Access control security management method, device, equipment, medium and product

Abstract

The embodiment of the application provides a kind of access control passage security management method, device, equipment, medium and product, involve artificial intelligence field or financial technology field.First, the original face image of access personnel is collected by image acquisition device;Then, the original face image collected is identified to determine the expression classification result of access personnel;Then, according to expression classification result, behavior discrimination processing is carried out, and the behavior level of access personnel is obtained;Finally, the corresponding security management strategy is determined based on behavior level, and the strategy is executed.The method combines the way of face image collection, expression recognition and behavior classification control, solves the problem of security risk and insufficient reliability of existing access control management mode, realizes intelligent monitoring and differentiated security control of access personnel state, and improves the security and control reliability of access control.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence or financial technology, and in particular to a method, device, equipment, medium and product for secure access control. Background Technology

[0002] Personnel access control is a crucial foundation for the smooth operation of bank branches. Without effective screening and control of personnel entering and exiting, various security risks can easily arise at bank branches.

[0003] Existing security control methods mainly rely on physical identification. Access control in key security areas of bank branches usually combines multiple verification methods such as keys, card swipes, passwords, and fingerprints to verify the identity of people entering and leaving.

[0004] However, this type of physical identification method has many drawbacks. First, if keys, access cards, or other credentials are lost, it will directly reduce the efficiency of business processing. Second, this method can only control access permissions and cannot provide early warnings for abnormal behavior of branch personnel. Third, the accuracy of identity recognition is insufficient and is easily affected by environmental factors such as lighting and facial posture, and there are even cases where former employees impersonate others to gain access. Summary of the Invention

[0005] This application provides a method, device, equipment, medium, and product for secure access control, in order to solve the problems of security risks and insufficient reliability in existing access control methods.

[0006] In a first aspect, embodiments of this application provide a security management method for access control, applied to access control equipment, wherein the access control equipment includes an image acquisition device, comprising:

[0007] The original facial images of passersby are captured using an image acquisition device;

[0008] The original facial image is processed to obtain the expression classification results of the person passing by;

[0009] The facial expression classification results are processed to determine the behavior level of the person passing through.

[0010] Determine the security management policy corresponding to the behavior level, and execute the security management policy.

[0011] In one possible implementation, the step of performing recognition processing on the original facial image to obtain the expression classification result of the person passing by includes:

[0012] The original face image is processed to obtain a modified face image;

[0013] The modified face image is input into a pre-built facial expression recognition model to obtain the expression classification result output by the facial expression recognition model.

[0014] In one possible implementation, the step of performing detection processing on the original face image to obtain the modified face image includes:

[0015] The original face image is processed to obtain the detected face image;

[0016] The detected face image is enhanced to obtain a modified face image.

[0017] In one possible implementation, the step of performing detection processing on the original face image to obtain a detected face image includes:

[0018] The original face image is subjected to grayscale normalization processing to obtain a grayscale face image;

[0019] The grayscale face image is subjected to histogram equalization to obtain an enhanced face image;

[0020] The enhanced face image is subjected to face detection processing to obtain a localized face image;

[0021] The measured face image is cropped to obtain the detected face image.

[0022] In one possible implementation, the enhancement processing of the detected face image to obtain a modified face image includes:

[0023] The detected face image is subjected to data augmentation processing to obtain an augmented face image;

[0024] The augmented face image is subjected to face re-examination processing to obtain candidate face images;

[0025] The candidate face image is processed by drawing bounding boxes to obtain the modified face image.

[0026] In one possible implementation, the step of performing behavior discrimination processing on the expression classification result to obtain the behavior level of the person passing by includes:

[0027] Obtain security control and evaluation criteria;

[0028] Based on the security control evaluation criteria, the facial expression classification results are processed for behavioral discrimination to obtain the behavioral level of the person passing through.

[0029] In one possible implementation, if the behavior level is classified as slightly abnormal, a security management policy corresponding to the behavior level is determined and executed, including:

[0030] Based on the aforementioned minor anomaly, a warning message is generated;

[0031] The display component of the access control device is activated, and the warning message is displayed on the display component.

[0032] In one possible implementation, if the behavior level is moderately abnormal, a security management policy corresponding to the behavior level is determined and the security management policy is executed, including:

[0033] The first warning light component of the access control device is activated;

[0034] The first warning light component is controlled to broadcast an abnormal light signal according to the first preset flashing frequency.

[0035] In one possible implementation, if the behavior level is severely abnormal, a security management policy corresponding to the behavior level is determined and executed, including:

[0036] Activate the second warning light component and the voice warning component of the access control device;

[0037] According to the second preset flashing frequency, the second warning light component is controlled to broadcast light abnormalities, and according to the preset sounding frequency, the voice warning component is controlled to broadcast sound abnormalities.

[0038] In one possible implementation, if the behavior level is "no anomaly," a security management policy corresponding to the behavior level is determined and executed, including...

[0039] The access control device's opening and closing components are driven to perform an opening action, so that the access control device's door is in an open state.

[0040] Secondly, embodiments of this application provide a security management device for access control, applied to access control equipment, wherein the access control equipment includes an image acquisition device, comprising:

[0041] The acquisition module is used to acquire raw facial images of people passing through the image acquisition device;

[0042] The processing module is used to perform recognition processing on the original facial image to obtain the expression classification result of the person passing by;

[0043] The acquisition module is used to acquire security control and evaluation criteria;

[0044] The processing module is also used to perform behavior discrimination processing on the expression classification results according to the security control evaluation criteria to obtain the behavior level of the person passing through;

[0045] The determination module is used to determine the security management strategy corresponding to the behavior level;

[0046] The execution module is used to execute the security management policy.

[0047] In one possible implementation, the processing module is further configured to perform detection processing on the original face image to obtain a modified face image;

[0048] The device further includes: an input module;

[0049] The input module is used to input the modified face image into a pre-built face expression recognition model to obtain the expression classification result output by the face expression recognition model.

[0050] In one possible implementation, the processing module is further configured to perform detection processing on the original face image to obtain a detected face image;

[0051] The processing module is specifically used to enhance the detected face image to obtain a modified face image.

[0052] In one possible implementation, the processing module is further configured to perform grayscale normalization processing on the original face image to obtain a grayscale face image;

[0053] The processing module is also used to perform histogram equalization processing on the grayscale face image to obtain an enhanced face image;

[0054] The processing module is also used to perform face detection processing on the enhanced face image to obtain a localized face image;

[0055] The processing module is specifically used to perform size cropping processing on the located face image to obtain the detected face image.

[0056] In one possible implementation, the processing module is further configured to perform data augmentation processing on the detected face image to obtain an augmented face image;

[0057] The processing module is also used to perform face re-examination processing on the augmented face image to obtain candidate face images;

[0058] The processing module is specifically used to draw bounding boxes on the candidate face image to obtain a modified face image.

[0059] In one possible implementation, the acquisition module is further configured to acquire security control evaluation criteria;

[0060] The processing module is specifically used to perform behavior discrimination processing on the expression classification results according to the security control evaluation criteria to obtain the behavior level of the person passing through.

[0061] In one possible implementation, the apparatus further includes: a generation module;

[0062] The generation module is used to generate warning information based on the minor anomaly;

[0063] The device further includes: a startup module;

[0064] The startup module is used to start the display component of the access control device;

[0065] The device further includes: a display module;

[0066] The display module is used to display the warning information on the display component.

[0067] In one possible implementation, the activation module is further configured to activate the first warning light component of the access control device;

[0068] The device further includes: a control module;

[0069] The control module is used to control the first warning light component to broadcast abnormal light signals according to a first preset flashing frequency.

[0070] In one possible implementation, the activation module is further configured to activate the second warning light component and the voice warning component of the access control device;

[0071] The control module is also used to control the second warning light component to broadcast abnormal light signals according to the second preset flashing frequency;

[0072] The activation module is also used to control the voice warning component to broadcast abnormal sounds according to a preset sound frequency.

[0073] In one possible implementation, the device further includes: a drive module;

[0074] The drive module is used to drive the access control opening and closing components of the access control device to perform an opening action, so that the door of the access control device is in an open state.

[0075] Thirdly, embodiments of this application provide a security management device for access control, including: a memory and a processor;

[0076] The memory stores computer-executed instructions;

[0077] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0078] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0079] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0080] The access control security management method, device, equipment, medium, and product provided in this application first acquires raw facial images of personnel using an image acquisition device; then, the acquired raw facial images are processed to determine the personnel's expression classification; next, behavior discrimination processing is performed based on the expression classification results to obtain the personnel's behavior level; finally, a corresponding security management strategy is determined based on the behavior level and executed. This method, combining facial image acquisition, expression recognition, and behavior-level control, solves the security risks and reliability issues of existing access control methods, achieving intelligent monitoring and differentiated security control of personnel entering and exiting, thus improving the security and reliability of access control. Attached Figure Description

[0081] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0082] Figure 1 Flowchart of the access control security management method provided in this application Figure 1 ;

[0083] Figure 2 Flowchart of the access control security management method provided in this application Figure 2 ;

[0084] Figure 3 The existing MobileNetV2 network architecture diagram provided for this application;

[0085] Figure 4 The implementation process of depthwise separable convolution for the existing MobileNetV2 network architecture graph provided in this application;

[0086] Figure 5 Image of an existing Relu6 activation function provided in this application;

[0087] Figure 6 A schematic diagram of the existing residual structure provided for this application;

[0088] Figure 7 A schematic diagram of the inverse residual structure and linear bottleneck structure of MobileNetV2 provided in this application;

[0089] Figure 8 A schematic diagram of the existing deep residual network multi-branch structure provided in this application;

[0090] Figure 9 This is a schematic diagram of the structure of an existing bridging attention module provided in this application;

[0091] Figure 10 A schematic diagram of the existing bridging attention module for integrating multiple output features provided in this application;

[0092] Figure 11 A schematic diagram of the multi-branch inverted residual structure provided in this application;

[0093] Figure 12 A schematic diagram of the multi-scale feature learning structure provided in this application;

[0094] Figure 13 A schematic diagram of the structure of the facial expression recognition model provided in this application;

[0095] Figure 14 Flowchart of the access control security management method provided in this application Figure 3 ;

[0096] Figure 15 A structural schematic diagram of the access control security management device provided in this application;

[0097] Figure 16 A structural diagram of the access control security management device provided in this application.

[0098] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0099] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0100] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, have taken necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation access points for users to choose to authorize or refuse.

[0101] Furthermore, the technical solution involved in this application, which involves big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.) and the use of artificial intelligence technology for automated decision-making, and makes decisions that have a significant impact on personal rights based on the results of automated decision-making, provides users with corresponding operation entry points for users to choose to agree to or reject the results of automated decision-making; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0102] It should be noted that the access control security management methods, devices, equipment, media and products provided in this application can be used in the fields of artificial intelligence or financial technology, or in any field other than artificial intelligence or financial technology. The application fields of the access control security management methods, devices, equipment, media and products in this application are not limited.

[0103] First, the terms used in this application will be explained.

[0104] The Adaptive Moment Estimation (Adam) algorithm combines the advantages of momentum gradient descent and root mean square propagation, enabling it to dynamically calculate the first and second moments of the gradient and adaptively adjust the learning rate for different parameters.

[0105] Cross-entropy loss (CEL) is a commonly used loss calculation method in classification tasks. It quantifies the error of the recognition result by calculating the difference between the probability distribution predicted by the model and the true label distribution of the sample.

[0106] Personnel access control is a crucial foundation for the smooth operation of bank branches. Without effective screening and control of personnel entering and exiting, various security risks can easily arise at bank branches.

[0107] Existing security control methods mainly rely on physical identification. Access control in key security areas of bank branches usually combines multiple verification methods such as keys, card swipes, passwords, and fingerprints to verify the identity of people entering and leaving.

[0108] However, this type of physical identification has many drawbacks. First, it affects business processing efficiency. Access control is typically handled by one employee and one operations supervisor or branch manager (such as deputy branch manager) who keeps the keys (or access cards). If an employee or operations supervisor loses their key during morning or evening baggage pickups, obtaining a spare key takes time. If the person keeping the spare key is on leave or out handling other business, it not only affects the efficiency of baggage pickup, delays pickups at the next branch, but also impacts cash transactions for customers at the current branch. This significantly damages the branch's reputation and also negatively affects customers.

[0109] Secondly, there is a lack of real-time early warning for abnormal employee behavior, and the labor costs are high. Traditional access control only solves the "can I enter?" problem, and cannot automatically warn of abnormal employee behavior. Furthermore, monitoring of key security areas such as cash areas and warehouses relies excessively on surveillance cameras, requiring 24-hour on-duty personnel, which increases human resource costs. If important items such as blank vouchers are missing from the warehouse due to abnormal employee behavior, reviewing surveillance footage afterwards is also time-consuming and labor-intensive.

[0110] Third, the identification of personnel is not intelligent enough. Although electronic access control (including facial recognition, fingerprint recognition, and password recognition) is relatively common, it can automatically record entry and exit times and personnel identities, and different access permissions can be set for different employees. Permissions can also be revoked immediately after leaving the company. However, it is greatly affected by the environment. For example, changes in lighting and facial posture can affect facial recognition, wet or peeling fingers can affect fingerprint recognition, and overly complex passwords are easy to forget. These factors make it difficult for employees to enter key security areas.

[0111] The access control security management method provided in this application first acquires the original facial images of personnel using an image acquisition device, then recognizes the images to obtain expression classification results, determines the level of personnel behavior based on the classification results, and finally matches and executes corresponding security management strategies. This method effectively improves the security and reliability of access control by recognizing the expressions of personnel and determining their behavior levels accordingly, thereby implementing appropriate security management measures. It achieves intelligent abnormal behavior early warning and precise execution of security management strategies, thus enhancing the overall level of security control.

[0112] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0113] Figure 1 Flowchart of the access control security management method provided in this application Figure 1 It is used in access control equipment, which includes image acquisition devices, such as... Figure 1 As shown, the method includes:

[0114] S101. Acquire the original facial images of the passersby using an image acquisition device.

[0115] The image acquisition device is used to acquire the original facial images of people passing through. It can be an imaging component such as a camera installed above, to the side or in the integrated structure of the door frame of the key area entrance. Its function is to acquire facial images that can be used for subsequent recognition and processing when people pass through the access control area.

[0116] The original face image is the input data directly acquired by the image acquisition device. It has not undergone preprocessing or feature extraction, so it can completely preserve the original information such as lighting, posture, expression and background of the scene.

[0117] Understandably, when the access control equipment is in standby mode, it can continuously receive video streams or continuous still images output in real time by the image acquisition device. This type of image data is generated by the image acquisition device taking fixed-point shots and collecting data within the preset acquisition area defined by the access control equipment. It can be stably transmitted back to the access control equipment, providing complete and reliable raw image data support for the operation of subsequent functions such as face detection, recognition, and behavior analysis.

[0118] For example, when a person enters the preset acquisition area of ​​the access control device, the image acquisition device can continuously acquire multiple frames of images at a preset frame rate. The access control device then selects one or more frames from these frames that meet the required clarity, facial orientation, and minimal occlusion as the original face image. Clarity is determined based on the Laplacian operator response value, edge gradient strength, and image contrast; occlusion is assessed by considering the integrity of facial landmarks and the proportion of the visible facial area; facial orientation is determined by estimating the yaw, pitch, and roll angles based on head posture, thereby selecting images within the recognizable angle range.

[0119] S102. Perform recognition processing on the original face image to obtain the expression classification results of the passersby.

[0120] Among them, the facial expression classification results are used to characterize the current facial expression type of the person passing through, and are the direct basis for subsequent behavior level determination, including but not limited to: seven types of expressions: disgust, surprise, anger, happiness, sadness, fear and neutral.

[0121] Understandably, raw facial images can reflect the facial expressions and real-time emotional state of passersby, containing rich facial feature information. In order to determine the current emotional expression of the passerby, it is necessary to perform recognition processing on the collected raw facial images to obtain the corresponding expression classification results.

[0122] S103. Perform behavioral discrimination processing on the facial expression classification results to obtain the behavioral level of the passerby.

[0123] Among them, the behavior level is used to classify and determine the risk of the behavior of people passing through the access control area, including but not limited to mild abnormality, moderate abnormality, severe abnormality, and no abnormality.

[0124] Understandably, access control devices can retrieve preset behavioral judgment criteria and combine these criteria to judge the facial expression classification results to obtain the behavioral level of the person passing through. For example, when the facial expression classification result is happy and the person approaches the access control normally on the designated passage path, it can be judged as no abnormality; when the facial expression classification result is neutral, but the dwell time is short and there is no repeated probing action, it can be judged as mild abnormality; when the facial expression classification result is disgust, and there are characteristics of prolonged dwell time, back and forth, and repeated approaching and leaving in front of the access control area, it can be judged as moderate abnormality; when the facial expression classification result is fear, and there are symptoms such as prolonged lingering in front of the door, abnormal intrusion actions, and repeated attempts to approach the door lock execution area, it can be judged as severe abnormality.

[0125] S104. Determine the security management strategy corresponding to the behavior level and implement the security management strategy.

[0126] Among them, the security management strategy is used to implement corresponding access control actions for different behavior levels, so as to achieve hierarchical management from normal access, risk warning, on-site early warning to mandatory intervention.

[0127] Understandably, the access control device first reads the behavior level and then retrieves the security management policy matching that behavior level from its local policy table. When the behavior level is normal, the access control device can maintain the normal passage process, such as driving the door lock actuator to open the door after identity verification. When the behavior level is slightly abnormal, the access control device can execute a prompt security management policy, such as displaying instructions on proper passage. When the behavior level is moderately abnormal, the access control device can execute an audio-based security management policy, such as activating an audio warning component and issuing warnings at a preset frequency. When the behavior level is severely abnormal, the access control device can execute a stronger audio-visual security management policy, such as activating both an audio warning component and a visual warning component, and issuing warnings at their respective frequencies.

[0128] The access control security management method provided in this application first uses an image acquisition device to acquire the original facial images of people passing through, then performs recognition and analysis on the images to classify people's facial expressions, then determines the level of people's behavior based on the expression classification results, and finally matches and executes the corresponding level of security management policy. This method, based on the linkage of facial image acquisition, expression recognition, and behavior level determination, overcomes the limitations of physical credentials, enables early warning of abnormal personnel behavior, improves the stability and accuracy of identity recognition, prevents misuse of permissions, and comprehensively strengthens the security management capabilities of the premises.

[0129] Figure 2 Flowchart of the access control security management method provided in this application Figure 2 ,like Figure 2 As shown, in this embodiment... Figure 1 Based on the embodiments, a detailed description of the security management method for access control is provided, which includes:

[0130] S201. Acquire the original facial images of passersby using an image acquisition device.

[0131] The explanation of step S201 is the same as that in the above embodiments, and will not be repeated here.

[0132] S202. Perform detection processing on the original face image to obtain the modified face image.

[0133] It is understandable that raw facial images are easily affected by external factors such as changes in lighting, shooting angle, and obstruction by external objects, resulting in image defects. Using them directly will reduce the accuracy of subsequent recognition.

[0134] Therefore, by detecting and processing the original face image, we can identify and improve the problems in the image, reduce environmental interference, ensure the integrity and clarity of key facial features, and effectively improve the reliability of subsequent expression recognition results.

[0135] S203. Input the modified face image into the pre-built face expression recognition model to obtain the expression classification result output by the face expression recognition model.

[0136] Among them, the facial expression recognition model is a model used to recognize and process modified facial images and output expression classification results. It can extract features from the eyebrows, eyes, corners of the mouth and facial muscle changes in the facial region, and make expression category judgments accordingly.

[0137] Understandably, by using a trained facial expression recognition model to analyze optimized facial images, the model can accurately extract facial features, determine emotions, and output corresponding expression classification results.

[0138] For example, the facial expression recognition model provided in this application is constructed by building upon the existing Inverted Residuals and LinearBottlenecks (MobileNetV2) lightweight convolutional neural network for mobile devices, and combining it with EfficientChannel Attention (ECA), Bridge Attention Module (BAM), and Deep Residual Network (ResNet) to improve MobileNetV2, aiming to achieve a relative balance between classification accuracy and training cost.

[0139] Understandably, the existing MobileNetV2 network architecture diagram, such as Figure 3 As shown in the figure, 3×3Conv represents a 3×3 convolutional layer, 1×1Conv represents a 1×1 convolutional layer, bottleneck represents a bottleneck layer, GAP represents a global average pooling layer, Dropout represents a Dropout layer, FC represents a fully connected layer, and Softmax represents a Softmax classifier.

[0140] The existing MobileNetV2 network architecture has the following three characteristics:

[0141] First, it employs depthwise separable convolution. Existing standard convolution uses multi-channel kernels to simultaneously perform convolution operations on all channels of the input feature map, enabling spatial feature extraction and channel dimension transformation in one operation, but with high parameter and computational costs. Depthwise separable convolution decomposes standard convolution into two operations: depthwise convolution and pointwise convolution. First, for the input feature map, standard convolution uses a multi-channel kernel to perform convolution operations on all channels, while depthwise convolution uses a separate kernel for each input channel, without performing dimension transformation. That is, the output feature map after depthwise convolution has the same number of channels as the input feature map. Then, pointwise convolution uses a 1×1 kernel to fuse the channels of the feature map output by depthwise convolution, adjusting the number of channels in the output feature map as needed to achieve dimension transformation. The MobileNetV2 network uses depthwise separable convolution instead of standard convolution, significantly reducing parameter and computational costs, making the network more lightweight.

[0142] The implementation process of depthwise separable convolution in the existing MobileNetV2 network architecture graph, such as Figure 4 As shown. Assume the size of the input feature map is [H, W, ...]. The number of channels in the output feature map is... The kernel size for depthwise convolution is K×K, and the kernel size for pointwise convolution is 1×1, where H is the height of the input feature map and W is the width of the input feature map. This represents the number of channels in the input feature map. Therefore, standard convolution uses a K×K size and has... Number of input channels and A convolution kernel with a certain number of output channels, the computational cost of a standard convolution. The expression is shown in equation (1):

[0143] (1)

[0144] computational cost of depthwise convolution The expression is shown in equation (2):

[0145] (2)

[0146] computational cost of pointwise convolution The expression for is shown in equation (3):

[0147] (3)

[0148] Since the computational cost of depthwise separable convolution is the sum of the computational costs of depthwise convolution and pointwise convolution, its computational cost is... The expression is shown in equation (4):

[0149] (4)

[0150] Comparing Equations (1) and (4), the computational cost of depthwise separable convolution is... Computational complexity compared to standard convolution The ratio is shown in equation (5):

[0151] (5)

[0152] From equation (5), it can be seen that when As the size increases, the computational cost of depthwise separable convolution is approximately one-third that of standard convolution. Secondly, we assume the ReLU6 activation function is used. The existing ReLU6 activation function graph is as follows: Figure 5 As shown, compared to the ReLU activation function, it avoids excessively large activation values ​​by limiting the range of output values, making the model's output values ​​more stable.

[0153] Thirdly, inverted residual structures and linear bottleneck structures are employed. Traditional residual structures typically reduce computation by first reducing dimensionality, then extracting features, and finally increasing dimensionality. This involves first using 1×1 convolutional layers to reduce the number of channels in the feature map, then using 3×3 convolutional layers for feature extraction, and finally using 1×1 convolutional layers to increase the number of channels, forming a bottleneck structure that is narrow in the middle and wide at both ends. A schematic diagram of the existing residual structure is shown below. Figure 6 As shown in the figure. The symbol ⊕ indicates element-wise addition, BN indicates a batch normalized layer, and ReLU indicates the ReLU activation function.

[0154] The inverted residual structure of the MobileNetV2 network differs from traditional residual structures. It employs a "first increase in dimensionality, then feature extraction, then dimensionality reduction" approach to reduce computation. Specifically, it first uses 1×1 convolutional layers to increase the number of channels in the feature map, then uses 3×3 deep convolutional layers for feature extraction, and finally uses 1×1 convolutional layers to reduce the number of channels, forming an inverted bottleneck structure that is wide in the middle and narrow at both ends. The linear bottleneck structure of the MobileNetV2 network is similar to the inverted residual structure, except that it lacks residual connections. The structure is as follows: Figure 7 As shown. Figure 7 (a) in the diagram represents the inverted residual structure. Figure 7 In the diagram, (b) represents a linear bottleneck structure, 3×3Dwise represents a 3×3 depth convolutional layer, and Linear represents a linear activation function.

[0155] The MobileNetV2 network uses a linear activation function in the last 1×1 convolutional layer of the inverted residual structure and the linear bottleneck structure in order to reduce information loss and thus retain more feature information. The other convolutional layers use the ReLU6 activation function to provide some non-linear transformation.

[0156] The existing efficient channel attention mechanism is a lightweight channel attention module that relies on one-dimensional convolution to complete cross-channel feature interaction. It has the characteristics of few parameters and fast computation speed, and can optimize the model feature extraction effect with low overhead.

[0157] Existing deep residual networks introduce the idea of ​​grouped convolution, dividing the number of channels into multiple groups to reduce computation. They expand the receptive field of the model by constructing a multi-branch structure and enhance the model's feature learning ability through layer-by-layer feature fusion within this multi-branch structure. The multi-branch structures of existing deep residual networks are as follows: Figure 8 As shown in the figure, the chunk operation represents dividing the feature map output by the 1×1 convolutional layer into four equal groups of feature maps with the same number of channels. , , All represent 3×3 ordinary convolutional layers. The concat operation means concatting the feature maps extracted from the four branches.

[0158] Existing bridging attention modules enhance attention weight generation by integrating the output features of preceding convolutional layers, thereby improving model performance. The structure of a bridging attention module is as follows: Figure 9 As shown, the process by which the BAM module integrates the output features of the preceding convolutional layers is as follows: Figure 10 As shown in the figure, the symbols Represented as element-wise multiplication, Sigmoid represents the Sigmoid activation function.

[0159] Depend on Figure 9 and Figure 10 As can be seen, the bridging attention module consists of two phases: integration and generation. Its main implementation process is as follows:

[0160] In the first step, during the integration phase, assume that the bridging attention module needs to integrate n convolutional layers, and that the output features of these convolutional layers are as follows: The feature dimension is ,in and Represented as the height and width of the feature. Let i represent the number of channels of the feature, where i ranges from 1 to n. First, for these output features F... i Perform global average pooling (GAP) operation, at which point the feature dimension of each channel is... The process is shown in equation (6):

[0161] (6)

[0162] The second step is to then examine these features. Feature compression is performed using a fully connected layer to obtain the compressed feature S. i The process is shown in equation (7):

[0163] (7)

[0164] In the formula, This is represented as the weight matrix of the fully connected layer, with a size of . , This represents the number of channels in the last convolutional layer after integration. This is represented as the dimensionality reduction factor.

[0165] The third step then requires processing each compressed feature. Perform batch normalization to obtain features The purpose is to ensure the features They have similar distributions, which makes the integration effect more effective. The process is shown in equation (8):

[0166] (8)

[0167] The fourth step is to define the features. The final integrated feature is obtained by fusing the components additively. and will integrate features The inputs are sequentially fed into the ReLU activation function, the fully connected layer, and the Sigmoid activation function in the generation stage to obtain the final attention weight W. The corresponding calculation formulas are shown in equations (9) and (10):

[0168] (9)

[0169] (10)

[0170] The fifth step is to finally generate the attention weights. With the output features of the last convolutional layer Perform element-wise multiplication to obtain the output features. The corresponding calculation formula is shown in equation (11):

[0171] (11)

[0172] As can be seen from the implementation process of the bridging attention module, the dimensionality reduction factor in the integration phase... Affecting the characteristics after compression The size of the feature affects the computational cost of integrating multiple output features in the bridging attention module. The larger the value, the stronger the feature. The smaller the value, the less computation is required for the bridging attention module to integrate multiple output features. However, this also limits the model's feature learning ability, leading to a decrease in model performance. The smaller the value, the more characteristic... The larger the value, the greater the computational cost for the B-bridge attention module to integrate multiple output features. Therefore, setting an appropriate value... The value is crucial, as it can ensure model performance while reducing computational cost. Considering both model performance and training cost, this invention uses the dimensionality reduction factor... The value is set to 2.

[0173] Based on this, the design of the facial expression recognition model provided in this application includes:

[0174] First, design a multi-branch inverted residual structure.

[0175] For the inverse residual structure with a stride of 1, the inverse residual structure of the MobileNetV2 network is improved into a multi-branch structure, and an efficient channel attention mechanism is introduced to enhance the model's ability to extract important features. The improved inverse residual structure with a stride of 1 is named the Multi-branch Inverse Residual Structure (MIRS), as follows: Figure 11 As shown, Figure 11 This is a schematic diagram of the multi-branch inverse residual structure provided in this application. In the figure, 1×1DConv represents a 1×1 dilated convolutional layer, Relu6 represents the Relu6 activation function, 3×3dK1, 3×3dK2, and 3×3dK3 all represent 3×3 depth convolutional layers, and ECA represents the ECA attention mechanism.

[0176] Specifically, the MIRS module has a Dropout layer and an identity mapping between the input feature map and the output feature map only when the step size is 1 and the number of input channels equals the number of output channels.

[0177] Secondly, it involves multi-scale feature learning structures.

[0178] For the linear bottleneck structure with a stride of 2, the linear bottleneck structure of the MobileNetV2 network is improved into a multi-branch structure. The improved linear bottleneck structure with a stride of 2 is named the ECA Multi-scale Feature Learning Structure (EMFLS), as follows: Figure 12 As shown, Figure 12 This is a schematic diagram of the multi-scale feature learning structure provided in this application. In the diagram, maxpool represents the max pooling layer.

[0179] The EMFLS module can perform max pooling operations on the feature map of the first feature branch, and does not perform layer-by-layer feature fusion on each feature branch, but only performs concat fusion, in order to reduce the amount of computation.

[0180] Finally, a lightweight network is designed to bridge attention and multi-scale feature learning.

[0181] Based on the MobileNetV2 network, the following improvements were made: the first 3×3 convolutional layer with a stride of 2 was replaced with a 3×3 dilated convolutional layer with a stride of 1 and a max pooling layer with a stride of 2, expanding the receptive field while retaining key feature information; the MIRS and EMFLS modules were designed as multi-branch structures, enhancing multi-scale feature extraction capabilities while reducing computational overhead, and an efficient channel attention mechanism was introduced to strengthen the model's focus on important features; the output features of multiple preceding MIRS modules were integrated through a bridging attention module to improve the model's feature learning ability; the last 1×1 convolutional layer of the original MobileNetV2 network was removed to reduce the number of model parameters; and the output features of multiple deep modules were upsampled to achieve the fusion of shallow and deep features, making the feature information extracted by the model richer and more comprehensive.

[0182] The improved MobileNetV2 network is named Lightweight Networks For Bridging Attention And Multi-scale Feature Learning (LBMNet), which is also known as a facial expression recognition model. Figure 13 This is a schematic diagram of the facial expression recognition model provided in this application. The 3×3 dilated convolutional layer in the diagram has a dilation rate of 2 and padding of 2, which is intended to ensure that the input and output feature map sizes remain unchanged. The kernel size of the max-pooling layer is 2×2. The LBMNet model consists of a 3×3 dilated convolutional layer, a max-pooling layer, a MIRS module, an EMFLS module, a BAM module, a 1×1 convolutional layer, a global average pooling layer, a Dropout layer, a fully connected layer, and a Softmax layer. Its structural parameters are shown in Table 1. In the table, t is the dilation factor used by this layer module, n is the number of times this layer module needs to be executed, s is the stride of this layer module, and f is the scaling factor used for upsampling by this layer module.

[0183] The structural parameters of the facial expression recognition model are shown in Table 1.

[0184] Table 1

[0185]

[0186] The implementation process of the facial expression recognition model is as follows: First, the modified input face image undergoes preliminary feature extraction through a 3×3 dilated convolutional layer, increasing the number of channels from 3 to 32. Then, it is downsampled through a max-pooling layer, reducing the modified face image from 224×224 to 112×112. Next, the modified face image is sequentially processed through multiple MIRS and EMFLS modules for deep feature extraction. The feature maps output by the deep modules contain rich semantic information, while the shallow modules retain clear texture and edge details. A bridging attention module integrates the output features of the shallow and deep MIRS modules across layers, enhancing the model's feature representation capability. Finally, the output features of multiple deep modules are upsampled and fused with the corresponding shallow module output features using a concat method, resulting in a fused feature map with 1280 channels and a size of 56×56. The feature map is then compressed to a 1×1 size by a global average pooling layer and converted into a one-dimensional feature vector by a Flatten operation. Finally, it is processed by a Dropout layer and a fully connected layer, and the classification results of each expression category are output by a Softmax layer.

[0187] Optionally, the completed facial expression recognition model can be trained to obtain a usable facial expression recognition model. The training process is as follows: First, acquire a historical facial image dataset and perform image processing on the dataset to obtain a preprocessed dataset; then, train an initial model using the preprocessed dataset to obtain candidate prediction models; finally, test the candidate prediction models using a test dataset and obtain the test results. If the test results are satisfactory, the candidate prediction model is determined as the final facial expression recognition model; if the test fails, the model training is repeated until the test results are satisfactory.

[0188] Understandably, after obtaining a historical face image dataset, a series of image processing operations are performed sequentially, such as grayscale normalization, histogram equalization, face detection, size cropping, and data augmentation. This can eliminate interference from external factors such as lighting, pose, and background, remove invalid image information, enrich sample diversity, unify image specifications and data distribution, and provide high-quality, standardized input data for subsequent model training, thus ensuring the foundation for training from the source.

[0189] The initial facial expression recognition model is trained using a preprocessed image dataset. Regularized sample data is input into the network model, and an adaptive moment estimation optimization algorithm is used for training. The model parameters are continuously updated through iterative computation, learning the mapping relationship between facial texture, expression features, and categories. The training process fully mines the effective features in the data, gradually improving the model's feature extraction and autonomous learning capabilities. Finally, a preliminary candidate prediction model is output, completing the transition from the initial state to a model with recognition capabilities.

[0190] The candidate prediction models are evaluated using an independent test dataset. The recognition error is calculated using the cross-entropy loss algorithm, and the model's suitability is determined based on the recognition results. If the test metrics meet the preset standards, the candidate model is selected as the final usable facial expression recognition model. If the test results fail to meet the standards, it indicates that the model's generalization ability or recognition accuracy is defective, and the training process needs to be restarted for repeated iterations and optimizations until all test metrics of the model are qualified, thereby ensuring the reliability of the model in practical applications.

[0191] S204. Obtain security control and evaluation criteria.

[0192] Among them, the security control evaluation criteria, as a behavior judgment rule, can establish a correspondence between expression categories and behavior levels, and map the identified expression classification results to behavior levels such as no abnormality, mild abnormality, moderate abnormality, or severe abnormality.

[0193] This step can be obtained locally, for example, from the parameter configuration module built into the access control device, or from a cloud server. This application does not impose any special restrictions on this.

[0194] S205. Based on the security control evaluation criteria, the facial expression classification results are processed to determine the behavior level of the passersby.

[0195] The purpose of this step is to analyze the facial expression classification results based on the established security control evaluation criteria, complete the identification of personnel behavior, and output the corresponding behavior level.

[0196] Understandably, the results of facial expression classification cannot directly guide access control for security management. However, security management evaluation criteria are the standards for judging behavior and can clearly define the correspondence between different expression categories and behavior levels.

[0197] Therefore, by establishing security control evaluation criteria as the basis for judgment and processing the facial expression classification results into behavioral judgments, the behavioral level of the passersby can be obtained, ensuring that the overall security judgment process is consistent and the results are objective and reliable.

[0198] S206. If the behavior level is mildly abnormal, generate a warning message based on the mildly abnormality.

[0199] Warning messages are used to alert pedestrians so that they can adjust their expressions, postures, or passage status in a timely manner.

[0200] Understandably, after receiving a minor anomaly assessment result, the access control device can generate corresponding text prompts based on that assessment result. The text prompts may include information such as "Please maintain normal passage status", "Please do not stay for a long time", or "Please pay attention to your current behavior status".

[0201] S207. Activate the display component of the access control device and display a warning message on the display component.

[0202] The display component is the information output unit on the access control device. It can be an LCD screen, an OLED (Organic Light-Emitting Diode) screen, or other visual interfaces, which can present prompts to passersby or other relevant personnel. The display component can be integrated into the front panel of the access control device housing, or it can be electrically connected to the access control controller through a separate display module. The two are usually connected by a bus, serial port, or network to facilitate receiving policy-issued instructions.

[0203] Understandably, after generating a warning message, the display component can be activated to present the warning message on the display interface with a preset font size, color, and display duration, so that passersby can clearly identify it at close range.

[0204] This method enables access control equipment to immediately output warning messages upon detecting minor anomalies and provide on-site prompts via display components, allowing users to correct abnormal states promptly without disrupting normal access control. Furthermore, this method shifts the handling of minor anomalies from backend recording to frontend prompts, shortening the risk feedback chain and reducing the probability of escalating anomalies. It also avoids the impact on access efficiency caused by directly imposing mandatory blocking measures on low-level anomalies, thus achieving a balance between security management and business continuity.

[0205] S208. If the behavior level is moderately abnormal, activate the first warning light component of the access control device.

[0206] The first warning light component can be installed on the top housing, side frame, or prominent position of the access control device, and can be composed of LED warning lights, light strips, or strobe modules. Its light emission color can be red or amber to create a strong visual warning effect in key security areas. In practical applications, other models of this component can also be selected, and this application embodiment does not limit this choice.

[0207] Understandably, when the behavior of a person passing through is determined to be moderately abnormal, the corresponding safety management policy is invoked to send an activation command to the corresponding first warning light component, which can put it into a standby state.

[0208] S209. Control the first warning light component to broadcast abnormal light signals according to the first preset flashing frequency.

[0209] The first preset flashing frequency can be set according to the lighting conditions, observation distance and environmental noise interference of the access control channel, so that the warning light can still be clearly identified in the bright light environment of the business hall.

[0210] Understandably, by controlling the operation of the first warning light component according to the first preset flashing frequency, the continuous flashing light can send an abnormality prompt to the personnel on site, allowing passers-by and surrounding security personnel to quickly identify the current access control status without having to read text information.

[0211] This method enables access control devices to provide visual warnings via lights when moderate anomalies are detected, avoiding the response delays caused by relying solely on manual review or text prompts. Because the flashing frequency matches the anomaly level, the warning information has stronger distinguishability and on-site dissemination capabilities, effectively alerting security personnel to abnormal access and providing a time window for subsequent intervention. This improves the real-time performance, identifiability, and handling efficiency of access control management in key security areas.

[0212] S210. If the behavior level is severe abnormality, activate the second warning light component and voice warning component of the access control device.

[0213] The second warning light component is used to generate a visual warning output on the access control device. It can consist of a high-brightness LED light group installed at the front of the access control housing. A light-transmitting warning cover can be provided on the outside of the light group to diffuse the light and improve long-distance visibility. In practical applications, other models of this component can also be selected, and this application embodiment does not limit this.

[0214] The voice warning component can consist of a speaker or a voice broadcast module, used to output pre-recorded or real-time synthesized prompts to provide immediate intervention to passersby. In practical applications, other models of this component can also be selected, and this application embodiment does not limit this.

[0215] Understandably, after the access control device determines that the behavior level of the person passing through is severely abnormal, it invokes the security management policy corresponding to that level and sends an activation command to the second warning light component and the voice warning component, which can put both into a standby state.

[0216] S211. Control the second warning light component to broadcast light abnormalities according to the second preset flashing frequency, and control the voice warning component to broadcast sound abnormalities according to the preset sounding frequency.

[0217] Understandably, the light drive signal is periodically controlled based on a second preset flashing frequency, causing the second warning light component to alternately light up and turn off according to a set duration, thus forming a continuous light anomaly broadcast. Simultaneously, the controller outputs a broadcast trigger signal to the voice warning component according to a preset sounding frequency, causing the voice warning component to intermittently repeat the anomaly warning content. The light anomaly broadcast and the sound anomaly broadcast can be executed synchronously or sequentially according to a preset time sequence to avoid a single warning method being obscured in complex environments.

[0218] By using a combination of sound and light to enhance the visibility of access control systems, abnormal alerts can simultaneously cover both visual and auditory channels, reducing the probability of overlooking unusual behavior. The coordination of a second preset flashing frequency and a preset sounding frequency enhances the consistency of the warning rhythm, enabling access control equipment to provide clear and stable real-time warnings even with limited computing power. This improves the security management response capabilities of key security areas and helps to provide immediate reminders and behavioral constraints to those passing through.

[0219] S212. If the behavior level is no abnormality, drive the access control device's access control opening and closing components to perform an opening action so that the access control device's door is in the open state.

[0220] Among them, the access control opening and closing component is used to drive the door to complete the opening action, thus ensuring the reliable opening and closing of the door.

[0221] The door can be a swing door, a sliding door, or a swing gate. After receiving the release signal, the access control opening and closing component outputs a drive electrical signal or mechanical drive force to make the door move out of the closing limit and enter the open state. In practical applications, other models of this component can also be selected, but this application embodiment does not limit this.

[0222] Understandably, when the behavior of the personnel is determined to be normal, the corresponding security management policy is mapped to a door opening command, and an execution signal is sent to the access control opening and closing component. This signal drives the access control device's opening and closing component to perform the opening action. The access control opening and closing component can activate a motor, relay, or electromagnetic lock unlocking mechanism based on this execution signal, keeping the door open for a limited time to allow personnel to pass through the access control area normally. After passage, the door can return to the closed state under the action of a return spring, door closer, or control reset mechanism, thereby maintaining the passage order in key security areas.

[0223] This method allows for direct access control when no abnormal behavior is observed in the access personnel, avoiding unnecessary alarm triggers and passage obstructions, thus balancing security management with passage efficiency. Because the door opening action is uniformly driven by the access control components, the access control equipment can respond promptly, reducing delays and errors caused by manual intervention, thereby improving the continuity, stability, and management reliability of access control in key security areas of the bank.

[0224] This application provides a security management method for access control. First, an image acquisition device captures the original facial image of the person passing through. Then, the original facial image is processed to obtain an optimized facial image, which is input into a facial expression recognition model to obtain an expression classification result. Subsequently, security control evaluation criteria are retrieved, and the behavior level of the person passing through is determined based on the expression classification result. Then, corresponding operations are performed according to different behavior levels: if a minor abnormality is determined, a warning message is generated and displayed through the access control's display component; if a moderate abnormality is determined, a warning light component is activated, providing a light warning at a preset frequency; if a moderate abnormality is determined, different warning light components and voice warning components are activated, providing light and voice warnings at preset frequencies; if no abnormality is determined, the access control opening and closing component is controlled to open the door and allow passage.

[0225] This method optimizes facial image recognition quality through preprocessing, combines facial expression recognition with security assessment standards to classify behavior levels, and integrates tiered audio-visual, graphic, and textual warnings with access control linkage mechanisms. It effectively solves various problems existing in traditional physical identity recognition methods, not only improving the anti-interference capability and recognition accuracy of facial recognition, avoiding the risk of unauthorized use, and eliminating the drawbacks of physical credentials, but also accurately distinguishing personnel behavior status and making tiered early warning and response, enriching security control methods, and comprehensively improving venue access management and security protection.

[0226] Figure 14 Flowchart of the access control security management method provided in this application Figure 3 ,like Figure 14 As shown, this embodiment, based on the above embodiments, provides a detailed explanation of the process for detecting and processing the original face image to obtain a modified face image. The method includes:

[0227] S301. Perform detection processing on the original face image to obtain the detected face image.

[0228] Understandably, raw facial images are susceptible to factors such as lighting, occlusion, and shooting angle, resulting in image quality defects. Direct use of these images would reduce the accuracy of subsequent recognition. Therefore, by performing detection processing on raw facial images, image quality can be improved, complete facial features can be preserved, and a reliable basis can be provided for subsequent image analysis.

[0229] Optionally, this application provides a possible implementation method, including: performing grayscale normalization processing on the original face image to obtain a grayscale face image; performing histogram equalization processing on the grayscale face image to obtain an enhanced face image; performing face detection processing on the enhanced face image to obtain a localized face image; and performing size cropping processing on the localized face image to obtain a detected face image.

[0230] Among them, grayscale normalization is used to map the pixel values ​​of the original face image to a preset grayscale range, so as to reduce the impact of changes in light intensity under different shooting environments on image quality and facilitate subsequent unified processing.

[0231] A grayscale face image is an image that has been normalized to the grayscale range. It retains facial contour and texture information and is used as input for histogram equalization processing.

[0232] Histogram equalization is used to adjust pixel distribution, making the contrast and details of facial areas clearer, thus creating an enhanced face image.

[0233] Face detection processing is used to locate facial regions in enhanced face images and output the corresponding location information. The localized face image is the image after the facial regions have been located.

[0234] Size cropping is used to uniformly crop the located face image according to preset size parameters so that the output face image meets the input size requirements of the subsequent expression recognition model.

[0235] Understandably, after obtaining the original facial images of the passersby, the images are first processed by grayscale normalization to reduce the impact of different lighting conditions on facial textures and convert the color image into a uniform grayscale image. Then, histogram equalization is performed on the grayscale facial image to improve the contrast of facial contours, eye, nose, and mouth areas, and key textures, resulting in an enhanced facial image. Next, face detection processing is performed based on the enhanced facial image, using a preset face detection model or cascaded classifier to locate the facial region, resulting in a localized facial image. Finally, the localized facial image is cropped to a preset size while maintaining the integrity of the main facial features, forming the detected facial image.

[0236] By sequentially performing grayscale normalization, histogram equalization, face localization, and size cropping on the original image, the detectability of face regions can be improved under complex lighting and fluctuating conditions, while ensuring uniform output image size. The resulting detected face images have more stable clarity and more consistent input specifications, which can reduce the probability of false positives and false negatives in subsequent expression classification and improve the stability and response efficiency of image processing at the access control device.

[0237] S302. Enhance the detected face image to obtain the modified face image.

[0238] Understandably, even after obvious flaws have been removed from the detected images, problems such as insufficient contrast and weak features may still exist. Therefore, enhancing the detected face images can effectively improve image quality, reduce external interference, and provide qualified image data for subsequent face recognition.

[0239] Optionally, this application provides a possible implementation method, including: performing data augmentation processing on the detected face image to obtain an augmented face image; performing face re-examination processing on the augmented face image to obtain a candidate face image; and performing bounding box drawing processing on the candidate face image to obtain a modified face image.

[0240] Data augmentation refers to applying geometric transformations and imaging perturbations to the detected face image to form augmented face images with diversity. Geometric transformations include rotation, translation, flipping, scaling, and mirroring, while imaging perturbations include brightness changes, contrast changes, and noise superposition.

[0241] Face re-examination processing refers to performing face validity verification again on augmented face images to eliminate non-face regions or images with incomplete face features, thereby retaining candidate face images that can be used for subsequent processing.

[0242] The bounding box drawing process refers to drawing bounding boxes around the face region of the candidate face image. The coordinates of the bounding boxes can be determined by the face position parameters obtained from the re-examination. The bounding boxes are used to highlight the main face region and suppress background interference.

[0243] The modified face image can be used as the input image for the face expression recognition model to improve the stability of subsequent recognition.

[0244] Understandably, after obtaining the image through grayscale normalization, histogram equalization, face localization, and cropping, data augmentation is first performed to increase the number of sample images. Then, face re-examination is performed on each augmented image to check facial contours, feature distribution, and region integrity, selecting all qualified images as candidate face images. Finally, bounding boxes are drawn for each candidate face image to complete face region localization and annotation, resulting in the final processed face image.

[0245] This application provides a security management method for access control. First, the acquired original face image is processed for detection. Then, the detected face image is enhanced to obtain an optimized face image. This method, by sequentially performing image detection and image enhancement operations, effectively reduces interference from external environmental factors such as lighting and face pose, improves the image quality of the face image, provides a clear and reliable image foundation for subsequent face recognition and analysis, and thus improves overall recognition accuracy and operational stability.

[0246] Figure 15 A structural schematic diagram of the access control security management device provided in this application is shown below. Figure 15 As shown, the access control security management device 400 provided in this embodiment includes:

[0247] The acquisition module 401 is used to acquire the original facial images of passersby through the image acquisition device;

[0248] The processing module 402 is used to perform recognition processing on the original face image to obtain the expression classification results of the passers-by;

[0249] Module 403 is used to acquire security control evaluation criteria;

[0250] The processing module 402 is also used to perform behavior discrimination processing on the expression classification results according to the security control evaluation criteria to obtain the behavior level of the passerby;

[0251] Module 404 is used to determine the security management policy corresponding to the behavior level;

[0252] Execution module 405 is used to execute security management policies.

[0253] In one possible implementation, the processing module 402 is further configured to perform detection processing on the original face image to obtain a modified face image;

[0254] The device also includes: input module 406;

[0255] The input module 406 is used to input the modified face image into the pre-built face expression recognition model to obtain the expression classification result output by the face expression recognition model.

[0256] In one possible implementation, the processing module 402 is further configured to perform detection processing on the original face image to obtain a detected face image;

[0257] The processing module 402 is specifically used to enhance the detected face image to obtain a modified face image.

[0258] In one possible implementation, the processing module 402 is further configured to perform grayscale normalization processing on the original face image to obtain a grayscale face image.

[0259] The processing module 402 is also used to perform histogram equalization processing on the grayscale face image to obtain an enhanced face image;

[0260] The processing module 402 is also used to perform face detection processing on the enhanced face image to obtain a localized face image;

[0261] The processing module 402 is specifically used to perform size cropping processing on the located face image to obtain the detected face image.

[0262] In one possible implementation, the processing module 402 is further configured to perform data augmentation processing on the detected face image to obtain an augmented face image;

[0263] The processing module 402 is also used to perform face re-examination processing on the amplified face image to obtain candidate face images;

[0264] The processing module 402 is specifically used to draw bounding boxes on the candidate face image to obtain the modified face image.

[0265] In one possible implementation, the acquisition module 403 is also used to acquire security control evaluation criteria;

[0266] The processing module 402 is specifically used to perform behavioral discrimination processing on the facial expression classification results according to the security control evaluation criteria, so as to obtain the behavior level of the passerby.

[0267] In one possible implementation, the apparatus further includes: a generation module 407;

[0268] Generation module 407 is used to generate warning messages based on minor anomalies;

[0269] The device also includes: a startup module 408;

[0270] Startup module 408 is used to start the display component of the access control device;

[0271] The device also includes: a display module 409;

[0272] Display module 409 is used to display warning messages on display components.

[0273] In one possible implementation, the activation module 408 is also used to activate the first warning light assembly of the access control device;

[0274] The device also includes: a control module 410;

[0275] The control module 410 is used to control the first warning light component to broadcast abnormal light signals according to the first preset flashing frequency.

[0276] In one possible implementation, the activation module 408 is also used to activate the second warning light component and the voice warning component of the access control device;

[0277] The control module 410 is also used to control the second warning light assembly to broadcast abnormal light signals according to the second preset flashing frequency;

[0278] The startup module 408 is also used to control the voice warning component to broadcast abnormal sounds according to the preset sound frequency.

[0279] In one possible implementation, the device further includes: a drive module 411;

[0280] The drive module 411 is used to drive the access control device's access control opening and closing components to perform an opening action, so that the access control device's door is in an open state.

[0281] The access control security management device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0282] Figure 16 A structural diagram of the access control security management device provided in this application. Figure 16 As shown, the electronic device 500 provided in this embodiment includes at least one processor 501 and a memory 502. Optionally, the device 500 further includes a communication component 503. The processor 501, memory 502, and communication component 503 are connected via a bus 504.

[0283] In a specific implementation, at least one processor 501 executes computer execution instructions stored in memory 502, causing at least one processor 501 to perform the above-described method.

[0284] The specific implementation process of processor 501 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0285] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0286] The memory may include high-speed memory (Random Access Memory, RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0287] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0288] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0289] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0290] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0291] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application-Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0292] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0293] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0294] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0295] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0296] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0297] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A security management method for access control, characterized in that, Applied to access control equipment, the access control equipment including an image acquisition device, the method includes: The original facial images of passersby are captured using an image acquisition device; The original facial image is processed to obtain the expression classification results of the person passing by; The facial expression classification results are processed to determine the behavior level of the person passing through. Determine the security management policy corresponding to the behavior level, and execute the security management policy.

2. The method according to claim 1, characterized in that, The process of recognizing and processing the original facial image to obtain the expression classification result of the person passing by includes: The original face image is processed to obtain a modified face image; The modified face image is input into a pre-built facial expression recognition model to obtain the expression classification result output by the facial expression recognition model.

3. The method according to claim 2, characterized in that, The step of detecting and processing the original face image to obtain the modified face image includes: The original face image is processed to obtain the detected face image; The detected face image is enhanced to obtain a modified face image.

4. The method according to claim 3, characterized in that, The step of performing detection processing on the original face image to obtain the detected face image includes: The original face image is subjected to grayscale normalization to obtain a grayscale face image; The grayscale face image is subjected to histogram equalization to obtain an enhanced face image; The enhanced face image is subjected to face detection processing to obtain a localized face image; The measured face image is cropped to obtain the detected face image.

5. The method according to claim 3, characterized in that, The enhancement process for the detected face image to obtain the modified face image includes: The detected face image is subjected to data augmentation processing to obtain an augmented face image; The augmented face image is subjected to face re-examination processing to obtain candidate face images; The candidate face image is processed by drawing bounding boxes to obtain the modified face image.

6. The method according to claim 1, characterized in that, The step of performing behavior discrimination processing on the expression classification results to obtain the behavior level of the person passing through includes: Obtain security control and evaluation criteria; Based on the security control evaluation criteria, the facial expression classification results are processed for behavioral discrimination to obtain the behavioral level of the person passing through.

7. The method according to claim 1, characterized in that, If the behavior level is classified as mildly abnormal, determine the security management policy corresponding to the behavior level and execute the security management policy, including: Based on the aforementioned minor anomaly, a warning message is generated; The display component of the access control device is activated, and the warning message is displayed on the display component.

8. The method according to claim 1, characterized in that, If the behavior level is moderately abnormal, determine the security management policy corresponding to the behavior level and execute the security management policy, including: The first warning light component of the access control device is activated; The first warning light component is controlled to broadcast an abnormal light signal according to the first preset flashing frequency.

9. The method according to claim 1, characterized in that, If the behavior level is classified as severe abnormality, determine the security management policy corresponding to the behavior level and execute the security management policy, including: Activate the second warning light component and the voice warning component of the access control device; According to the second preset flashing frequency, the second warning light component is controlled to broadcast light abnormalities, and according to the preset sounding frequency, the voice warning component is controlled to broadcast sound abnormalities.

10. The method according to claim 1, characterized in that, If the behavior level is no anomaly, determine the security management policy corresponding to the behavior level and execute the security management policy, including... The access control device's opening and closing components are driven to perform an opening action, so that the access control device's door is in an open state.

11. A security management device for access control, characterized in that, Applied to access control equipment, the access control equipment includes an image acquisition device, comprising: The acquisition module is used to acquire raw facial images of people passing through the image acquisition device; The processing module is used to perform recognition processing on the original facial image to obtain the expression classification result of the person passing by; The acquisition module is used to acquire security control and evaluation criteria; The processing module is also used to perform behavior discrimination processing on the expression classification results according to the security control evaluation criteria to obtain the behavior level of the person passing through; The determination module is used to determine the security management strategy corresponding to the behavior level; The execution module is used to execute the security management policy.

12. A security management device for access control, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-10.

14. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-10.

Images