An intelligent access control system for face recognition

Abstract

This invention discloses an intelligent access control system for facial recognition, relating to the field of facial recognition technology. It optimizes the facial recognition performance of the access control system under dynamic conditions through five modules: a resource monitoring module records peak memory usage data for each step under normal recognition conditions in static scenes; a multi-angle facial recognition module records images from different angles and obtains recognition coverage in dynamic scenes; a real-time analysis module analyzes memory fluctuations and abnormal recognition coverage to evaluate the system's real-time performance; a real-time performance level classification module classifies the system into high, medium, and low real-time performance levels based on the real-time performance evaluation results and takes corresponding measures; and an alarm signal management module dynamically adjusts the alarm trigger threshold by predicting future recognition performance at the medium real-time performance level, effectively reducing recognition latency, false alarms, and system overload, thus improving the security and stability of the access control system.

Description

Technical Field

[0001] The present invention relates to the technical field of face recognition, and in particular to an intelligent access control system for face recognition. Background Art

[0002] Smart access control for face recognition refers to the use of cameras to collect facial images of people, and use artificial intelligence algorithms to identify and match them, thereby automatically controlling the opening or rejection of access control. Compared with traditional access control systems, smart access control does not require keys, door cards or passwords. Users only need to point their faces at the camera, and the system can verify them through the stored facial data in the database. This method improves the convenience and security of the access control system, and is especially suitable for office buildings, communities and other places with high security requirements.

[0003] In the prior art, the verification steps of the intelligent access control system for face recognition include: first, the camera captures the real-time facial image of the person entering, and removes noise and enhances the image quality through preprocessing technology. Then, the system extracts key facial feature points in the image, such as eyes, nose, mouth, etc., and generates a corresponding digital feature vector. Next, the system compares the feature vector with the user's facial features stored in the database and calculates the similarity through an algorithm. If the similarity exceeds the set threshold, the system confirms the identity of the person and authorizes the access control to open; otherwise, it denies access or triggers a security alarm.

[0004] The prior art has the following deficiencies:

[0005] In the prior art, when the camera captures the real-time facial image of the person entering, frequent dynamic changes (such as micro-expressions, sudden head movements, and rapid access control) may require the access control system to reorganize and compare facial features multiple times in a very short time, thereby increasing the computational burden. If the camera frequently captures images at different angles, with different expressions, or blurry images, the system needs to constantly readjust the image, extract features, and compare the database, resulting in delays in the recognition process. At the same time, when recognition delays or failures occur frequently, the system may continuously trigger security alarms by mistake. When legitimate personnel are frequently denied access or the system misidentifies, it may trigger an emergency response from the security system, such as locking the door, activating protective measures, or sending an alarm signal to security managers. Continuous false alarms not only consume a lot of human resources to respond, but may also cause real emergencies to be covered up or ignored, further increasing safety risks. Repeated false triggers may also cause the system to shut down or be forced to reset, affecting the overall access control operation. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent access control system for face recognition to solve the shortcomings of the background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solutions: an intelligent access control system for face recognition, comprising a resource monitoring module, a multi-angle face recognition module, a real-time analysis module, a real-time level classification module and an alarm signal management module;

[0008] Resource monitoring module: sets the initial parameters of the access control system in a static scene, captures the facial image of the target user in real time through several cameras, and records the peak memory usage data of each facial image processing step under normal recognition conditions;

[0009] Multi-angle facial recognition module: Set the dynamic conditions in the test scene, record the facial images captured by the camera at different angles, mark the angle range of the facial images, pass the images at different angles to the facial recognition system, and obtain image recognition coverage data at different angles;

[0010] Real-time analysis module: Under dynamic conditions, analyze the fluctuation of the peak memory usage data of each facial image processing step and the abnormality of the image recognition coverage data at different angles to evaluate the real-time performance of the access control system in recognizing the target user's facial image under dynamic conditions;

[0011] Real-time level classification module: divides the real-time recognition of the target user's facial image by the access control system under dynamic conditions into different levels, namely, high real-time recognition level, medium real-time recognition level and low real-time recognition level, and performs corresponding processing;

[0012] Alarm signal management module: When the real-time recognition of the target user's facial image by the access control system under dynamic conditions is at a medium real-time recognition level, a predictive analysis is performed on the real-time recognition of the target user's facial image by the access control system under dynamic conditions within a fixed time period, and the triggering threshold of the security alarm is dynamically adjusted according to the analysis results.

[0013] Preferably, in the real-time analysis module, under dynamic conditions, the fluctuation of the peak memory usage data occupied by each facial image processing step is analyzed to generate a memory usage peak fluctuation index, and the method for obtaining the memory usage peak fluctuation index is:

[0014] First, collect the data of the peak memory usage, record the peak memory usage at each time point in the M time period and construct the time series data e(t), which represents the memory usage value at each time point t. Convert the time series signal e(t) into a signal X(f) in the frequency domain, where f is the frequency. For discrete time series data e[n], the expression is: ; where X[g] is the spectral component corresponding to frequency g, n is the discrete time point, N is the total number of sample points, j is the imaginary unit, and g is the frequency index, indicating the gth frequency component. For each frequency component X[g], the amplitude spectrum is calculated , the expression is: ; In the formula, Re(X[g]) is the real part of X[g], Im(X[g]) is the imaginary part of X[g]; By calculating the amplitude spectrum , determine the main frequency component, and calculate the amplitude of the main frequency component , the expression is: ; Use the sum of squared amplitudes to quantify overall volatility: ; Calculate the peak fluctuation index of memory usage, the expression is: ; Where DF is the peak fluctuation index of memory usage.

[0015] Preferably, under dynamic conditions, the abnormal degree of image recognition coverage data at different angles is analyzed to generate a recognition coverage abnormality index, and the method for obtaining the recognition coverage abnormality index is:

[0016] Collect recognition coverage data sets at m angles, and use the coverage data at each angle Indicates that c is the total number of coverage data, and for each data point, its neighborhood range is determined, that is, the number of neighbors k is selected, and the k-distance of each data point is calculated: ; In the formula, Yes The distance to the kth nearest neighbor of Yes and ; calculate the reachable distance of each point to its neighbors, ; Yes arrive If the reachable distance The k-distance is greater than , then take the k-distance and calculate the local reachability density. The calculation expression is: ; In the formula, Yes The local reachable density of The k-neighborhood of Yes The number of neighbors in the neighborhood, by comparing the points The local reachability density of and its neighbors is compared to calculate the recognition coverage anomaly index, and the expression is: ; Where LK is the recognition coverage anomaly index, Yes The local reachable density of .

[0017] Preferably, the memory occupancy peak fluctuation index and the recognition coverage anomaly index are converted into a first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model uses each group of first eigenvectors to predict the real-time value label of the target user's facial image recognition under dynamic conditions as the prediction target, and takes minimizing the sum of the prediction errors of all real-time value labels of the target user's facial image recognition under dynamic conditions as the training target. The machine learning model is trained until the sum of the prediction errors converges and the model training is stopped. The real-time value of the target user's facial image recognition under dynamic conditions is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0018] Preferably, in the real-time level classification module, the obtained real-time value of the target user's facial image recognition under dynamic conditions by the access control system is compared with a gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the real-time value of the target user's facial image recognition under dynamic conditions by the access control system is compared with the first standard threshold and the second standard threshold respectively;

[0019] If the real-time value of the access control system for the target user's facial image recognition under dynamic conditions is greater than the second standard threshold, it means that the access control system has high real-time recognition of the target user's facial image under dynamic conditions, and a high real-time signal is generated at this time, and it is classified as a high real-time recognition level, indicating that the access control system can quickly and accurately identify the user without additional adjustment;

[0020] If the real-time value of the access control system for target user facial image recognition under dynamic conditions is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the real-time performance of the access control system for target user facial image recognition under dynamic conditions is average, and a medium real-time signal is generated at this time, and it is classified as a medium real-time recognition level, indicating that the recognition accuracy and response speed of the access control system are reduced, and further analysis is needed;

[0021] If the real-time value of the access control system for target user facial image recognition under dynamic conditions is less than the first standard threshold, it means that the real-time performance of the access control system for target user facial image recognition under dynamic conditions is low. At this time, a low real-time signal is generated and it is classified as a low real-time recognition level, indicating that the recognition ability of the access control system is seriously reduced, and the management personnel is notified immediately for maintenance.

[0022] Preferably, in the alarm signal management module, at the medium real-time level, a prediction analysis is performed on the real-time performance within a fixed time in the future, wherein the prediction expression is: ; In the formula, To predict the time The real-time value at time q, R(q) is the actual real-time value at time q, is a model parameter, which indicates the influence weight of historical data on future prediction values, q is time, v is the total number of real-time values, and the predicted real-time values ​​are , dynamically adjust the trigger threshold of the security alarm, the alarm trigger adjustment formula is: ; In the formula, is the adjusted alarm trigger threshold, is the basic alarm trigger threshold, α is the adjustment coefficient; when the real-time prediction When it drops, immediately raise the alarm trigger threshold When the predicted real-time performance remains stable and increases, the alarm threshold remains unchanged; if the real-time performance of the access control system is predicted to drop below the low real-time recognition level, the alarm trigger threshold is reset.

[0023] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0024] 1. The present invention effectively solves the deficiencies in the prior art through the collaborative work of multiple modules. Through the resource monitoring module and the multi-angle facial recognition module, the system can record the peak memory usage and recognition coverage data at different angles in real time, and deeply analyze the memory fluctuations and coverage anomalies through the real-time analysis module, and accurately evaluate the recognition real-time of the access control system under dynamic conditions. Subsequently, the real-time level classification module divides the system into high, medium, and low real-time levels according to its real-time value, ensuring that the system can respond to different situations in a timely manner and take appropriate processing measures. For the medium real-time level, the alarm signal management module dynamically adjusts the alarm trigger threshold by predicting and analyzing the recognition performance in the future, so as to avoid false alarms and unnecessary system interruptions.

[0025] 2. The present invention utilizes multi-module collaboration, machine learning algorithms and dynamic adjustment mechanisms, and the system can adapt to frequent dynamic changes, such as micro-expressions, head movements, etc., thereby reducing the computational burden, reducing recognition delays and falsely triggered security alarms. At the same time, through in-depth analysis of the memory fluctuation index and the recognition coverage anomaly index, the system can predict the trend of performance degradation in advance, optimize the alarm triggering mechanism, and ensure that management personnel are promptly reminded to intervene in low real-time situations, thereby improving the stability, security and overall performance of the access control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0027] Figure 1 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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.

[0029] For examples, see Figure 1 As shown, the intelligent access control system for face recognition described in this embodiment includes a resource monitoring module, a multi-angle face recognition module, a real-time analysis module, a real-time level classification module and an alarm signal management module;

[0030] Resource monitoring module: sets the initial parameters of the access control system in a static scene, captures the facial image of the target user in real time through several cameras, and records the peak memory usage data of each facial image processing step under normal recognition conditions;

[0031] Multi-angle facial recognition module: Set the dynamic conditions in the test scene, record the facial images captured by the camera at different angles, mark the angle range of the facial images, pass the images at different angles to the facial recognition system, and obtain image recognition coverage data at different angles;

[0032] Real-time analysis module: Under dynamic conditions, analyze the fluctuation of the peak memory usage data of each facial image processing step and the abnormality of the image recognition coverage data at different angles to evaluate the real-time performance of the access control system in recognizing the target user's facial image under dynamic conditions;

[0033] Real-time level classification module: divides the real-time recognition of the target user's facial image by the access control system under dynamic conditions into different levels, namely, high real-time recognition level, medium real-time recognition level and low real-time recognition level, and performs corresponding processing;

[0034] Alarm signal management module: When the real-time recognition of the target user's facial image by the access control system under dynamic conditions is at a medium real-time recognition level, a predictive analysis is performed on the real-time recognition of the target user's facial image by the access control system under dynamic conditions within a fixed time period, and the triggering threshold of the security alarm is dynamically adjusted according to the analysis results.

[0035] In the resource monitoring module, the steps to initialize the access control system and static scene settings include: 1.1 Camera position calibration: Ensure that the camera is installed in a suitable position and can stably capture the front facial image of the target user. Adjust the height, angle and focal length of the camera to ensure clear capture of facial features. 1.2 Light control: In static scenes, ensure that the ambient light conditions are stable (such as no backlight or extreme light and dark differences). If necessary, an auxiliary light source can be set near the camera to ensure a clear image. 1.3 Network and system check: Ensure that the network connection of the access control system is stable, and check whether the hardware resources such as the processor and memory are operating normally to avoid affecting the recognition performance due to system hardware or network problems.

[0036] The steps for the camera to capture the target user's facial image in real time include: 2.1 Start the camera: Turn on the camera and start capturing the target user's frontal facial image in real time. During this process, ensure that the user remains still and avoids rapid changes in expression and posture. 2.2 Image acquisition parameter setting: Set the camera's resolution (such as 1080p or 4K), frame rate (such as 30 frames per second) and other image acquisition parameters to ensure that the captured image is clear and rich in details. 2.3 Image preprocessing (optional): Enable automatic image preprocessing in the system, including denoising, brightness adjustment, etc., to ensure that the captured image can achieve the best effect in the subsequent recognition steps.

[0037] The initialization steps of the facial recognition system processing flow include: 3.1 Image preprocessing: The system preprocesses the captured image, such as denoising, grayscale adjustment, image cropping, etc. The purpose is to optimize image quality and reduce unnecessary background interference. 3.2 Facial detection and feature extraction: The system uses an algorithm to detect the facial area in the image and extract facial feature points (such as eyes, nose, mouth, etc.). These feature points are converted into digital feature vectors for subsequent comparison. 3.3 Database comparison: The extracted feature vector is compared with the facial data stored in the database to calculate the match. The system records the comparison time and the comparison success rate.

[0038] The steps for real-time monitoring of peak memory usage data include: 4.1 Start the performance monitoring tool: Use system performance monitoring tools (such as top, htop on Linux or Task Manager on Windows) or dedicated monitoring software (such as Prometheus, Zabbix) to monitor memory usage. 4.2 Set memory monitoring parameters: Set the sampling frequency of the monitoring tool (for example, collect data once per second) to ensure that the peak memory usage changes of each step are captured frequently. 4.3 Mark each processing step: Mark each processing step (such as image preprocessing, feature extraction, database comparison, etc.) and record the memory usage of these steps synchronously during execution. Image preprocessing memory usage: Record the system's memory usage when performing preprocessing operations such as denoising, brightness adjustment, and cropping. Record the system's memory consumption when extracting facial features, including the process of feature point recognition and generating feature vectors. Record the system's memory consumption when performing feature comparison, especially when searching and comparing existing feature data in the database. 4.4 Capture memory peaks: When each step is completed, capture the peak memory usage of the step in real time and save it to the log or database for subsequent analysis.

[0039] In the multi-angle facial recognition module, determine the dynamic change conditions that need to be simulated in the test, such as: head angle: different angles of the user's face relative to the camera (front, side, slightly tilted, etc.), such as: 0° (front), ±30° (slightly side face), ±60° (significant side face), ±90° (completely side face), head movement: whether the speed of head movement (such as fast head turn, slow movement) will affect recognition. The impact of facial expression changes (such as smiling, frowning, etc.) on recognition accuracy during the test. Select multiple people of different genders, ages, and skin colors for testing to ensure data diversity. In the camera setting, ensure the same resolution and frame rate as the static test to ensure consistent image quality. The camera should be fixed in a stable position to ensure the reliability of the angle test data.

[0040] In the test site, set up clear angle marks (such as 0°, 30°, 60°, 90°) to ensure that the testers can stand or turn accurately at different angles. 2.2 Start image capture: Let the testers stand in front of the camera in turn, simulating different angles from 0° (front) to ±90° (complete side face), and keep the head still for a few seconds at each angle. The camera captures facial images at each angle in real time. Movement and expression change test: At each angle, the tester can also make slight head movements (such as slow head turns, fast head turns) or facial expression changes (such as smiling, frowning) to observe the impact of these changes on recognition. If possible, the camera can capture the entire process of dynamic changes in video form for subsequent analysis.

[0041] In each captured facial image file or video frame, manually or automatically annotate the angle range corresponding to the image (for example: 0°, ±30°, ±60°, ±90°). This can be achieved by using angle markers at the test site, or automatically annotating using the angle detection function of the camera. If video data is captured, the timestamp in the video can be used to record the specific time point when each angle changes to ensure that the image can accurately correspond to each angle during analysis. 3.3 Storage and classification: Image and video data with annotated angle ranges are stored by angle classification to ensure the convenience of subsequent data processing and analysis.

[0042] The facial images or video frame data with the annotated angles are passed to the facial recognition system for recognition. This usually involves inputting the data into the recognition engine and calculating the degree of match through feature extraction and comparison algorithms. The recognition system extracts facial feature points from each image, generates feature vectors, and compares them with the facial data stored in the database to obtain recognition results and matching scores. The recognition results at each angle are recorded, including key information such as whether the match is successful or not, matching score, recognition time, etc. The recognition coverage is defined as the proportion of successfully recognized images to the total number of images within a certain angle range, and the expression is: ; Classify and count the recognition results at different angles, and calculate the recognition coverage at each angle (such as 0°, ±30°, ±60°, ±90°). For scenes with dynamic changes (such as movement, expression changes), the recognition coverage can also be calculated separately. 5.3 Compare coverage: Compare and analyze the recognition coverage data at each angle to find out the differences in the system's recognition capabilities under different angles and dynamic conditions. 5.4 Record abnormal situations: Mark situations where the recognition failure rate is high or the recognition delay is obvious at certain angles or under dynamic change conditions, providing a basis for subsequent optimization of the system.

[0043] In the real-time analysis module, under dynamic conditions, the fluctuation of the peak memory usage data occupied by each facial image processing step is analyzed to generate a memory usage peak fluctuation index. The method for obtaining the memory usage peak fluctuation index is:

[0044] First, collect the data of the peak memory usage, record the peak memory usage at each time point in the M time period and construct the time series data e(t), which represents the memory usage value at each time point t. For example, the sampling frequency is once per second, and a total of 1024 sample points are collected. Convert the time series signal e(t) into a signal X(f) in the frequency domain, where f is the frequency. For discrete time series data e[n], the expression is: ; where X[g] is the spectral component corresponding to frequency g (in complex form), n is the discrete time point (i.e., sample), N is the total number of sample points, j is the imaginary unit, and g is the frequency index, indicating the gth frequency component. For each frequency component X[g], the amplitude spectrum is calculated , the expression is: ; Where Re(X[g]) is the real part of X[g], and Im(X[g]) is the imaginary part of X[g];

[0045] By calculating the amplitude spectrum , we can analyze which frequency components dominate. The frequencies corresponding to larger amplitudes represent the main fluctuation periods of system memory usage. Lower frequencies represent longer-term changes; higher frequencies represent short-term fluctuations. Calculate the amplitude of the main frequency component , the expression is: ; Use the sum of the squares of the amplitudes (energy) to quantify the overall fluctuation: ; Calculate the peak fluctuation index of memory usage, the expression is: ; Where DF is the peak fluctuation index of memory usage.

[0046] The larger the peak fluctuation index of memory usage, the more dramatic the memory usage fluctuation of the access control system under dynamic conditions. This fluctuation may be due to the frequent occupation and release of system resources when processing complex facial recognition tasks (such as facial images at different angles, rapidly changing expressions, and head movements). A large fluctuation index indicates that the system may encounter load pressure when processing dynamic changes, resulting in increased recognition delays, affecting the real-time and response speed of recognition, and may even cause a brief system performance degradation or freeze.

[0047] On the contrary, the smaller the peak fluctuation index of memory usage, the more stable the system memory usage and the more evenly distributed the load of processing facial images. A smaller fluctuation index means that the system can stably cope with changes under dynamic conditions, the allocation of resources is more efficient, and the real-time performance of recognition is higher. In this case, the system can process image data more quickly in dynamic scenes, ensuring the accuracy and smoothness of facial recognition, and is less likely to encounter performance bottlenecks or delays.

[0048] Under dynamic conditions, the abnormal degree of image recognition coverage data at different angles is analyzed to generate the recognition coverage abnormality index. The method for obtaining the recognition coverage abnormality index is:

[0049] Collect recognition coverage data sets at m angles, and use the coverage data at each angle Indicates that c is the total number of coverage data. The coverage data is standardized or normalized and mapped to a uniform scale (such as [0,1]). The neighborhood range is determined for each data point, that is, the number of neighbors k is selected (usually 5 to 20 neighbors are selected). The number of neighbors determines the calculation range of the local density. Select an appropriate k value, which is generally adjusted according to the size and density of the data set. Larger k values ​​are used for more uniform data sets, and smaller k values ​​are suitable for processing sparsely distributed data sets. The k-distance of each data point is calculated as follows: ; Yes The distance to the kth nearest neighbor of Yes and The distance between them (usually using Euclidean distance). Calculate the reachable distance of each point relative to its neighbors, ; Yes arrive If the reachable distance The k-distance is greater than , then take the k-distance, reflecting Influence in the local area. Calculate the local reachability density, which represents the density of a point relative to its neighbors, that is, the extent to which the point is covered by its neighbors. The lower the density, the more abnormal the point may be. The calculation expression is: ; In the formula, Yes The local reachable density of The k-neighborhood of Yes The number of neighbors in the neighborhood, by comparing the points The local reachability density of and its neighbors is compared to calculate the recognition coverage anomaly index, and the expression is: ; Where LK is the recognition coverage anomaly index, Yes The local reachable density of .

[0050] The larger the recognition coverage anomaly index is, the more the facial image recognition coverage at certain angles under dynamic conditions deviates significantly from the normal range, indicating that the system's recognition effect at these angles is poor, with more recognition failures or errors. This situation usually reflects the system's insufficient processing capabilities for certain angles, expression changes, or head movements in dynamic scenes, resulting in reduced real-time recognition. Reduced real-time performance not only means longer recognition time, but may also increase the frequency of false alarms, affecting the user's travel experience and system safety.

[0051] On the contrary, the smaller the recognition coverage anomaly index is, the more consistent and stable the recognition coverage is at different angles, the system can effectively cope with various changes under dynamic conditions, and the recognition effect is better. This means that the system processes facial images more evenly at different angles and under changing conditions, with higher recognition accuracy and better real-time performance. With high real-time performance, the system can quickly respond to and process recognition tasks, reduce delays and false alarms, and ensure the stability and efficiency of the access control system in a dynamic environment.

[0052] The memory usage peak fluctuation index and the recognition coverage anomaly index are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model predicts the real-time value label of the target user's facial image recognition under dynamic conditions by each group of first eigenvectors as the prediction target, and takes minimizing the sum of prediction errors of all real-time value labels of the target user's facial image recognition under dynamic conditions as the training target. The machine learning model is trained until the sum of prediction errors converges and the model training is stopped. The real-time value of the target user's facial image recognition under dynamic conditions is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0053] The method for obtaining the real-time value of the target user's facial image recognition under dynamic conditions by the access control system is as follows: from the first feature vector training data of the trained machine learning model, the corresponding function expression is obtained: ; In the formula, is the output function of the model, DF is the peak fluctuation index of memory usage, LK is the recognition coverage anomaly index, It is the real-time value of the access control system for target user facial image recognition under dynamic conditions.

[0054] In the real-time level classification module, the obtained real-time value of the access control system for the target user's facial image recognition under dynamic conditions is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the real-time value of the access control system for the target user's facial image recognition under dynamic conditions is compared with the first standard threshold and the second standard threshold respectively;

[0055] If the real-time value of the access control system for the target user's facial image recognition under dynamic conditions is greater than the second standard threshold, it means that the access control system has high real-time recognition of the target user's facial image under dynamic conditions, and a high real-time signal is generated at this time, and it is classified as a high real-time recognition level, indicating that the access control system can quickly and accurately identify the user without additional adjustment;

[0056] If the real-time value of the access control system for target user facial image recognition under dynamic conditions is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the real-time performance of the access control system for target user facial image recognition under dynamic conditions is average, and a medium real-time signal is generated at this time, and it is classified as a medium real-time recognition level, indicating that the recognition accuracy and response speed of the access control system have decreased, and further analysis is needed;

[0057] If the real-time value of the access control system for target user facial image recognition under dynamic conditions is less than the first standard threshold, it means that the real-time performance of the access control system for target user facial image recognition under dynamic conditions is low. At this time, a low real-time signal is generated and it is classified as a low real-time recognition level, indicating that the recognition ability of the access control system is seriously reduced. The management personnel is notified immediately, indicating that the system performance is poor and manual intervention or maintenance may be required.

[0058] In the alarm signal management module, at the medium real-time level, the real-time performance within a fixed time in the future is predicted and analyzed to ensure that corresponding measures are taken in advance when the system load increases or the recognition delay occurs. The prediction expression is: ; In the formula, To predict the time The real-time value at time q, R(q) is the actual real-time value at time q, is a model parameter, which indicates the weight of the impact of historical data on future prediction values, q is time, and v is the total number of real-time values. By fitting and predicting historical real-time data, the system can obtain future real-time change trends.

[0059] According to the predicted real-time value , the system can dynamically adjust the trigger threshold of the security alarm. When the predicted real-time performance may be close to the low real-time performance, the alarm threshold can be appropriately raised to avoid premature alarm triggering due to short-term performance fluctuations. The alarm trigger adjustment formula is: ; In the formula, is the adjusted alarm trigger threshold, is the basic alarm trigger threshold, which is usually a pre-set fixed value, and α is the adjustment coefficient (ranging from 0 to 1), which controls the dynamic adjustment range of the alarm threshold. The larger the α, the larger the adjustment range; when the real-time prediction When it drops, immediately raise the alarm trigger threshold When the predicted real-time performance remains stable and increases, the alarm threshold remains unchanged; if the real-time performance of the access control system is predicted to drop below the low real-time recognition level, the alarm trigger threshold is reset.

[0060] In this embodiment, the resource monitoring module is responsible for setting initial parameters in static scenes, capturing facial images through a camera and recording the peak memory usage data of each processing step; the multi-angle facial recognition module records facial images at different angles in dynamic scenes to obtain recognition coverage data; the real-time analysis module evaluates the real-time performance of the system under dynamic conditions by analyzing memory peak fluctuations and recognition coverage anomalies; the real-time level classification module divides the system into high, medium and low real-time levels based on the real-time analysis results, and takes corresponding processing measures; the alarm signal management module predicts and analyzes the future recognition real-time at the medium real-time level, and dynamically adjusts the alarm trigger threshold according to the results to optimize safety performance.

[0061] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. An intelligent access control system for face recognition, characterized by: It includes resource monitoring module, multi-angle facial recognition module, real-time analysis module, real-time classification module and alarm signal management module; Resource monitoring module: sets the initial parameters of the access control system in a static scene, captures the facial image of the target user in real time through several cameras, and records the peak memory usage data of each facial image processing step under normal recognition conditions; Multi-angle facial recognition module: Set the dynamic conditions in the test scene, record the facial images captured by the camera at different angles, mark the angle range of the facial images, pass the images at different angles to the facial recognition system, and obtain image recognition coverage data at different angles; Real-time analysis module: Under dynamic conditions, analyze the fluctuation of the peak memory usage data of each facial image processing step and the abnormality of the image recognition coverage data at different angles to evaluate the real-time performance of the access control system in recognizing the target user's facial image under dynamic conditions; Real-time level classification module: divides the real-time recognition of the target user's facial image by the access control system under dynamic conditions into different levels, namely, high real-time recognition level, medium real-time recognition level and low real-time recognition level, and performs corresponding processing; Alarm signal management module: When the real-time recognition of the target user's facial image by the access control system under dynamic conditions is at a medium real-time recognition level, a predictive analysis is performed on the real-time recognition of the target user's facial image by the access control system under dynamic conditions within a fixed time period, and the triggering threshold of the security alarm is dynamically adjusted according to the analysis results.

2. The intelligent access control system for face recognition according to claim 1, characterized in that: In the real-time analysis module, under dynamic conditions, the fluctuation of the peak memory usage data occupied by each facial image processing step is analyzed to generate a memory usage peak fluctuation index. The method for obtaining the memory usage peak fluctuation index is: First, collect the data of the peak memory usage, record the peak memory usage at each time point in the M time period and construct the time series data e(t), which represents the memory usage value at each time point t. Convert the time series signal e(t) into a signal X(f) in the frequency domain, where f is the frequency. For discrete time series data e[n], the expression is: ; where X[g] is the spectral component corresponding to frequency g, n is the discrete time point, N is the total number of sample points, j is the imaginary unit, and g is the frequency index, indicating the gth frequency component. For each frequency component X[g], the amplitude spectrum is calculated , the expression is: ; In the formula, Re(X[g]) is the real part of X[g], Im(X[g]) is the imaginary part of X[g]; By calculating the amplitude spectrum , determine the main frequency component, and calculate the amplitude of the main frequency component , the expression is: ; Use the sum of squared amplitudes to quantify overall volatility: ; Calculate the peak fluctuation index of memory usage, the expression is: ; Where DF is the peak fluctuation index of memory usage.

3. The intelligent access control system for face recognition according to claim 2, characterized in that: Under dynamic conditions, the abnormal degree of image recognition coverage data at different angles is analyzed to generate the recognition coverage abnormality index. The method for obtaining the recognition coverage abnormality index is: Collect recognition coverage data sets at m angles, and use the coverage data at each angle Indicates that c is the total number of coverage data, and for each data point, its neighborhood range is determined, that is, the number of neighbors k is selected, and the k-distance of each data point is calculated: ; In the formula, Yes The distance to the kth nearest neighbor of Yes and The distance between Calculate the reachable distance of each point to its neighbors, ； Yes arrive If the reachable distance The k-distance is greater than , then take the k-distance and calculate the local reachability density. The calculation expression is: ; In the formula, Yes The local reachable density of The k-neighborhood of Yes The number of neighbors in the neighborhood, by comparing the points The local reachability density of and its neighbors is compared to calculate the recognition coverage anomaly index, and the expression is: ; Where LK is the recognition coverage anomaly index, Yes The local reachable density of .

4. The intelligent access control system for face recognition according to claim 3 is characterized by: The memory usage peak fluctuation index and the recognition coverage anomaly index are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model predicts the real-time value label of the target user's facial image recognition under dynamic conditions by each group of first eigenvectors as the prediction target, and takes minimizing the sum of prediction errors of all real-time value labels of the target user's facial image recognition under dynamic conditions as the training target. The machine learning model is trained until the sum of prediction errors converges and the model training is stopped. The real-time value of the target user's facial image recognition under dynamic conditions is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

5. The intelligent access control system for face recognition according to claim 4 is characterized in that: In the real-time level classification module, the obtained real-time value of the access control system for the target user's facial image recognition under dynamic conditions is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the real-time value of the access control system for the target user's facial image recognition under dynamic conditions is compared with the first standard threshold and the second standard threshold respectively; If the real-time value of the access control system for the target user's facial image recognition under dynamic conditions is greater than the second standard threshold, it means that the access control system has high real-time recognition of the target user's facial image under dynamic conditions, and a high real-time signal is generated at this time, and it is classified as a high real-time recognition level, indicating that the access control system can quickly and accurately identify the user without additional adjustment; If the real-time value of the access control system for target user facial image recognition under dynamic conditions is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the real-time performance of the access control system for target user facial image recognition under dynamic conditions is average, and a medium real-time signal is generated at this time, and it is classified as a medium real-time recognition level, indicating that the recognition accuracy and response speed of the access control system are reduced, and further analysis is needed; If the real-time value of the access control system for target user facial image recognition under dynamic conditions is less than the first standard threshold, it means that the real-time performance of the access control system for target user facial image recognition under dynamic conditions is low. At this time, a low real-time signal is generated and it is classified as a low real-time recognition level, indicating that the recognition ability of the access control system is seriously reduced, and the management personnel is notified immediately for maintenance.

6. The intelligent access control system for face recognition according to claim 1, characterized in that: In the alarm signal management module, at the medium real-time level, the real-time performance within a fixed time in the future is predicted and analyzed, where the prediction expression is: ; In the formula, To predict the time The real-time value at time q, R(q) is the actual real-time value at time q, is a model parameter, which indicates the influence weight of historical data on future prediction values, q is time, v is the total number of real-time values, and the predicted real-time values ​​are , dynamically adjust the trigger threshold of the security alarm, the alarm trigger adjustment formula is: ; In the formula, is the adjusted alarm trigger threshold, is the basic alarm trigger threshold, α is the adjustment coefficient; when the real-time prediction When it drops, immediately raise the alarm trigger threshold When the predicted real-time performance remains stable and increases, the alarm threshold remains unchanged; if the real-time performance of the access control system is predicted to drop below the low real-time recognition level, the alarm trigger threshold is reset.

Images