Access control system for analyzing personnel entering and exiting unmanned billiard hall based on AI data information acquisition

Abstract

The invention relates to the technical field of access control, in particular to an access control system for analyzing persons entering and exiting an unmanned billiard hall based on AI data information acquisition. The method comprises the steps that a sensing unit captures original data of persons entering and exiting the unmanned billiard hall through a sensor; the analysis and recognition unit is used for receiving the original data of the sensing unit, performing face recognition by using a face detection algorithm and performing behavior analysis through a multi-joint-point human body posture estimation model; the verification unit compares the processed data with information stored in a database by adopting multi-mode biological feature recognition to verify the identity of a person; and the control execution unit sends an instruction according to a result provided by the verification unit and a preset authority to control the opening and closing of the unmanned billiard hall access control. According to the design of the invention, by integrating various biological characteristic recognition technologies such as face detection, fingerprint recognition and human body behavior analysis, the access control system can accurately recognize identities of people entering and exiting the access control system, and effectively prevent unauthorized people from entering the access control system.

Description

Technical Field

[0001] The present invention relates to the field of access control technology, and in particular to an access control system for analyzing people entering and exiting an unmanned billiard hall based on AI data information collection. Background Art

[0002] Traditional access control systems may rely on simple methods such as cards and passwords for identity authentication, which are prone to the risk of loss, theft or cracking, and cannot effectively prevent unauthorized entry, posing a security risk; manual management and traditional electronic access control systems are often inefficient in managing personnel entry and exit, especially during peak hours, requiring a lot of time for identity authentication, increasing customer waiting time and dissatisfaction; traditional access control systems usually do not have intelligent analysis functions, and cannot effectively monitor and analyze the behavior patterns of people entering and leaving, making it difficult to detect potential security threats or abnormal behaviors in a timely manner; traditional identity authentication methods such as swiping cards and entering passwords are relatively cumbersome, affecting the user experience, especially in poor environmental conditions (such as wet hands and dim light), these methods may cause verification failures, further reducing the user experience. Therefore, an access control system based on AI data information collection for personnel analysis in and out of unmanned billiard halls is designed. Summary of the invention

[0003] The purpose of the present invention is to provide an access control system for analyzing people entering and leaving an unmanned billiard hall based on AI data information collection, so as to solve the problems of insufficient security, low management efficiency and lack of intelligent behavior monitoring raised in the above-mentioned background technology.

[0004] To achieve the above object, the present invention aims to provide an access control system for analyzing personnel entering and leaving an unmanned billiard hall based on AI data information collection, comprising:

[0005] A sensing unit, which captures raw data of people entering and leaving the unmanned billiard hall through sensors;

[0006] An analysis and recognition unit receives the raw data of the perception unit and performs face recognition using a face detection algorithm and behavior analysis using a multi-joint human posture estimation model. Since the facial features of people entering and exiting the unmanned billiard hall are different when viewed from the front and side, a face posture angle is introduced into the face loss function.

[0007] a verification unit that uses multimodal biometrics to compare the processed data with information stored in a database to verify the identity of the individual;

[0008] A control execution unit, wherein the control execution unit issues instructions according to the result provided by the verification unit and the preset authority to control the switch of the unmanned billiard hall entrance.

[0009] As a further improvement of the technical solution, the raw data of people entering and leaving the unmanned billiard hall captured by the perception unit include: visual images, faces and fingerprints.

[0010] As a further improvement of the technical solution, the analysis and recognition unit uses a face detection algorithm to perform face recognition and a multi-joint human posture estimation model to perform behavior analysis, including the following steps:

[0011] S1.1, preprocess the raw data;

[0012] S1.2, receiving fingerprint image data obtained from the fingerprint reader, and extracting fingerprint features using a fingerprint extraction algorithm;

[0013] S1.3. Use the face detection algorithm to locate the face area in the image, use the pre-trained face extraction model to extract the face feature vector from the detected face, use the multi-joint human posture estimation model to detect the human joints, and analyze the behavior patterns.

[0014] As a further improvement of the technical solution, in S1.2, the fingerprint feature is extracted using a fingerprint extraction algorithm, comprising the following steps:

[0015] S1.21, preprocessing the fingerprint image;

[0016] S1.22, calculating the local direction of each pixel in the fingerprint image;

[0017] S1.23, using the ridge thinning algorithm to thin the ridges of the fingerprint to a single pixel width;

[0018] S1.24. Find where the ridge ends and where the ridge splits into two.

[0019] S1.25. Search for endpoints and bifurcation points on the refined image, and calculate the position and direction of each suspected feature point;

[0020] S1.26. Verify whether the suspected feature points are indeed endpoints or bifurcation points through neighborhood analysis.

[0021] As a further improvement of the technical solution, in S1.3, extracting a feature vector from the detected face using a pre-trained face extraction model includes the following steps:

[0022] S1.31. Define a face loss function to train the face extraction model. Since the facial features of people entering and exiting the unmanned billiard hall are different from the front and side, the face posture angle is introduced into the face loss function, and the face loss function is further optimized according to the differences in facial features under different lighting conditions.

[0023] S1.32, cropping the located face area, and preprocessing the cropped face image;

[0024] S1.33, inputting the preprocessed face image into the trained face extraction model;

[0025] S1.34. Through the forward propagation calculation of the face extraction model, the feature vector of the face is obtained.

[0026] As a further improvement of the technical solution, in S1.31, the face loss function is:

[0027]

[0028] Since the facial features of people entering and exiting the unmanned billiard hall are different from the front and side, the facial posture angle is introduced into the face loss function:

[0029]

[0030] In view of the differences in facial features under different lighting angles, the incident angle and azimuth of the lighting are combined with the facial posture angle, and the face loss function is further optimized as follows:

[0031]

[0032] Among them, L(A, P, N) represents the face loss function; L1(A, P, N; θ) represents the face loss function for the different facial features of people entering and leaving the unmanned billiard hall from the front and side; L2(A, P, N; θ, θ a ,θ o ) represents the face loss function for different facial features under different illumination angles; A represents a face image sample of a specific individual selected from a set of face images; P represents a face image sample belonging to the same person as sample A; N represents a face image sample that does not belong to the same person as sample A; f represents the embedding function; w A represents the weight of sample A; w N represents the weight of sample N; α represents the interval; θ represents the face posture angle; θ a Represents the incident angle of light; θ o represents the azimuth of the light; α(θ,θ a ,θ o ) represents the interval value that is dynamically adjusted according to the face posture angle, light incident angle, and light azimuth.

[0033] As a further improvement of the technical solution, in S1.3, a multi-joint human posture estimation model is used to detect human joints and analyze behavior patterns, including the following steps:

[0034] S1.35. Collect behavioral data of people entering and leaving the unmanned billiard hall, and pre-process the behavioral data;

[0035] S1.36, use a multi-joint human posture estimation model to process the visual image, identify and mark the positions of human joints;

[0036] S1.37, obtain the coordinate position of each joint point from the recognition results, calculate the angle between adjacent joint points based on the coordinate position, and track the changes of various parts of the body over time;

[0037] S1.38, combine the position and angle information of different joints to form fusion features;

[0038] S1.39. Use the extracted joint coordinates, angles and changes of body parts over time to train a multi-joint human posture estimation model to predict behavior patterns.

[0039] As a further improvement of the technical solution, the verification unit uses multimodal biometric recognition to compare the processed data with the information stored in the database to verify the identity of the individual, including the following steps:

[0040] S2.1. Compare the extracted facial feature vector with the feature vector registered in the database and calculate the matching score; compare using fingerprint feature points and calculate the matching score;

[0041] S2.2. Normalize the matching scores of facial features and fingerprint features, fuse the comparison results of different biometric features through a multimodal fusion algorithm, and generate a comprehensive matching score. Since the wetness and stains of the finger affect the clarity of the fingerprint feature, the multimodal fusion algorithm is optimized;

[0042] S2.3. Compare the comprehensive matching score with the preset threshold m to determine whether the identity authentication standard is met.

[0043] As a further improvement of the technical solution, in S2.2, the multimodal fusion algorithm is:

[0044] S z =q j ·S j +q i ·S i ;

[0045] Since the wetness and stains of the finger affect the clarity of fingerprint features, the multimodal fusion algorithm is optimized:

[0046] S z1 =q j ·S j +qi ·e i ·S i ;

[0047] Among them, S z represents the comprehensive matching score; S z1 Represents the optimized comprehensive matching score; S j Indicates the matching score of facial features; S i Represents the matching score of fingerprint features; q j Represents the weight of facial features; q i The weight of the fingerprint feature; e i Represents the quality factor of the fingerprint feature, which is used to adjust the weight of the fingerprint feature when the finger is wet and stained.

[0048] As a further improvement of the technical solution, the method of issuing instructions according to the result provided by the verification unit and the preset authority to control the switch of the unmanned billiard hall door includes the following steps:

[0049] S3.1. Based on the results provided by the verification unit and the individual's access rights, the system makes a decision on whether to allow entry. If the individual is authorized to enter, it proceeds to the next step; if not, it triggers an alarm and sends a notification to the administrator;

[0050] S3.2. For any unauthorized entry attempts, record the time, location, and biometric information of the attempter and send the above information to the security team;

[0051] S3.3, after making a decision to allow entry, a control instruction is generated, and the control instruction is sent to the access switch control system;

[0052] S3.4. After receiving the control command, the access switch control system executes the corresponding action.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. The access control system for analyzing people entering and leaving the unmanned billiard hall based on AI data information collection can accurately identify the identities of people entering and leaving the hall by integrating multiple biometric recognition technologies such as face detection, fingerprint recognition, and human behavior analysis, effectively preventing unauthorized people from entering. In addition, the system can also monitor and analyze people's behavior patterns in real time, detect abnormal behavior in a timely manner, and alert management personnel, greatly improving the safety and management efficiency of the venue. At the same time, through the application of multimodal biometric recognition technology, high-precision identity authentication can be achieved even under complex lighting conditions or different facial postures, further enhancing the reliability and security of the system.

[0055] 2. This access control system, which analyzes people entering and exiting the unmanned billiard hall based on AI data information collection, takes into account various factors in actual applications, such as the impact of wet or stained fingers on fingerprint recognition. By introducing quality factors and balance factors to optimize the multimodal fusion algorithm, it ensures that stable and efficient recognition services can be provided even under adverse conditions, greatly improving the user experience. In addition, the fast and accurate identity authentication process reduces user waiting time, simplifies the entry procedure, makes the entry and exit of legitimate users smoother and more convenient, and also provides managers with a more flexible permission management solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is the overall flow chart of the present invention;

[0057] The meaning of each number in the figure is:

[0058] 1. Perception unit; 2. Analysis and identification unit; 3. Verification unit; 4. Control execution unit. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] Example: See Figure 1 As shown, a door access control system based on AI data information collection for personnel analysis of entering and leaving the unmanned billiard hall is provided, including:

[0061] The sensing unit 1 captures the raw data of people entering and leaving the unmanned billiard hall through sensors;

[0062] In this embodiment, the raw data captured of people entering and exiting the unmanned billiard hall includes: visual images, faces, and fingerprints.

[0063] The analysis and recognition unit 2 receives the raw data of the perception unit 1 and uses the face detection algorithm to perform face recognition and the multi-joint human posture estimation model to perform behavior analysis. Since the facial features of people entering and leaving the unmanned billiard hall are different from the front and side, the facial posture angle is introduced into the face loss function;

[0064] In this embodiment, the face detection algorithm is a computer vision technology used to locate and detect the position of a face in an image or video; compared with traditional identity authentication methods (such as passwords or cards), face recognition and human behavior analysis can provide higher accuracy and security because they are based on a person's unique biometrics; the face detection algorithm is used to ensure that only authorized personnel can enter the unmanned billiard hall, thereby improving the security and privacy protection level of the venue; combining the identity authentication results and the individual's access rights, refined access control is achieved to improve management efficiency; using the face detection algorithm to perform face recognition and behavior analysis through a multi-joint human posture estimation model, including the following steps:

[0065] S1.1. Preprocess the raw data, including image scaling, cropping, grayscale conversion, and standardization;

[0066] S1.2, receiving fingerprint image data obtained from a fingerprint reader, and extracting fingerprint features using a fingerprint extraction algorithm, the fingerprint features including features such as ridge patterns, endpoints and bifurcation points of the fingerprint;

[0067] Among them, the fingerprint extraction algorithm is a method for extracting feature points (such as ridge patterns, endpoints, and bifurcation points) from fingerprint images; everyone's fingerprints are unique, even twins are no exception. This high degree of uniqueness makes fingerprints a very reliable biometric feature; fingerprints remain basically unchanged throughout a person's life cycle unless they are severely physically damaged, which makes fingerprint recognition stable and reliable over the long term; the fingerprint extraction algorithm can identify and extract multiple feature points (such as ridge endpoints and bifurcation points), and the combination of these feature points greatly improves the accuracy of recognition; the fingerprint recognition algorithm can work in a variety of environments, including indoors and outdoors, day and night; using the fingerprint extraction algorithm to extract fingerprint features includes the following steps:

[0068] S1.21. Preprocess the fingerprint image, including image scaling, cropping, grayscale, standardization, etc.;

[0069] S1.22, calculate the local direction of each pixel in the fingerprint image (use a sliding window to calculate the gradient direction around each pixel, and then find the average direction), the direction field helps the subsequent refinement and feature point detection process;

[0070] S1.23, using the ridge thinning algorithm to thin the ridges of the fingerprint to a single pixel width;

[0071] S1.24. Find the place where the ridge line ends, that is, the point where no continuous ridge line extends out, and find the place where the ridge line splits into two or more ridge lines (check the 8-neighborhood of each pixel. If there is only one continuous ridge line around a pixel, it is considered an endpoint. Check the 8-neighborhood of each pixel. If there are multiple ridge lines intersecting around a pixel, it is considered a bifurcation point);

[0072] S1.25. Search for endpoints and bifurcation points on the refined image, and calculate the position and direction (i.e., the direction of the ridge line) of each suspected feature point.

[0073] S1.26. Verify whether the suspected feature point is indeed an endpoint or bifurcation point through neighborhood analysis;

[0074] S1.3. Use the face detection algorithm to locate the face area in the image, use the pre-trained face extraction model to extract the face feature vector from the detected face, use the multi-joint human posture estimation model to detect the human joints, and analyze the behavior pattern;

[0075] Further, extracting a feature vector from the detected face using a pre-trained face extraction model includes the following steps:

[0076] S1.31. Define a face loss function to train the face extraction model. Since the facial features of people entering and exiting the unmanned billiard hall are different from the front and side, the face posture angle is introduced into the face loss function, and the face loss function is further optimized according to the differences in facial features under different lighting conditions.

[0077] Furthermore, the face loss function is:

[0078]

[0079] Introducing the face posture angle can make the model better adapt to face images with different postures, thereby improving the accuracy and robustness of recognition. By considering the posture angle of the face, the misrecognition rate caused by posture changes can be reduced and the reliability of the system can be improved. The introduction of the posture angle can enable the model to be exposed to more diverse face data during the training process, thereby enhancing the generalization ability of the model and making it perform better in practical applications. In the actual unmanned billiard hall, people may have different postures when entering and exiting. The introduction of the posture angle can better handle these complex scenes. By introducing the posture angle in the loss function, the loss weight can be dynamically adjusted according to different postures, so that the model pays more attention to key features during the training process. Whether it is a front or side face image, the system can accurately identify it, which enhances the adaptability and flexibility of the system. Since the facial features of people entering and exiting the unmanned billiard hall are different when they are in the front and side, the face posture angle is introduced in the face loss function:

[0080]

[0081] Under different lighting conditions, the shadow and brightness distribution of the face will be significantly different, resulting in inconsistent feature extraction. By introducing the lighting angle in the loss function, the model can better adapt to these changes and improve the accuracy and robustness of recognition. Lighting changes may cause large differences in the features of the same person under different lighting conditions. The optimized loss function can reduce the misrecognition rate caused by lighting changes and improve the reliability of the system. In practical applications, lighting conditions are very complex, including natural light, lighting, shadows, etc. Changes in these lighting conditions will have a significant impact on the extraction of facial features. Facial features may be distorted under different lighting conditions, especially under extreme lighting conditions (such as backlight, direct strong light). Different incident angles will lead to brightness differences in different parts of the face, thereby affecting feature extraction. The azimuth angle will cause different shadow distribution of the face, affecting the stability and consistency of the features. By introducing the incident angle and azimuth of the light, the system can better process images under different lighting conditions and improve the accuracy of recognition. By calculating the incident angle and azimuth of the light, the weight of feature extraction is adjusted to reduce the impact of lighting changes; the incident angle and azimuth of the light are combined with the facial posture angle so that users can pass through the access control smoothly under any lighting and posture conditions, which improves the comfort and convenience of use. By considering the diversity of lighting and posture, the system can better adapt to various environmental conditions, including insufficient indoor light, strong direct light, etc.; in view of the differences in facial features under different lighting angles, the incident angle and azimuth of the light are combined with the facial posture angle, and the face loss function is further optimized as follows:

[0082]

[0083] Among them, L(A, P, N) represents the face loss function; L1(A, P, N; θ) represents the face loss function for the different facial features of people entering and leaving the unmanned billiard hall from the front and side; L2(A, P, N; θ, θ a ,θ o ) represents the face loss function for different facial features under different illumination angles; A represents a facial image sample of a specific individual selected from a set of facial images (anchor sample); P represents a facial image sample belonging to the same person as sample A (positive sample), which is used to measure the consistency of the model between different instances of the same person; N represents a facial image sample that does not belong to the same person as sample A (negative sample), which is used to measure the consistency of the model between different people; f represents the embedding function, which is a neural network model that converts the input facial image into a feature vector of fixed length, which is used to represent the characteristics of the face; w A represents the weight of sample A; w Nrepresents the weight of sample N; α represents the interval, which ensures that the distance between the positive sample and the anchor sample is less than the distance between the negative sample and the anchor sample plus a fixed interval; θ represents the face posture angle, that is, the face posture angle, which is used to measure the position of the face relative to the camera, including the front and side; θ a Represents the incident angle of light (estimates the direction of light by analyzing the grayscale value of the face image); θ o represents the azimuth of illumination (by analyzing the distribution of shadows in the face image, the azimuth of illumination is estimated, and the shadow will appear on the opposite side of the illumination direction); α(θ, θ a ,θ o ) represents the interval value that is dynamically adjusted according to the face pose angle, light incident angle, and light azimuth. The purpose is to allow the model to maintain good discrimination ability under different poses and light angles;

[0084] Furthermore, the multi-joint human posture estimation model can identify and track key parts (joints) of the human body, and analyze and identify specific behavior patterns based on the positions and motion trajectories of these joints; by detecting human joints and analyzing behavior patterns, the system can monitor the behavior of people entering and leaving in real time, detect abnormal situations in time, and identify and warn potential security threats, such as violent behavior, falls, long-term detention, etc.; the multi-joint human posture estimation model can detect multiple joints (such as head, shoulders, elbows, etc.), provide rich behavior information, and improve the accuracy of behavior analysis. By processing video streams in real time, the system can quickly and accurately analyze behavior patterns and detect abnormalities in time; by detecting multiple joints, the system can handle different angles of view to improve the flexibility and reliability of recognition; using the multi-joint human posture estimation model to detect human joints and analyze behavior patterns includes the following steps:

[0085] S1.35. Collect behavioral data of people entering and leaving the unmanned billiard hall, using images or videos recorded by cameras installed inside and outside the unmanned billiard hall, and pre-process the behavioral data (including format conversion, data cleaning, and data enhancement);

[0086] S1.36. Use a multi-joint human posture estimation model to process visual images, identify and mark the positions of human joints (such as head, shoulders, elbows, etc.);

[0087] S1.37, obtain the coordinate position of each joint point from the recognition results, calculate the angle between adjacent joint points based on the coordinate position, and track the changes of various parts of the body over time;

[0088] Among them, if there are two joint points A(x 1 ,y 1 ) and B(x 2 ,y 2), the vector between them can be expressed as: If there is a third joint point C(x 3 ,y 3 ),So The angle θ between two vectors is calculated using the dot product formula:

[0089] S1.38, combine the position and angle information of different joints (join the position and angle information of different joints into a long feature vector. For example, suppose we have n joints, each with x and y coordinates and angle information, and join these information into a 3n-dimensional feature vector) to form a fusion feature;

[0090] S1.39, use the extracted joint coordinates, angles and changes of body parts over time to train a multi-joint human posture estimation model to predict behavior patterns;

[0091] S1.32. Crop the located face area (the face detection algorithm will return a bounding box containing the face, including the coordinates of the upper left corner and the lower right corner. According to these coordinates, the face area can be directly cropped from the original image), and preprocess the cropped face image, including resizing, color space conversion, normalization, and dimension expansion;

[0092] S1.33, inputting the preprocessed face image into the trained face extraction model;

[0093] S1.34. Through the forward propagation calculation of the face extraction model, the feature vector of the face is obtained.

[0094] Verification unit 3 uses multimodal biometrics to compare the processed data with information stored in a database to verify the identity of the individual;

[0095] In this embodiment, the complementarity between different biometric features can significantly improve the accuracy of recognition. For example, facial features can provide high-resolution facial details, while fingerprint features can provide unique skin texture information; multimodal recognition can significantly reduce the false recognition rate and rejection rate through the combination of multiple features, improve the reliability of the system, and multimodal recognition can effectively prevent deception by counterfeiting means such as photos and videos, and improve the security of the system. For example, combining face and fingerprint recognition can effectively prevent photo deception; through the combination of multiple biometric features, multi-level verification can be achieved to improve the anti-counterfeiting ability of the system. For example, combining face, fingerprint and iris recognition can provide a higher level of security; using multimodal biometric recognition to compare the processed data with the information stored in the database to verify the identity of an individual includes the following steps:

[0096] S2.1. Compare the extracted facial feature vector with the feature vector registered in the database and calculate the matching score; use fingerprint feature points (endpoints, bifurcation points, etc.) for comparison and calculate the matching score;

[0097] S2.2. Normalize the matching scores of facial features and fingerprint features, fuse the comparison results of different biometric features through a multimodal fusion algorithm, and generate a comprehensive matching score. Since the wetness and stains of the finger affect the clarity of the fingerprint feature, the multimodal fusion algorithm is optimized;

[0098] Among them, the multimodal fusion algorithm is:

[0099] S z =q j ·S j +q i ·S i ;

[0100] The moisture and stains of fingers will affect the extraction and comparison of fingerprint features. The optimized multimodal fusion algorithm can reduce the misrecognition rate caused by unclear fingerprint features and improve the recognition accuracy of the system. By optimizing the multimodal fusion algorithm, it can be ensured that other biometric features (such as face) can still be accurately recognized when the fingerprint features are affected, thereby improving the overall reliability of the system. The optimized algorithm can reduce repeated recognition caused by unclear fingerprint features and improve the user experience. The optimized multimodal fusion algorithm can better utilize the complementarity between different biometric features and improve the accuracy of recognition. By introducing a quality factor to adjust the weight of fingerprint features, when the fingerprint quality is detected to be low (for example, the finger is wet or stained), the weight of the fingerprint feature can be appropriately reduced, thereby reducing its impact on the final comprehensive matching score. Since the degree of moisture and stains of the finger affect the clarity of the fingerprint features, the multimodal fusion algorithm is optimized:

[0101] S z1 =q j ·S j +q i ·e i ·S i ;

[0102] Among them, S z represents the comprehensive matching score, i.e., the final multimodal fusion result; S z1 Represents the optimized comprehensive matching score; S j Indicates the matching score of facial features; S i Represents the matching score of fingerprint features; q j Represents the weight of facial features; q iThe weight of the fingerprint feature; e i The quality factor of the fingerprint feature is used to adjust the weight of the fingerprint feature when the finger is wet or stained. When the fingerprint quality is low (for example, the finger is wet or stained), the weight of the fingerprint feature can be appropriately reduced to reduce its impact on the final comprehensive matching score;

[0103] S2.3. Compare the comprehensive matching score with the preset threshold m to determine whether the identity authentication standard is met. If the comprehensive score exceeds the threshold, the identity authentication is considered successful, otherwise, it is considered a verification failure.

[0104] Specifically, the quality factor e of the fingerprint feature i The calculation is performed using the following method:

[0105] Among them, the gradient amplitude is used to calculate the clarity:

[0106]

[0107] Where I(x,y) is the fingerprint image, and Represent the gradient of the image in the x and y directions respectively. Then calculate an average value G avg :

[0108]

[0109] N is the total number of pixels in the image. A high-resolution fingerprint image will have a larger average gradient.

[0110] Then convert the input color fingerprint image into a grayscale image:

[0111] I'(x,y)=0.2989·R(x,y)+0.5870·G(x,y)+0.1140·B(x,y);

[0112] Among them, R(x,y), G(x,y), B(x,y) are the red, green, and blue channel values ​​of the (x,y) pixel in the image, and I′(x,y) is the pixel value of the grayscale image.

[0113] Calculating the local grayscale difference represents the difference between adjacent pixels and reflects the local fluctuation degree of the image:

[0114] ΔI(x,y)=|I′(x,y)-I(x+1,y)|+|I(x,y)-I(x,y+1)|;

[0115] Compute the noise variance of an image:

[0116]

[0117] Where, μ is the average gray value of the image;

[0118]

[0119] variance The larger it is, the higher the noise level is and the worse the fingerprint image quality is.

[0120] Normalize the noise level to between 0 and 1 for use with other factors (e.g., clarity, wetness);

[0121]

[0122] in, and are the minimum and maximum values ​​of the noise variance, respectively, indicating the lowest and highest levels of image noise.

[0123] The influence of moisture and dirt on the fingerprint image is local, which is often manifested as blurring of fingerprint texture or uneven color in a local area. In order to quantitatively analyze the degree of moisture / stain, this embodiment uses local texture analysis and color analysis to calculate its influence.

[0124] The grayscale value range of the fingerprint image I(x,y) is [0,255]. In one direction, GLCM(i,j) is calculated, where (i,j) is the grayscale value, which represents the grayscale value relationship between the pixel (x,y) and its neighboring pixels.

[0125] Extract multiple statistical features from GLCM,

[0126] Contrast: Contrast = ∑ i,j (ij) 2 GLCM(i,j);

[0127] Describes the grayscale difference of a local area of ​​an image.

[0128] Uniformity:

[0129] Describes the smoothness of texture in a local area of ​​an image. Wetness or stains often cause uneven images.

[0130] entropy:

[0131] Quantification of wetness / stains:

[0132] W=α·(1-Homogeneity)+β·Contrast;

[0133] Among them, α and β are adjustment weight coefficients.

[0134] Normalization of wetness / stain level

[0135]

[0136] Among them, W min and W max are the minimum and maximum wetness / stain index, respectively.

[0137] Combining the noise level and the degree of wetness / staining, the quality factor is finally calculated:

[0138] γ,δ are weight coefficients.

[0139] In this embodiment, α=0.5-0.7, β=0.3-0.5, γ=0.5-0.7, δ=0.3-0.5.

[0140] The control execution unit 4 issues instructions according to the result provided by the verification unit 3 and the preset authority to control the switch of the door of the unmanned billiard hall;

[0141] In this embodiment, issuing instructions according to the result provided by the verification unit 3 and the preset authority to control the switch of the door of the unmanned billiard hall includes the following steps:

[0142] S3.1. Based on the results provided by verification unit 3 and the individual's access rights, the system makes a decision on whether to allow access. If the individual is authorized to enter, it proceeds to the next step; if the individual does not have permission, an alarm is triggered and a notification is sent to the administrator;

[0143] Among them, users are divided into different categories and set different access rights, such as staff (according to work shifts, they have access rights during specific time periods, such as the morning shift (8 am to 4 pm), the evening shift (4 pm to 2 am)), temporary visitors, etc. (usually can only access during specific time periods, such as a few hours or half a day); temporary visitors are only allowed to enter specific areas (each area has access control), such as the front desk, lounge area, designated billiard room, etc., and are not allowed to enter the backstage or private areas; different levels of access rights are assigned to users according to their roles and needs, and high-level users can access more areas or operate within a wider time range; administrators can adjust user access rights at any time according to actual conditions, such as adding or revoking a user's access rights to a specific area; the system records the time, location and access rights of each user entering and leaving;

[0144] S3.2. For any unauthorized entry attempts, record the time, location, and biometric information of the attempter and send the above information to the security team;

[0145] Furthermore, once an unauthorized entry attempt is identified, the access control system will refuse to open the door lock, prevent the person from entering, and record the specific time of the attempted entry in order to track the time point of the incident, determine and record the location of the entrance attempted, which is helpful for analyzing security vulnerabilities; when multiple people are detected entering at the same time but there is only one authorization verification, the system automatically marks this behavior as suspicious. Once tailing behavior is detected, the system immediately triggers an alarm and sends relevant information (including time, location, and image of the attempter) to the security team;

[0146] S3.3, after making a decision to allow entry, a control instruction is generated, and the control instruction is sent to the access switch control system;

[0147] S3.4. After receiving the control command, the access switch control system performs corresponding actions, such as unlocking the door lock, opening the automatic door, etc., to allow people to enter.

[0148] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention, and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. An access control system based on AI data information collection to analyze people entering and leaving the unmanned billiard hall, characterized by: include: A sensing unit (1), wherein the sensing unit (1) captures raw data of people entering and exiting the unmanned billiard hall through a sensor; An analysis and recognition unit (2), wherein the analysis and recognition unit (2) receives the raw data of the perception unit (1) and uses a face detection algorithm to perform face recognition and a multi-joint human posture estimation model to perform behavior analysis. Since the facial features of people entering and exiting the unmanned billiard hall are different when viewed from the front and side, a face posture angle is introduced into the face loss function; A verification unit (3), wherein the verification unit (3) uses multimodal biometric recognition to compare the processed data with information stored in a database to verify the identity of the individual; A control execution unit (4) is configured to issue instructions based on the result provided by the verification unit (3) and preset authority to control the opening and closing of the door access control system of the unmanned billiard hall.

2. The access control system for analyzing people entering and leaving an unmanned billiard hall based on AI data information collection according to claim 1 is characterized by: The raw data captured by the perception unit (1) regarding people entering and exiting the unmanned billiard hall include visual images, faces, and fingerprints.

3. The access control system for analyzing people entering and leaving an unmanned billiard hall based on AI data information collection according to claim 2 is characterized by: The analysis and recognition unit (2) performs face recognition using a face detection algorithm and performs behavior analysis using a multi-joint human posture estimation model, including the following steps: S1.1, preprocess the raw data; S1.2, receiving fingerprint image data obtained from the fingerprint reader, and extracting fingerprint features using a fingerprint extraction algorithm; S1.

3. Use the face detection algorithm to locate the face area in the image, use the pre-trained face extraction model to extract the face feature vector from the detected face, use the multi-joint human posture estimation model to detect the human joints, and analyze the behavior patterns.

4. The access control system for analyzing people entering and leaving an unmanned billiard hall based on AI data information collection according to claim 3 is characterized by: In S1.2, the fingerprint feature is extracted using a fingerprint extraction algorithm, including the following steps: S1.21, preprocessing the fingerprint image; S1.22, calculating the local direction of each pixel in the fingerprint image; S1.23, using the ridge thinning algorithm to thin the ridges of the fingerprint to a single pixel width; S1.

24. Find where the ridge ends and where the ridge splits into two. S1.

25. Search for endpoints and bifurcation points on the refined image, and calculate the position and direction of each suspected feature point; S1.

26. Verify whether the suspected feature points are indeed endpoints or bifurcation points through neighborhood analysis.

5. The access control system for analyzing people entering and leaving an unmanned billiard hall based on AI data information collection according to claim 4 is characterized by: In S1.3, extracting a feature vector from a detected face using a pre-trained face extraction model includes the following steps: S1.

31. Define a face loss function to train the face extraction model. Since the facial features of people entering and exiting the unmanned billiard hall are different from the front and side, the face posture angle is introduced into the face loss function, and the face loss function is further optimized according to the differences in facial features under different lighting conditions. S1.32, cropping the located face area, and preprocessing the cropped face image; S1.33, inputting the preprocessed face image into the trained face extraction model; S1.

34. Through the forward propagation calculation of the face extraction model, the feature vector of the face is obtained.

6. The access control system for analyzing people entering and leaving an unmanned billiard hall based on AI data information collection according to claim 5 is characterized by: In S1.31, the face loss function is: Since the facial features of people entering and exiting the unmanned billiard hall are different from the front and side, the facial posture angle is introduced into the face loss function: In view of the differences in facial features under different lighting angles, the incident angle and azimuth of the lighting are combined with the facial posture angle, and the face loss function is further optimized as follows: Among them, L(A, P, N) represents the face loss function; L1(A, P, N; θ) represents the face loss function for the different facial features of people entering and leaving the unmanned billiard hall from the front and side; L2(A, P, N; θ, θ a ,θ o ) represents the face loss function for different facial features under different illumination angles; A represents a face image sample of a specific individual selected from a set of face images; P represents a face image sample belonging to the same person as sample A; N represents a face image sample that does not belong to the same person as sample A; f represents the embedding function; w A represents the weight of sample A; w N represents the weight of sample N; α represents the interval; θ represents the face posture angle; θ a Represents the incident angle of light; θ o represents the azimuth of the light; α(θ,θ a ,θ o ) represents the interval value that is dynamically adjusted according to the face posture angle, light incident angle, and light azimuth.

7. The access control system for analyzing people entering and leaving an unmanned billiard hall based on AI data information collection according to claim 6 is characterized by: In S1.3, a multi-joint human posture estimation model is used to detect human joints and analyze behavior patterns, including the following steps: S1.

35. Collect behavioral data of people entering and leaving the unmanned billiard hall, and pre-process the behavioral data; S1.36, use a multi-joint human posture estimation model to process the visual image, identify and mark the positions of human joints; S1.37, obtain the coordinate position of each joint point from the recognition results, calculate the angle between adjacent joint points based on the coordinate position, and track the changes of various parts of the body over time; S1.38, combine the position and angle information of different joints to form fusion features; S1.

39. Use the extracted joint coordinates, angles and changes of body parts over time to train a multi-joint human posture estimation model to predict behavior patterns.

8. The access control system for analyzing people entering and leaving an unmanned billiard hall based on AI data information collection according to claim 7 is characterized by: The verification unit (3) uses multimodal biometric recognition to compare the processed data with the information stored in the database to verify the identity of the individual, including the following steps: S2.

1. Compare the extracted facial feature vector with the feature vector registered in the database and calculate the matching score; compare using fingerprint feature points and calculate the matching score; S2.

2. Normalize the matching scores of facial features and fingerprint features, fuse the comparison results of different biometric features through a multimodal fusion algorithm, and generate a comprehensive matching score. Since the wetness and stains of the finger affect the clarity of the fingerprint feature, the multimodal fusion algorithm is optimized; S2.

3. Compare the comprehensive matching score with the preset threshold m to determine whether the identity authentication standard is met.

9. The access control system for analyzing people entering and leaving an unmanned billiard hall based on AI data information collection according to claim 8 is characterized by: In S2.2, the multimodal fusion algorithm is: S z =q j ·S j +q i ·S i ; Since the wetness and stains of the finger affect the clarity of fingerprint features, the multimodal fusion algorithm is optimized: S z1 =q j ·S j +q i ·e i ·S i ; Among them, S z represents the comprehensive matching score; S z1 Represents the optimized comprehensive matching score; S j Indicates the matching score of facial features; S i Represents the matching score of fingerprint features; q j Represents the weight of facial features; q i The weight of the fingerprint feature; e i Represents the quality factor of the fingerprint feature, which is used to adjust the weight of the fingerprint feature when the finger is wet and stained.

10. The access control system for analyzing people entering and leaving an unmanned billiard hall based on AI data information collection according to claim 9 is characterized in that: The method of issuing instructions based on the result provided by the verification unit (3) and the preset authority to control the switch of the door of the unmanned billiard hall includes the following steps: S3.

1. Based on the result provided by the verification unit (3) and the individual's access rights, the system decides whether to allow access. If the individual is authorized to enter, it proceeds to the next step; if the individual is not authorized, an alarm is triggered and a notification is sent to the administrator. S3.

2. For any unauthorized entry attempts, record the time, location, and biometric information of the attempter and send the above information to the security team; S3.3, after making a decision to allow entry, a control instruction is generated, and the control instruction is sent to the access switch control system; S3.

4. After receiving the control command, the access switch control system executes the corresponding action.

Images