A method of ab door identity verification

Abstract

The application discloses a kind of ab door identity verification method, applied to the AB door passage with interlocking mechanism, AB door passage includes A door and B door, including the following steps: step one, camera is respectively arranged on the top of A door and B door, the shooting range of two cameras respectively covers the complete passage area of corresponding door body;Step two, the construction of face and head database and the number statistics of entry: through B door side camera to execute entry processing;Step three, exit identity verification and abnormal identification: through A door side camera to execute exit processing;Step four, number check and abnormal positioning.The method of the application constructs the whole-process identity verification closed loop of entry and exit, realizes personnel non-sensing passage and accurate number control, can effectively identify and locate abnormal personnel, reduce labor cost, improve AB door passage efficiency and security control precision.

Description

Technical Field

[0001] This invention relates to the field of security access control technology, and more specifically to a method for AB door identity verification. Background Technology

[0002] AB doors, as core security facilities with dual protection, are widely used in various high-security scenarios. Their core function is to prevent unauthorized personnel from entering and ensure traceability of personnel access through the interlocking of the two doors and personnel identification. In scenarios where a large number of people pass through AB doors, the efficiency and accuracy of personnel identification directly determine the security effectiveness of the AB door. Existing technologies for personnel identification through AB doors have several shortcomings, as detailed below: Manual verification method: Relying on management personnel to verify the identity of each person entering and exiting the A and B gates (such as checking documents and manually comparing faces) not only requires a large number of management personnel, resulting in high labor costs, but also requires professional training for management personnel in advance, making the preparation work cumbersome; at the same time, manual verification is prone to fatigue errors, subjective judgment biases, and problems such as omissions and misjudgments of identities when cross-checking, which may lead to major safety accidents in serious cases and cannot meet the needs of rapid passage of a large number of people.

[0003] Traditional facial recognition gate verification methods require pre-registering facial information for all passersby, resulting in a huge workload and long preparation period. Gate verification requires personnel to actively cooperate and align themselves with the camera, leading to low efficiency and unsuitability for scenarios with large numbers of people passing through. Furthermore, this method can only verify the identity of a single person and cannot automatically check the number of people entering and exiting, nor can it locate abnormal personnel (such as unauthorized individuals) in real time, resulting in security vulnerabilities.

[0004] Existing image recognition verification methods: Some solutions use image recognition technology to assist in verification, but most only achieve recognition of a single link (such as only recognizing faces or only counting people), and do not form a complete closed loop of "entry database construction - exit verification - people verification - anomaly location"; and lack accurate head tracking and identity association mechanisms, which are prone to identity matching errors and people counting deviations, and cannot meet the high-precision and full-process control requirements of A / B gate.

[0005] Inadequate control over abnormal personnel: Existing solutions can only identify unauthorized personnel and cannot track and output the location of abnormal personnel in real time, making it difficult for managers to respond quickly; at the same time, they lack scientific abnormal judgment logic, which can easily lead to misjudgment or omission of abnormalities, and cannot effectively prevent security risks such as personnel infiltration or omission.

[0006] In summary, existing A / B gate identity verification solutions cannot simultaneously guarantee efficiency, accuracy, and security in scenarios involving large numbers of people passing through. They suffer from problems such as large upfront preparation workload, high labor costs, susceptibility to errors, and insufficient anomaly control. There is an urgent need for an A / B gate identity verification method and system that can achieve seamless verification, accurate cancellation, headcount verification, and anomaly location. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the present invention aims to provide a method for A / B gate identity verification. Through a closed-loop image recognition control process, it achieves accurate identity verification, headcount checking, and anomaly location for seamless personnel passage. This method is suitable for high-traffic personnel passage scenarios, reduces labor costs, and improves the accuracy of security control.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for AB door identity verification, applied to an AB door passage with an interlocking mechanism, the AB door passage including door A and door B, comprising the following steps: Step 1: Deploy cameras above doors A and B respectively, with each camera covering the entire passage area of ​​its corresponding door. Step 2, Construction of the database of faces and heads of people entering the venue and counting of people: The entry process is performed through the side camera of Gate B; Step 3, Exit Identity Verification and Anomaly Detection: Exit processing is performed via the camera on the side of Gate A; Step 4, headcount verification and anomaly location: Compare the number of people leaving with the number of people entering. If they do not match, an anomaly alert will be triggered. At the same time, for targets identified as abnormal or unknown persons, their real-time location on the monitoring screen will be output.

[0009] As a further improvement to the present invention, the specific steps for constructing the entry base warehouse in step two are as follows: Step 21: The camera at Gate B is activated, capturing surveillance footage at a frame rate of 30 frames per second to ensure that the captured head and facial features are clear and comprehensive. Step 22: First, start the head tracking algorithm to track each head in the image in real time and output a unique tracking ID. At the same time, crop the head according to the head area corresponding to each tracking ID and enter the cropped head features into the head database A. Steps two and three involve further processing the head features corresponding to each tracking ID in the head database A, extracting face features, and inputting them into the face database B. The features in both the head database A and the face database B are named and categorized according to the tracking ID to ensure a one-to-one correspondence. Step 24: The backend system outputs tracking IDs for those who have not entered 5 facial features in real time, triggering manual reminders until the person corresponding to the tracking ID completes the entry of 5 facial features, ensuring the integrity and accuracy of the facial database B, and providing reliable support for subsequent exit verification.

[0010] As a further improvement to the present invention, the specific steps for counting the number of people in step two are as follows: Step 25: The backend system performs head recognition on each frame captured by the B-door camera, counts the number of heads in each frame, and sums them up. Step 26: Select all frames within a recent period and summarize the frequency of occurrences based on the same number of heads; Step 27: Output the number of heads that appear most frequently as the final number of people entering the venue, to ensure the accuracy of the number of people entering the venue and avoid statistical deviations caused by people momentarily obscuring each other; Step 28: Establish a one-to-one mapping relationship between "tracking ID = face ID = personnel ID" and store the mapping relationship in the backend system to achieve accurate association between personnel identity and tracking ID and face ID, providing a basis for identity verification upon exit.

[0011] As a further improvement to the present invention, the specific steps of exit identity verification and anomaly identification in step three are as follows: Step 31: The A-door camera is activated, capturing the monitoring screen at a frame rate of 30 frames per second. The head tracking algorithm is started to track each head in the screen in real time and output a unique exit tracking ID. Then, at a frequency of taking 1 frame every 30 frames, the head area corresponding to each exit tracking ID is cropped to obtain head X, which is used for subsequent face comparison. Step 32: Input head X into the face recognition algorithm, call the face features in the face database B for comparison, and the backend system outputs one of three comparison results: cannot enter the database, meaning that head X cannot extract valid face features; match face ID, which is the tracking ID at the time of entry, to achieve identity matching; or newID, meaning that no face features were matched in the face database B. Record the comparison result under the corresponding exit tracking ID to form an association record between the exit tracking ID and the comparison result, which is used for subsequent identification of abnormal personnel. Step 33: Based on the comparison results under the exit tracking ID, determine the personnel type according to the following logic to ensure accurate identification of abnormal personnel: If the number of times a face ID is matched is greater than the number of times a new ID is matched under the same exit tracking ID, it is determined to be a normal person, that is, an authorized person who has been entered into the database when entering the venue; If the number of times a face ID is matched is less than the number of times a new ID is matched under the same exit tracking ID, it is determined to be an abnormal person, that is, a person whose identity may be confused or impersonated; If all comparison results under the same exit tracking ID are "unable to enter the base database" or "matches the face base database B less than 3 times", it is determined to be an unknown person, that is, an unauthorized person who has not been entered into the base database.

[0012] As a further improvement of the present invention, step three also includes an exit count step, as follows: First, the background system performs human figure recognition on each frame captured by the A-gate camera, counts the number of human figures in each frame and sums them up; then, it selects all frames within a recent period that have the same duration as the entry count, summarizes the occurrence frequency of the same number of people, and outputs the number of people with the highest occurrence frequency as the final exit count, ensuring the accuracy of the exit count.

[0013] As a further improvement of the present invention, the specific method for outputting the real-time position of the target in the monitoring screen in step four is as follows: for targets determined to be abnormal persons or unknown persons, the position of the target in the monitoring screen is output in real time through a head tracking algorithm.

[0014] The beneficial effects of the present invention Seamless and efficient, reducing upfront preparation costs: No need for personnel to actively cooperate with verification, no need for large-scale pre-entry of personnel facial information, automatic completion of head and face database construction upon entry, significantly reducing the amount of upfront preparation work; single camera + algorithm combination solution, no need to purchase additional complex hardware, reducing equipment and labor costs, adaptable to scenarios with large numbers of people passing through quickly.

[0015] Accurate identification reduces security vulnerabilities: A 4K camera ensures clear feature capture, and three core algorithms enable uninterrupted head tracking and high-precision face comparison. Through the mapping relationship of "tracking ID = face ID = personnel ID", identity verification is achieved one by one upon exit, avoiding misjudgments and omissions caused by manual verification and traditional gate verification, and reducing the risk of major security incidents.

[0016] Full-process closed-loop management: forming a complete closed loop of "entry database construction - entry personnel statistics - exit tracking and comparison - identity verification - anomaly judgment - personnel verification - anomaly location", which can not only realize identity verification, but also verify the number of people entering and leaving in real time and locate abnormal personnel, solving the pain point of incomplete management of existing solutions.

[0017] Strong scene adaptability: It can be widely used in various A / B gate scenarios such as prisons, factories, and parks. The frame extraction frequency and statistical duration can be flexibly adjusted according to the number of people passing through and the size of the venue. There is no need to make large-scale modifications to the existing A / B gates. It is convenient to deploy and easy to promote.

[0018] Convenient anomaly handling: Real-time output of the monitoring location of abnormal and unknown personnel, allowing managers to quickly locate and handle the situation, improving the response speed and effectiveness of AB gate security control; Scientific anomaly judgment logic reduces false and missed anomalies, further enhancing the security level. Detailed Implementation

[0019] The present invention will be further described in detail below with reference to the given embodiments.

[0020] The AB door identity verification method of this embodiment is applied to an AB door passage with an interlocking mechanism. The AB door passage includes door A and door B, and includes the following steps: Step 1: Deploy cameras above doors A and B respectively, with each camera covering the entire passage area of ​​its corresponding door. Step 2, Construction of the database of faces and heads of people entering the venue and counting of people: The entry process is performed through the side camera of Gate B; Step 3, Exit Identity Verification and Anomaly Detection: Exit processing is performed via the camera on the side of Gate A; Step 4, headcount verification and anomaly location: Compare the number of people leaving with the number of people entering. If they do not match, an anomaly alert will be triggered. At the same time, for targets identified as abnormal or unknown persons, their real-time location on the monitoring screen will be output.

[0021] When personnel pass through, they enter the passage through Gate B. The camera at Gate B simultaneously collects personnel features, builds a database, and counts the number of people entering. After passing through the passage, personnel exit through Gate A. The camera at Gate A verifies identity, identifies anomalies, and counts the number of people leaving. The backend system compares the number of people leaving with the number entering. If the numbers do not match, an anomaly alert is immediately triggered, and the real-time location of identified anomalies and unknown individuals is output to management personnel. This solution eliminates the need for manual identity verification and requires no voluntary stopping for gate data collection, achieving seamless and rapid passage. It solves the problems of low efficiency and large errors in traditional manual verification, as well as slow gate verification and heavy initial data entry workload. The closed-loop management of the entire process also fills the security loopholes in traditional solutions, such as the inability to automatically verify the number of people and the inability to locate anomalies.

[0022] Furthermore, the specific steps for constructing the base warehouse in step two are as follows: Step 21: The camera at Gate B is activated, capturing surveillance footage at a frame rate of 30 frames per second to ensure that the captured head and facial features are clear and comprehensive. Step 22: First, start the head tracking algorithm to track each head in the image in real time and output a unique tracking ID. At the same time, crop the head according to the head area corresponding to each tracking ID and enter the cropped head features into the head database A. Steps two and three involve further processing the head features corresponding to each tracking ID in the head database A, extracting face features, and inputting them into the face database B. The features in both the head database A and the face database B are named and categorized according to the tracking ID to ensure a one-to-one correspondence. Step 24: The backend system outputs tracking IDs for those who have not entered 5 facial features in real time, triggering manual reminders until the person corresponding to the tracking ID completes the entry of 5 facial features, ensuring the integrity and accuracy of the facial database B, and providing reliable support for subsequent exit verification.

[0023] This step fully collects facial features of people passing through by high frame rate frame extraction. First, each person passing through is assigned a unique identifier by head tracking. Then, facial features are extracted based on the identifier to avoid feature misalignment in multi-person scenarios. Through the input standard of 5 facial features and the manual reminder mechanism, the integrity of the feature database is ensured, the failure rate of subsequent exit comparison is reduced, and a reliable data foundation is provided for accurate identity verification.

[0024] Furthermore, the specific steps for the headcount statistics in step two are as follows: Step 25: The backend system performs head recognition on each frame captured by the B-door camera, counts the number of heads in each frame, and sums them up. Step 26: Select all frames within a recent period and summarize the frequency of occurrences based on the same number of heads; Step 27: Output the number of heads that appear most frequently as the final number of people entering the venue, to ensure the accuracy of the number of people entering the venue and avoid statistical deviations caused by people momentarily obscuring each other; Step 28: Establish a one-to-one mapping relationship between "tracking ID = face ID = personnel ID" and store the mapping relationship in the backend system to achieve accurate association between personnel identity and tracking ID and face ID, providing a basis for identity verification upon exit.

[0025] This step employs a multi-frame statistical mode counting logic, which can offset counting errors caused by people walking in different directions and momentary occlusion, thus improving the accuracy of the number of people entering the venue. At the same time, it establishes a mapping relationship between tracking ID, face ID and personnel ID, deeply binding the image tracking target with the personnel identity, avoiding identity matching mismatch when exiting the venue, and helping to improve the accuracy of identity control throughout the entire process.

[0026] Furthermore, the specific steps for exit identity verification and anomaly identification in step three are as follows: Step 31: The A-door camera is activated, capturing the monitoring screen at a frame rate of 30 frames per second. The head tracking algorithm is started to track each head in the screen in real time and output a unique exit tracking ID. Then, at a frequency of taking 1 frame every 30 frames, the head area corresponding to each exit tracking ID is cropped to obtain head X, which is used for subsequent face comparison. Step 32: Input head X into the face recognition algorithm, call the face features in the face database B for comparison, and the backend system outputs one of three comparison results: cannot enter the database, meaning that head X cannot extract valid face features; match face ID, which is the tracking ID at the time of entry, to achieve identity matching; or newID, meaning that no face features were matched in the face database B. Record the comparison result under the corresponding exit tracking ID to form an association record between the exit tracking ID and the comparison result, which is used for subsequent identification of abnormal personnel. Step 33: Based on the comparison results under the exit tracking ID, determine the personnel type according to the following logic to ensure accurate identification of abnormal personnel: If the number of times a face ID is matched is greater than the number of times a new ID is matched under the same exit tracking ID, it is determined to be a normal person, that is, an authorized person who has been entered into the database when entering the venue; If the number of times a face ID is matched is less than the number of times a new ID is matched under the same exit tracking ID, it is determined to be an abnormal person, that is, a person whose identity may be confused or impersonated; If all comparison results under the same exit tracking ID are "unable to enter the base database" or "matches the face base database B less than 3 times", it is determined to be an unknown person, that is, an unauthorized person who has not been entered into the base database.

[0027] This step also assigns independent IDs based on head tracking during the exit process, and uses a reduced-frequency frame extraction method for face comparison to reduce computing power consumption while ensuring recognition accuracy. By statistically analyzing the frequency of multiple rounds of comparison results, personnel type classification is determined, which effectively reduces the probability of false positives and false negatives compared to single face comparisons. It can accurately distinguish between normal personnel, suspicious personnel, and unauthorized unknown personnel, thus improving the hierarchical mechanism for anomaly control and helping to enhance the precision of security management.

[0028] Furthermore, step three also includes a step for counting the number of people leaving the venue, as follows: First, the backend system performs human figure recognition on each frame captured by the camera at Gate A, counts the number of human figures in each frame and sums them up; then, it selects all frames within a recent period that have the same duration as the entry count, sums the number of times the same number of people appears, and outputs the number of people appearing most frequently as the final number of people leaving the venue, ensuring the accuracy of the exit count.

[0029] This step uses the same duration and mode statistics rules as the entry count to ensure a consistent statistical benchmark for entry and exit, avoiding distortion in the headcount comparison due to differences in statistical rules. At the same time, the exit step uses human figure recognition to match the posture characteristics of people leaving, further improving the counting accuracy and ensuring the reliability of the headcount verification process.

[0030] Furthermore, the specific method for outputting the real-time position of the target in the monitoring screen in step four is as follows: for targets identified as abnormal or unknown persons, the position of the target in the monitoring screen is output in real time through a head tracking algorithm.

[0031] This step relies on the real-time positioning capability of head tracking to directly output the location of abnormal targets in the monitoring screen to the management personnel. Without the need for manual investigation and locking, it can significantly shorten the emergency response time, improve the efficiency of abnormal handling, and extend abnormal control from "identification" to "location and handling", further strengthening the security control capabilities of AB gates.

[0032] Based on the above solution, the following example is provided in this embodiment: A prison has an A / B gate passageway, with gate A being the outer gate and gate B the inner gate. A buffer zone is formed between the two gates (following an A / B gate interlocking mechanism to ensure that the other gate is locked when one gate is open). This passageway needs to meet the daily control requirements for a large number of inmates entering and exiting. A 4K high-definition camera is installed above each gate, at a height of 2.8 meters, with a downward viewing angle of 12° to ensure that the camera's field of view completely covers the passage area of ​​gates A and B. The detection range is limited based on face size (ensuring that facial features can be clearly captured when the person is 1.5-3 meters away from the camera). A data processing terminal is provided in the background to run head recognition, head tracking, and facial recognition algorithms, enabling database storage, headcount statistics, comparison and verification, anomaly detection, and result output. Camera deployment and debugging: Install two 4K cameras above Gate A and Gate B respectively, adjust the height to 2.8 meters and the downward angle to 12°, and adjust the camera angle to ensure that there are no blind spots or obstructions, and that the heads and faces of people passing through can be clearly captured; debug the algorithm parameters in the back-end terminal to ensure that the frame rate is stable at 30 frames / second and the algorithm processing speed meets the real-time requirements.

[0033] Entry phase (executed by camera at Gate B): When inmates enter through Gate B, the Gate B camera is activated, capturing images at 30 frames per second. The head tracking algorithm is also activated to track the head of each inmate in real time, assigning each person a unique tracking ID (such as track001, track002).

[0034] The head region corresponding to each tracking ID is cropped, and the cropped head features are entered into the head database A. At the same time, the face features corresponding to each head are extracted and entered into the face database B. The features in both the head database A and the face database B are named after the tracking ID (e.g., the head features and face features corresponding to track001 are both named track001).

[0035] The back-end terminal monitors the face database B in real time. If a tracking ID (such as track005) has only 3 facial features entered, which is less than 5, a manual reminder is triggered. The administrator will prompt the inmate to adjust his posture until 5 facial features are entered.

[0036] Entry personnel statistics: The back-end terminal performs head recognition and sums the images of each frame of the B-gate camera within the past 20 seconds, summing the number of times the same number of heads appears. If "15 people" appears the most times (appearing in 380 frames), then the number of people entering is output as 15. At the same time, a mapping relationship of "tracking id = face id = inmate identity id" is established (e.g., track001 corresponds to face id001 and inmate id001), and stored in the back-end system.

[0037] Exit phase (executed by camera at door A): When an inmate exits through Gate A, the Gate A camera is activated, capturing images at 30 frames per second. The head tracking algorithm is also activated to track the head of each person exiting in real time and output a unique exit tracking ID (such as outtrack001 or outtrack002).

[0038] At a frequency of 1 frame every 30 frames, the head corresponding to each exit tracking ID is cropped to obtain head X (e.g., outtrack001 corresponds to head X001).

[0039] Face matching: Input a person's head X into the face recognition algorithm, call the face database B. If person X001 matches face ID001 (i.e., entry tracking ID001), record the matching result under outtrack001; if person X008 does not match any face database feature, record it as newID under outtrack008; if person X010 cannot extract valid face features, record it as "cannot enter database" under outtrack010.

[0040] Abnormal personnel identification: Under outtrack001, the number of times the face ID was matched was 8 and the number of times the new ID was matched was 2, which is determined to be a normal person; under outtrack008, the number of times the face ID was matched was 3 and the number of times the new ID was matched was 5, which is determined to be an abnormal person; under outtrack010, all comparison results were "unable to enter the base database", which is determined to be an unknown person.

[0041] Number of people present: The back-end terminal performs human figure recognition on each frame of the A-door camera within the past 20 seconds and sums them up, summing the number of times the same number of people appears. If "14 people" appears the most times (appearing in 375 frames), then the number of people present is output as 14.

[0042] Headcount verification and anomaly location: The back-end terminal compares the number of people leaving (14) with the number of people entering (15). If a discrepancy is found, an anomaly alert is triggered. At the same time, the location of the abnormal person (outtrack008) and the unknown person (outtrack010) in the monitoring screen of Gate A (e.g., "the left 1 / 3 area of ​​the screen, 1.2 meters away from Gate A") is output in real time, and the management personnel can quickly go to locate and handle the situation.

[0043] Results recording and query: The back-end terminal automatically stores the entry and exit data, number of people entering and exiting, comparison results, information on abnormal personnel and their locations, forming a traceable A and B gate identity verification ledger. Managers can query and compile statistics at any time to achieve refined security control.

[0044] In the above scheme, head recognition, head tracking, and face recognition all adopt lightweight convolutional neural network models. Head recognition can effectively filter out interference such as clothing and hats, with an accuracy rate of no less than 98%. The head tracking algorithm can achieve uninterrupted tracking even when people are moving quickly or experiencing momentary occlusion, with a tracking accuracy rate of no less than 97%. The face recognition algorithm can adapt to face comparison with different postures and slight occlusion, with a comparison accuracy rate of no less than 96%. The processing speed meets the real-time requirement of 30 frames per second (processing time per frame ≤ 30ms). The 4K camera resolution ensures clear capture of face and head features, and the 30 frames per second frame sampling frequency avoids feature omission. The installation height of 2.8 meters and the downward angle of 12° not only meet the optimal installation standards but also ensure that the facial features of passersby are within the best capture range, reducing recognition errors. Entry is counted using head recognition, and exit is counted using human shape recognition. This, combined with the "optimal result from multiple frames over a recent period," effectively avoids errors in headcount due to personnel obstruction or instantaneous entry and exit, ensuring accurate headcount statistics. The counting time is set at 20 seconds, which can be flexibly adjusted according to the flow of people (extended to 30 seconds when people are densely packed, and shortened to 10 seconds when people are less numerous). The number of faces to be entered is set to 5, ensuring the integrity of the face database. The matching comparison logic and matching threshold (3 times) effectively distinguish between normal, abnormal, and unknown individuals, reducing false positives and false negatives. This threshold can be flexibly adjusted according to the actual scenario to ensure the one-to-one mapping of "tracking ID = face ID = person ID," ensuring the accuracy of identity verification upon exit, achieving "one person enters, one person verifies," avoiding identity confusion and impersonation, and providing support for tracing abnormal personnel.

[0045] This embodiment's solution is not limited to prison AB gates; it can be widely applied to various AB gate scenarios requiring strict personnel access control, such as factories, train stations, industrial parks, office buildings, and airports. Simply adjust the camera's installation height, frame rate, people counting time, and anomaly detection threshold according to the scenario's needs. The camera specifications can be adjusted based on actual recognition accuracy requirements, but should not be lower than 4K. The installation height can be adjusted within the range of 2.5-3.5 meters, with the downward angle controlled within 15° to ensure the detection range and recognition accuracy are adapted to the scene. The frame rate, number of faces entered, anomaly detection threshold, and people counting time can all be flexibly adjusted. For example, in scenarios with high personnel traffic, the frame rate can be increased to 60 frames / second, and the number of faces entered can be adjusted to 3-8, adapting to different scenario requirements. This system can be used in conjunction with existing AB gate interlocking systems and security alarm systems without requiring large-scale modifications to existing AB gate facilities; only cameras and backend processing terminals need to be deployed, resulting in low deployment costs and ease of promotion. This method can be enhanced with anomaly alarm functionality as needed. For example, when unknown individuals are detected or the number of people entering and exiting is inconsistent, the backend system can automatically issue an audible and visual alarm and push the anomaly information to the administrator's mobile phone, further improving the timeliness of security control. The head database A and face database B can be regularly cleaned and updated according to actual needs to avoid database redundancy and ensure comparison efficiency; they also support database expansion to meet the needs of long-term passage for a large number of people.

[0046] In summary, the A / B gate identity verification method of this invention constructs a closed-loop control system covering the entire process, including entry database construction and personnel statistics, exit identity verification and personnel statistics, entry and exit personnel verification, and real-time location of abnormal personnel, by deploying cameras covering the passage area at both A and B gates. Employing technologies such as head tracking and facial feature binding, multi-frame mode counting, and multi-round comparison and hierarchical judgment, it achieves seamless and accurate verification in high-traffic scenarios without requiring active personnel cooperation or pre-entry of facial information, effectively reducing labor costs and pre-preparation workload. Simultaneously, it can accurately identify abnormal and unknown personnel and output their real-time location, improving emergency response efficiency and comprehensively preventing security risks such as unauthorized entry, identity theft, and personnel omissions, significantly enhancing the security control level of A / B gate access.

[0047] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for AB door identity verification, applied to an AB door passageway with an interlocking mechanism, wherein the AB door passageway includes door A and door B, characterized in that: Includes the following steps: Step 1: Deploy cameras above doors A and B respectively, with each camera covering the entire passage area of ​​its corresponding door. Step 2, Construction of the database of faces and heads of people entering the venue and counting of people: The entry process is performed through the side camera of Gate B; Step 3, Exit Identity Verification and Anomaly Detection: Exit processing is performed via the camera on the side of Gate A; Step 4, headcount verification and anomaly location: Compare the number of people leaving with the number of people entering. If they do not match, an anomaly alert will be triggered. At the same time, for targets identified as abnormal or unknown persons, their real-time location on the monitoring screen will be output.

2. The A / B gate identity verification method according to claim 1, characterized in that: The specific steps for constructing the base warehouse in step two are as follows: Step 21: The camera at Gate B is activated, capturing surveillance footage at a frame rate of 30 frames per second to ensure that the captured head and facial features are clear and comprehensive. Step 22: First, start the head tracking algorithm to track each head in the image in real time and output a unique tracking ID. At the same time, crop the head according to the head area corresponding to each tracking ID and enter the cropped head features into the head database A. Steps two and three involve further processing the head features corresponding to each tracking ID in the head database A, extracting face features, and inputting them into the face database B. The features in both the head database A and the face database B are named and categorized according to the tracking ID to ensure a one-to-one correspondence. Step 24: The backend system outputs tracking IDs for those who have not entered 5 facial features in real time, triggering manual reminders until the person corresponding to the tracking ID completes the entry of 5 facial features, ensuring the integrity and accuracy of the facial database B, and providing reliable support for subsequent exit verification.

3. The method for A / B gate identity verification according to claim 2, characterized in that: The specific steps for the headcount statistics in step two are as follows: Step 25: The backend system performs head recognition on each frame captured by the B-door camera, counts the number of heads in each frame, and sums them up. Step 26: Select all frames within a recent period and summarize the frequency of occurrences based on the same number of heads; Step 27: Output the number of heads that appear most frequently as the final number of people entering the venue, to ensure the accuracy of the number of people entering the venue and avoid statistical deviations caused by people momentarily obscuring each other; Step 28: Establish a one-to-one mapping relationship between "tracking ID = face ID = personnel ID" and store the mapping relationship in the backend system to realize the accurate association between personnel identity and tracking ID and face ID, providing a basis for identity verification upon exit.

4. The A / B door identity verification method according to claim 3, characterized in that: The specific steps for exit identity verification and anomaly identification in step three are as follows: Step 31: The A-door camera is activated, capturing the monitoring screen at a frame rate of 30 frames per second. The head tracking algorithm is started to track each head in the screen in real time and output a unique exit tracking ID. Then, at a frequency of taking 1 frame every 30 frames, the head area corresponding to each exit tracking ID is cropped to obtain head X, which is used for subsequent face comparison. Step 32: Input head X into the face recognition algorithm, call the face features in the face database B for comparison, and the backend system outputs one of three comparison results: cannot enter the database, meaning that head X cannot extract valid face features; match face ID, which is the tracking ID at the time of entry, to achieve identity matching; or newID, meaning that no face features were matched in the face database B. Record the comparison result under the corresponding exit tracking ID to form an association record between the exit tracking ID and the comparison result, which is used for subsequent identification of abnormal personnel. Step 33: Based on the comparison results under the exit tracking ID, determine the personnel type according to the following logic to ensure accurate identification of abnormal personnel: If the number of times a face ID is matched is greater than the number of times a new ID is matched under the same exit tracking ID, it is determined to be a normal person, that is, an authorized person who has been entered into the database when entering the venue; If the number of times a face ID is matched is less than the number of times a new ID is matched under the same exit tracking ID, it is determined to be an abnormal person, that is, a person whose identity may be confused or impersonated; If all comparison results under the same exit tracking ID are "unable to enter the base database" or "matches the face base database B less than 3 times", it is determined to be an unknown person, that is, an unauthorized person who has not been entered into the base database.

5. The A / B door identity verification method according to claim 4, characterized in that: Step three also includes a step for counting the number of people leaving the venue, as follows: First, the backend system performs human figure recognition on each frame captured by the A-door camera, counts the number of human figures in each frame and sums them up; then, it selects all frames within a recent period that have the same duration as the entry count, sums the number of occurrences of the same number of people, and outputs the number of people with the most occurrences as the final number of people leaving the venue, ensuring the accuracy of the exit count.

6. The A / B door identity verification method according to claim 5, characterized in that: The specific method for outputting the real-time position of the target in step four is as follows: for targets identified as abnormal or unknown persons, the position of the target in the monitoring screen is output in real time through a head tracking algorithm.