Intelligent identity identification and authentication system based on examinee identity data

Abstract

The invention belongs to the technical field of education examination management, and particularly discloses and provides an intelligent identity recognition and authentication system based on examinee identity data, which comprises the steps of: monitoring and collecting examinee waiting images in a whole process through an entrance authentication module, and restoring image quality by combining with real-time face comparison of a pre-stored identity database so as to improve authentication accuracy; the identity classification processing module performs cross comparison on face data of examinees failed in secondary verification, identifies examinees not in the examination site, generates navigation guidance, triggers early warning or alarm for examinees lacking examination and non-examinees, and performs cooperative management on abnormal identities in the whole network; behavior characteristics and writing posture data of examinees are collected through a departure monitoring module, identity credibility weight is reduced in combination with a departure duration index, a dual-channel authentication mechanism is started during returning, identity authenticity is verified in a multi-dimensional manner, a trunk micro-motion identity spectrum and iris living body authentication technology is introduced, a high-security identity authentication scheme is provided, and the safety of the examinees is improved. And accurate management of the whole process is realized.

Description

Technical Field

[0001] This invention belongs to the field of educational examination management technology and relates to an intelligent identity recognition and authentication system based on candidate identity data. Background Technology

[0002] In the field of educational examination management, the accuracy and security of candidate identity authentication are crucial. With the expansion of examination scale and the increasing demand for intelligent systems, traditional identity authentication methods face numerous challenges. Existing systems have significant limitations in handling diverse examination identity authentication needs, thus necessitating a more efficient and accurate intelligent identity recognition and authentication system to address these issues.

[0003] Traditional candidate identification and authentication mainly relies on manual verification of admission tickets and ID cards, or the use of a single facial recognition technology for entry verification. Manual verification involves invigilators visually comparing the candidate's appearance with the photo on their ID card; single facial recognition involves a simple match between the facial image captured on-site and pre-stored candidate facial data to complete the identity verification.

[0004] The aforementioned manual verification methods have significant shortcomings, including: a lack of technology for dynamic monitoring of candidates' behavior throughout the entire process, particularly the absence of a mechanism for re-authenticating the identity of candidates returning from leaving the examination room, making it difficult to identify the risk of proxy testing based on behavioral differences; the inability to achieve automated classification and handling based on identity data for scenarios such as candidates from other examination centers mistakenly entering or being absent from the examination, resulting in a difficulty in balancing examination management efficiency and security; and the lack of effective identity re-authentication methods for candidates re-entering the examination room after leaving, making it impossible to detect proxy testing and other violations in a timely manner, thus posing a significant security vulnerability. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above objectives, the present invention proposes the following technical solution: an intelligent identity recognition and authentication system based on candidate identity data, comprising the following: an entry authentication module, which collects candidate waiting images through a full-process monitoring terminal deployed in the waiting area of ​​the examination site, performs real-time face comparison verification based on a pre-stored candidate identity database, and outputs verification pass or verification failure results.

[0006] The verification failure handling module receives information about candidates whose verification failed, analyzes the reasons for the failure (including image quality issues and identity verification issues), and re-captures qualified images for secondary verification.

[0007] The identity classification and handling module identifies candidates with identity verification problems based on the secondary verification results, classifying them as candidates from other test centers, non-examinees, or absent candidates, and then executes corresponding handling instructions based on the identity classification category.

[0008] The departure monitoring module monitors the departure status of candidates in real time through the full-process monitoring terminal, and initiates the return identity re-authentication process when a candidate who has left re-enters, marking high-risk return candidates.

[0009] The return-to-exam authentication module initiates dual-channel authentication for high-risk return-to-exam candidates. Once dual-channel authentication is successful, the credibility weight is reset; otherwise, the candidate's status is locked and an alarm is triggered.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention collects the candidate waiting images through the full-process monitoring terminal of the entry authentication module, combines the pre-stored identity database for real-time face comparison, and repairs the image quality for problems such as insufficient lighting and low pixel count through the verification failure processing module, thereby improving the authentication accuracy.

[0011] (2) This invention uses an identity classification and processing module to cross-compare the facial data of candidates who fail secondary verification with the identity database of all test centers, automatically identifying candidates who are not from the test center and generating dynamic guidance containing actual test center navigation information; it triggers tiered warnings for absent candidates based on their entry timestamp and the reserved time before the start of the exam; and it executes departure alarms for non-examinees based on their loitering behavior characteristics and time thresholds. At the same time, it disseminates biometric summaries of candidates from different test centers through the education network, realizing network-wide collaborative management of abnormal identity status and improving the refined control capabilities of the entire process of exam identity authentication.

[0012] (3) This invention collects the historical entry behavior characteristics and writing posture data of candidates through the departure monitoring module, and reduces the credibility weight of the departing candidate's identity by combining the departure time index, so as to realize the quantitative assessment of the changes in the candidate's status after leaving the venue, and then activates the dual-channel authentication mechanism when returning to the venue to realize the multi-dimensional verification of the authenticity of the returning candidate's identity, effectively identify the risk of taking the exam on behalf of another candidate and trigger the graded disposal process.

[0013] (4) By introducing torso micro-motion identity spectrum analysis technology, this invention constructs a multi-level anti-counterfeiting system from behavioral features to biological features based on point cloud reconstruction and frequency domain feature comparison, combined with the multi-level verification mechanism of iris liveness authentication, providing a highly secure identity authentication solution for educational examination scenarios. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the system module connections of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 As shown, the present invention proposes an intelligent identity recognition and authentication system based on candidate identity data, which includes: an entry authentication module, a verification failure handling module, an identity classification and handling module, an exit monitoring module, and a return authentication module.

[0018] The entry authentication module, verification failure handling module, identity classification and handling module, departure monitoring module, and return authentication module are connected in sequence.

[0019] The entry authentication module collects images of candidates waiting in the waiting area through a full-process monitoring terminal deployed in the examination site, and performs real-time facial comparison verification based on a pre-stored candidate identity database, outputting a verification pass or verification failure result.

[0020] Specifically, the end-to-end monitoring terminal uses cameras deployed in the waiting area of ​​the examination center to collect a series of videos of each candidate's behavior while waiting to enter the examination room.

[0021] The pre-stored candidate identity database includes candidate ID, test site area, facial image, and candidate iris biometric database, which is used to store the candidate's iris code.

[0022] In a preferred embodiment, the real-time face comparison verification based on the pre-stored candidate identity database, and the output of verification pass or verification failure results, includes: delineating the face region of each candidate from the collected candidate waiting images, matching them one by one with the corresponding candidate face images at the test center in the pre-stored candidate identity database, marking the candidates who successfully match as verified, and marking the remaining candidates as verified.

[0023] The verification failure processing module receives the candidate information that failed the verification, analyzes the type of verification failure, including image quality issues and identity verification issues, and re-captures qualified images for secondary verification.

[0024] In a preferred embodiment, the verification failure processing module includes: acquiring the corresponding image features of the face image of the candidate who failed verification, and analyzing the reasons for the failure as insufficient lighting or low pixel count leading to image quality problems or identity verification problems where the face region image features do not match the corresponding candidates at the test center in the pre-stored candidate identity database.

[0025] Specifically, image recognition technology is used to examine the corresponding lighting and pixel features of the acquired face images to analyze the reasons for failure: when the average value of the V channel in the HSV space is less than 40 due to insufficient lighting, the face image is determined to have image quality problems caused by insufficient lighting. HSV includes the H channel representing hue, the S channel representing saturation, and the V channel representing brightness; when the average number of pixels per millimeter in length or width of the face image is less than 8, the face image is determined to have image quality problems caused by too few pixels.

[0026] To address image quality issues, facial areas with problems caused by insufficient lighting or low pixel count were marked in the collected images of candidates waiting in the waiting area, and an image acquisition reset mechanism was activated for the corresponding candidates.

[0027] The image acquisition reset mechanism involves controlling the supplementary lighting array for face images caused by insufficient lighting and increasing the pixel resolution for face images with low pixel density, thereby re-acquiring qualified images and performing secondary verification. For example, for the problem of insufficient lighting, the supplementary lighting array is controlled to output diffused light with a color temperature of 5500K and an illuminance of 800 lux to improve lighting conditions; for the problem of low pixel density, a super-resolution model is invoked to enlarge the size of the input low pixel density image by 4 times.

[0028] This invention collects images of candidates waiting in the waiting area through a full-process monitoring terminal of the entry authentication module, combines them with a pre-stored identity database for real-time face comparison, and uses a verification failure handling module to repair image quality for problems such as insufficient lighting and low pixel count, thereby improving the authentication accuracy and reducing misjudgments by single face recognition in complex environments.

[0029] The identity classification and handling module identifies candidates with identity verification problems based on the secondary verification results, classifying them as candidates from other test centers, non-examinees, or absent candidates, and then executes corresponding handling instructions based on the identity classification category.

[0030] In a preferred embodiment, the step of identifying candidates with identity verification problems based on secondary verification results and classifying them as candidates from other test centers, non-examinees, or absent candidates includes: counting all candidates who failed verification in the waiting images of candidates collected within the test center, and searching their facial images in the candidate identity database of other test centers within the test center, thereby classifying candidates whose faces are successfully matched as candidates from other test centers, and counting the candidates who did not match successfully.

[0031] Get the length of time that unmatched candidates stayed in the waiting area of ​​the test center. If their length of stay exceeds the preset length of stay, they are classified as non-candidates.

[0032] If a candidate's face image is missing from the candidate identity database stored at the test center, the candidate with the missing face image will be recorded as an absent candidate.

[0033] In a further preferred embodiment, the step of executing the corresponding processing instruction based on the identity classification category includes: obtaining the actual test center area of ​​candidates who are not at this test center, comparing it with the current test center area, obtaining the target test center navigation information, and then outputting target guidance information containing the actual test center navigation information, such as sending the actual test center navigation information to the test center or broadcasting the actual test center navigation information.

[0034] The system can detect absent candidates at the current test center in real time and obtain the reserved time between the timestamp of the detected absent candidate and the start time of the test. When the timetamp is less than the preset reserved time, the system can send a start time warning message to the absent candidate.

[0035] The system obtains the lingering behavior of non-examinees in the current test center's waiting area. The lingering behavior includes interacting with examinees at the test center or lingering alone. The system then obtains the reserved time. When the reserved time is less than the preset reserved time, the system triggers a departure alarm voice for non-examinees.

[0036] In a further preferred embodiment, the step of executing the corresponding processing instruction based on the identity classification category further includes: obtaining the corresponding biometric summary of non-examination site candidates who have been verified at each examination site, and transmitting it to the actual examination site area of ​​the corresponding candidate through the education network. The biometric summary may be such as facial feature points or clothing feature points.

[0037] When a candidate from another test center moves to a different test center, the biometric summary of the candidate entering the test center is captured in real time. When the biometric summary matches the biometric summary of the candidate from another test center retrieved from the actual test center area, the candidate's classification as a candidate from another test center is removed.

[0038] This invention uses an identity classification and processing module to cross-compare the facial data of candidates who fail secondary verification with the identity database of all test centers, automatically identifying candidates who are not from the test center and generating dynamic guidance containing actual test center navigation information; it triggers tiered warnings for absent candidates based on their entry timestamp and the reserved time before the exam; and it executes departure alarms for non-examinees based on their loitering behavior characteristics and time thresholds. Simultaneously, it disseminates biometric summaries of candidates from different test centers through the education network, achieving network-wide collaborative management of abnormal identity statuses and improving the refined control capabilities of the entire examination identity authentication process.

[0039] The departure monitoring module monitors the departure status of candidates in real time through the full-process monitoring terminal, and initiates the return identity re-authentication process when a candidate who has left re-enters, marking high-risk return candidates.

[0040] In a preferred embodiment, the departure monitoring module includes: initializing the candidate's identity credibility weight to a full score when the candidate enters the venue.

[0041] The system monitors candidates' departure behavior through a full-process monitoring terminal, and simultaneously retrieves historical entry videos of departing candidates collected by the terminal.

[0042] Specifically, the end-to-end monitoring terminal uses cameras deployed in classrooms to collect video footage of each candidate's behavior after entering the examination center.

[0043] Based on the historical entry videos of departing candidates, a departure duration index is calculated to trigger a credibility decay engine, which reduces the candidate's identity credibility weight. When the candidate's initial identity credibility weight drops to a preset weight threshold, the candidate is marked as a high-risk returnee.

[0044] In a further preferred embodiment, the step of calculating the departure duration index based on the historical entrance examination videos of departing candidates to trigger the credibility decay engine and reduce the credibility weight of the candidate's identity is specifically: extracting the entrance behavior feature set and historical writing posture feature set of departing candidates from their historical entrance examination videos. The entrance behavior feature set includes the seating position and departure timestamp, and the writing posture feature set includes, but is not limited to, eye movement deflection frequency and page turning frequency.

[0045] Specifically, through a full-process monitoring terminal deployed in the classroom, the historical entrance examination videos of departing candidates are analyzed frame by frame. Using computer vision technology, the spatial coordinate information of the candidates is identified from the video stream, and the two-dimensional plane coordinates of their seating positions and departure timestamps are extracted as core parameters of the entrance behavior feature set. At the same time, eye-tracking devices are used to collect the eye movement trajectories of candidates during the writing process, and the sequence of candidates' page-turning actions is collected by analyzing the video frame by frame to obtain the candidates' eye movement deflection frequency and page-turning frequency, thus constructing a historical writing posture feature set.

[0046] Obtain a pre-set baseline feature set of candidates' writing postures, which represents the expected writing postures. Compare this feature set with the baseline feature set of candidates who have left the examination room one by one to evaluate the degree of difference in writing postures between the two.

[0047] Specifically, the pre-set baseline feature set of candidates' writing posture is formulated by the education management department based on typical behavioral patterns in standardized testing scenarios. It includes standard threshold ranges for features in each dimension. Based on this, the historical writing posture feature set of candidates leaving the test is compared with the baseline feature set dimension by dimension, and the deviation of each dimension of writing posture features is calculated using a relative deviation formula. Where i represents the number of the writing posture feature dimension, i = 1, 2, ..., n, A iWrite the i-th dimension of the posture feature value for the departing candidate, such as the actual eye movement deflection frequency, B. i Let the i-th dimension of the writing posture feature value in the benchmark feature set be the benchmark value. By accumulating the deviation of all dimensions and taking the mean, we obtain the overall writing posture difference index μ0. The larger the value, the more significant the deviation between the candidate's actual posture and the standard pattern.

[0048] Based on the departure timestamps of the departing candidates, the departure duration t is calculated. - Based on the index of differences in writing posture, the departure time index T is calculated using the following formula: This formula normalizes the exit duration to avoid excessive amplification of the index by long exit durations; at the same time, it uses μ... - As a weighting factor, candidates with abnormal writing postures receive higher index values ​​for the same amount of time spent away from the exam.

[0049] A nonlinear decay function is used to map the departure time exponent to a credibility decay coefficient D, which is used to adjust the candidate's initial identity credibility weight, thereby obtaining the reduced identity credibility weight.

[0050] Specifically, the formula for the credibility decay coefficient is: D = e - β*T, where β is a preset attenuation factor used to adjust the attenuation rate. For example, if the preset value is 1.2, this function causes the confidence weight to decrease exponentially as T increases. For instance, when T = 0.5, the attenuation coefficient is D = e -0 ·6≈0.549, indicating that the credibility weight has dropped to 54.9% of the initial value, where e is the natural constant.

[0051] The return-to-exam authentication module initiates dual-channel authentication for high-risk return-to-exam candidates. After successful dual-channel authentication, the credibility weight is reset; otherwise, the candidate's status is locked and an alarm is triggered.

[0052] In a preferred embodiment, the return authentication module includes dual-channel authentication: channel one performs a spatiotemporal consistency comparison between the face and historical entry images, and channel two performs a comparison between behavioral features and pre-stored behavioral entropy templates.

[0053] If the comparison results of Channel 1 are consistent, it indicates that the candidate's entry record matches the historical entry record in terms of time and space, thus increasing the credibility of the identity verification.

[0054] Channel 2, based on the historical entrance exam videos of departing candidates, scans their seated torso point cloud to construct a baseline torso micro-movement identity spectrum of the departing candidate before departure.

[0055] Specifically, the method for constructing the torso micro-motion identity spectrum is as follows: extract the torso region image sequence of the candidates in a seated state from the historical entrance examination videos of the candidates leaving the examination room, and use three-dimensional point cloud reconstruction technology to sample the static contour and subtle movements of the candidates' torso in the torso region image sequence to generate a point cloud data set containing the spatial position information of each part of the torso.

[0056] The dynamic time warping algorithm is used to perform time series analysis on point cloud datasets to identify the minute displacement trajectories of the torso in natural states such as breathing and limb fine-tuning. The time-domain torso micro-movement trajectory is converted into frequency domain features through Fourier transform, and the energy distribution features under different frequency components are extracted to form a basic spectrum matrix characterizing the micro-movement characteristics of the examinee's torso.

[0057] Different frequency components represent different characteristics or patterns of trunk micro-movements. For example, some frequency components correspond to the candidate's natural breathing or slight body swaying, while other frequency components correspond to unnatural or deliberate micro-movements.

[0058] By comparing the frequency domain features with the pre-set candidate behavior entropy template in a spatiotemporal consistency manner, a benchmark identity spectrum reflecting the candidate's true torso micro-movement features before leaving the venue is constructed. This spectrum serves as the benchmark for biometric comparison during subsequent return authentication, thereby achieving a unique representation of the candidate's individual identity.

[0059] The torso micro-motion spectrum of candidates who returned from leaving the examination hall was collected within a set time period. The difference between the two spectra in terms of frequency distribution and energy proportion was calculated using the Euclidean distance formula. The normalized offset distance was then obtained. in, The frequency distribution characteristics of the j-th dimension in the torso micro-movement identity spectrum, based on the baseline, are the distribution parameters of the torso micro-movements pre-stored by the examinee at the j-th frequency component. The energy proportion feature of the j-th dimension in the baseline torso micro-motion identity spectrum is, that is, the proportion parameter of the j-th frequency component in the overall energy of the baseline spectrum. These represent the frequency distribution feature and energy proportion feature of the j-th dimension corresponding to the changed torso micro-motion identity spectrum, respectively. j represents the number of the spectrum feature dimension, j = 1, 2, ..., m, which is the position index of the corresponding frequency distribution or energy proportion feature parameters in the spectrum matrix, and m represents the total number of feature dimensions.

[0060] When the regularized offset distance exceeds the preset offset distance threshold, high-risk returnees are marked as suspected test-takers, and iris authentication is initiated against them. If iris authentication fails, an alarm is triggered on the proctoring end.

[0061] Specifically, for those suspected of taking the exam on behalf of others, the binocular iris acquisition device is activated. The iris texture image is obtained by irradiation with a near-infrared light source, and the real-time iris code is extracted. This is then compared with the candidate's pre-stored iris code using Hamming distance. If the comparison threshold is greater than 0.3, the iris authentication is deemed to have failed. An alarm message containing the candidate's ID, entry time, and abnormal characteristics is then sent to the proctoring terminal, and the candidate's exam status is simultaneously locked.

[0062] This invention collects historical entry behavior characteristics and writing posture data of candidates through an exit monitoring module, and reduces the credibility weight of the departing candidate's identity by combining the exit duration index, thereby achieving a quantitative assessment of the changes in the candidate's status after leaving the venue. Then, a dual-channel authentication mechanism is activated when the candidate returns to the venue to achieve multi-dimensional verification of the authenticity of the returning candidate's identity, effectively identify the risk of proxy test taking and trigger a graded handling process.

[0063] This invention introduces torso micro-motion identity spectrum analysis technology, based on point cloud reconstruction and frequency domain feature comparison, combined with a multi-level verification mechanism of iris liveness authentication, to construct a multi-level anti-counterfeiting system from behavioral features to biometric features, providing a highly secure identity authentication solution for educational examination scenarios.

[0064] It should be noted that the formulas described above, through the principle of dimensional consistency and mathematical standardization methods (such as normalization, dimensionless parameter conversion, or unit system unification), can translate physical quantities with different properties into unitless standard values ​​or superimposed parameters of the same dimension. This eliminates the interference of different dimensions on the computational logic, allowing the formulas to retain the original data distribution characteristics while possessing mathematical rationality and adaptability to objective laws. The descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the invention.

Claims

1. An intelligent identity recognition and authentication system based on examinee identity data, characterized in that, include: The entry authentication module collects images of candidates waiting in the waiting area through a full-process monitoring terminal deployed in the waiting area of ​​the test center, and performs real-time face comparison verification based on a pre-stored candidate identity database, outputting the verification result as either successful or unsuccessful. The verification failure handling module receives information about candidates whose verification failed, analyzes the reasons for the failure (including image quality issues and identity verification issues), and re-captures qualified images for secondary verification. The identity classification and handling module identifies candidates with identity verification problems based on the secondary verification results, classifying them as candidates who are not from this test center, non-examinees, or absent candidates, and then executes corresponding handling instructions based on the identity classification category. The departure monitoring module monitors the departure status of candidates in real time through the full-process monitoring terminal, and initiates the return identity re-authentication process when a candidate who has left re-enters, marking high-risk return candidates. The return-to-exam authentication module initiates dual-channel authentication for high-risk return-to-exam candidates. Once dual-channel authentication is successful, the credibility weight is reset; otherwise, the candidate's status is locked and an alarm is triggered.

2. The intelligent identity recognition and authentication system based on candidate identity data according to claim 1, characterized in that, The real-time face comparison verification based on the pre-stored candidate identity database, and the output of verification pass or verification failure results, includes: delineating the face region of each candidate from the collected candidate waiting images, matching it one by one with the corresponding candidate face images of the test center in the pre-stored candidate identity database, marking the candidates who successfully match as verified, and marking the remaining candidates as verified.

3. The intelligent identity recognition and authentication system based on candidate identity data according to claim 2, characterized in that, The verification failure handling module includes: Obtain the corresponding image features of the face images of candidates who failed verification, and analyze the reasons for failure, such as insufficient lighting or low pixel count leading to image quality problems, or identity verification problems where the face region image features do not match the corresponding candidates at the test center in the pre-stored candidate identity database. In response to image quality issues, facial areas with image quality problems caused by insufficient lighting or low pixel count were marked in the collected images of candidates waiting in the waiting area, and the image acquisition reset mechanism was activated for candidates in the corresponding waiting areas. The image acquisition reset mechanism controls the supplementary lighting array for face images caused by insufficient lighting and increases the pixel resolution for face images caused by low pixel count, thereby re-acquiring qualified images and performing secondary verification.

4. The intelligent identity recognition and authentication system based on candidate identity data according to claim 1, characterized in that, The process of identifying candidates with identity verification issues based on secondary verification results, categorizing them as candidates from outside the test center, non-examinees, or absent candidates, includes the following: The system collects images of candidates waiting in the examination center and identifies all candidates whose facial images failed verification. It then searches for their facial images in the candidate identity database stored at other examination centers within the examination center. Candidates whose facial images are successfully matched are classified as candidates from other examination centers, and candidates whose facial images are not successfully matched are also identified. Get the length of time that unmatched candidates have stayed in the waiting area of ​​the test center. If their length of stay exceeds the preset length of stay, they are classified as non-candidates. If a candidate's face image is missing from the candidate identity database stored at the test center, the candidate with the missing face image will be recorded as an absent candidate.

5. The intelligent identity recognition and authentication system based on candidate identity data according to claim 1, characterized in that, The execution of corresponding processing instructions based on identity classification categories includes: Obtain the actual test center area for candidates who are not at this test center, compare it with the current test center area, obtain the target test center navigation information, and then output the target guidance information containing the actual test center navigation information; Real-time detection of absent candidates at the current test center, and acquisition of the reserved time between the detected absent candidate's timestamp and the start timetamp. When the timetamp is less than the preset reserved time, an exam start warning message is sent to the absent candidate. The system obtains the lingering behavior of non-examinees in the current test center's waiting area. The lingering behavior includes interacting with examinees at the test center or lingering alone. The system then obtains the reserved time. When the reserved time is less than the preset reserved time, the system triggers a departure alarm voice for non-examinees.

6. The intelligent identity recognition and authentication system based on candidate identity data according to claim 5, characterized in that, The execution of corresponding processing instructions based on identity classification categories also includes: Obtain the biometric summaries of verified non-test center candidates from each test center and transmit them to the actual test center area of ​​the corresponding candidates through the education network; When a candidate from another test center moves to a different test center, the biometric summary of the candidate entering the test center is captured in real time. When the biometric summary matches the biometric summary of the candidate from another test center retrieved from the actual test center area, the candidate's classification as a candidate from another test center is removed.

7. The intelligent identity recognition and authentication system based on candidate identity data according to claim 1, characterized in that, The departure monitoring module includes: When a candidate enters the examination room, their identity credibility weight is initialized to a full score. The system detects candidates leaving the examination room through a full-process monitoring terminal, and simultaneously retrieves historical entry videos of candidates leaving the examination room collected by the full-process monitoring terminal. Based on the historical entry videos of departing candidates, a departure duration index is calculated to trigger a credibility decay engine, which reduces the candidate's identity credibility weight. When the candidate's initial identity credibility weight drops to a preset weight threshold, the candidate is marked as a high-risk returnee.

8. The intelligent identity recognition and authentication system based on candidate identity data according to claim 7, characterized in that, The departure duration index, calculated based on the historical entry videos of departing candidates, is used to trigger the credibility decay engine, reducing the credibility weight of the candidate's identity. Specifically: Extract the entry behavior feature set and historical writing posture feature set of the departing candidates from their historical entry examination videos; Obtain a pre-set set of reference features for candidates' writing postures, compare it with the reference feature set of candidates who have left the examination room, and evaluate the degree of difference in writing posture between the two. Based on the departure timestamps of departing candidates, the departure duration is obtained, and combined with the index of differences in writing posture, the departure duration index is calculated. A nonlinear decay function is used to map the departure time exponent to a credibility decay coefficient, which is used to adjust the candidate's initial identity credibility weight, thereby obtaining the reduced identity credibility weight.

9. The intelligent identity recognition and authentication system based on candidate identity data according to claim 1, characterized in that, The return authentication module, wherein: Dual-channel authentication includes: Channel 1 performs spatiotemporal consistency comparison between the face and historical entry images, and Channel 2 performs comparison between behavioral features and pre-stored behavioral entropy templates; If the comparison results for Channel 1 are consistent, it means that the candidate's entry time and location match the historical entry records. Channel 2, based on the historical entrance exam video of departing candidates, scans their seated torso point cloud to construct the torso micro-movement identity spectrum of departing candidates before leaving the exam. The spectrum of torso micro-movement identity changes during a set time period after the candidates who left the venue returned was collected. The difference between the two spectra in the corresponding characteristic dimensions of frequency distribution and energy proportion was calculated using the Euclidean distance formula. The normalized offset distance was then obtained. When the regularized offset distance exceeds the preset offset distance threshold, high-risk returnees are marked as suspected test-takers, and iris authentication is initiated against them. If iris authentication fails, an alarm is triggered on the proctoring end.

Images