A self-service visitor intelligent authorization method and system

Abstract

The present invention discloses a self-service visitor intelligent authorization method and system, which relates to the field of identity authentication management technology. The self-service visitor intelligent authorization method and system include the following steps: Step 1, collecting multimodal visitor data, assigning a unique session identifier to the visitor request, performing data preprocessing, and analyzing modal consistency; Step 2, performing concurrent conflict identification detection on the concurrent visitor identification request sessions entering the buffer queue, and after detection, performing conflict marking and separation processing on the visitor identification request sessions; Step 3, extracting the core information of the visitor identification request session, formulating a matching authorization strategy, performing risk assessment and behavior risk control, generating a decision and feeding back the decision result; Step 4, performing real-time feedback control, recording behavior logs and monitoring information, detecting and adjusting the load status in real time, and performing data visualization analysis. The system solves the problem of system confusion caused by multiple visitors scanning codes and performing face verification recognition at the same time in high-density channels.

Description

Technical Field

[0001] The present invention relates to the technical field of identity authentication management, and in particular to a self-service visitor intelligent authorization method and system. Background Art

[0002] With the rapidly growing demand for intelligent, secure, and automated visitor access management in scenarios such as office buildings, apartments, and large convention centers, traditional manual registration or single-point identification methods are gradually being replaced by digital identification systems. Currently, widely used visitor identification methods typically rely on single-modal identity verification methods such as QR code scanning, facial recognition, and ID card scanning to control visitor access.

[0003] For example, the invention patent with publication number CN119180355A discloses a visitor reservation management system, which involves the field of monitoring and controlling access. It includes a data acquisition module, a special processing module, a data analysis module, a signature security module, and a visualization port, and the modules are connected by signals. By collecting visitor information and QR code information, the number of times a visitor enters and exits the venue, the arrival rate, the security of the digital signature, etc. are obtained through processing. A data analysis model is established to generate a cache evaluation coefficient, which is compared with the cache threshold to obtain a cache judgment result and an alarm word, which are sent to the signature security module and the visualization port respectively. The signature security module uses fuzzy logic to determine the cache solution and sends it to the visualization port, effectively reducing the burden of real-time processing of the system, improving the efficiency and security of QR code verification, reducing the calculation delay during visitor verification, and making visitor entry and exit more convenient.

[0004] For example, the invention patent with announcement number CN115512527B announces a campus visitor management method and management system based on visitor identity identification, including the steps of obtaining the positioning information of the tag terminal in real time when receiving a start signal from a tag terminal worn by the visiting personnel; issuing a start instruction to the sensing terminal located at the visiting destination so that the sensing terminal can detect in real time whether the tag terminal bound to it is within the sensing range of the sensing terminal; stopping obtaining the positioning information of the tag terminal when receiving confirmation information from the sensing terminal indicating that the bound tag terminal is within the sensing range, and when the tag terminal is within the sensing range, the sensing terminal sends a confirmation information every preset time; and re-acquiring the positioning information of the tag terminal in real time when the tag terminal leaves the sensing range of the sensing terminal bound to it.

[0005] However, in the process of implementing the technical solutions of the invention in the embodiments of the present application, the present application found that the above technology has at least the following technical problems:

[0006] During the actual deployment process, especially during peak traffic hours or in entrance and exit environments with high population density and limited operating space, the existing visitor systems generally expose problems such as low operating efficiency, poor anti-interference ability, weak risk identification, and inflexible strategies. They are unable to meet the comprehensive requirements of the new generation of intelligent access systems for high concurrency processing and precise control. The systems usually rely on single-modal recognition such as face or QR code, lack a modal fusion mechanism, and are easily interfered by factors such as occlusion, lighting, and forgery. When multiple people operate at the same time, it is difficult to accurately distinguish the modal ownership, and "multi-code mixed scanning" or "person-document mismatch" are prone to occur.

[0007] Therefore, in response to the above problems, there is an urgent need for a self-service visitor intelligent authorization method and system. Summary of the Invention

[0008] Technical problems solved

[0009] In response to the shortcomings of the existing technology, the present invention provides a self-service visitor intelligent authorization method and system, which solves the problem of system confusion caused by multiple visitors scanning codes and performing facial verification and recognition at the same time in high-density channels.

[0010] Technical solution. To achieve the above objectives, the present invention is implemented through the following technical solution: a self-service visitor intelligent authorization method and system, comprising the following steps: step one, collecting multimodal visitor data, assigning a unique session identifier to the visitor request, performing data preprocessing, and analyzing modal consistency; step two, performing concurrent conflict identification detection on concurrent visitor identification request sessions entering the buffer queue, and after detection, performing conflict marking and separation processing on the visitor identification request sessions; step three, extracting the core information of the visitor identification request session, formulating a matching authorization strategy, performing risk assessment and behavioral risk control, generating a decision and feeding back the decision result; step four, performing real-time feedback control, recording behavior logs and monitoring information, detecting and adjusting the load status in real time, and performing data visualization analysis.

[0011] Furthermore, multimodal visitor data is collected, a unique session identifier is assigned to the visitor request, and the specific process of data preprocessing is as follows: when the visitor approaches the collection terminal, the infrared sensor device detects the human body approaching in real time, triggering the terminal to enter the working state and wake up the multimodal recognition process, and collects multimodal visitor data through the sensing device: the face recognition camera sensor collects face images and key points and extracts face features, and extracts face feature vectors based on the face image through the face recognition model based on angle interval; the identity information reading module reads the identity information and the visitor's identity photo, and extracts the identity photo feature vector based on the visitor's identity photo through the face recognition model based on angle interval; the infrared sensor captures the visitor's three-dimensional spatial position information; the QR code scanner reads the electronic pass code information, including the visitor's appointment time and appointment target area; a unique session identifier is assigned to the visitor request, including a timestamp and channel number; a retry and low-weight processing strategy is adopted for missing and abnormal multimodal visitor data, and the multimodal visitor data is standardized and normalized.

[0012] Furthermore, the specific process of modal consistency is analyzed as follows: obtaining facial feature vectors, and calculating the normalized cosine similarity of facial feature vectors based on the face recognition model of angle interval to obtain facial recognition confidence; obtaining identity information accuracy by the number of correct identity information characters recognized in the multimodal visitor data as the total number of identity information characters; based on the electronic pass code read in the multimodal visitor data, identifying whether the identity token built into the electronic pass code information matches the digital signature to obtain the QR code validity; multiplying the square of facial recognition confidence by the face reliability weight factor to obtain face matching, multiplying the square of identity information accuracy by the identity reliability weight factor to obtain identity information matching, multiplying the square of QR code validity by the code modal reliability weight factor to obtain QR code validity matching, multiplying facial recognition confidence, identity information accuracy and QR code validity by two-by-two, then summing them and multiplying them by the modal interaction weight factor to obtain modal interaction correlation terms, and performing face matching and identity information matching. The modal consistency evaluation value is obtained by summing the matching degree, QR code validity matching degree, and modal interaction correlation terms and taking the square root. When the modal consistency evaluation value is greater than or equal to the modal consistency threshold, the multimodal recognition results are considered highly consistent and the authorization is credible. The information in this recognition is saved in a temporary cache, and the permission engine is used for logical authorization judgment. The door opening command is issued to the gate controller, prompting that the authorization is passed. The modal consistency evaluation value, timestamp, and channel number are written to the backend database. When the modal consistency evaluation value is less than the modal consistency threshold, it is considered that there is a certain degree of modal conflict and identity ambiguity, and direct release should not be given. The unique session identifier is marked as a conflict label, and the user is required to re-scan the QR code and stand firmly for face re-recognition. The original multimodal visitor data is cached and the modal consistency evaluation is re-performed. If it is still less than the threshold, the pass is locked and prompted to go to the manual registration counter. The cloud platform is linked to notify the backend, and the low-scoring modal consistency evaluation value, abnormal modality, and visitor identity photo are encrypted and recorded.

[0013] Furthermore, the specific process of concurrent conflict identification detection for concurrent visitor identification request sessions entering the buffer queue is as follows: encapsulate the unique identification session identifier generated for each visitor into an identification request session object, add it to the concurrent buffer queue sorted by timestamp and channel number, and set the maximum concurrent capacity limit; perform a uniqueness check and preliminary spatial conflict prediction before adding a new identification request session, compare the time overlap and three-dimensional position proximity to mark the suspected conflict state in advance, and enter the subsequent conflict detection module for processing; calculate the spatial distance between the two people through the three-dimensional spatial position information in the multimodal visitor data, if the distance is less than the spatial distance threshold, it is determined to be the suspected same person, and then perform facial similarity judgment; obtain facial feature vectors, and calculate the normalized cosine similarity of facial feature vectors based on the face recognition model of angle interval; obtain the face image and key points collected by the face recognition camera sensor and perform real-time analysis and normalization based on the convolutional neural network to obtain active Live image quality score; using facial images and key points collected from multimodal visitor data, identify and calculate the loss ratio of key points in the facial image to obtain the degree of facial occlusion; use cosine similarity plus the product of the live image quality score and the quality enhancement weight factor minus the product of the facial occlusion degree and the occlusion degree weight factor, and multiply the result by the crowd density weight factor to obtain the facial similarity judgment value; when the facial similarity judgment value is less than the facial similarity threshold, it is marked as a low-trust identity and immediately jumps to conflict marking and separation processing, suspending subsequent modal fusion, and prompting the user to re-position the recognition through voice prompts. The user is allowed to retry up to two times, while recording the live image quality score and the degree of facial occlusion. If the threshold is still not met, it will automatically transfer to manual review and prompt the visitor to go to the registration counter for processing; when the facial similarity judgment value is greater than or equal to the facial similarity threshold, the facial modality is considered trustworthy and directly enters the authorization strategy and decision control process.

[0014] Furthermore, the specific process of conflict marking and separating visitor identification request sessions after detection is as follows: when the camera sensor module detects that two or more identification request sessions are highly overlapping in time and close in spatial location, that is, when it is determined to be a suspected conflict, the relevant session status is marked as conflict pending, and the visitor is prompted through voice and screen to rescan the code later. At the same time, detailed information of the conflict event, including the unique session identifier, timestamp, spatial distance, and facial features, is recorded in the background, and the session is paused to enter the subsequent identification process. Then, a short time window is entered for re-identification, and the facial image, three-dimensional spatial location information, and electronic pass code information are re-collected. If it is determined that the session is significantly different from the other sessions, the conflict mark is removed and the status is updated to conflict resolved. If the conflict cannot be resolved after multiple retries, it is marked for manual review and the visitor is prompted to go to the manual registration counter. If the session is confirmed to be spatially independent and feature-independent individuals in the initial judgment, its status is directly marked as independent, conflict processing is skipped, and the authorization module is entered to complete the normal passage process.

[0015] Furthermore, the specific process of extracting the core information of the visitor identification request session, formulating a matching authorization strategy, and conducting risk assessment and behavioral risk control is as follows: obtaining the core information of the visitor identification request session, including identity information, appointment time, and appointment target area; allowing visitors to enter thirty minutes before and after the appointment time, not allowing visitors to visit specific floors, and allowing visitors to scan the code repeatedly at most three times a day; obtaining the number of repeated scans by the visitor by identifying the frequency of occurrence of the unique session identifier of the same visitor per unit time; obtaining the number of attempts to obtain the number of unauthorized area attempts by identifying the number of attempts in which the appointment target area of ​​the electronic pass code information scanned by the visitor in the multimodal visitor data collection phase does not match its authority; obtaining the face feature vector and the identity photo feature vector, and using the face recognition model based on the angle interval The model is normalized by the cosine similarity between the two vectors to obtain the inconsistency between the person and the certificate; the channel congestion is obtained by identifying the ratio of the number of visitor identification request sessions of the channel per unit time to the maximum number of visitor identification request sessions of the channel in the recent period; the face similarity judgment value is calculated; the inverse of the person and the certificate inconsistency is exponentially calculated and then added with a constant 1 to obtain the person and the certificate consistency adjustment item; the number of repeated code scans and the number of attempts in the unauthorized area are added and then divided by the person and the certificate consistency adjustment item to obtain the behavior anomaly score item; the face similarity judgment value is subtracted from the constant 1 and then squared to obtain the face image untrustworthiness penalty item; the sum of the behavior anomaly score item, the face image untrustworthiness penalty item and the channel congestion is multiplied by the anomaly adjustment weight factor to obtain the anomaly assessment value.

[0016] Furthermore, the specific process of generating decisions and feeding back decision results is as follows: when the abnormal assessment value is less than the risk threshold, it is judged to be normal and reliable, directly released, enters the authorization module, and writes a normal log without review; when the abnormal assessment value is greater than or equal to the risk threshold, it is judged to be abnormal risk, and the user is prompted to adjust the position and re-scan the code, retaining the first failed data, allowing the visitor to identify again. If it is still abnormal after the retry, the visitor identification request session will be automatically paused, and it will be displayed to go to manual registration, generate a high-risk log and capture the face image, and link the background security management system to push abnormal personnel reminders; when the identification process is completed and the authorization decision result is generated, the result and related identification status will be pushed to the local terminal interface and background in real time.

[0017] Furthermore, real-time feedback control is carried out, behavior logs and monitoring information are recorded, and the specific process of real-time detection of load status and adjustment is as follows: if the authorization step is passed, the green indicator light will be on and the gate will be opened automatically; if the authorization step fails, the red indicator light will flash, the voice prompt information will be abnormal, and the error code will be returned and recorded; the visitor's passage time, channel number, result status, response delay, and failure rate information will be recorded; the channel congestion is obtained by identifying the ratio of the number of visitor identification request sessions in the channel per unit time to the maximum number of visitor identification request sessions in the channel in the recent period; the system response delay is obtained by automatically calculating the time from the visitor scanning the code to the feedback of the actual identification response; the current window identification failure rate is obtained by identifying the historical multimodal visitor data within five minutes and calculating the number of all visitor request failures and the total number of identifications; the current window identification failure rate is negated and multiplied by the constant The suppression factor term is obtained by adding the numbers together, the sum of the channel congestion and the system response delay is divided by the suppression factor term, and the result is multiplied by the global adjustment weight factor to obtain the pressure adjustment value; the pressure adjustment value is compared with the pressure threshold in real time. When the pressure adjustment value is less than the pressure threshold, it is considered to be a normal load state, and the recognition process is executed according to the default strategy, all modes are enabled, the original recognition threshold is maintained, complete voice and screen prompt feedback, standard logs are recorded and the background visualization status is maintained; when the pressure adjustment value is greater than or equal to the pressure threshold, it is considered to be a high load state, and the policy compensation mechanism is automatically enabled, including raising the face and consistency judgment threshold, temporarily closing some low-priority modes, limiting the number of user retries, simplifying the voice and screen prompt frequency, adjusting the modal fusion weight, compressing the recognition process response time, and at the same time issuing a high-pressure warning to the background and strengthening log recording.

[0018] Furthermore, the specific process of data visualization analysis is as follows: construct a risk score trend line chart to display the abnormal assessment values ​​in different time periods during the visitor identification process, and simultaneously visualize three key influencing factors: the inconsistency between person and document, the penalty item of unreliability of facial image, and the channel congestion, to identify the source of risk and the changing trend; with time as the horizontal axis and the abnormal assessment value as the vertical axis, the main line highlights the risk score, and the remaining curves use different marks to distinguish the modal influencing factors, and the visualization is embedded in the management background to generate analysis reports regularly to identify high-pressure channels, abnormal time periods and modal failures.

[0019] Furthermore, it includes a multimodal acquisition and identification processing module, a concurrent conflict detection and resolution module, an authorization strategy and decision control module, and an interactive feedback and monitoring optimization module: the multimodal acquisition and identification processing module is used to collect multimodal visitor data, assign a unique session identifier to the visitor request, perform data preprocessing, and analyze modal consistency; the concurrent conflict detection and resolution module is used to perform concurrent conflict identification detection on the concurrent visitor identification request session entering the buffer queue, and after detection, perform conflict marking and separation processing on the visitor identification request session; the authorization strategy and decision control module is used to extract the core information of the visitor identification request session, formulate matching authorization strategies, perform risk assessment and behavior risk control, generate decisions and feedback decision results; the interactive feedback and monitoring optimization module is used to perform real-time feedback control, record behavior logs and monitoring information, detect and adjust the load status in real time, and perform data visualization analysis.

[0020] Beneficial effects

[0021] The present invention has the following beneficial effects:

[0022] (1) The present invention constructs a unified modal consistency evaluation value by integrating multimodal information such as facial images, identity information, QR code data, and spatial position, performs standardization and anomaly compensation in the data preprocessing stage, and enhances the identity judgment ability through modal confidence weighting and cross-modal interaction terms, effectively avoiding the risk of misidentification by a single recognition method in occlusion, lighting or blurred scenes, and significantly improving the recognition accuracy and environmental adaptability.

[0023] (2) The present invention, by introducing a unique session identifier mechanism, spatial position analysis, facial feature similarity calculation and live image quality assessment, and combining crowd density weight factors with spatial overlap rules, achieves refined recognition and processing of typical concurrent conflict scenarios such as "multiple people on the same screen", "multiple code mixed scanning", and "alternative recognition", effectively ensuring the identity independence and security in the process of simultaneous multi-user recognition.

[0024] (3) The present invention constructs a comprehensive risk scoring model, integrates multi-dimensional factors such as visitor behavior (repeated code scanning, unauthorized access), modal quality (inconsistency between person and ID card, unclear face image) and environmental status (channel congestion), generates a dynamic abnormality assessment value, and combines the global adjustment factor to achieve threshold adaptive control, thus realizing an intelligent passage decision-making mechanism with "quantifiable risk and adjustable strategy" in the identification process, greatly improving the system's risk prevention and control capabilities.

[0025] (4) The present invention generates multi-dimensional visual analysis results including risk score trend charts, identity inconsistency trends, channel load fluctuations, etc. through state feedback control, real-time behavior log recording and identification load calculation, and embeds them into the background management platform to identify high-risk sessions and system bottlenecks, support on-demand adjustment of system strategies and security warning linkage, and improve overall controllability, observability and maintainability.

[0026] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A flowchart of a self-service visitor intelligent authorization method;

[0028] Figure 2 This is a structural diagram of a self-service visitor intelligent authorization system;

[0029] Figure 3 A line chart showing the risk score trend of a self-service visitor intelligent authorization method and system. DETAILED DESCRIPTION

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0031] See also Figure 1-Figure 3The embodiment of the present invention provides a technical solution: a self-service visitor intelligent authorization method and system, comprising the following steps: step 1, collecting multimodal visitor data, assigning a unique session identifier to the visitor request, performing data preprocessing, and analyzing modal consistency; step 2, performing concurrent conflict identification detection on the concurrent visitor identification request sessions entering the buffer queue, and after detection, performing conflict marking and separation processing on the visitor identification request sessions; step 3, extracting the core information of the visitor identification request session, formulating a matching authorization strategy, performing risk assessment and behavior risk control, generating a decision and feeding back the decision result; step 4, performing real-time feedback control, recording behavior logs and monitoring information, detecting and adjusting the load status in real time, and performing data visualization analysis.

[0032] Specifically, the specific process of collecting multimodal visitor data, assigning a unique session identifier to the visitor request, and performing data preprocessing is as follows: when the visitor approaches the collection terminal, the infrared sensor device detects the human body approaching in real time, triggering the terminal to enter the working state and wake up the multimodal recognition process, and collecting multimodal visitor data through the sensing device: the face recognition camera sensor collects face images and key points and extracts face features, and extracts face feature vectors based on the face image through the face recognition model based on the angle interval, thereby achieving high-precision face identity information acquisition and image quality control; the identity information reading module reads the identity information and the visitor's identity photo, and extracts the identity photo feature vector based on the face recognition model based on the angle interval, thereby enhancing the document image. The modal comparison capability with on-site images facilitates subsequent consistency judgment of identity verification; the infrared sensor captures the three-dimensional spatial location information of visitors, ensuring the spatial independence of each visitor in concurrent scenarios, helping to prevent mixed scanning of multiple codes and interference in identification; the QR code scanner reads the electronic pass code information, including the visitor's appointment time and target area, providing a structured spatiotemporal label basis for authorization decisions, improving the timeliness and regional accuracy of access control; a unique session identifier is assigned to the visitor request, including a timestamp and channel number, ensuring the uniqueness of data binding and traceability of session management during the multi-visitor identification process; a retry and low-weight processing strategy is adopted for missing and abnormal multimodal visitor data, and multimodal visitor data is standardized and normalized.

[0033] In this implementation plan, through the multimodal visitor data collection and preprocessing process, the system achieves high-precision face recognition and image quality control, enhances the consistency comparison capability between documents and on-site images, and improves spatial independence recognition and anti-interference performance in concurrent scenarios; at the same time, structured spatiotemporal information and unique session identifiers ensure data traceability and authorization accuracy, combined with fault-tolerant processing of abnormal modalities and data standardization operations, significantly enhancing the system's recognition stability and fusion robustness in complex environments.

[0034] Specifically, the specific process of modal consistency analysis is as follows: obtain the facial feature vector, and calculate the normalized cosine similarity of the facial feature vector based on the face recognition model of the angle interval to obtain the facial recognition confidence; obtain the identity information accuracy by the number of correct identity information characters recognized in the multimodal visitor data to the total number of identity information characters; based on the electronic pass code read in the multimodal visitor data, identify whether the identity token built into the electronic pass code information matches the digital signature to obtain the QR code validity; multiply the square of the facial recognition confidence by the facial reliability weight factor to obtain the face matching degree, multiply the square of the identity information accuracy by the identity reliability weight factor to obtain the identity information matching degree, multiply the square of the QR code validity by the code modal reliability weight factor to obtain the QR code validity matching degree, multiply the facial recognition confidence, identity information accuracy and QR code validity by two-by-two, then sum and multiply by the modal interaction weight factor to obtain the modal interaction correlation term, and perform face matching and identity information matching. The modal consistency evaluation value is obtained by summing the matching degree, QR code validity matching degree, and modal interaction correlation terms and taking the square root. When the modal consistency evaluation value is greater than or equal to the modal consistency threshold, the multimodal recognition results are considered highly consistent and the authorization is credible. The information in this recognition is saved in a temporary cache, and the permission engine is used for logical authorization judgment. The door opening command is issued to the gate controller, prompting that the authorization is passed. The modal consistency evaluation value, timestamp, and channel number are written to the backend database. When the modal consistency evaluation value is less than the modal consistency threshold, it is considered that there is a certain degree of modal conflict and identity ambiguity, and direct release should not be given. The unique session identifier is marked as a conflict label, and the user is required to re-scan the QR code and stand firmly for face re-recognition. The original multimodal visitor data is cached and the modal consistency evaluation is re-performed. If it is still less than the threshold, the pass is locked and prompted to go to the manual registration counter. The cloud platform is linked to notify the backend, and the low-scoring modal consistency evaluation value, abnormal modality, and visitor identity photo are encrypted and recorded.

[0035] Among them, the specific calculation formula of the modal consistency evaluation value is:

[0036] ;

[0037] Where, is the modal consistency evaluation value; Confidence of face recognition; is the accuracy of identity information; The validity of the QR code; The face reliability weight factor is obtained by fitting the success rate of the last 500 face recognitions calculated through system monitoring logs, and ranges from 0.5 to 1; is the identity reliability weight factor, which is obtained by fitting the successful resolution rate of the last 500 identity information calculated through system monitoring logs and ranges from 0.5 to 1; is the code modality reliability weight factor, which is obtained by fitting the pass rate of the last 500 electronic pass code verifications calculated through system monitoring logs and ranges from 0.5 to 1; is the modal interaction weight factor, which is obtained by collecting several sets of face recognition confidence, identity information accuracy and QR code validity, and cross-validated using a grid search algorithm, and ranges from 0 to 0.3.

[0038] In this implementation plan, the system integrates facial recognition confidence, identity information accuracy and QR code validity, and introduces modal weights and interaction factors to construct modal consistency evaluation values, thereby achieving accurate judgment of visitor identity consistency under multimodal data; when the evaluation value meets the standard, the system automatically authorizes and caches the recognition results to improve traffic efficiency and recognition credibility; when the evaluation value is insufficient, the re-identification and conflict marking mechanism is triggered to ensure risk control and process closure in cases of identity ambiguity or data anomalies, thereby effectively improving the system's intelligent judgment capabilities and security in complex scenarios.

[0039] Specifically, the specific process of concurrent conflict identification detection for concurrent visitor identification request sessions entering the buffer queue is as follows: encapsulate the unique identification session identifier generated for each visitor into an identification request session object, add it to the concurrent buffer queue sorted by timestamp and channel number, and set the maximum concurrent capacity limit. Before adding a new identification request session, perform a uniqueness check and preliminary spatial conflict prediction, compare the time overlap and three-dimensional position proximity to mark the suspected conflict state in advance, and enter the subsequent conflict detection module for processing; calculate the spatial distance between the two people through the three-dimensional spatial position information in the multimodal visitor data. If the distance is less than the spatial distance threshold, it is determined to be the suspected same person, and then perform facial similarity judgment; obtain facial feature vectors, and calculate the normalized cosine similarity of facial feature vectors based on the face recognition model of angle interval; obtain the face image and key points collected by the face recognition camera sensor and perform real-time analysis and normalization based on the convolutional neural network to obtain living body Image quality score; using facial images and key points collected from multimodal visitor data, identify and calculate the loss ratio of key points in the facial image to obtain the degree of facial occlusion; use cosine similarity plus the product of the live image quality score and the quality enhancement weight factor minus the product of the facial occlusion degree and the occlusion degree weight factor, and multiply the result by the crowd density weight factor to obtain the facial similarity judgment value; when the facial similarity judgment value is less than the facial similarity threshold, it is marked as a low-trust identity and immediately jumps to conflict marking and separation processing, suspending subsequent modal fusion, and prompting a voice prompt that the image is unclear or occluded, please reposition the recognition, allowing the user to retry up to two times. At the same time, the live image quality score and the degree of facial occlusion are recorded. If the threshold is still not met, it will automatically transfer to manual review and prompt the visitor to go to the registration counter for processing; when the facial similarity judgment value is greater than or equal to the facial similarity threshold, the facial modality is considered trustworthy and directly enters the authorization strategy and decision control process.

[0040] Among them, the specific calculation formula for the face similarity judgment value is:

[0041] ;

[0042] Where, is the face similarity judgment value; Calculate the normalized cosine similarity of facial feature vectors for the face recognition model based on angle separation; Score for in vivo image quality; The degree of facial occlusion is obtained by calculating the loss ratio of facial key points. The higher the ratio, the heavier the occlusion. The crowd density weight factor is calculated in real time using the channel load monitoring module based on the ratio of the number of requests per unit time in the current channel to the number of people detected by the camera near the current gate, and ranges from 0.8 to 1.2. The quality enhancement weight factor is obtained by simulating a large number of recognition scenarios and recording the correlation between the live image quality score and the face recognition accuracy. In the offline training samples, the live image quality score is used as the independent variable and the recognition accuracy is the dependent variable. The range is between 0.1 and 0.2. is the occlusion degree weight factor, which is based on the degree of facial occlusion, such as the ratio of the number of missing facial key points to the total number. It is obtained by real-time analysis of the relationship curve between the offline statistical occlusion rate and the misrecognition probability, and ranges from 0.2 to 0.4.

[0043] In this implementation plan, by constructing a buffer queue sorted by timestamp and channel number and setting concurrent capacity limits, orderly management and uniqueness guarantee of multiple visitor identification requests are achieved; combining three-dimensional spatial position judgment and face similarity calculation, the system can accurately identify suspected conflicting individuals in concurrent scenarios; introducing a live image quality scoring and occlusion degree correction mechanism to effectively improve the accuracy and stability of recognition credibility assessment; when it is judged to be a low-trust identity, the system can prompt the user to retry in real time, and automatically transfer to manual review after the retry fails, realizing closed-loop control of conflict identification and exception handling in a concurrent environment, significantly enhancing the system's anti-interference ability and security.

[0044] Specifically, the specific process of conflict marking and separation of visitor identification request sessions after the above detection is as follows: when the camera sensor module detects that two or more identification request sessions are highly overlapped in time and close in spatial position, that is, when it is determined to be a suspected conflict, the relevant session status is marked as conflict pending, and the visitor is prompted by voice and screen to re-scan the code later, effectively avoiding data confusion caused by simultaneous identification of multiple visitors, improving system identification accuracy and user guidance interaction experience, and recording detailed information of the conflict event in the background, including unique session identifier, timestamp, spatial distance and facial features, to enhance the system's traceability and operation and maintenance intervention capabilities for conflicting behaviors, and suspending the session to enter the subsequent identification process; then entering a short time window for re-identification, Re-collect facial images, three-dimensional spatial location information, and electronic passcode information. If it is determined that the session is significantly different from other sessions, the conflict mark is removed and the status is updated to conflict resolved, realizing dynamic correction and autonomous resolution of conflict identification, and improving the system's adaptability and resolution capabilities to concurrent multi-user behaviors. If the conflict cannot be eliminated after multiple retries, it is marked for manual review and the visitor is prompted to go to the manual registration desk to ensure that potential risks or abnormal sessions are promptly transferred to the manual intervention path to ensure a safe closed-loop identification process. If the session is confirmed to be a spatially independent and feature-independent individual in the initial judgment, its status is directly marked as independent, skipping conflict processing and entering the authorization module to complete the normal passage process, improving system processing efficiency and ensuring smooth passage for non-conflicting individuals.

[0045] In this implementation plan, by marking conflicts and providing prompts for visitor sessions that overlap in time and are close in space, combined with background recording and re-identification mechanisms, accurate identification and dynamic resolution of concurrent conflicts can be achieved; when retrying is effective, the conflict can be automatically resolved, improving the system's resolution capability and interactive experience; if the retry fails, manual review is required to ensure a secure closed-loop identification system; sessions with independent spatial features are directly authorized to pass, effectively improving processing efficiency and channel smoothness, and achieving highly reliable identification and risk closed-loop control in a concurrent environment.

[0046] Specifically, the specific process of extracting the core information of the visitor identification request session, formulating a matching authorization strategy, and conducting risk assessment and behavioral risk control is as follows: obtaining the core information of the visitor identification request session, including identity information, appointment time, and appointment target area; allowing visitors to enter thirty minutes before and after the appointment time, not allowing visitors to visit specific floors, and allowing visitors to scan the code repeatedly at most three times a day; obtaining the number of repeated scans by the visitor by identifying the frequency of occurrence of the unique session identifier of the same visitor per unit time; obtaining the number of attempts to obtain the number of unauthorized area attempts by identifying the number of attempts in which the appointment target area of ​​the electronic pass code information scanned by the visitor in the multimodal visitor data collection phase does not match its authority; obtaining the face feature vector and the identity photo feature vector, and using the face recognition model based on the angle interval The model is normalized by the cosine similarity between the two vectors to obtain the inconsistency between the person and the certificate; the channel congestion is obtained by identifying the ratio of the number of visitor identification request sessions of the channel per unit time to the maximum number of visitor identification request sessions of the channel in the recent period; the face similarity judgment value is calculated; the inverse of the person and the certificate inconsistency is exponentially calculated and then added with a constant 1 to obtain the person and the certificate consistency adjustment item; the number of repeated code scans and the number of attempts in the unauthorized area are added and then divided by the person and the certificate consistency adjustment item to obtain the behavior anomaly score item; the face similarity judgment value is subtracted from the constant 1 and then squared to obtain the face image untrustworthiness penalty item; the sum of the behavior anomaly score item, the face image untrustworthiness penalty item and the channel congestion is multiplied by the anomaly adjustment weight factor to obtain the anomaly assessment value.

[0047] The specific calculation formula for the abnormal evaluation value is:

[0048] ;

[0049] Where, is the abnormal assessment value; The number of times the code is scanned repeatedly; Number of attempts for unauthorized areas; The inconsistency of witnesses; is the face similarity judgment value; is the channel congestion; To adjust the weight factor for abnormality, the value is 1 if the security is in an alert state and 0 if not, based on real-time data such as time sensitivity, such as whether it is in the morning and evening rush hours. The value is obtained through real-time regression analysis and dynamic adjustment through weighted calculation, and the range is between 0 and 1.

[0050] With the anomaly adjustment weight factor set to 1, and the anomaly adjustment weight factor unchanged, the anomaly assessment value is calculated over time based on the number of repeated scans, number of unauthorized area attempts, identity discrepancy, facial similarity judgment value, and channel congestion for different visitor identifications. Table 1 shows the anomaly assessment value data.

[0051] Table 1 Abnormal evaluation value data table

[0052]

[0053] like Figure 3 As shown in Table 1 and Figure 3 It can be seen that when the abnormal adjustment weight factor is 1 and remains unchanged, different risk score trends are reflected through different abnormal assessment values ​​when different visitors have different times of repeated code scanning, number of attempts in unauthorized areas, inconsistency between identity and ID, facial similarity judgment value, and channel congestion.

[0054] In this implementation plan, a multi-factor risk assessment model is constructed by extracting core data such as identity information, appointment time period and target area from visitor identification requests, combining key behavioral characteristics such as the number of repeated code scans, attempts in unauthorized areas and consistency between person and ID. Facial image credibility and channel congestion are introduced as supplementary criteria, and abnormal assessment values ​​are calculated through index adjustment and weighted combination to achieve accurate quantification of visitor behavior risks, effectively support dynamic authorization decision-making and identification security control, and enhance the system's risk control capabilities and the level of intelligent access management.

[0055] Specifically, the specific process of generating decisions and feeding back decision results is as follows: when the abnormal assessment value is less than the risk threshold, it is judged to be normal and reliable, and is directly released to enter the authorization module and write a normal log without review, thereby achieving rapid passage of low-risk visitors and efficient use of identification resources; when the abnormal assessment value is greater than or equal to the risk threshold, it is judged to be abnormal risk, and the user is prompted to adjust the position and re-scan the code, retaining the first failed data, allowing the visitor to identify again, enhancing the fault tolerance and user experience for occasional identification errors, and if it is still abnormal after retrying, the visitor identification request session will be automatically suspended, and the manual registration will be displayed, a high-risk log will be generated and a facial image will be captured, and the background security management system will be linked to push abnormal personnel reminders, build a closed-loop response mechanism for security incidents, and enhance the system's active early warning capabilities for potential risks; when the identification process is completed and the authorization decision result is generated, the result and the relevant identification status will be pushed to the local terminal interface and background in real time, realizing real-time linkage between front-end user guidance and background data synchronization, and improving the visualization and control efficiency of identification decisions.

[0056] In this implementation plan, risk grading decisions are made through the judgment of abnormal assessment values. Low-risk visitors can be quickly released and enter the authorization process, improving passage efficiency and system resource utilization; abnormal risk visitors are guided to retry and linked to the security system for processing, realizing fault tolerance and safe closed-loop management and control; at the same time, the recognition results and status information are pushed synchronously to the front-end and back-end to ensure smooth user interaction and real-time visualization of the management platform, comprehensively improving the response efficiency, security and intelligence level of the recognition system.

[0057] Specifically, the specific process of real-time feedback control, recording behavior logs and monitoring information, and real-time detection of load status and adjustment is as follows: if the authorization step is passed, the green indicator light will be on and the gate will be opened automatically; if the authorization step fails, the red indicator light will flash, the voice prompt information will be abnormal, and the error code will be returned and recorded; the visitor's passage time, channel number, result status, response delay, and failure rate information will be recorded; the channel congestion is obtained by identifying the ratio of the number of visitor identification request sessions in the channel per unit time to the maximum number of visitor identification request sessions in the channel in the recent period; the system response delay is obtained by automatically calculating the time from the visitor scanning the code to the feedback of the actual identification response; the current window identification failure rate is obtained by identifying the historical multimodal visitor data within five minutes and calculating the number of all visitor request failures and the total number of identifications; the current window identification failure rate is negated and divided by a constant The suppression factor term is obtained by adding them together, the sum of the channel congestion and the system response delay is divided by the suppression factor term, and the result is multiplied by the global adjustment weight factor to obtain the pressure adjustment value; the pressure adjustment value is compared with the pressure threshold in real time. When the pressure adjustment value is less than the pressure threshold, it is considered to be a normal load state, and the recognition process is executed according to the default strategy, all modes are enabled, the original recognition threshold is maintained, complete voice and screen prompt feedback, standard logs are recorded and the background visualization status is maintained; when the pressure adjustment value is greater than or equal to the pressure threshold, it is considered to be a high load state, and the policy compensation mechanism is automatically enabled, including raising the face and consistency judgment threshold, temporarily closing some low-priority modes, limiting the number of user retries, simplifying the voice and screen prompt frequency, adjusting the modal fusion weight, compressing the recognition process response time, and at the same time issuing a high-pressure warning to the background and strengthening log recording.

[0058] Among them, the specific calculation formula of the pressure adjustment value is:

[0059] ;

[0060] Where, is the pressure adjustment value; is the current channel congestion; The current system response delay; Identify the failure rate for the current window; To globally adjust the weight factor, a linear regression algorithm is used to fit the traffic records within one month, including channel congestion, system response delay, and average window recognition failure rate, and the range is between 0 and 1.

[0061] In this implementation plan, real-time feedback control of the recognition process is achieved through traffic status indication, log recording and load monitoring; pressure adjustment values ​​are dynamically generated by calculating channel congestion, response delay and recognition failure rate, and the current system load status is intelligently judged; full-modal recognition and complete interactive prompts are maintained under normal conditions to ensure traffic experience and data integrity; the strategy compensation mechanism is automatically started under high load conditions to dynamically adjust the recognition threshold, mode activation and prompt frequency, effectively alleviating system pressure and ensuring recognition stability, thereby improving overall recognition efficiency and operational resilience.

[0062] Specifically, the specific process of data visualization analysis is as follows: construct a risk score trend line chart to display the abnormal assessment values ​​in different time periods during the visitor identification process, realize intuitive monitoring of the identification status and trend insights into risk changes, and simultaneously visualize three key influencing factors: inconsistency between person and certificate, penalty item for unreliability of facial image, and channel congestion, so that managers can trace the source of risk composition and accurately locate the problem mode; with time as the horizontal axis and abnormal assessment value as the vertical axis, the main line highlights the risk score, and the remaining curves use different marks to distinguish the modal influencing factors, enhance the readability of the chart and the ability of multi-indicator correlation analysis; and embed the visualization into the management background, generate analysis reports regularly, support system operation status evaluation and strategy optimization decision-making, identify high-pressure channels, abnormal time periods and modal failure conditions, and improve the interpretability, transparency and tuning efficiency of the overall system operation.

[0063] In this implementation plan, by constructing a risk score trend line chart, the abnormal assessment values ​​and changes in key influencing factors during the identification process are intuitively displayed to help managers identify risk sources and problem modes; the chart uses time as the axis to highlight the main risk line and distinguish multi-dimensional factors, improving readability and analysis depth; the visualization results are embedded in the background and reports are generated regularly to achieve identification status monitoring, strategy optimization support and high-risk channel positioning.

[0064] Reference Figure 2As shown, the second aspect of the present invention provides a self-service visitor intelligent authorization system, which is applied to the above-mentioned self-service visitor intelligent authorization method, including a multimodal acquisition and identification processing module, a concurrent conflict detection and resolution module, an authorization strategy and decision control module and an interactive feedback and monitoring optimization module: wherein the multimodal acquisition and identification processing module is used to collect multimodal visitor data, assign a unique session identifier to the visitor request, perform data preprocessing, and analyze modal consistency; the concurrent conflict detection and resolution module is used to perform concurrent conflict identification detection on the concurrent visitor identification request session entering the buffer queue, and after detection, conflict marking and separation processing on the visitor identification request session; the authorization strategy and decision control module is used to extract the core information of the visitor identification request session, formulate matching authorization strategies, perform risk assessment and behavior risk control, generate decisions and feedback decision results; the interactive feedback and monitoring optimization module is used to perform real-time feedback control, record behavior logs and monitoring information, detect and adjust the load status in real time, and perform data visualization analysis.

[0065] In this implementation plan, the self-service visitor intelligent authorization system provided realizes the accurate identification of visitor identities, efficient separation of conflicting behaviors, intelligent decision-making of risk passage, and visual management of system status through the collaborative operation of modules such as multimodal collection and recognition processing, concurrent conflict detection, risk assessment and dynamic authorization control, real-time feedback and visual monitoring, thereby comprehensively improving recognition accuracy, passage efficiency, and the security and controllability of system operation.

[0066] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0067] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A self-service visitor intelligent authorization method, characterized in that: The following steps are involved: Step 1: Collect multimodal visitor data, assign unique session identifiers to visitor requests, perform data preprocessing, and analyze modality consistency; Step 2: Perform concurrent conflict identification detection on the concurrent visitor identification request sessions entering the buffer queue, and after detection, perform conflict marking and separation processing on the visitor identification request sessions; Step 3: Extract the core information of the visitor identification request session, formulate a matching authorization strategy, conduct risk assessment and behavior risk control, generate a decision and provide feedback on the decision result; Step 4: Perform real-time feedback control, record behavior logs and monitoring information, detect and adjust load status in real time, and perform data visualization analysis; The specific process of analyzing modal consistency is as follows: obtaining a facial feature vector and calculating the normalized cosine similarity of the facial feature vector based on an angular interval face recognition model to obtain the face recognition confidence; obtaining the identity information accuracy rate by calculating the number of correct identity information characters recognized in the multimodal visitor data relative to the total number of identity information characters; and obtaining the QR code validity by identifying whether the identity token embedded in the electronic pass code information matches the digital signature based on the electronic pass code read from the multimodal visitor data. Multiply the square of the face recognition confidence by the face reliability weight factor to obtain the face matching degree, multiply the square of the identity information accuracy by the identity reliability weight factor to obtain the identity information matching degree, multiply the square of the QR code validity by the code modality reliability weight factor to obtain the QR code validity matching degree, multiply the face recognition confidence, identity information accuracy and QR code validity by two, then sum and multiply by the modal interaction weight factor to obtain the modal interaction correlation term, sum the face matching degree, identity information matching degree, QR code validity matching degree and modal interaction correlation term and take the square root to obtain the modal consistency evaluation value; The specific process of performing concurrent conflict identification detection on concurrent visitor identification request sessions entering the buffer queue is as follows: obtaining a facial feature vector, calculating the normalized cosine similarity of the facial feature vector based on an angular interval face recognition model; obtaining a facial image and key points collected by a facial recognition camera sensor, performing real-time analysis and normalization based on a convolutional neural network to obtain a live image quality score; identifying and calculating the loss ratio of key points in the facial image using facial images and key points collected from multimodal visitor data to obtain a facial occlusion degree; subtracting the product of the facial occlusion degree and the occlusion degree weight factor from the product of the cosine similarity plus the live image quality score and the quality enhancement weight factor, and multiplying the result by the crowd density weight factor to obtain a facial similarity judgment value; The specific process of extracting the core information of the visitor identification request session, formulating a matching authorization strategy, and conducting risk assessment and behavior risk control is as follows: obtaining the number of repeated scans by the visitor by identifying the frequency of the unique session identifier of the same visitor within a unit time; obtaining the number of attempts of the electronic pass code information scanned by the visitor in the multimodal visitor data collection phase not matching the reservation target area with the visitor's authority; obtaining the face feature vector and the identity photo feature vector, and obtaining the person-document inconsistency degree by normalizing the cosine similarity between the two vectors using the face recognition model based on the angle interval; and obtaining the number of attempts of the visitor to the channel within a unit time. The channel congestion is calculated by the ratio of the number of visitor identification request sessions to the maximum number of visitor identification request sessions of the channel in the recent period. The facial similarity judgment value is calculated. The inverse of the person-document inconsistency is exponentially calculated and then added with a constant of one to obtain the person-document consistency adjustment item. The number of repeated code scans and the number of unauthorized area attempts are added and then divided by the person-document consistency adjustment item to obtain the behavior anomaly score item. The facial similarity judgment value is subtracted from the constant of one and then squared to obtain the facial image untrustworthiness penalty item. The sum of the behavior anomaly score item, the facial image untrustworthiness penalty item, and the channel congestion is multiplied by the anomaly adjustment weight factor to obtain the anomaly assessment value.

2. A self-service visitor intelligent authorization method according to claim 1, characterized in that: The specific process of collecting multimodal visitor data, assigning a unique session identifier to a visitor request, and performing data preprocessing is as follows: When a visitor approaches the collection terminal, the infrared sensor device detects the human body approaching in real time, triggering the terminal to enter the working state and waking up the multimodal recognition process. The sensing device collects multimodal visitor data: the face recognition camera sensor collects facial images and key points and extracts facial features. The face feature vector is extracted from the face image through the face recognition model based on the angle interval. The identity information reading module reads the identity information and visitor's identity photo. The identity photo feature vector is extracted from the visitor's identity photo through the face recognition model based on the angle interval. The infrared sensor captures the visitor's three-dimensional spatial location information. The QR code scanner reads the electronic pass code information, including the visitor's appointment time and appointment target area. A unique session identifier is assigned to the visitor request, including a timestamp and channel number. A retry and low-weight processing strategy is adopted for missing and abnormal multimodal visitor data, and the multimodal visitor data is standardized and normalized.

3. A self-service visitor intelligent authorization method according to claim 1, characterized in that: The analysis of modal consistency further includes: When the modal consistency evaluation value is greater than or equal to the modal consistency threshold, the multimodal recognition results are considered highly consistent and the authorization is credible. The information in this recognition is saved in a temporary cache, and the authorization engine is used to make a logical authorization judgment. The door opening command is sent to the gate controller, prompting that the authorization is passed. The modal consistency evaluation value, timestamp, and channel number are written to the background database. When the modal consistency evaluation value is less than the modal consistency threshold, it is considered that there is a certain degree of modal conflict and identity ambiguity, and the visitor should not be released directly. The unique session identifier should be marked as a conflict label, and the user is required to re-scan the QR code and stand firmly for face re-recognition. The original multimodal visitor data is cached and the modal consistency evaluation is re-performed. If it is still less than the threshold, the visitor is locked and prompted to go to the manual registration desk. The cloud platform is linked to notify the backend, and the low-scoring modal consistency evaluation value, abnormal modality and visitor identity photo are encrypted and recorded.

4. A self-service visitor intelligent authorization method according to claim 1, characterized in that: The performing concurrent conflict identification detection on the concurrent visitor identification request sessions entering the buffer queue further includes: The unique identification session identifier generated for each visitor is encapsulated into an identification request session object and added to a concurrent buffer queue sorted by timestamp and channel number. A maximum concurrent capacity limit is set. Before adding a new identification request session, a uniqueness check and preliminary spatial conflict prediction are performed. The suspected conflict status is marked in advance by comparing the time overlap and 3D position proximity, and then enters the subsequent conflict detection module for processing; The spatial distance between two people is calculated using the three-dimensional spatial location information in the multimodal visitor data. If the distance is less than the spatial distance threshold, they are considered to be the same person, and facial similarity is then determined. When the face similarity judgment value is less than the face similarity threshold, it is marked as a low-confidence identity and immediately jumps to conflict marking and separation processing. Subsequent modal fusion is suspended, and a voice prompt is issued to indicate that the image is unclear or blocked, asking the user to reposition the recognition. The user is allowed to retry up to two times. At the same time, the live image quality score and the degree of facial occlusion are recorded. If the threshold is still not met, the system automatically transfers to manual review and prompts the visitor to go to the registration counter for processing. When the face similarity judgment value is greater than or equal to the face similarity threshold, the face modality is considered credible and the authorization strategy and decision control process is directly entered.

5. A self-service visitor intelligent authorization method according to claim 1, characterized in that: The specific process of performing conflict marking and separation processing on the visitor identification request session after the detection is as follows: When the camera sensor module detects that two or more recognition request sessions are highly overlapping in time and close in spatial location, that is, when it is determined to be a suspected conflict, the relevant session status is marked as pending conflict, and the visitor is prompted through voice and screen to re-scan the code later. At the same time, detailed information of the conflict event is recorded in the background, including the unique session identifier, timestamp, spatial distance and facial features, and the session is suspended to enter the subsequent recognition process. Then, a short time window is entered for re-recognition, and the facial image, three-dimensional spatial location information and electronic pass code information are re-collected. If it is determined that the session is significantly different from the other sessions, the conflict mark is removed and the status is updated to conflict resolved. If the conflict cannot be resolved after multiple retries, it is marked for manual review and the visitor is prompted to go to the manual registration counter. If the session is confirmed to be spatially independent and feature-independent individuals in the initial judgment, its status is directly marked as independent, and the conflict processing is skipped. It enters the authorization module to complete the normal passage process.

6. A self-service visitor intelligent authorization method according to claim 1, characterized in that: Extracting the core information of the visitor identification request session, formulating a matching authorization strategy, and performing risk assessment and behavior risk control also includes: Obtain the core information of the visitor identification request session, including identity information, appointment time, and appointment target area; allow visitors to enter thirty minutes before and after the appointment time, not allow visitors to access specific floors, and visitors can scan the code up to three times a day.

7. A self-service visitor intelligent authorization method according to claim 1, characterized in that: The specific process of generating a decision and feeding back the decision result is as follows: When the abnormality assessment value is less than the risk threshold, it is judged as normal and credible, and is directly released to enter the authorization module and written into the normal log without the need for review; When the abnormality assessment value is greater than or equal to the risk threshold, it is determined to be an abnormal risk and the user is prompted to adjust their position and scan the code again. The first failed data is retained and the visitor is allowed to be identified again. If the abnormality is still present after the retry, the visitor identification request session is automatically suspended and the user is prompted to go to manual registration. A high-risk log is generated and a facial image is captured. The backend security management system is linked to push abnormal personnel reminders. When the recognition process is completed and the authorization decision result is generated, the result and related recognition status are pushed to the local terminal interface and background in real time.

8. A self-service visitor intelligent authorization method according to claim 1, characterized in that: The specific process of performing real-time feedback control, recording behavior logs and monitoring information, and detecting and adjusting load status in real time is as follows: If the authorization step is successful, the green indicator light will light up and the gate will open automatically; if the authorization step fails, the red indicator light will flash, a voice prompt message will be given, and an error code will be returned and recorded; Record visitor passage time, channel number, result status, response delay, and failure rate information; The channel congestion is determined by the ratio of the number of visitor identification request sessions per unit time to the maximum number of visitor identification request sessions per unit time. The system response delay is calculated by automatically calculating the time from when a visitor scans the QR code to when the actual identification response is received. The current window identification failure rate is calculated by identifying historical multimodal visitor data within five minutes and calculating the number of all visitor request failures and the total number of identifications. The suppression factor term is obtained by adding the inverse of the current window identification failure rate to a constant of one. The suppression factor term is divided by the sum of the channel congestion and the system response delay, and the result is multiplied by the global adjustment weight factor to obtain the pressure adjustment value. The pressure adjustment value is compared with the pressure threshold in real time. When the pressure adjustment value is less than the pressure threshold, it is considered to be a normal load state. The recognition process is executed according to the default strategy, enabling all modes, maintaining the original recognition threshold, and providing complete voice and screen prompt feedback. Standard logs are recorded and background visualization is maintained. When the pressure adjustment value is greater than or equal to the pressure threshold, it is considered a high-load state, and the policy compensation mechanism is automatically enabled, including raising the face and consistency judgment thresholds, temporarily closing some low-priority modes, limiting the number of user retries, simplifying the frequency of voice and screen prompts, adjusting the modal fusion weight, compressing the recognition process response time, and issuing a high-pressure warning to the background and strengthening log recording.

9. A self-service visitor intelligent authorization method according to claim 1, characterized in that: The specific process of performing data visualization analysis is as follows: A risk score trend line chart is constructed to display the abnormal assessment values ​​in different time periods during the visitor identification process, and three key influencing factors are simultaneously visualized: the inconsistency between the person and the ID, the penalty for the unreliability of the facial image, and the channel congestion, to identify the source of risk and the changing trend. With time as the horizontal axis and the abnormal assessment value as the vertical axis, the main line highlights the risk score, and the remaining curves use different marks to distinguish the modal influencing factors. This visualization is embedded in the management background, and analysis reports are generated regularly to identify high-pressure channels, abnormal time periods, and modal failures.

10. A self-service visitor intelligent authorization system, characterized in that: include: Multimodal acquisition and recognition processing module, concurrent conflict detection and resolution module, authorization strategy and decision control module, and interactive feedback and monitoring optimization module: The multimodal acquisition and identification processing module is used to collect multimodal visitor data, assign unique session identifiers to visitor requests, perform data preprocessing, and analyze modality consistency; The concurrent conflict detection and resolution module is used to perform concurrent conflict identification detection on concurrent visitor identification request sessions entering the buffer queue, and after detection, perform conflict marking and separation processing on the visitor identification request sessions; The authorization policy and decision control module is used to extract the core information of the visitor identification request session, formulate matching authorization policies, conduct risk assessment and behavior risk control, generate decisions and feedback decision results; The interactive feedback and monitoring optimization module is used to perform real-time feedback control, record behavior logs and monitoring information, detect and adjust load status in real time, and perform data visualization analysis; The specific process of analyzing modal consistency is as follows: obtaining a facial feature vector and calculating the normalized cosine similarity of the facial feature vector based on an angular interval face recognition model to obtain the face recognition confidence; obtaining the identity information accuracy rate by calculating the number of correct identity information characters recognized in the multimodal visitor data relative to the total number of identity information characters; and obtaining the QR code validity by identifying whether the identity token embedded in the electronic pass code information matches the digital signature based on the electronic pass code read from the multimodal visitor data. Multiply the square of the face recognition confidence by the face reliability weight factor to obtain the face matching degree, multiply the square of the identity information accuracy by the identity reliability weight factor to obtain the identity information matching degree, multiply the square of the QR code validity by the code modality reliability weight factor to obtain the QR code validity matching degree, multiply the face recognition confidence, identity information accuracy and QR code validity by two, then sum and multiply by the modal interaction weight factor to obtain the modal interaction correlation term, sum the face matching degree, identity information matching degree, QR code validity matching degree and modal interaction correlation term and take the square root to obtain the modal consistency evaluation value; The specific process of performing concurrent conflict identification detection on concurrent visitor identification request sessions entering the buffer queue is as follows: obtaining a facial feature vector, calculating the normalized cosine similarity of the facial feature vector based on an angular interval face recognition model; obtaining a facial image and key points collected by a facial recognition camera sensor, performing real-time analysis and normalization based on a convolutional neural network to obtain a live image quality score; identifying and calculating the loss ratio of key points in the facial image using facial images and key points collected from multimodal visitor data to obtain a facial occlusion degree; subtracting the product of the facial occlusion degree and the occlusion degree weight factor from the product of the cosine similarity plus the live image quality score and the quality enhancement weight factor, and multiplying the result by the crowd density weight factor to obtain a facial similarity judgment value; The specific process of extracting the core information of the visitor identification request session, formulating a matching authorization strategy, and conducting risk assessment and behavior risk control is as follows: obtaining the number of repeated scans by the visitor by identifying the frequency of the unique session identifier of the same visitor within a unit time; obtaining the number of attempts of the electronic pass code information scanned by the visitor in the multimodal visitor data collection phase not matching the reservation target area with the visitor's authority; obtaining the face feature vector and the identity photo feature vector, and obtaining the person-document inconsistency degree by normalizing the cosine similarity between the two vectors using the face recognition model based on the angle interval; and obtaining the number of attempts of the visitor to the channel within a unit time. The channel congestion is calculated by the ratio of the number of visitor identification request sessions to the maximum number of visitor identification request sessions of the channel in the recent period. The facial similarity judgment value is calculated. The inverse of the person-document inconsistency is exponentially calculated and then added with a constant of one to obtain the person-document consistency adjustment item. The number of repeated code scans and the number of unauthorized area attempts are added and then divided by the person-document consistency adjustment item to obtain the behavior anomaly score item. The facial similarity judgment value is subtracted from the constant of one and then squared to obtain the facial image untrustworthiness penalty item. The sum of the behavior anomaly score item, the facial image untrustworthiness penalty item, and the channel congestion is multiplied by the anomaly adjustment weight factor to obtain the anomaly assessment value.

Images