risk and high risk.

[0106] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A door access control system based on face recognition, characterized in that: Includes the following modules: Image acquisition module: collects face images, iris images and surveillance images of entrants; Database module: stores the facial images and iris templates of authorized personnel; Image processing module: pre-processing based on the collected face images, iris images and surveillance images, including: denoising and image enhancement; Feature extraction module: extracts feature parameters of face images and iris images based on the collected and preprocessed face images and iris images; Image analysis module: perform face comparison detection based on the feature parameters of the extracted face image; analyze the iris image of the entrant based on the face comparison result; analyze the abnormal behavior of strangers based on the collected and processed surveillance images; analyze and calculate the recognition failure rate and potential abnormality coefficient; Comprehensive analysis module: Comprehensively analyze and calculate the risk coefficient, and evaluate the risk level of the access control system based on the calculation results.

2. The access control system based on face recognition according to claim 1, characterized in that: Extracting characteristic parameters of the face image and the iris image according to the collected and preprocessed face image and the iris image includes the following steps: Based on the collected and pre-processed face image, the face position is located from the face image and the face area is cropped by using a face detection algorithm; the feature points of the face area are extracted by using a feature point detection algorithm; Extracting feature parameters based on the feature points of the detected facial area of ​​the human face, including: feature parameters of geometric shapes, color feature parameters and texture feature parameters; the feature parameters of the geometric shapes include: the X coordinate and Y coordinate of the center point of the eye, the X coordinate and Y coordinate of the center point of the nose, and the X coordinate and Y coordinate of the center point of the mouth; the color feature parameters include: the RGB value of the skin color of the human face; the texture feature parameters include: the length and depth of the wrinkles on the human face, and the area and number of the spots; Iris features are extracted from the processed iris image, including shape features, texture features and color features, and an iris template is generated based on the extracted features.

3. The access control system based on face recognition according to claim 2 is characterized in that: Performing face comparison detection based on the feature parameters of the extracted face image includes the following steps: Extract features from the facial images of authorized personnel in the database, and perform normalization and cleaning based on the extracted feature parameters of authorized personnel and entrants; According to the extracted and processed feature parameters, the feature parameters of the facial area of ​​the entrant are compared with the feature parameters of the facial image of the authorized person in the database; By analyzing the formula: Get the geometric shape feature parameter similarity J, color feature parameter similarity Y and texture feature parameter similarity W; where (X 1k , Y 1k ) represents the coordinate point of the geometric shape feature parameter of the authorized person; (X 2k , Y 2k ) represents the coordinate point of the geometric shape feature parameter of the entrant, when k=1, it represents the coordinate of the center point of the eye, when k=2, it represents the coordinate of the center point of the nose, and when k=3, it represents the coordinate of the center point of the mouth; R1, G1 and B1 represent the RGB values ​​of the facial skin color of the authorized person, and R2, G2 and B2 represent the RGB values ​​of the facial skin color of the entrant; M L 、M S 、M A and M Q They represent the wrinkle length, wrinkle depth, spot area and number of spots of the entrant, respectively. L 、N S 、N A and N Q Respectively indicate the length of wrinkles, the depth of wrinkles, the area of ​​spots and the number of spots of the authorized personnel; By analyzing the formula: R = (J*ε1+Y*ε2+W*ε3)×100% Obtain the facial similarity R of the human face; wherein J represents the similarity of geometric shape feature parameters, Y represents the similarity of color feature parameters, W represents the similarity of texture feature parameters, and ε1, ε2 and ε3 represent the weight coefficients of the similarity of geometric shape feature parameters, color feature parameters and texture feature parameters respectively; Compare the facial similarity with the threshold, analyze and determine whether the face of the entrant matches the facial image of the authorized person in the database, and obtain the face comparison result; the comparison result includes: face comparison result match success and match failure; When the face comparison result is a successful match, the entrant is an authorized person and the access control is opened; when the face comparison result is a failed match, the face image of the entrant is re-collected and the face comparison is performed again to determine whether the face comparison result is a successful match; when the re-collected face image of the entrant matches successfully, the access control is opened, otherwise the iris image of the entrant is analyzed to determine whether the entrant is an authorized person.

4. The access control system based on face recognition according to claim 3 is characterized in that: According to the face comparison result, the iris image of the entrant is analyzed, including the following steps: When the re-collected face image of the entrant fails to match, the iris image of the entrant is analyzed; an iris template is generated based on the iris features extracted from the iris image; The iris features include: shape features, texture features and color features; wherein the feature values ​​of the shape features include: the inner and outer circle radii and eccentricity of the iris; the feature values ​​of the texture features include: decomposing the iris image into LL channel, HL channel and LH channel by Haar wavelet transform, and calculating the mean and variance of the energy of the decomposed LL channel, HL channel and LH channel to obtain the feature values ​​of the texture features; the feature values ​​of the color features include: the mean and standard deviation of the RGB values ​​of the iris area; The extracted eigenvalues ​​are combined into a eigenvector, which is a multidimensional array, in which each element corresponds to an eigenvalue, and the eigenvalue of the i-th dimension represents the eigenvalue corresponding to the i-th feature, and the constructed eigenvector is used as an iris template; Compare the iris template of the entrant with the iris template of the authorized person stored in the database; By analyzing the formula: Get the iris template similarity H; where, A represents the dot product of the iris template of the entrant and the iris template of the authorized person. i represents the eigenvalue of the iris template of the entrant in the i-th dimension, B i represents the feature value of the iris template of the authorized person in the i-th dimension, n represents the number of dimensions of the iris template, and the number of dimensions represents the number of features; The iris template similarity is compared with the threshold. When the iris template similarity exceeds the threshold, it means the verification is successful; otherwise, it means the verification fails. When the verification fails, the entrant is marked as a stranger.

5. The access control system based on face recognition according to claim 4 is characterized in that: Analyzing the abnormal behavior of strangers based on the collected and processed surveillance images includes the following steps: Based on the collected and processed surveillance images, abnormal behaviors of strangers in the surveillance images are annotated, and the annotated abnormal behaviors are input into the deep learning model for training; the abnormal behaviors include: wandering, following others into the surveillance area, staying in the surveillance area for a long time, and covering the face; According to the trained deep learning model, the access control system based on face recognition collects and processes surveillance images in real time, inputs them into the trained deep learning model to analyze the abnormal behavior of strangers and outputs the analysis results and confidence levels.

6. The access control system based on face recognition according to claim 1, characterized in that: The analysis and calculation of recognition failure rate and potential abnormality coefficient includes the following steps: By analyzing the formula: Get the recognition failure rate F; where S represents the total number of recognitions, S f Indicates the total number of recognition failures; By analyzing the formula: Get the potential abnormal coefficient Q; where C j represents the confidence of the deep learning model output of the i-th abnormal behavior, δ j represents the impact weight of the i-th abnormal behavior, and n represents the total number of abnormal behaviors.

7. The access control system based on face recognition according to claim 1, characterized in that: Comprehensively analyze and calculate the risk coefficient, and evaluate the risk level of the access control system based on the calculation results, including the following steps: By analyzing the formula: β=(F*W1+Q*W2)×100% The risk coefficient β is obtained; wherein F represents the recognition failure rate, Q represents the potential abnormality coefficient, and W1 and W2 represent the weights of the recognition failure rate and the potential abnormality coefficient respectively; a threshold is set according to the risk coefficient to evaluate the risk level of the access control system; When β≤20%, the risk level is low; When 20%<β<50%, the risk level is moderate; When β ≥ 50%, the risk level is high.

Images