Access control equipment data management method and system based on multi-source fusion

Abstract

The invention discloses an access control equipment data management method and system based on multi-source fusion. The method comprises the following steps: collecting a multi-source access control data stream in real time; based on a preset feature extraction rule set, extracting a multi-modal biological feature vector, a voucher legality identifier, an abnormal behavior probability value and an equipment health degree index, and inputting the multi-modal biological feature vector, the voucher legality identifier, the abnormal behavior probability value and the equipment health degree index into a dynamic security assessment matrix generation model to generate a real-time security assessment matrix; matching the dimension safety score of the real-time safety evaluation matrix with a preset threshold strategy library, and dynamically generating an access control strategy instruction set; and issuing the access control strategy instruction set to the target access control equipment execution terminal. The method has the following advantages and effects: the fault tolerance bottleneck of a single-dimensional decision chain is broken through, and the system misjudgment rate is reduced by at least one order of magnitude on the premise of ensuring the security by establishing a dynamic coupling mechanism of the multi-source data stream.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a door access device data management method and system based on multi-source fusion. BACKGROUND

[0002] In the field of door access security, the prior art generally adopts a single-dimensional authentication mechanism. This single-source data processing method has the core defect of dimension fragmentation leading to decision-making errors: when transient anomalies of a specific data source occur, the system cannot compensate for errors through data in other dimensions, resulting in a significant increase in the probability of misjudgment.

[0003] For example, when video monitoring detects behavior anomalies such as loitering or nervous expressions, even if the biological feature matching degree is "qualified", its credibility should be significantly reduced, but the traditional system cannot reflect this dynamic weight reduction; when the door access device itself is in poor condition, such as response delay or communication packet loss, the reliability of its authentication results and the degree of threat to overall security are not included in real-time evaluation; This isolated, static, and lack of multi-dimensional correlation dynamic evaluation decision-making mode causes an irreconcilable contradiction between security and access efficiency. The core technical problem to be solved by the present application is precisely this: how to break through the fault tolerance bottleneck of single-dimensional decision chain and reduce the system misjudgment rate by at least an order of magnitude by establishing a dynamic coupling mechanism of multi-source data flow while ensuring security. SUMMARY

[0004] The purpose of the present application is to provide a door access device data management method and system based on multi-source fusion to solve the problems raised in the background art.

[0005] The above technical purpose of the present application is achieved by the following technical solution: The present application provides a door access device data management method based on multi-source fusion, comprising the following steps: S100, real-time acquisition of multi-source door access data flow; wherein the multi-source door access data flow includes biological feature recognition data flow, door access pass credential data flow, video monitoring behavior analysis data flow, and door access device running state data flow; S200, based on a pre-set feature extraction rule set, respectively extracting a multi-modal biological feature vector from the biological feature recognition data flow, a credential legality identifier from the door access pass credential data flow, an abnormal behavior probability value from the video monitoring behavior analysis data flow, and a device health degree index from the door access device running state data flow; S300, input the multi-modal biological feature vector, the legal identifier of the certificate, the abnormal behavior probability value and the device health index into a dynamic security evaluation matrix generation model to generate a real-time security evaluation matrix; wherein the real-time security evaluation matrix contains multiple dimension security scores; S400, match the dimension security scores of the real-time security evaluation matrix with a preset threshold strategy library to dynamically generate a set of access control strategy instructions; the set of access control strategy instructions includes an access permission instruction and a device maintenance alarm instruction; S500, issue the set of access control strategy instructions to a target access control device execution terminal.

[0006] By adopting the above technical solution, the collaborative processing mechanism of multi-source data streams is established to break the limitations of traditional access control system single data source decision-making; real-time collection and fusion analysis of biological feature recognition data streams, access pass certificate data streams, video monitoring behavior analysis data streams and access control device running state data streams generate a real-time security evaluation matrix containing multiple dimension security scores; this multi-dimensional coupled evaluation model significantly enhances the environmental adaptability and decision reliability of the system; the strategy instruction set generated based on the dynamic security evaluation matrix can maintain accurate judgment in high interference scenarios, effectively preventing systematic misjudgment caused by single source data anomalies; at the same time, through the linkage triggering of the device maintenance alarm instruction, the integration management of security efficiency and device operation and maintenance is realized, greatly improving the overall operation robustness of the access control system.

[0007] Further, the S200 includes extracting a multi-modal biological feature vector from the biological feature recognition data stream, which includes: for the biological feature recognition data stream, performing multi-modal feature fusion processing based on a preset feature extraction rule set to obtain the multi-modal biological feature vector; wherein the multi-modal feature fusion processing includes: performing principal component analysis dimension reduction processing on the fingerprint feature sequence in the biological feature recognition data stream to generate a fingerprint feature subvector; performing local binary pattern coding processing on the iris feature sequence in the biological feature recognition data stream to generate an iris feature subvector; performing surface fitting processing on the face three-dimensional point cloud data in the biological feature recognition data stream to generate a face three-dimensional feature subvector; combine the fingerprint feature subvector, the iris feature subvector and the face three-dimensional feature subvector into the multi-modal biological feature vector through a weighted series fusion algorithm; wherein the weighted series fusion algorithm dynamically adjusts the contribution of each feature subvector based on a preset feature weight coefficient; wherein the fingerprint feature weight coefficient is dynamically calibrated by the historical authentication accuracy rate; the iris feature weight coefficient is negatively correlated with the ambient light intensity; the face three-dimensional feature weight coefficient is adaptively scaled according to the point cloud data completeness.

[0008] By adopting the technical scheme, the fingerprint feature sub-vector, the iris feature sub-vector and the face three-dimensional feature sub-vector are respectively generated through principal component analysis dimension reduction processing, local binary pattern coding processing and surface fitting processing, and adaptive fusion is performed based on a dynamic weight coefficient; a historical authentication accuracy calibration mechanism of the fingerprint feature weight coefficient ensures continuous optimization of the biometric authentication parameter; the negative correlation characteristic of the iris feature weight coefficient and the environmental light intensity can actively compensate for the influence of optical interference; the face three-dimensional feature weight coefficient is dynamically scaled according to the point cloud data integrity, effectively solving the recognition blind area of the face shielding scene; and the multi-modal feature fusion mode significantly improves the fault tolerance of biometric authentication and provides high-robustness feature input for subsequent security evaluation.

[0009] Further, the S200 includes extracting a credential legality identifier from the access control pass credential data stream, including: performing a credential legality verification process on the access control pass credential data stream based on a preset feature extraction rule set to obtain the credential legality identifier; wherein the credential legality verification process includes: parsing the encrypted electronic credential ID in the access control pass credential data stream and performing authenticity verification through a preset asymmetric decryption algorithm library; verifying whether the credential validity period timestamp in the access control pass credential data stream is within a preset valid time window; calculating a deviation value of the historical usage frequency statistical value in the access control pass credential data stream from a preset usage frequency threshold; if the electronic credential is authentic, the credential validity period timestamp is valid, and the deviation value of the historical usage frequency statistical value from the preset usage frequency threshold is less than a preset deviation threshold, a legal identifier is generated as the credential legality identifier; otherwise, an illegal identifier is generated as the credential legality identifier. By adopting the technical scheme, the authenticity of the encrypted electronic credential ID is verified through the asymmetric decryption algorithm library, the timeliness of the credential validity period timestamp is verified based on the valid time window, and the usage behavior compliance detection is realized in combination with the historical usage frequency statistical value deviation analysis; the synergistic operation of the triple verification mechanism can accurately identify security threats such as credential forgery, overtime abuse, and high-frequency theft; and the interception capability for high-level security risks is further strengthened.

[0010] Further, the S200 includes extracting an abnormal behavior probability value from the video monitoring behavior analysis data stream, including: performing real-time behavior analysis processing on the video monitoring behavior analysis data stream based on a preset feature extraction rule set to obtain the abnormal behavior probability value; wherein the real-time behavior analysis processing includes: The original video frame sequence in the video monitoring behavior analysis data stream is detected frame by frame to obtain a human skeleton joint coordinate sequence corresponding to a continuous timestamp; based on the human skeleton joint coordinate sequence, a joint motion vector between adjacent frames is calculated, and a joint motion trajectory vector within a preset time window is constructed; the joint motion trajectory vector is input into a pre-trained abnormal posture classification model to output a posture deviation score; wherein the posture deviation score reflects the matching degree of the current posture and a preset standard passing posture; Micro-expression feature extraction is performed on the face region sub-picture in the video monitoring behavior analysis data stream to generate micro-expression intensity indicators including eye region muscle contraction frequency, mouth corner asymmetry index and forehead wrinkle change intensity; similarity matching is performed between the micro-expression intensity indicators and a preset micro-expression reference library to calculate an emotional abnormality confidence; According to the posture deviation score and the emotional abnormality confidence, an abnormal behavior probability value is generated through a weighted fusion formula.

[0011] By adopting the above technical solution, the posture deviation score is generated by matching the joint motion trajectory vector with the standard passing posture, and the emotional abnormality confidence is calculated based on the similarity matching of the micro-expression intensity indicators; the abnormal behavior probability value is formed by weighted fusion of the two analysis results, which completely covers the physical characteristics and psychological characteristics of behavior abnormalities; the human skeleton joint detection technology has precise sensing accuracy for abnormal limb movements, and micro-expression feature analysis such as eye muscle contraction frequency and mouth corner asymmetry can identify potential coercion behaviors; this dual monitoring mechanism of physiological behavior and psychological state greatly improves the early warning ability of hidden security threats.

[0012] Further, the S200 extracts the device health index from the access control device running state data stream, including: for the access control device running state data stream, performing health quantization processing based on a preset feature extraction rule set to obtain the device health index; wherein the health quantization processing includes: Monitoring the motor current fluctuation value in the access control device running state data stream, calculating the standard deviation of the motor current fluctuation value and a preset current reference value; Measuring the gate response delay time in the access control device running state data stream, calculating the ratio of the gate response delay time and a preset upper limit of response time; Statistical network communication packet loss rate in the access control device running state data stream; The standard deviation of the motor current fluctuation value and the preset current reference value, the ratio of the gate response delay time and the preset upper limit of response time, and the network communication packet loss rate are calculated by a weighted summation formula to obtain the device health index.

[0013] By adopting the technical scheme, the motor current fluctuation value standard deviation is monitored to detect the abnormality of the power system, the mechanical transmission efficiency is evaluated based on the gate response delay time ratio, and the control system stability is analyzed in combination with the network communication packet loss rate; the three-way device state parameters are weighted and summed to form a unified health degree index, thereby realizing the quantifiable diagnosis of the device operation state; the current fluctuation analysis can predict the motor winding aging risk, the response delay detection can identify the gate track stuck fault, and the network packet loss statistics can warn the control link interruption hidden danger; the multi-parameter fusion evaluation mechanism can improve the device state sensing accuracy and provide accurate data support for system maintenance decision-making.

[0014] Further, the S300 specifically includes the following sub-steps: S310, constructing a standardized feature input layer based on the multi-modal biological feature vector, the certificate legality identifier, the abnormal behavior probability value, and the device health degree index; wherein the standardized feature input layer performs the following processing: performing standardization processing on the multi-modal biological feature vector to generate a biological feature standardized vector; mapping the certificate legality identifier into a binary value; performing linear scaling on the abnormal behavior probability value; performing negative conversion on the device health degree index to generate a device risk index; S320, calculating the Pearson correlation coefficient of the biological feature standardized vector and the linearly scaled abnormal behavior probability value to generate a biological-behavior correlation factor; calculating the conditional probability value of the certificate legality identifier and the device risk index to generate a certificate-device coupling weight; and fusing the biological-behavior correlation factor and the certificate-device coupling weight to generate a dynamic fusion weight; S330, based on the biological feature standardized vector, the binary value of the certificate legality identifier, the linearly scaled abnormal behavior probability value, and the device risk index, respectively calculating a biological authentication security score, a certificate security score, a behavior risk score, and a device risk score, and performing weighted combination on the biological authentication security score, the certificate security score, the behavior risk score, and the device risk score and the dynamic fusion weight to generate a real-time security evaluation matrix containing multiple dimension security scores and corresponding fusion weight coefficients.

[0015] By adopting the technical scheme, the scale normalization of multi-source heterogeneous data is realized through the standardized feature input layer; the Pearson correlation coefficient analysis of the biological-behavior correlation factor reveals the potential correlation rule between physiological features and behavior features, and the conditional probability calculation of the certificate-device coupling weight quantifies the mutual influence degree of certificate security and device risk; the fusion of the two-way dynamic weight generates the final evaluation weight, so that the security evaluation matrix has the environment adaptive ability; the matrix construction mechanism breaks through the limitation of the traditional static score model and can still maintain reliable decision-making in the scenario where the biological authentication fails but the behavior is normal, thereby significantly reducing the system false rejection rate.

[0016] Further settings are that the S330 includes the following steps: The biological authentication security score is obtained by calculating the cosine similarity between the biological feature standardized vector and the pre-registered biological feature template; The credential security score is directly mapped to the binary value of the credential legitimacy identifier; The behavior risk score is obtained by converting the linearly scaled abnormal behavior probability value into a percentage system; The device risk score is obtained by reverse standardization of the device risk index.

[0017] By adopting the above technical solution, the matching degree between the real-time biological feature and the registered template is accurately quantified by cosine similarity calculation, which improves the accuracy of the biological authentication security score. The percentage conversion of the abnormal behavior probability value realizes the visual classification of the behavior risk, and the reverse standardization of the device risk index establishes a risk threshold warning mechanism. The standardized conversion of the four-dimensional score not only unifies the dimension, but also retains the physical meaning of each data source, providing a data basis for strategy generation.

[0018] Further settings are that the S400 specifically includes the following sub-steps: S410, generating a comprehensive security score based on the dimension security score in the real-time security evaluation matrix and the fusion weight coefficient ; S420, matching the comprehensive security score with a preset threshold strategy library to generate an initial access control strategy instruction set: If , an access permission instruction is generated; If , a device maintenance alarm instruction is generated; Wherein, are the hierarchical thresholds of the threshold strategy library, respectively; S430, calculating a strategy confidence based on the comprehensive security score, and if the strategy confidence is less than a set threshold, an artificial review forced flag is attached to the access control strategy instruction set.

[0019] By adopting the technical scheme, the multi-dimensional evaluation results are integrated through the comprehensive safety score, and the threshold strategy library is used to realize the hierarchical response of the access control; the quantitative calculation of the strategy confidence degree can identify the low reliability evaluation scene, and the introduction of the artificial review mandatory sign provides a review channel for the edge decision; when the system is in an extreme interference environment or an abnormal device state, the confidence degree checking mechanism can actively avoid the risk of automatic decision; and the balance between safety and passing efficiency can be realized on the premise of ensuring efficient passing in a conventional scene.

[0020] The application also provides a multi-source fusion-based access control equipment data management system, comprising the following modules: A multi-source data acquisition module is configured to acquire real-time multi-source access control data streams, and comprises: A biological sensor array is configured to acquire biological feature recognition data streams; An RFID reader cluster is configured to acquire access control credential data streams; An intelligent camera group is deployed in an access control area and is configured to acquire video monitoring behavior analysis data streams; An Internet of Things monitoring module is configured to acquire access control equipment operation state data streams; A feature extraction module is configured to execute the following processing based on a preset feature extraction rule set: A multi-modal feature fusion unit is configured to extract multi-modal biological feature vectors from the biological feature recognition data streams; A credential legality verification unit is configured to extract a credential legality identifier from the access control credential data streams; A real-time behavior analysis unit is configured to extract an abnormal behavior probability value from the video monitoring behavior analysis data streams; A health quantification unit is configured to extract an equipment health degree index from the access control equipment operation state data streams; A safety evaluation matrix generation module comprises: A standardized feature input layer is configured to perform standardization processing on the multi-modal biological feature vectors to generate biological feature standardized vectors, map the credential legality identifier to a binary value, perform linear scaling on the abnormal behavior probability value, and perform negative conversion on the equipment health degree index to generate an equipment risk index; A dynamic weight calculation unit is configured to calculate the Pearson correlation coefficient of the biological feature standardized vectors and the abnormal behavior probability value after linear scaling to generate a biological-behavior correlation factor, calculate the conditional probability value of the credential legality identifier and the equipment risk index to generate a credential-equipment coupling weight, and fuse to generate a dynamic fusion weight; The matrix construction unit is configured to calculate a biological authentication security score, a credential security score, a behavior risk score and a device risk score based on the biological feature standardized vector, the credential legitimacy identifier binary value, the linearly scaled abnormal behavior probability value and the device risk index respectively, and generate a real-time security evaluation matrix by combining the dynamic fusion weight and allocating the dimension fusion weight coefficient. The strategy generation module comprises: The comprehensive score unit is configured to generate a comprehensive security score based on the dimension security score of the real-time security evaluation matrix and the fusion weight coefficient. The threshold matching unit is configured to match the comprehensive security score with a preset threshold strategy library to dynamically generate a set of access control strategy instructions. The confidence degree checking unit is configured to calculate a strategy confidence degree based on the comprehensive security score, and if the strategy confidence degree is lower than a set threshold, an artificial review forced flag is added in the instruction set. The instruction execution module is configured to issue the set of access control strategy instructions to a target access control device execution terminal.

[0021] In summary, the present application has the following advantages: The fault tolerance bottleneck of single-dimension decision chain is broken, and the system misjudgment rate is reduced by at least one order of magnitude under the premise of ensuring security by establishing a dynamic coupling mechanism of multi-source data flow. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 The figure is a main flow block diagram of the embodiment. Figure 2 The figure is a flowchart of extracting a multi-modal biological feature vector from biological feature recognition data flow in S200 in the embodiment. Figure 3 The figure is a flowchart of extracting a credential legitimacy identifier from access control pass credential data flow in S200 in the embodiment. Figure 4 The figure is a flowchart of extracting an abnormal behavior probability value from video monitoring behavior analysis data flow in S200 in the embodiment. Figure 5 The figure is a flowchart of extracting a device health degree index from access control device running state data flow in S200 in the embodiment. Figure 6 The figure is a flowchart of S300 in the embodiment. Figure 7 The figure is a flowchart of S400 in the embodiment. Figure 8 The figure is a system block diagram of the embodiment. DETAILED DESCRIPTION

[0023] The present application will be further described in detail below with reference to the accompanying drawings.

[0024] As shown in the accompanying Figures 1 to 8 ; The embodiment discloses a gate control equipment data management method based on multi-source fusion, comprising the following steps: S100, real-time acquisition of multi-source gate control data stream; wherein the multi-source gate control data stream includes biological feature recognition data stream, gate control pass voucher data stream, video monitoring behavior analysis data stream and gate control equipment running state data stream; Specifically, biological feature recognition data stream is collected through a biological sensor array, and the biological feature recognition data stream includes fingerprint feature sequence, iris feature sequence and face three-dimensional point cloud data; The gate control pass voucher data stream is collected through an RFID reader cluster, and the gate control pass voucher data stream contains encrypted electronic voucher ID, voucher validity period timestamp and historical use frequency statistical value; The video monitoring behavior analysis data stream is collected through a group of intelligent cameras deployed in the gate control area, and real-time human body posture trajectory tracking and micro-expression analysis are performed on the original video stream; The gate control equipment running state data stream is collected through an Internet of Things monitoring module, and the gate control equipment running state data stream includes motor current fluctuation value, gate response delay time and network communication packet loss rate.

[0025] More specifically, the biological sensor array includes an integrated optical fingerprint instrument, a near-infrared iris camera and a structured light face scanner; the RFID reader cluster includes a 13.56 MHz high-frequency reader for reading employee cards to obtain encrypted electronic voucher ID, voucher validity period timestamp and historical use frequency statistical value containing SHA-256 encryption; the group of intelligent cameras includes a 4K resolution wide-angle camera array deployed 3 meters above the gate machine, which synchronously extracts eye muscle contraction frequency and mouth corner asymmetry index by real-time analyzing human skeleton joint coordinate sequence; the Internet of Things monitoring module includes collecting motor current fluctuation value through a Hall sensor, measuring gate response delay time through a laser range finder, and statistically analyzing communication packet loss rate through a network probe.

[0026] S200, based on the preset feature extraction rule set, respectively extracting multi-modal biological feature vectors from the biological feature recognition data stream, extracting voucher legality identifiers from the gate control pass voucher data stream, extracting abnormal behavior probability values from the video monitoring behavior analysis data stream, and extracting equipment health degree indicators from the gate control equipment running state data stream; S300, inputting the multi-modal biological feature vectors, the voucher legality identifiers, the abnormal behavior probability values and the equipment health degree indicators into a dynamic security evaluation matrix generation model to generate a real-time security evaluation matrix; wherein the real-time security evaluation matrix contains multiple dimension security scores; S400, matching the dimension security score according to the real-time security evaluation matrix with a preset threshold strategy library, dynamically generating a door access control strategy instruction set; the door access control strategy instruction set includes an access permission instruction and a device maintenance alarm instruction; S500, issuing the door access control strategy instruction set to a target door access device execution terminal.

[0027] In a possible implementation, the step S200 of extracting the multi-modal biometric feature vector from the biometric feature recognition data stream includes: performing multi-modal feature fusion processing on the biometric feature recognition data stream based on a preset feature extraction rule set to obtain the multi-modal biometric feature vector; wherein the multi-modal feature fusion processing includes: performing principal component analysis dimension reduction processing on the fingerprint feature sequence in the biometric feature recognition data stream to generate a fingerprint feature sub-vector; Specifically, the principal component analysis dimension reduction processing extracts the dominant feature components of the fingerprint ridge direction field based on a preset fingerprint feature dimension reduction dimension, discards the secondary components with a variance contribution rate lower than a preset contribution rate threshold, and forms the dimension-reduced fingerprint feature sub-vector; performing local binary pattern coding processing on the iris feature sequence in the biometric feature recognition data stream to generate an iris feature sub-vector; Specifically, the local binary pattern coding processing divides the iris annular region into several equal sectors based on a preset iris texture block rule, calculates the difference between the pixel gray value in each sector and the center pixel and performs binary coding, and aggregates to generate the iris feature sub-vector; performing surface fitting processing on the face three-dimensional point cloud data in the biometric feature recognition data stream to generate a face three-dimensional feature sub-vector; The surface fitting processing adopts a non-uniform rational B-spline surface reconstruction algorithm, extracts the curvature extreme points of the nose tip, eye corner and mouth corner based on a preset key feature point positioning rule, generates a smooth parameterized surface by minimizing the distance error from the point cloud to the surface, and finally extracts the surface normal vector direction histogram as the face three-dimensional feature sub-vector; The fingerprint feature sub-vector, the iris feature sub-vector and the face three-dimensional feature sub-vector are combined into a multi-modal biometric feature vector through a weighted serial fusion algorithm; wherein the weighted serial fusion algorithm dynamically adjusts the contribution degree of each feature sub-vector based on a preset feature weight coefficient; wherein the fingerprint feature weight coefficient is dynamically calibrated by the historical authentication accuracy rate; the iris feature weight coefficient is negatively correlated with the ambient light intensity; and the face three-dimensional feature weight coefficient is adaptively scaled according to the completeness of the point cloud data.

[0028] The process of outputting the fused multi-modal biometric feature vector is as follows:

[0029] wherein, is a post-fusion multi-modal biometric feature vector; is a fingerprint feature sub-vector; is an iris feature sub-vector; is a face three-dimensional feature sub-vector; , and are a fingerprint feature weight coefficient, an iris feature weight coefficient and a face feature weight coefficient, respectively.

[0030] Specifically, the fingerprint feature sub-vector coefficient is calibrated as follows:

[0031] wherein, is a preset base weight value; is a historical fingerprint authentication accuracy rate; is an accuracy rate threshold value; is a set learning rate coefficient; The calibration process of the iris feature weight coefficient is as follows:

[0032] wherein, is a maximum theoretical weight of the iris feature; is a set decay coefficient; is a normalized ambient light intensity, which is obtained by a light sensor; The calibration process of the face three-dimensional feature weight coefficient is as follows:

[0033] wherein, is a maximum weight of the face feature; is an effective point cloud proportion; is a preset posture compensation coefficient; is a real-time distance from the face to the camera; is a maximum effective recognition distance.

[0034] Embodiment 1 Extracting the fingerprint feature sub-vector: performing principal component analysis dimension reduction on a 120-dimensional original fingerprint feature sequence, compressing the dimension number to 30, and setting a contribution rate threshold value of 5%; Extracting the iris feature sub-vector: dividing an iris annular region into 32 sectors, and performing local binary pattern coding to generate a 256-bit vector; Extracting the face three-dimensional feature sub-vector: based on the curvature extreme points of the tip of the nose and the corners of the eyes, generating a normal vector histogram through a NURBS surface reconstruction method; The fingerprint feature sub-vector coefficient (0.92 - 0.85); wherein, the historical fingerprint authentication accuracy rate is 92%, the accuracy rate threshold is 85%, the learning rate coefficient is 0.05, the preset basic weight value is 0.4; the iris feature weight coefficient ; wherein, the maximum theoretical weight of the iris feature is 0.5, the decay coefficient is 0.6, the normalized ambient light intensity is 0.8; the face three-dimensional feature weight coefficient , the effective point cloud proportion is 93%, the ratio of is 0.8, the maximum weight of the face feature is 0.6; the process of outputting the fused multi-modal biometric feature vector is: .

[0035] In one possible implementation, the S200 includes extracting the credential legality identifier from the access control pass credential data stream, which includes: performing a credential legality verification process based on a preset feature extraction rule set on the access control pass credential data stream to obtain the credential legality identifier; wherein, the credential legality verification process includes: parsing the encrypted electronic credential ID in the access control pass credential data stream, and performing authenticity verification through a preset asymmetric decryption algorithm library; Specifically, the asymmetric decryption algorithm uses the RSA-2048 algorithm to decrypt the encrypted electronic credential ID, generates a decrypted credential digest value, and calculates the bit matching degree of the decrypted credential digest value and the pre-stored legal credential digest value. If the bit matching degree is greater than or equal to 0.95, a true authenticity verification pass flag is generated; otherwise, a true authenticity verification failure flag is generated.

[0036] verifying whether the credential validity period timestamp in the access control pass credential data stream is within a preset valid time window; Specifically, in the process of verifying whether the credential validity period timestamp in the access control pass credential data stream is within a preset valid time window, the preset valid time window is obtained; then, the timestamp compliance is verified:

[0037] wherein, is the timeliness verification flag; is the start time of the valid time window; is the validity time window deadline; is the validity period timestamp of the credential; if , a timeliness check pass flag is generated; if , a timeliness check failure flag is generated.

[0038] The deviation value of the historical usage frequency statistical value in the access control pass credential data stream and the preset usage frequency threshold value is calculated. Specifically, in the process of calculating the deviation value of the historical usage frequency statistical value in the access control pass credential data stream and the preset usage frequency threshold value:

[0039] wherein, is the historical usage frequency statistical value; is the preset usage frequency threshold value; is the absolute deviation value; if is less than or equal to the preset deviation threshold value , a frequency compliance flag is generated; otherwise, a frequency anomaly flag is generated; If the electronic credential is real, the validity period timestamp of the credential is valid, and the deviation value of the historical usage frequency statistical value and the preset usage frequency threshold value is less than the preset deviation threshold value, that is, the authenticity verification pass flag, the timeliness check pass flag and the frequency compliance flag are all in the valid state, a legal identifier is generated as the credential legality identifier; otherwise, an illegal identifier is generated as the credential legality identifier. Embodiment 2 The RSA-2048 decryption encrypted electronic credential ID has a bit matching degree of 97.3%, which is greater than the 95% threshold value, and a timeliness check pass flag is generated; the validity period timestamp of the credential is within the preset valid time window, and ; the historical usage frequency statistical value is less than the preset deviation threshold value , and a frequency compliance flag is generated.

[0040] In a possible implementation, the step of extracting an abnormal behavior probability value from the video monitoring behavior analysis data stream in S200 includes: Based on the preset feature extraction rule set, real-time behavior analysis processing is performed on the video monitoring behavior analysis data stream to obtain an abnormal behavior probability value; wherein, the real-time behavior analysis processing includes: The original video frame sequence in the video monitoring behavior analysis data stream is detected frame by frame to generate a human skeleton joint coordinate sequence corresponding to a continuous timestamp; based on the human skeleton joint coordinate sequence, a joint motion vector between adjacent frames is calculated, and a joint motion trajectory vector within a preset time window is constructed; the joint motion trajectory vector is input into a pre-trained abnormal posture classification model to output a posture deviation score; wherein the posture deviation score reflects the matching degree of the current posture and the preset standard passing posture; The process of outputting the posture deviation score by the abnormal posture classification model is: The real-time acquired joint motion trajectory vector is matched with the preset standard passing posture database for matching degree calculation; wherein the standard passing posture database stores the reference skeleton sequence of legal passing behavior; then the key motion features are extracted, including the number of abnormal joint motion accelerations, the proportion of abnormal joint angle duration, and the degree of left-right symmetric limb action deviation; The key motion features are analyzed by the pre-trained abnormal posture classification model to output a posture deviation score ranging from 0 to 1; wherein: The score value tends to 0, indicating that the current posture is highly matched with the standard passing posture; The score value tends to 1, indicating that the current posture is seriously deviated from the standard passing posture.

[0041] Further optimization can be made according to the real-time collected environmental light intensity and camera angle parameters to dynamically calibrate the posture deviation score.

[0042] The micro-expression feature extraction is performed on the face region sub-picture in the video monitoring behavior analysis data stream to generate micro-expression intensity indicators including eye region muscle contraction frequency, mouth corner asymmetry index, and forehead wrinkle change intensity; similarity matching is performed between the micro-expression intensity indicators and the preset micro-expression reference library to calculate an emotional abnormality confidence; According to the posture deviation score and the emotional abnormality confidence, an abnormal behavior probability value is generated through a weighted fusion formula.

[0043] The process of generating the abnormal behavior probability value is:

[0044] wherein, is the abnormal behavior probability value; is the normalized posture deviation score; is the emotional abnormality confidence; and are the set posture analysis weight coefficient and micro-expression analysis weight coefficient, respectively.

[0045] Specifically, the posture analysis weight coefficient is dynamically adjusted according to the ambient light intensity; the micro-expression analysis weight coefficient is positively correlated with the facial region image resolution; Preferably, if the ambient light intensity is less than a preset light intensity threshold , then , wherein is a reference weight value of the posture analysis weight; otherwise, ; Preferably, wherein is a reference weight value of the micro-expression analysis weight; is the facial region image resolution, which is measured in terms of vertical pixel number, and the scaling coefficient value is scaled when .

[0046] Embodiment 3 abnormal behavior probability value ; wherein the normalized posture deviation score is 0.67, the emotional abnormality confidence is 0.28, and are 0.35 and 0.45, respectively.

[0047] In one possible implementation, the step of extracting the device health degree index from the access control device running state data stream in S200 includes: performing health quantization processing based on a preset feature extraction rule set on the access control device running state data stream to obtain the device health degree index; wherein the health quantization processing includes: monitoring the motor current fluctuation value in the access control device running state data stream, calculating the standard deviation of the motor current fluctuation value and a preset current reference value; specifically:

[0048] wherein, is the standard deviation of the motor current fluctuation value and the preset current reference value; is the i th current sampling value; is the preset current reference value; is the sampling number; measuring the gate response delay time in the access control device running state data stream, calculating the ratio of the gate response delay time and a preset upper limit of response time; specifically:

[0049] wherein, is the ratio of the gate response delay time and the preset upper limit of response time; is the actual response time; ​a preset upper limit of response time; statistically counting a network communication packet loss rate in the running state data stream of the access control device; specifically,

[0050] wherein, the network communication packet loss rate; the total number of sent data packets; the number of successfully received data packets; and can be obtained from the communication log or record of the access control device and the control terminal; the motor current fluctuation value, the standard deviation of the preset current reference value, the ratio of the gate response delay time to the preset upper limit of response time, and the network communication packet loss rate are calculated by a weighted summation formula to obtain the device health index; specifically,

[0051] wherein, the device health index, , and are the current fluctuation weight, the response delay weight, and the network packet loss weight, respectively; the default values of the three are 0.5, 0.3, and 0.2, respectively.

[0052] As a preferred, in a high load scenario, it is increased to 0.6.

[0053] Embodiment 4 the standard deviation of the motor current fluctuation value and the preset current reference value , the ratio of the gate response delay time to the preset upper limit of response time , and the network communication packet loss rate The calculated values are 0.12, 0.84, and 0.053, respectively; in a high load scenario increased to 0.6; then .

[0054] In one possible implementation, S300 specifically includes the following sub-steps: S310, constructing a standardized feature input layer based on the multi-modal biometric feature vector, the credential legitimacy identifier, the abnormal behavior probability value and the device health degree index; wherein the standardized feature input layer performs the following processing: performing standardization processing on the multi-modal biometric feature vector to generate a biometric feature standardized vector; preferably, the standardization processing adopts Z-score; mapping the credential legitimacy identifier into a binary numerical value; preferably, the legal identifier is set as 1 and the illegal identifier is set as 0; performing linear scaling on the abnormal behavior probability value; preferably, the linear scaling is in the interval [0, 100] after the linear scaling; performing negative conversion on the device health degree index to generate a device risk index; S320, calculating a Pearson correlation coefficient of the biometric feature standardized vector and the abnormal behavior probability value after the linear scaling to generate a biological-behavior correlation factor; calculating a conditional probability value of the credential legitimacy identifier and the device risk index to generate a credential-device coupling weight; fusing the biological-behavior correlation factor and the credential-device coupling weight to generate a dynamic fusion weight; The process of generating the dynamic fusion weight is as follows:

[0055] wherein, is the dynamic fusion weight; is the biological-behavior correlation factor; is the credential-device coupling weight; and is a preset balance coefficient; Preferably, the process of calculating the conditional probability value of the credential legitimacy identifier and the device risk index to generate the credential-device coupling weight is as follows:

[0056] wherein, is the credential legitimacy identifier, 1 represents a legal identifier and 0 represents an illegal identifier; is the device risk index; is a preset device risk threshold; When , the device low-risk probability is higher, i.e. is smaller, and is larger; When , the device high-risk probability is higher, i.e. is larger, and is larger.

[0057] S330, based on the biometric feature standardization vector, the binary value of the credential legitimacy identifier, the linearly scaled abnormal behavior probability value and the device risk index, respectively calculate the biometric authentication security score, the credential security score, the behavior risk score and the device risk score, and combine the biometric authentication security score, the credential security score, the behavior risk score and the device risk score with the dynamic fusion weight to generate a real-time security evaluation matrix containing multiple dimension security scores and corresponding fusion weight coefficients.

[0058] The constructed real-time security evaluation matrix is specifically:

[0059] Among them, is the biometric authentication security score; is the credential security score; is the behavior risk score; is the device risk score; , , and are the corresponding fusion weight coefficients of each dimension after allocation; The allocation of the fusion weight coefficient is based on the dynamic fusion weight ; Preferably, the allocation rule of the fusion weight coefficient is:

[0060] Embodiment 5 Biological-behavior correlation factor , credential-device coupling weight ; then ; wherein, and The values of and are respectively set to 0.6 and 0.4.

[0061] The generation of the biometric authentication security score, the credential security score, the behavior risk score and the device risk score is shown in the following table: Table 1 Generation of biometric authentication security score, credential security score, behavior risk score and device risk score

[0062] In one possible implementation, the S330 based on the biometric feature standardization vector, the binary value of the credential legitimacy identifier, the linearly scaled abnormal behavior probability value and the device risk index, respectively calculate the biometric authentication security score, the credential security score, the behavior risk score and the device risk score include: The biometric authentication security score is obtained by calculating the cosine similarity between the biometric feature standardization vector and the pre-registered biometric feature template, and the score range is in the interval [0, 1]; The credential security score is directly mapped to a binary value of the credential legitimacy identifier, the legitimate identifier corresponds to a score of 1, and the illegitimate identifier corresponds to a score of 0; The behavior risk score is obtained by converting the linearly scaled abnormal behavior probability value into a percentage system; correspondingly, , wherein is the behavior risk score; The device risk score is obtained by inversely normalizing the device risk index.

[0063] Correspondingly, , is the device risk score; is the device risk index, which is obtained by negatively converting the device health index .

[0064] In one possible implementation, S400 specifically includes the following sub-steps: S410, generating a comprehensive security score based on the dimension security scores in the real-time security evaluation matrix and the fusion weight coefficients .The generation process of the comprehensive security score is as follows:

[0065] S420, matching the comprehensive security score with a preset threshold strategy library to generate an initial access control strategy instruction set: If , an access permission instruction is generated; If , a device maintenance alarm instruction is generated; wherein are hierarchical thresholds of the threshold strategy library, respectively; S430, calculating a strategy confidence based on the comprehensive security score, and if the strategy confidence is less than a set threshold, an artificial review forced flag is attached to the access control strategy instruction set; Preferably, the generation process of the strategy confidence is as follows:

[0066] wherein is the strategy confidence; is a theoretical maximum score.

[0067] Embodiment 6 The comprehensive security score ; , an access permission instruction is generated; The strategy confidence , greater than a set threshold 80, no manual review is triggered.

[0068] The embodiment also discloses a multi-source fusion-based access control equipment data management system, comprising the following modules: A multi-source data acquisition module is configured to acquire multi-source access control data streams in real time, and comprises: A biosensor array is configured to acquire a biometric feature recognition data stream comprising a fingerprint feature sequence, an iris feature sequence, and a three-dimensional point cloud data of a face; An RFID reader cluster is configured to acquire an access control pass data stream comprising an encrypted electronic pass ID, a pass validity period timestamp, and a historical use frequency statistical value; An intelligent camera group is deployed in an access control area and is configured to acquire a video monitoring behavior analysis data stream, and perform real-time human posture trajectory tracking and micro-expression analysis on an original video stream; An Internet of Things monitoring module is configured to acquire an access control equipment operating state data stream comprising a motor current fluctuation value, a gate response delay time, and a network communication packet loss rate; A feature extraction module is configured to perform the following processing based on a preset feature extraction rule set: A multi-modal feature fusion unit is configured to extract a multi-modal biometric feature vector from the biometric feature recognition data stream; A pass legality verification unit is configured to extract a pass legality identifier from the access control pass data stream; A real-time behavior analysis unit is configured to extract an abnormal behavior probability value from the video monitoring behavior analysis data stream; A health quantification unit is configured to extract an equipment health degree index from the access control equipment operating state data stream; A security evaluation matrix generation module comprises: A standardized feature input layer is configured to perform standardization processing on the multi-modal biometric feature vector to generate a biometric feature standardized vector, map the pass legality identifier to a binary value, perform linear scaling on the abnormal behavior probability value, and perform negative conversion on the equipment health degree index to generate an equipment risk index; A dynamic weight calculation unit is configured to calculate a Pearson correlation coefficient between the biometric feature standardized vector and the linearly scaled abnormal behavior probability value to generate a biometric-behavior correlation factor, calculate a conditional probability value between the pass legality identifier and the equipment risk index to generate a pass-equipment coupling weight, and fuse to generate a dynamic fusion weight; The matrix construction unit is configured to calculate a biological authentication security score, a credential security score, a behavior risk score and a device risk score based on the biological feature standardized vector, the credential legitimacy identifier binary value, the linearly scaled abnormal behavior probability value and the device risk index respectively; and generate a real-time security evaluation matrix by combining dynamic fusion weight distribution dimension fusion weight coefficients. The strategy generation module comprises: The comprehensive score unit is configured to generate a comprehensive security score based on the dimension security scores of the real-time security evaluation matrix and the fusion weight coefficients. The threshold matching unit is configured to match the comprehensive security score with a preset threshold strategy library to dynamically generate a set of access control strategy instructions. The confidence degree checking unit is configured to calculate a strategy confidence degree based on the comprehensive security score, and if the strategy confidence degree is lower than a set threshold, an artificial review forced flag is added in the instruction set. The instruction execution module is configured to issue the set of access control strategy instructions to a target access control device execution terminal.

[0069] The embodiments are merely illustrative of the present application, and are not intended to limit the present application. Those skilled in the art can make modifications to the embodiments according to the needs without creative contribution, and the modifications are protected by the patent law as long as they are within the scope of the claims of the present application.

Claims

1. A data management method for access control devices based on multi-source fusion, characterized in that, It includes the following steps: S100. Real-time acquisition of multi-source access control data streams; wherein, the multi-source access control data streams include biometric recognition data streams, access control pass data streams, video surveillance behavior analysis data streams, and access control equipment operation status data streams; S200. Based on a preset feature extraction rule set, extract multimodal biometric vectors from the biometric recognition data stream, extract credential validity identifiers from the access control credential data stream, extract abnormal behavior probability values ​​from the video surveillance behavior analysis data stream, and extract device health indicators from the access control device operating status data stream. S300. Input the multimodal biometric vector, credential validity identifier, abnormal behavior probability value and device health index into the dynamic security assessment matrix generation model to generate a real-time security assessment matrix; wherein, the real-time security assessment matrix contains multiple dimensions of security scores; S400. Match the dimensional security score of the real-time security assessment matrix with the preset threshold policy library to dynamically generate an access control policy instruction set; the access control policy instruction set includes access permission instructions and equipment maintenance alarm instructions. S500: The access control strategy instruction set is sent to the target access control device execution terminal.

2. The access control device data management method based on multi-source fusion according to claim 1, characterized in that: The step S200 of extracting multimodal biometric vectors from a biometric identification data stream includes: performing multimodal feature fusion processing on the biometric identification data stream based on a preset feature extraction rule set to obtain the multimodal biometric vectors; wherein the multimodal feature fusion processing includes: Principal component analysis is performed on the fingerprint feature sequence in the biometric identification data stream to reduce its dimensionality and generate fingerprint feature sub-vectors. The iris feature sequence in the biometric identification data stream is subjected to local binary pattern encoding to generate iris feature sub-vectors; The facial 3D point cloud data in the biometric recognition data stream is subjected to surface fitting processing to generate facial 3D feature sub-vectors. The fingerprint feature sub-vector, iris feature sub-vector, and facial 3D feature sub-vector are combined into the multimodal biometric vector using a weighted concatenation fusion algorithm. The weighted concatenation fusion algorithm dynamically adjusts the contribution of each feature sub-vector based on preset feature weight coefficients. The fingerprint feature weight coefficient is dynamically calibrated by the historical authentication accuracy. The iris feature weight coefficient is negatively correlated with the ambient light intensity. The facial 3D feature weight coefficient is adaptively scaled according to the completeness of the point cloud data.

3. The access control device data management method based on multi-source fusion according to claim 1, characterized in that: The extraction of the credential validity identifier from the access control credential data stream in step S200 includes: For the access control pass data stream, a pass validity verification process is performed based on a preset feature extraction rule set to obtain the pass validity identifier; wherein, the pass validity verification process includes: The encrypted electronic credential ID in the access control pass data stream is parsed and its authenticity is verified using a preset asymmetric decryption algorithm library; Verify whether the validity period timestamp of the access control pass data stream is within a preset valid time window; Calculate the deviation between the historical usage frequency statistics in the access control pass data stream and the preset usage frequency threshold; If the electronic voucher is authentic, the voucher's validity period timestamp is valid, and the deviation between the historical usage frequency statistics and the preset usage frequency threshold is less than the preset deviation threshold, then a valid identifier is generated as the voucher's validity identifier; otherwise, an invalid identifier is generated as the voucher's validity identifier.

4. The access control device data management method based on multi-source fusion according to claim 1, characterized in that: The extraction of abnormal behavior probability values ​​from the video surveillance behavior analysis data stream in step S200 includes: Based on a preset feature extraction rule set, real-time behavior analysis processing is performed on the video surveillance behavior analysis data stream to obtain the probability value of the abnormal behavior; wherein, the real-time behavior analysis processing includes: The original video frame sequence in the video surveillance behavior analysis data stream is subjected to frame-by-frame human skeleton joint point detection to generate a human skeleton joint point coordinate sequence corresponding to consecutive timestamps; based on the human skeleton joint point coordinate sequence, the joint point motion vector between adjacent frames is calculated, and the joint point motion trajectory vector within a preset time window is constructed; the joint point motion trajectory vector is input into a pre-trained abnormal posture classification model to output a posture deviation score; wherein, the posture deviation score reflects the matching degree between the current posture and the preset standard passing posture; Micro-expression features are extracted from the facial sub-images in the video surveillance behavior analysis data stream to generate micro-expression intensity indices, including the frequency of muscle contraction in the eye area, the asymmetry index of the corners of the mouth, and the intensity of changes in forehead wrinkles. Based on the micro-expression intensity indices, similarity matching is performed with a preset micro-expression benchmark library to calculate the confidence level of emotional abnormality. The probability value of the abnormal behavior is generated by a weighted fusion formula based on the posture deviation score and the confidence level of the emotional abnormality.

5. The access control device data management method based on multi-source fusion according to claim 1, characterized in that: The step S200 of extracting device health indicators from the access control device operation status data stream includes: performing health metric processing on the access control device operation status data stream based on a preset feature extraction rule set to obtain the device health indicators; wherein, the health metric processing includes: Monitor the motor current fluctuation value in the data stream of the access control device's operating status, and calculate its standard deviation from the preset current reference value; Measure the gate response delay time in the access control device's operating status data stream and calculate its ratio to a preset upper limit for response time; Statistically analyze the network communication packet loss rate in the data stream of the access control device's operating status; The equipment health index is calculated using a weighted summation formula, which is based on the standard deviation of the motor current fluctuation value from the preset current reference value, the ratio of the gate response delay time to the preset response time upper limit, and the network communication packet loss rate.

6. The access control device data management method based on multi-source fusion according to claim 1, characterized in that: S300 specifically includes the following sub-steps: S310. A standardized feature input layer is constructed based on multimodal biometric vectors, credential validity identifiers, abnormal behavior probability values, and equipment health indicators; wherein, the standardized feature input layer performs the following processing: standardizing the multimodal biometric vectors to generate standardized biometric vectors; mapping the credential validity identifiers to binary values; linearly scaling the abnormal behavior probability values; and performing a negative transformation on the equipment health indicators to generate an equipment risk index; S320. Calculate the Pearson correlation coefficient between the standardized biometric vector and the linearly scaled probability value of abnormal behavior to generate a biometric-behavioral association factor; calculate the conditional probability value between the credential validity identifier and the equipment risk index to generate a credential-equipment coupling weight; fuse the biometric-behavioral association factor and the credential-equipment coupling weight to generate a dynamic fusion weight. S330. Based on the biometric standardized vector, the binary value of the credential validity identifier, the linearly scaled abnormal behavior probability value, and the device risk index, calculate the biometric authentication security score, credential security score, behavior risk score, and device risk score respectively. Then, weight the biometric authentication security score, credential security score, behavior risk score, and device risk score with the dynamic fusion weight to generate a real-time security assessment matrix containing multiple dimensions of security scores and corresponding fusion weight coefficients.

7. The access control device data management method based on multi-source fusion according to claim 6, characterized in that: In step S330, the calculation of biometric authentication security score, credential security score, behavioral risk score, and device risk score based on the standardized biometric vector, the binary numerical value of the credential validity identifier, the linearly scaled abnormal behavior probability value, and the device risk index includes: The biometric authentication security score is obtained by calculating the cosine similarity between the standardized biometric vector and the pre-registered biometric template; The security score of the credential is directly mapped to a binary value of the credential validity identifier; The behavioral risk score is obtained by converting the linearly scaled probability value of abnormal behavior into a percentage system. The equipment risk score is obtained by inversely standardizing the equipment risk index.

8. The access control device data management method based on multi-source fusion according to claim 1, characterized in that: S400 specifically includes the following sub-steps: S410. Based on the dimensional security scores and fusion weight coefficients in the real-time security assessment matrix, generate a comprehensive security score. ; S420. Match the comprehensive security score with a preset threshold policy library to generate an initial access control policy instruction set: like Generate access permission instructions; like Generate equipment maintenance alarm commands; in, These are the tiered thresholds of the threshold strategy library; S430. Calculate the policy confidence level based on the comprehensive security score. If the policy confidence level is less than the set threshold, add a manual review mandatory flag to the access control policy instruction set.

9. A data management system for access control devices based on multi-source fusion, applied to the data management method for access control devices based on multi-source fusion as described in any one of claims 1-8, characterized in that, It includes the following modules: Multi-source data acquisition module: used for real-time acquisition of multi-source access control data streams; includes: A biosensor array configured to acquire biometric identification data streams; An RFID reader cluster is configured to collect access control pass data streams. Intelligent camera arrays, deployed in access control areas, are configured to collect video surveillance behavior analysis data streams; The IoT monitoring module is configured to collect data streams of the operating status of access control equipment; Feature extraction module: Based on a preset feature extraction rule set, it performs the following processing: The multimodal feature fusion unit is configured to extract multimodal biometric vectors from the biometric recognition data stream; The credential validity verification unit is configured to extract the credential validity identifier from the access control credential data stream; The real-time behavior analysis unit is configured to extract abnormal behavior probability values ​​from the video surveillance behavior analysis data stream. The health measurement unit is configured to extract device health indicators from the access control device's operating status data stream. Security assessment matrix generation module; including: The standardized feature input layer is configured to perform standardized processing on multimodal biometric feature vectors to generate standardized biometric feature vectors; map credential validity identifiers to binary values; perform linear scaling on abnormal behavior probability values; and perform negative transformation on equipment health indicators to generate equipment risk indices. The dynamic weight calculation unit is configured to calculate the Pearson correlation coefficient between the standardized vector of biometrics and the linearly scaled probability value of abnormal behavior to generate a biometric-behavioral association factor; calculate the conditional probability value between the credential validity identifier and the equipment risk index to generate credential-equipment coupling weights; and fuse them to generate dynamic fusion weights. The matrix construction unit is configured to calculate the biometric authentication security score, credential security score, behavioral risk score, and device risk score based on the biometric standardized vector, the binary numerical value of the credential validity identifier, the linearly scaled abnormal behavior probability value, and the device risk index, respectively; and generate a real-time security assessment matrix by combining the dynamic fusion weight allocation of the fusion weight coefficients of each dimension. The strategy generation module includes: The comprehensive scoring unit is configured to generate a comprehensive security score based on the dimensional security scores and fusion weight coefficients of the real-time security assessment matrix. The threshold matching unit is configured to match the comprehensive security score with a preset threshold policy library and dynamically generate an access control policy instruction set. The confidence verification unit is configured to calculate the confidence of the strategy based on the comprehensive security score. If it is lower than the set threshold, a manual review mandatory flag will be added to the instruction set. Instruction execution module: Distributes the access control policy instruction set to the target access control device execution terminal.

Images