A Smart Monitoring Method and System for Preventing Cheating in Driver's Exams
By establishing a test session context in driver's license examinations, collecting and aligning multi-source events, generating risk features and quality values, calculating risk scores by combining scenario weights, and forming a signed evidence package, the problem of inconsistent multi-source data integration in existing technologies is solved, thus achieving security and reliability in the examination process.
Patent Information
- Application Number
- CN202611316883.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-28
- Publication Date
- 2026-11-17
AI Technical Summary
In driver's license examinations, existing technologies struggle to effectively integrate data from multiple monitoring devices, leading to inconsistent judgments of abnormal events, affecting the continuity and reliability of the examinations. Furthermore, equipment malfunctions or network interruptions may result in data loss or duplication, impacting the accuracy of supervision.
By establishing exam session context, collecting and aligning multi-source events, generating risk characteristics and quality values, calculating risk scores by combining scenario weights, and forming a signed evidence package, the data is graded and manually reviewed to ensure data continuity and accuracy.
It enables unified management of multi-source data, reduces false triggering caused by equipment failure or network interruption, ensures the security and reliability of the examination process, and provides reliable evidence records.
Smart Images

Figure CN123117797A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent examination monitoring technology, specifically relating to an intelligent monitoring method and system for preventing cheating in driver's examinations. Background Technology
[0002] The driver's theory test and the practical test are conducted in fixed test stations and test vehicles, respectively. The theory test can be monitored through entry identity verification, communication jamming, camera recording, and test terminal logs. The practical test can be recorded through in-vehicle audio and video, vehicle identification, and location data. However, abnormal situations may occur during the driver's test, such as remote control, miniature earpieces, impersonation, and voice or gesture prompts from accompanying personnel. Therefore, it is necessary to integrate information from different sources, such as identity, liveness, operation, posture, audio, gesture, location, and equipment health, for supervision. The data collection cycles, time bases, event formats, and availability of various monitoring devices are not the same; if each device alarms separately, data from different candidates, test seats, vehicles, or test sessions may be difficult to consistently attribute to the same monitored object; when there is a lack of time correlation between weak anomalies, a single signal is also difficult to serve as a sufficient basis for graded handling; and sensor disconnection, clock out-of-sync, or network interruption can also cause data loss, out-of-order, and duplication. If missing data is directly interpreted as high risk, it is easy to cause the handling of anomalies to deviate from the actual data quality, affecting the continuity and reliability of examination supervision. In addition, the theoretical examination terminal can lock the examination interface when conditions are met, while the monitoring system in the practical examination should maintain the alarm and evidence recording link; after the triggering event occurs, the handling action, the data before the trigger, the data after the trigger, the time synchronization status and the reason for the missing should also form a consistent evidence record for the monitoring center and authorized personnel to review.
[0003] In summary, this application provides an intelligent monitoring method and system for preventing cheating in driver's license examinations. Summary of the Invention
[0004] The purpose of this invention is to provide an intelligent monitoring method and system for preventing cheating in driver's license examinations, in order to solve the problems of the prior art.
[0005] The specific technical solution adopted by this invention is as follows: A smart monitoring method for preventing cheating in driver's license examinations includes the following steps: Establish an exam session, receive the candidate's identity verification result, and generate an exam session context when the identity verification is successful and the corresponding resources are not occupied by other valid exam sessions; the exam session context is bound to the candidate token, exam mode, exam resource identifier, and exam acquisition device; Based on the exam session context, multi-source events corresponding to the exam mode are collected, each multi-source event is assigned an event identifier and stored in a circular cache, and aligned according to session and time sequence to form an aligned event batch and a circular cache index. Based on the risk characteristics of multiple events with consistent generation directions and the quality values of each risk characteristic, a feature vector is formed; Scenario weights are selected based on the examination mode, and a quality-weighted basic risk score is calculated based on the feature vector. The final risk score is obtained by combining the associated incremental execution state transition and generating decision events. The decision-making events and examination models are classified and handled in a tiered manner, and action receipts representing the handling results are obtained. Based on the available raw fragments in the fixed circular cache of decision events, the available raw fragments, feature vectors, decision events, and action receipts are combined to form a signed evidence package; Receive manual review instructions for the signed evidence package, verify the manual review instructions; generate a review decision after the verification is passed, and write the review decision and corresponding action receipt into the audit log.
[0006] Preferably, the step of collecting multi-source events corresponding to the examination mode based on the examination session context includes: In the theoretical examination mode, multi-source events include face events, liveness events, keyboard and mouse events, answering events, head posture events, and shielding device health events; in the theoretical examination mode, resource identifiers correspond to examination seats or examination terminals. In the practical examination mode, multi-source events include video events from the driver's seat and passenger seat, audio events, operational behavior events, gesture events, GPS events, and vehicle-related events; in the practical examination mode, resource identifiers are formed by the controlled mapping values of the vehicle's VIN.
[0007] Preferably, the step of attaching event identifiers to each multi-source event and storing them in a circular cache, aligning them by session and time sequence to form an aligned event batch and a circular cache index, includes: The event identifier is formed by combining the device identifier, serial number and session identifier; duplicate events with the same event identifier and the same payload digest are only retained as one payload, while events with the same event identifier but different payload digests generate security alarms and retain their respective received information; Duplicate payloads are excluded by event identifier, events are ordered by source time, sequence number and reception time, and aligned event batches are formed by session grouping within the associated time window; for duplicate events that are late, skipped, out of sync, or have inconsistent digests, the reason code and device health status are retained, and the original received records are not overwritten.
[0008] Preferably, the state transition includes normal state, suspicious state, alarm state, intervention state, manual review state, downgrade state, and closed state; When the theoretical exam reaches the intervention condition under alarm status, it enters the intervention state. When the practical exam is under alarm status, it enters the manual review state. After the equipment is restored, the backlog data is verified and the timeline is rebuilt, the reviewer decides whether to control the system to return from the degraded state to the monitoring state or to maintain the manual review state.
[0009] Preferably, the available original fragments in the fixed circular cache based on the decision event include: Centered on the triggering time of the decision event, the available original fragments within a preset time window before and after the triggering time are fixed; When the trigger time is close to the start of the exam session, the exam session ends early, or the data acquisition equipment malfunctions, resulting in insufficient available raw fragments within the preset time window, the actual start and end times and the reason for the missing data are recorded in the evidence list.
[0010] Preferably, the step of forming a signature evidence package from available original fragments, feature vectors, decision events, and action receipts includes: For each piece of evidence ordered by time, a fragment digest is calculated. The total digest is calculated in the order of fixed fields: session identifier, time boundary, fragment digest, decision event, action receipt, and reason for absence. The total digest is signed using the organization's controlled private key. In the event of a network interruption, the signed evidence packet with a fixed packet identifier is written to the local outbound queue. After the network is restored, it is retransmitted idempotently according to the fixed packet identifier.
[0011] Preferably, the step of tiered handling based on decision events and examination modes, and obtaining action receipts representing the handling results, includes: In normal conditions, data collection continues; in suspicious conditions, the recording level is increased and a prompt is displayed on the invigilator's terminal; in alarm conditions, the trigger time and risk contribution are pushed to the invigilator's terminal or the monitoring center. When the theoretical exam meets the intervention conditions, a lock interface request is sent to the exam terminal bound to the current exam session and the invigilator is notified; when the theoretical terminal returns a failure code, a receipt for escalating to a manual alarm is generated. When the practical test meets the alarm conditions, an alarm and push request are generated. The monitoring link remains isolated from the vehicle's steering, braking and power actuators, and an action receipt is obtained to characterize the actual execution result of the handling.
[0012] Preferably, the risk characteristics based on the consistent generation direction of multi-source events and the quality values of each risk characteristic include: The quality values for each risk characteristic are calculated based on data completeness, the deviation of the source time from the platform time, and equipment health. No. Quality value of risk characteristics The calculation formula is: ; in, This is a limiting function used to restrict the calculation result to between 0 and 1. If the calculation result is less than 0, it is set to 0; if it is greater than 1, it is set to 1. For data integrity, The deviation of the source time from the platform time. The alignment attenuation constant is positive. For equipment health; When the source is inapplicable, verification fails, or the device is offline, the corresponding quality value is set to zero and the reason is recorded; when the quality values of all applicable risk characteristics are zero, the regulatory status is changed to a downgraded status and the original fragment is retained.
[0013] Preferably, the step of selecting scenario weights based on the examination mode, calculating a quality-weighted basic risk score based on feature vectors, and combining the associated incremental execution state transition to obtain the final risk score includes: The quality-weighted basic risk score is generated based on the quality-weighted basic risk, which is calculated according to the scenario weights, each risk feature and its quality value. Quality-weighted basic risk score The calculation formula is: ; in, Assigning scenario weights to correspond to the exam mode. For the first The summation range is for risk characteristics that are applicable under the current examination model and have a quality value greater than zero. The final risk score is the sum of the quality-weighted basic risk score and the associated increment, after being narrowed to a range of zero to one hundred points. Final risk score The calculation formula is: ; in, This is a limiting function used to restrict the calculation result to between 0 and 100. If the calculation result is less than 0, it is taken as 0; if it is greater than 100, it is taken as 100. This is the associated increment.
[0014] A driver's license examination anti-cheating intelligent monitoring system includes the following units: The session binding unit is used to establish an exam session, receive the candidate's identity verification result, and generate an exam session context when the identity verification is successful and the corresponding resources are not occupied by other valid exam sessions. The acquisition and alignment unit is used to acquire multi-source events corresponding to the exam mode based on the exam session context, attach event identifiers to each multi-source event and store them in a circular cache, align them according to the session and time sequence, and form an aligned event batch and a circular cache index. The feature quality unit is used to generate risk features with consistent direction and quality values of each risk feature based on multi-source events, and to form a feature vector; The risk state unit is used to select scenario weights according to the examination mode, calculate the quality-weighted basic risk score based on the feature vector, and combine the associated incremental execution state transition to obtain the final risk score and generate decision events; The handling and scheduling unit is used to classify and handle events according to decision-making events and examination modes, and to obtain action receipts that represent the handling results. The evidence signing unit is used to form a signed evidence package by combining the available original fragments, feature vectors, decision events, and action receipts based on the available original fragments in the fixed circular cache of decision events. The manual review unit is used to receive manual review instructions for the signed evidence package, verify the manual review instructions, generate a review decision after the verification is passed, and write the review decision and corresponding action receipt into the audit log. A circular buffer unit is used to store the original fragments and for the acquisition and alignment unit and the evidence signing unit to read them; The audit storage unit is used to store signed evidence packages, review decisions, and action receipts. Beneficial effects
[0015] This invention binds the examination session context with the candidate token, examination mode, examination resource identifier or vehicle identifier, and examination data collection device, and ensures that the session identifier, examination resource identifier, sequence number, and source time in a unified event are continuously transmitted during the collection, alignment, calculation, and evidence fixation stages. It maintains the attribution of heterogeneous events such as face, liveness, keyboard and mouse, posture, audio, gesture, and location, and combines event identifier deduplication and corresponding resource occupancy verification to locate duplicates, out-of-order events, and cross-session conflicts, preventing cross-session data confusion and reducing the number of data entries into erroneous monitored objects when multiple sessions are conducted in parallel.
[0016] This invention forms a quality value by combining time alignment results, data integrity, and device health. This quality value is used simultaneously in the risk weighting numerator and effective denominator. Combined with the correlation increment and state transition conditions when independent modalities are jointly abnormal within the associated time window, the contribution of data to the risk result changes with the actual quality. Through feature combination, mutually corroborating anomalies obtain a consistent state response, thereby reducing false triggers caused by a single low-quality frame, device disconnection, or isolated weak signal.
[0017] This invention is based on the selection of examination mode, which involves locking the theoretical examination interface or pushing alarms for the practical examination. It writes the decision event, action receipt, available original fragments for 30 seconds before and after, synchronization status, and reasons for missing information into a summary list. The summary is then processed by sequential summarization and institutional signature to form a signed evidence package. The manual review results drive the recovery or closure of the system. This ensures that the regulatory actions are consistent with the examination security boundaries and makes the triggering basis, execution results, and data integrity verifiable. Attached Figure Description
[0018] Figure 1 This is one of the method flowcharts of the present invention; Figure 2 This is the second flowchart of the method of the present invention; Figure 3 This is the third flowchart of the method of the present invention; Figure 4 This is a system module diagram of the present invention. Detailed Implementation
[0019] The technical solution of the present invention will be further described in detail below with reference to specific embodiments. The following embodiments are used to illustrate the present invention, but should not be used to limit the scope of protection of the present invention. The conditions in the embodiments can be further adjusted according to specific conditions. Simple improvements to the method of the present invention under the premise of the concept of the present invention are all within the scope of protection claimed by the present invention. Example 1
[0020] Please refer to Figures 1-3 This embodiment provides an intelligent monitoring method for preventing cheating in driver's license examinations, including the following steps: S1. Session establishment and object binding; Receive application records, candidate tokens, facial recognition gate verification results, exam mode, and theoretical exam slot identifier or practical vehicle VIN; first verify the identity verification result, then check whether the exam slot or vehicle is occupied by other open sessions; generate a unique session identifier for this exam when identity verification is successful and resources are available; When identity verification is successful and the corresponding resources are not occupied by other valid examination sessions, the candidate token, examination session, examination mode, resource identifier, and examination data collection device list are written into the examination session context, and the session start time is recorded; in the theoretical examination mode, the resource identifier corresponds to the examination seat or examination terminal; in the practical examination mode, the resource identifier is formed by the controlled mapping value of the vehicle VIN; If identity verification fails or resources are still occupied, a binding rejection record is generated and submitted to manual verification to maintain the effective monitoring session in a state where it has not been started.
[0021] S2, multi-source acquisition, serialization and time alignment; Read the exam session context and collect and receive multi-source data corresponding to the exam mode; The theoretical exam data includes facial events, liveness events, keyboard and mouse events, answering events, head posture events, and shielding device health events; the practical exam data includes video events, audio events, operational behavior events, gesture events, GPS events, and vehicle-related events in the driver's and passenger's seats. Add an event identifier to each event. The event identifier includes session identifier, exam mode, resource identifier, source time, reception time, monotonically increasing sequence number within the same device, etc. The original audio and video and structured logs are written to the circular cache at the same time. Duplicate events with the same event identifier and the same payload digest are retained only once; events with the same event identifier but different payload digests generate security alarms and retain their respective received information. Duplicate payloads are excluded by event identifier, events are ordered by source time, sequence number and reception time, and aligned event batches are formed by session grouping within the associated time window; for duplicate events that are late, skipped, out of sync, or have inconsistent digests, the reason code and device health status are retained, and the original records that have been received are not overwritten. Output alignment event batches, circular cache indexes, and device health events.
[0022] S3, Feature computation and data quality gating; It receives multi-source events such as alignment event batches, packet loss information, device heartbeats and synchronization deviations, and extracts risk characteristics such as identity inconsistency, liveness anomaly, operation anomaly, posture or gesture anomaly, voice prompts and external communication or remote control clues. Based on the risk characteristics generated by multiple events with consistent direction and the quality values of each risk characteristic, a feature vector is formed; where consistent direction of risk characteristics means that the direction of increase or decrease of all risk characteristic values is consistent with the direction of change of risk severity, the larger the value, the higher the risk, and all characteristics are uniformly uniform in the same monotonic relationship.
[0023] The quality values for each risk characteristic are calculated based on data completeness, the deviation of the source time from the platform time, and equipment health. No. Quality value of risk characteristics The calculation formula is: ; in, This is a limiting function used to restrict the calculation result to between 0 and 1. If the calculation result is less than 0, it is set to 0; if it is greater than 1, it is set to 1. For data integrity, The deviation of the source time from the platform time. The alignment attenuation constant is positive. For equipment health; This is an alignment decay constant used to measure the rate at which the credibility of risk characteristics decays due to time deviations during the time alignment process of multi-source events. The value is greater than zero, and the range is from 0.5 seconds to 2.0 seconds. The specific value can be dynamically adjusted according to the real-time requirements of the examination scenario and the accuracy of the device clock. When the source is not applicable, verification fails, or the device is offline, the corresponding quality value is set to zero and the reason is recorded; when the quality values of all applicable risk characteristics are zero, the regulatory status is changed to a downgraded status and the original fragment is retained. If valid quality features exist, the output will include... Feature vectors of source category and statistical window identifier; Among them, the inconsistency of identity features Normalized face similarity calculate, abnormal characteristics in living organisms Normalized live score calculate, ; , , and All are dimensionless numbers between zero and one; The theoretical answering operation anomaly is calculated by combining the duration of continuous no input and the answering rate after input is restored; the current duration of continuous no input is... The base duration for seconds and positive values is Seconds, the answering speed within the statistical window is The reference rate for the question is per second, which is positive. If the problem occurs every second, then the operation exhibits abnormal characteristics. for: ; in, A dimensionless number between zero and one; , The units are all seconds; , All units are per second. When any observation does not exceed the corresponding benchmark, its corresponding term is set to zero. Abnormal posture or gestures Calculation, where To calculate the cumulative duration of time the window is outside the configured focus area or when a notification gesture is triggered, The duration of the statistical window is positive, and both are in seconds. ; A dimensionless number between zero and one; Voice prompts and external communication detection results are labeled and calibrated within the same implementation to form zero-to-one features. Inapplicable sources are identified through... Exit fusion.
[0024] S4. Scenario-based risk integration and state transition; Receive the formed feature vector, the exam mode in the exam session context, the state of the previous calculation cycle, and the rule version; Scenario weights are selected based on the examination mode, and a quality-weighted basic risk score is calculated based on the feature vector. The final risk score is obtained by combining the associated incremental execution state transition and generating decision events. Assign weights to each exam mode ,in A dimensionless number between zero and one, and the sum of the weights of all applicable features under the same mode is one; The quality-weighted basic risk score is generated based on the quality-weighted basic risk, which is calculated according to the scenario weights, each risk feature and its quality value. Quality-weighted basic risk score The calculation formula is: ; in, Assigning scenario weights to correspond to the exam mode. For the first Risk characteristics are summed within the range of risk characteristics applicable under the current examination model and with a quality value greater than zero; when the denominator is zero, the regulatory status is changed to a downgraded status, and the original fragment is retained without quality-weighted basic risk scoring. Calculation; The final risk score is the sum of the quality-weighted basic risk score and the associated increment, after being narrowed to a range of zero to one hundred points. Final risk score The calculation formula is: ; in, This is a limiting function used to restrict the calculation result to between 0 and 100. If the calculation result is less than 0, it is taken as 0; if it is greater than 100, it is taken as 100. For related increments; Among them, when associated time window At least two independent sources simultaneously satisfy and hour, Take 15 points, otherwise Take zero points; among which, the associated time window A positive second value is used to determine whether multi-source risk events are correlated in time. The value is greater than zero and ranges from 3 to 10 seconds. The specific value can be dynamically adjusted according to the real-time requirements of the examination scenario and the sensitivity of the risk response. Identity results and liveness results derived from the same camera can be marked as the same source to avoid the formation of correlation increments based on only one collection device. Decision-generating events include: based on the final risk score The system is divided into zones: 0-40 points are configured as the normal zone, 40-70 points as the suspicious zone, 70-85 points as the alarm zone, and 85-100 points as the intervention zone. The zones are configured with left-closed and right-open intervals, and the last interval contains 100 points. Final risk score under normal conditions When a user enters the suspicious zone for two consecutive cycles, they are transferred to a suspicious state. Final risk score When the score is not lower than 70 points and cross-modal correlation is met, or when the configured compound rules of proxy testing and remote control are hit, the system will move to the alarm state; when the theoretical test score is not lower than 85 points or the compound rules are hit continuously, the system can enter the intervention state; when the practical test score is lower than the alarm state, the system will move to the manual review state.
[0025] S5. Classify and handle cases and obtain action receipts; The decision-making events and examination models are classified and handled in a tiered manner, and action receipts representing the handling results are obtained. Receive the output decision event and read the exam mode and current exam business status from the exam session context; In normal conditions, data collection continues; in suspicious conditions, the recording level is increased and a prompt is displayed on the invigilator's terminal; in alarm conditions, the trigger time and risk contribution are pushed to the invigilator's terminal or the monitoring center. When the theoretical exam meets the intervention conditions, a lock interface request is sent to the exam terminal bound to the current exam session and the invigilator is notified; when the theoretical terminal returns a failure code, a receipt for escalating to a manual alarm is generated. When the practical examination meets the alarm conditions, an alarm and push request are generated, and the monitoring link remains isolated from the vehicle's steering, braking and power actuators; each request carries a session identifier and an idempotent action identifier, and the execution end returns an action receipt that represents the actual execution result, such as the request time, completion time and success or failure code; When the theoretical terminal returns a failure code, a receipt for escalating to a manual alarm is generated, maintaining the request sent and the interface locked as two distinguishable states.
[0026] S6. Front and rear window evidence fixation, summary and signature push; Based on the available raw fragments in the fixed circular cache of decision events, the available raw fragments, feature vectors, decision events, and action receipts are combined to form a signed evidence package; Simultaneously receive the circular cache index, quality and missing information, decision events, and action receipts; center on the trigger time of the decision event, fix the available raw fragments within the preset time window before and after the trigger time; the preset time window before and after can be 30 seconds before and after the trigger time; the available data within the fixed interval of the circular cache 30 seconds before and after the trigger time of the current exam session are used as the available raw fragments, the fixed camera position audio and video and operation logs for theoretical exams, and the fixed video, audio and GPS trajectory for practical exams; When the trigger time is close to the start of the exam session, the exam session ends early, or the acquisition equipment malfunctions, resulting in insufficient available raw fragments within the preset time window, the actual start and end times and the reason for the missing information are recorded in the evidence list. Calculate fragment summaries for each piece of evidence ordered by time, and calculate the total summary in the fixed field order of session identifier, time boundary, fragment summary, decision event, action receipt and reason for absence, and sign the total summary using the organization’s controlled private key; Arranged chronologically for the first Fragments of evidence calculate Then the summary and signature for: ; ; in, For segment number, It is a positive integer. For the first A fragment of evidence, Provide a summary; For session identification, , This serves as evidence of the actual start and end timestamps in milliseconds. , These are the standardized codes for decision events and action feedback, respectively. Code the reason for the missing information; For cryptographic digest functions, For unambiguous encoding connections in the order shown, This is a summary. For the organization's controlled private key, For digital signatures; When the network is interrupted, the signed evidence packets with fixed packet identifiers are written into the local outbound queue. The signed evidence packets are queued with the same packet identifier and retransmitted idempotently after the network is restored. When the network is available, they are retransmitted idempotently to the monitoring center and audit storage according to the fixed packet identifiers.
[0027] S7. Manual review, recovery, and session closure; Receive manual review instructions for the signed evidence package, verify the manual review instructions; generate a review decision after the verification is passed, and write the review decision and corresponding action receipt into the audit log; Read the output signed evidence package, action receipts and device health events, and display the original fragments, synchronization status, contribution of each feature and actual handling results to authorized personnel; Upon receiving instructions to maintain alarms, clear alarms, restore the theory exam interface, retake or close the session, first compare the session identifier in the instruction with the current valid session, and then verify the operator's permissions and exam status. When the maintenance procedure or examination concludes, output the closure decision, write the end time, and seal the audit record; When the alarm is rejected and the session verification is successful, a recovery decision is output, the theoretical interface is unlocked or the practical flag is removed, a recovery receipt is generated, and the monitoring continues after multi-source collection, serialization and time alignment are performed. The review result becomes the explicit input for subsequent recovery or shutdown actions, the automatic risk result is kept as the basis for triggering regulatory actions, and the review decision of authorized personnel forms the final disposal record. Example 2
[0028] Please refer to Figure 4 This embodiment provides an intelligent monitoring system for preventing cheating in driver's license examinations, comprising the following units: The session binding unit receives the identity verification result, exam mode, and exam seat or vehicle identifier to establish an exam session. It receives the candidate's identity verification result and generates an exam session context when the identity verification is successful and the corresponding resources are not occupied by other valid exam sessions. The acquisition and alignment unit is used to acquire multi-source events corresponding to the exam mode based on the exam session context, attach event identifiers to each multi-source event and store them in a circular cache, align them according to the session and time sequence, and form an aligned event batch and a circular cache index. The feature quality unit is used to generate risk features with consistent direction and quality values of each risk feature based on multi-source events, and to form a feature vector; The risk state unit is used to select scenario weights according to the examination mode, calculate the quality-weighted basic risk score based on the feature vector, and combine the associated incremental execution state transition to obtain the final risk score and generate decision events; The handling and scheduling unit is used to classify and handle events according to decision-making events and examination modes, and to obtain action receipts that represent the handling results. The evidence signing unit is used to form a signed evidence package by combining the available original fragments, feature vectors, decision events, and action receipts based on the available original fragments in the fixed circular cache of decision events. The manual review unit is used to receive manual review instructions for the signed evidence package, verify the manual review instructions, generate a review decision after the verification is passed, and write the review decision and corresponding action receipt into the audit log. A circular buffer unit is used to store the original fragments and for the acquisition and alignment unit and the evidence signing unit to read them; The audit storage unit is used to store signed evidence packages, review decisions, and action receipts.
[0029] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A smart monitoring method for preventing cheating in driver's license examinations, characterized in that, Includes the following steps: Establish an exam session, receive the candidate's identity verification result, and generate an exam session context when the identity verification is successful and the corresponding resources are not occupied by other valid exam sessions; The exam session context is bound to the candidate token, exam mode, exam resource identifier, and exam data acquisition device; Based on the exam session context, multi-source events corresponding to the exam mode are collected, each multi-source event is assigned an event identifier and stored in a circular cache, and aligned according to session and time sequence to form an aligned event batch and a circular cache index. Based on the risk characteristics of multiple events with consistent generation directions and the quality values of each risk characteristic, a feature vector is formed; Scenario weights are selected based on the examination mode, and a quality-weighted basic risk score is calculated based on the feature vector. The final risk score is obtained by combining the associated incremental execution state transition and generating decision events. The decision-making events and examination models are classified and handled in a tiered manner, and action receipts representing the handling results are obtained. Based on the available raw fragments in the fixed circular cache of decision events, the available raw fragments, feature vectors, decision events, and action receipts are combined to form a signed evidence package; Receive manual review instructions for the signed evidence package, verify the manual review instructions; generate a review decision after the verification is passed, and write the review decision and corresponding action receipt into the audit log.
2. The intelligent monitoring method for preventing cheating in driver's license examinations according to claim 1, characterized in that, The collection of multi-source events corresponding to the examination mode based on the examination session context includes: In the theoretical examination mode, multi-source events include face events, liveness events, keyboard and mouse events, answering events, head posture events, and shielding device health events; in the theoretical examination mode, resource identifiers correspond to examination seats or examination terminals. In the practical examination mode, multi-source events include video events from the driver's seat and passenger seat, audio events, operational behavior events, gesture events, GPS events, and vehicle-related events; in the practical examination mode, resource identifiers are formed by the controlled mapping values of the vehicle's VIN.
3. The intelligent monitoring method for preventing cheating in driver's license examinations according to claim 1, characterized in that, The step of attaching event identifiers to each multi-source event and storing them in a circular cache, aligning them by session and time sequence to form an aligned event batch and a circular cache index, includes: The event identifier is formed by combining the device identifier, serial number and session identifier; duplicate events with the same event identifier and the same payload digest are only retained as one payload, while events with the same event identifier but different payload digests generate security alarms and retain their respective received information; Duplicate payloads are excluded by event identifier, events are ordered by source time, sequence number and reception time, and aligned event batches are formed by session grouping within the associated time window; for duplicate events that are late, skipped, out of sync, or have inconsistent digests, the reason code and device health status are retained, and the original received records are not overwritten.
4. The intelligent monitoring method for preventing cheating in driver's license examinations according to claim 1, characterized in that: The state transitions include normal state, suspicious state, alarm state, intervention state, manual review state, downgraded state, and closed state; When the theoretical exam reaches the intervention condition under alarm status, it enters the intervention state. When the practical exam is under alarm status, it enters the manual review state. After the equipment is restored, the backlog data is verified and the timeline is rebuilt, the reviewer decides whether to control the system to return from the degraded state to the monitoring state or to maintain the manual review state.
5. The intelligent monitoring method for preventing cheating in driver's license examinations according to claim 1, characterized in that, The available raw fragments in the fixed circular cache based on decision events include: Centered on the triggering time of the decision event, the available original fragments within a preset time window before and after the triggering time are fixed; When the trigger time is close to the start of the exam session, the exam session ends early, or the data acquisition equipment malfunctions, resulting in insufficient available raw fragments within the preset time window, the actual start and end times and the reason for the missing data are recorded in the evidence list.
6. The intelligent monitoring method for preventing cheating in driver's license examinations according to claim 1, characterized in that, The method of forming a signature evidence package from available original fragments, feature vectors, decision events, and action receipts includes: For each piece of evidence ordered by time, a fragment digest is calculated. The total digest is calculated in the order of fixed fields: session identifier, time boundary, fragment digest, decision event, action receipt, and reason for absence. The total digest is signed using the organization's controlled private key. In the event of a network interruption, the signed evidence packet with a fixed packet identifier is written to the local outbound queue. After the network is restored, it is retransmitted idempotently according to the fixed packet identifier.
7. The intelligent monitoring method for preventing cheating in driver's license examinations according to claim 1, characterized in that, The step of tiered handling based on decision events and examination modes, and obtaining action receipts representing the handling results, includes: In normal conditions, data collection continues; in suspicious conditions, the recording level is increased and a prompt is displayed on the invigilator's terminal; in alarm conditions, the trigger time and risk contribution are pushed to the invigilator's terminal or the monitoring center. When the theoretical exam meets the intervention conditions, a lock interface request is sent to the exam terminal bound to the current exam session and the invigilator is notified; when the theoretical terminal returns a failure code, a receipt for escalating to a manual alarm is generated. When the practical test meets the alarm conditions, an alarm and push request are generated. The monitoring link remains isolated from the vehicle's steering, braking and power actuators, and an action receipt is obtained to characterize the actual execution result of the handling.
8. The intelligent monitoring method for preventing cheating in driver's license examinations according to claim 1, characterized in that, The risk characteristics based on the consistent generation direction of multi-source events and the quality values of each risk characteristic include: The quality values for each risk characteristic are calculated based on data completeness, the deviation of the source time from the platform time, and equipment health. No. Quality value of risk characteristics The calculation formula is: ; in, This is a limiting function used to restrict the calculation result to between 0 and 1. If the calculation result is less than 0, it is set to 0; if it is greater than 1, it is set to 1. For data integrity, The deviation of the source time from the platform time. The alignment attenuation constant is positive. For equipment health; When the source is inapplicable, verification fails, or the device is offline, the corresponding quality value is set to zero and the reason is recorded; when the quality values of all applicable risk characteristics are zero, the regulatory status is changed to a downgraded status and the original fragment is retained.
9. The intelligent monitoring method for preventing cheating in driver's license examinations according to claim 1, characterized in that, The process of selecting scenario weights based on the examination mode, calculating a quality-weighted basic risk score based on feature vectors, and combining this with the associated incremental execution state transition to obtain the final risk score includes: The quality-weighted basic risk score is generated based on the quality-weighted basic risk, which is calculated according to the scenario weights, each risk feature and its quality value. Quality-weighted basic risk score The calculation formula is: ; in, Assigning scenario weights to correspond to the exam mode. For the first The summation range is for risk characteristics that are applicable under the current examination model and have a quality value greater than zero. The final risk score is the sum of the quality-weighted basic risk score and the associated increment, after being narrowed to a range of zero to one hundred points. Final risk score The calculation formula is: ; in, This is a limiting function used to restrict the calculation result to between 0 and 100. If the calculation result is less than 0, it is taken as 0; if it is greater than 100, it is taken as 100. This is the associated increment.
10. A driver's license examination anti-cheating intelligent monitoring system, characterized in that, Includes the following units: The session binding unit is used to establish an exam session, receive the candidate's identity verification result, and generate an exam session context when the identity verification is successful and the corresponding resources are not occupied by other valid exam sessions. The acquisition and alignment unit is used to acquire multi-source events corresponding to the exam mode based on the exam session context, attach event identifiers to each multi-source event and store them in a circular cache, align them according to the session and time sequence, and form an aligned event batch and a circular cache index. The feature quality unit is used to generate risk features with consistent direction and quality values of each risk feature based on multi-source events, and to form a feature vector; The risk state unit is used to select scenario weights according to the examination mode, calculate the quality-weighted basic risk score based on the feature vector, and combine the associated incremental execution state transition to obtain the final risk score and generate decision events; The handling and scheduling unit is used to classify and handle events according to decision-making events and examination modes, and to obtain action receipts that represent the handling results. The evidence signing unit is used to form a signed evidence package by combining the available original fragments, feature vectors, decision events, and action receipts based on the available original fragments in the fixed circular cache of decision events. The manual review unit is used to receive manual review instructions for the signed evidence package, verify the manual review instructions, generate a review decision after the verification is passed, and write the review decision and corresponding action receipt into the audit log. A circular buffer unit is used to store the original fragments and for the acquisition and alignment unit and the evidence signing unit to read them; The audit storage unit is used to store signed evidence packages, review decisions, and action receipts.
