Intelligent construction site face recognition attendance supervision and management system and management method

Abstract

The invention relates to the technical field of face recognition, in particular to an intelligent construction site face recognition attendance supervision and management system and management method. Comprising a binocular living body detection camera, an identity card reader, an IC / ID card reader, an infrared monitoring sensor and a gate acceleration sensor; the dynamic weight calculation module is used for dynamically adjusting the weight parameter of the face matching degree according to the living body detection result, the human-certificate comparison data and the environment illumination condition; the malicious gate rushing judgment module is used for generating a gate rushing risk value based on gate acceleration sensor data, infrared monitoring border crossing behaviors and continuous face recognition failure times; the attendance data optimization module is used for analyzing and correcting the confidence coefficient of attendance records through a historical recognition success rate and a time sequence; according to the invention, safety management requirements in complex scenes can be met.

Description

Technical Field

[0001] The present invention relates to the field of face recognition technology, and specifically to a smart construction site face recognition attendance supervision management system and management method. Background Art

[0002] Safety and personnel management at construction sites have always been core challenges in the engineering field. Traditional construction site attendance systems often rely on a single verification method, such as contactless IC cards or facial recognition technology based on 2D cameras. However, as construction sites expand in size and the construction environment becomes more complex, these systems have exposed significant shortcomings in reliability, safety, and management efficiency: Inadequate environmental adaptability: Construction sites often face strong light, backlight, dust, or low-light conditions. Conventional cameras, which rely on traditional facial recognition, have limited dynamic range and struggle to accurately capture facial features. Limited authentication methods: Existing systems often use independent verification modules, such as card swiping and facial recognition, allowing attackers to bypass checks using forged documents or stolen IC cards. Crude gate-crossing detection: Traditional gates use infrared sensors to detect intrusions, but they cannot distinguish between normal rapid passage and malicious attempts.

[0003] In recent years, some improvement plans have attempted to introduce liveness detection, multi-sensor fusion, or cloud platform integration, but limitations still exist: for example, the accuracy of liveness detection drops sharply under extreme lighting; multi-sensor data is simply superimposed without dynamic weight allocation; and there is insufficient coordination between localized systems and cloud business platforms.

[0004] Based on the above problems, there is an urgent need for a construction site attendance system that is environmentally robust, accurately verified, and data collaborative to meet the safety management needs in complex scenarios. Summary of the Invention

[0005] The current attendance supervision and management system simply superimposes multi-sensor data and lacks dynamic weight allocation; the local system and cloud business platform lack coordination; this solution provides a smart construction site facial recognition attendance supervision and management system and management method.

[0006] The present application provides an automatic torque spot check system and control method to solve the above problems.

[0007] To achieve the above object, the present invention is implemented through the following technical solutions: The present application discloses a smart construction site facial recognition attendance supervision and management system, including: Multimodal data acquisition module: including binocular liveness detection camera, ID card reader, IC / ID card reader, infrared monitoring sensor and gate acceleration sensor; Dynamic weight calculation module: used to dynamically adjust the weight parameters of face matching based on liveness detection results, person-document comparison data, and ambient lighting conditions; Malicious gate-breaking determination module: Generates a gate-breaking risk value based on gate acceleration sensor data, infrared monitoring of cross-border behavior, and the number of consecutive facial recognition failures; Attendance data optimization module: Modify the confidence level of attendance records through historical recognition success rate and time series analysis; Data linkage management module: used to upload attendance data to the Ministry of Housing and Urban-Rural Development platform in real time, and to link with the construction site LED display screen, mobile terminal and alarm device.

[0008] Adopting the above technical solution: This solution can solve the problem that traditional construction site attendance systems usually rely on a single sensor, have a high failure rate of facial recognition in strong light, backlight or low light conditions, and only detect cross-border behavior through infrared sensors, which cannot distinguish between normal passage and violent breaking into the gate; this solution integrates binocular cameras, ID card readers, accelerometers and other hardware through multimodal data acquisition modules. The wide dynamic camera adapts to strong light / low light scenes and combines liveness detection to prevent photo attacks; it integrates accelerometer data and infrared cross-border time to distinguish normal passage from malicious intrusion.

[0009] Preferably, the face matching weight adjustment formula in the dynamic weight calculation module is: , in, is the final matching weight value, is the confidence score of binocular liveness detection, To compare the similarity of human evidence, is the ambient light intensity and the unit is lux, =500 is the illumination reference value, =0.6, =0.3, =0.1 is the weight coefficient, =0.01 is the illumination compensation coefficient; it is judged to be legal at that time, otherwise it will trigger secondary verification.

[0010] Adopting the above technical solution: This solution introduces a light intensity compensation item, which can automatically reduce the weight of light impact in strong light / weak light; solves the problem of difficulty in extracting facial features in backlight environment, mistakenly judging legitimate people as strangers, and failing to dynamically adjust the weights of the two according to the actual scene, resulting in the failure of liveness detection under extreme light.

[0011] Further preferably, the gate-breaking risk calculation formula of the malicious gate-breaking determination module is: , in, is the gate-breaking risk value, The peak acceleration detected by the gate acceleration sensor, in units of , is the acceleration threshold, is the number of consecutive face recognition failures, The crossing time of infrared monitoring, in seconds, =2S is the safe passage time, =0.5, =0.3, =0.2 is the risk factor; when The sound and light alarm is triggered and the gate is locked.

[0012] Adopting the above technical solution: This solution can solve the problem that the traditional gate system relies on a single sensor to determine the gate-breaking behavior, which easily causes the infrared sensor to be blocked or falsely triggered, and misjudges normal and fast passage as gate-breaking. The setting of the acceleration threshold in this solution can effectively distinguish normal passage from violent gate-breaking.

[0013] Further preferably, the confidence correction formula of the attendance data optimization module is: , in, The confidence level of the corrected attendance ranges from 0 to 1. For the The success flag of the historical recognition is 1 for success and 0 for failure. is the current time, =0,1 is the time attenuation coefficient; when It is marked as abnormal attendance and manual review is initiated.

[0014] The above technical solution solves the problem of traditional attendance systems relying on single recognition results, which may lead to temporary camera damage or network interruption, resulting in single recognition failure and misjudgment as absence. This solution provides a corresponding confidence correction formula to achieve attenuated weight distribution: recent successful recognition records contribute more to the current confidence, reducing early data interference; automatic anomaly marking: when C < 0.7, manual review is triggered to reduce misjudgment due to occasional equipment failure.

[0015] Further preferably, the data linkage management module includes a data interface protocol for docking with the Ministry of Housing and Urban-Rural Development platform, which uploads personnel names, types of work, attendance times and gate-breaking events in real time; dynamically displays the number of people present, high-risk behavior statistics and temperature abnormality alarm information through an LED display screen; pushes alarm notifications containing the location of malicious gate-breaking incidents, on-site video clips and handling suggestions to security personnel through a mobile terminal APP; at the same time, synchronizes abnormal attendance data to the enterprise ERP system for wage accounting and contract management, and links with the construction site real-name training platform to automatically prohibit personnel who have not completed safety training from passing.

[0016] Adopting the above technical solution: This solution can solve the problem of data silos in traditional construction site management systems, which results in attendance data having to be manually exported and uploaded to the housing and construction platform, and cannot meet real-time supervision needs. This solution can meet government supervision requirements by automatically encrypting and uploading personnel pass data to the housing and construction platform.

[0017] A management method, applied to a smart construction site face recognition supervision and management system as described in any one of the above, is characterized by comprising: S1: Capture facial images through binocular cameras and simultaneously collect ambient light intensity. It also reads ID card or IC card information and matches it with a pre-stored whitelist, and monitors gate acceleration and infrared cross-border signals in real time. S2: Calculates the liveness detection score and the person-document comparison similarity, adjusts the weight parameters according to the ambient light intensity, and uses a dynamic weight formula to determine the legitimacy of the identity. If the verification passes, access is authorized and attendance is recorded. Otherwise, IC card-assisted verification is initiated. S3: When it is detected that the gate acceleration exceeds the threshold or the infrared crossing time is lower than the safety value, the facial area image is extracted and the gate-breaking risk value is calculated. If the risk value exceeds the threshold, an alarm is triggered and the gate-breaking video evidence is saved; S4: Based on historical recognition records and the time decay model, the confidence level of attendance data is corrected. Low-confidence data is marked as abnormal and the administrator is notified for review. The corrected attendance data is encrypted and uploaded to the Ministry of Housing and Urban-Rural Development platform and the enterprise management system. S5: Display the real-time number of people present and gate-breaking event statistics on the LED screen, push event details and processing priorities to the security personnel's mobile terminal, and lock the access rights of the team to which the malicious gate-breaking person belongs.

[0018] Adopting the above technical solution: This solution can solve the problem that the traditional construction site attendance method has a scattered process, resulting in a single verification step, and cannot cope with complex scenarios due to only relying on face or card swiping. This solution can realize the simultaneous collection of face, ID and ambient light data through multimodal verification, and can dynamically adjust the verification strategy.

[0019] Further preferably, the IC card auxiliary verification includes comparing the IC card number with the facial features associated with the pre-stored whitelist. If the card number is valid but the facial features do not match, the number of consecutive recognition failures is accumulated and the gate-breaking risk assessment process is triggered. At the same time, the abnormal card number is marked as a suspicious document and an audit log is generated.

[0020] Adopting the above technical solution: This solution can solve the problem of traditional IC card verification that someone can pass through by stealing other people's IC cards and cannot be associated with the person himself. This solution can prevent theft by comparing the IC card number with the facial features of the pre-stored whitelist.

[0021] Further preferably, the method of preserving the gate-breaking video evidence is to intercept the high-definition video stream from 10 seconds before to 5 seconds after the incident, and append the gate acceleration data, infrared crossing time and face matching calculation results as metadata to the video file, and at the same time bind the video clips with the associated attendance records and personnel identity information and store them in the security audit database.

[0022] Adopting the above technical solution: This solution can solve the problem of insufficient traditional food evidence preservation methods. Traditional recorded video clips are not associated with acceleration, recognition records and other data, making it difficult to restore the full picture of the event; this solution only saves 15 seconds of video before and after the event, reducing storage costs; at the same time, additional data such as acceleration and face matching degree are added to form a complete chain of evidence.

[0023] Further preferably, the encrypted upload uses the national secret SM4 algorithm to encrypt the attendance time, work number and facial feature vector, and performs batch data upload every 30 minutes. When the network is interrupted, local storage is automatically enabled and the retransmission flag is marked. After the network is restored, the encrypted data packets that have not been uploaded are transmitted first.

[0024] Adopting the above technical solution: The above solution can solve the problem that the traditional data uploading method easily causes the attendance data to contain sensitive information and be easily intercepted and leaked. This solution can encrypt sensitive information such as facial features and work numbers by adopting the SM4 algorithm.

[0025] Further preferably, the method for determining the processing priority is to divide the risk value of breaking through the gate into high-risk, medium-risk and low-risk levels. High-risk incidents must be handled on-site by security personnel within 10 minutes and reported to the management department. Medium-risk incidents generate standardized work orders and are automatically distributed to relevant responsible persons for processing within a time limit. Low-risk incidents are recorded in the system log and a weekly safety report is generated.

[0026] Adopting the above technical solution: This solution can solve the problem of low efficiency of traditional event handling mechanism. Since all alarms are handled in a unified manner, it is easy to cause corresponding delays in high-risk events. This application divides the risk level (R) into high, medium and low risk levels, and gives priority to handling high-risk events; medium-risk events automatically generate work orders and distribute them to the responsible person, shortening the processing cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0028] Figure 1 This is the block diagram of the facial recognition attendance supervision and management system for smart construction sites; Figure 2 This is a flow chart of the application management method. DETAILED DESCRIPTION

[0029] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0030] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, operations, elements, components and / or groups thereof.

[0031] See also Figure 1 and Figure 2 For example, traditional construction site attendance systems usually rely on a single sensor and have the following problems in complex construction site environments: Poor environmental adaptability: Face recognition has a high failure rate in strong light, backlight, or low light conditions; Single identity verification: Only supports card swiping or face recognition, and cannot handle ID fraud, fake cards, or photo attacks; Difficulty in determining malicious gate crossings: Only infrared sensors are used to detect cross-border behavior, and it is impossible to distinguish between normal passage and violent gate crossings; Data isolation: Attendance data is separated from systems such as salary accounting and safety training, resulting in low management efficiency. Based on this, the embodiment of the present application discloses a smart construction site face recognition attendance supervision and management system, including: Multimodal data acquisition module: including binocular liveness detection camera, ID card reader, IC / ID card reader, infrared monitoring sensor and gate acceleration sensor; Dynamic weight calculation module: used to dynamically adjust the weight parameters of face matching based on liveness detection results, person-document comparison data, and ambient lighting conditions; Malicious gate-breaking determination module: Generates a gate-breaking risk value based on gate acceleration sensor data, infrared monitoring of cross-border behavior, and the number of consecutive facial recognition failures; Attendance data optimization module: Modify the confidence level of attendance records through historical recognition success rate and time series analysis; Data linkage management module: used to upload attendance data to the Ministry of Housing and Urban-Rural Development platform in real time, and to link with the construction site LED display screen, mobile terminal and alarm device.

[0032] It is worth mentioning that this solution has the following beneficial effects by integrating binocular cameras, ID card readers, accelerometers and other hardware through a multimodal data acquisition module: the wide dynamic camera adapts to strong / low light scenes, and combines with liveness detection to prevent photo attacks; supports person-document matching, person-card matching and dynamic weight adjustment to block identity fraud; integrates accelerometer data with infrared cross-border time to distinguish between normal passage and malicious intrusion; and links with the housing and construction platform, enterprise ERP and training system in real time to achieve closed-loop management of the entire process from entry to salary.

[0033] For example, traditional face recognition systems have fixed weight distribution under complex lighting conditions, which can easily lead to the following defects: High misidentification rate: Facial feature extraction is difficult in backlit environments, resulting in legitimate individuals being mistakenly identified as strangers. Liveness detection and ID verification are disconnected: The weights of the two are not dynamically adjusted based on the actual scenario, resulting in liveness detection failure under extreme lighting conditions. Redundant secondary verification frequently triggers card swiping due to a fixed threshold, reducing traffic efficiency.

[0034] The face matching weight adjustment formula in the dynamic weight calculation module is: , in, is the final matching weight value, is the confidence score of binocular liveness detection, To compare the similarity of human evidence, is the ambient light intensity and the unit is lux, =500 is the illumination reference value, =0.6, =0.3, =0.1 is the weight coefficient, =0.01 is the illumination compensation coefficient; when If the authentication is successful, it is considered as legal. Otherwise, secondary verification is triggered.

[0035] In the above scheme, by setting =0.6, giving the highest weight to liveness detection to prevent attackers from using fake biometrics, setting =0.3 can ensure that the legitimacy of the province is further verified under the premise of credible liveness detection, through the sigmoid function Dynamically adjust the lighting influence weight. In bright light, or When in the low light state, the above function value approaches 0, thereby reducing the interference of light. When , the function value is 0.5, retaining some compensation.

[0036] The comprehensive weight of the above scheme When , it is determined to be legal to pass. The threshold is optimized through experimental data. =0.9, =0.8 and normal lighting, that is, when L=500, W=0.6×0.9+0.3×0.8+0.1×0.5=0.85, which can ensure a balance between verification pass rate and security in typical scenarios.

[0037] It is worth mentioning that the dynamic weight formula of this solution optimizes the verification logic through the following mechanism, which can achieve light adaptation and introduce light intensity compensation items , which can automatically reduce the weight of light impact under strong light; setting the basic weight of liveness detection α=0.6 can ensure that non-live attacks are intercepted; when the comprehensive weight W≥0.85, it is directly released, reducing the frequency of secondary verification and improving traffic efficiency.

[0038] For example, traditional gate systems rely on a single sensor to determine gate-breaking behavior, which has the following vulnerabilities: Infrared sensors are easily blocked or triggered by mistake, misjudging normal rapid passage as breaking through the gate; the acceleration suddenly increases when the gate is violently broken, but it is not associated with the number of recognition failures, resulting in delayed judgment; only the time of breaking through the gate is recorded, and key data such as acceleration and face matching are lacking, making it difficult to trace back afterwards.

[0039] The calculation formula of the gate-breaking risk of the malicious gate-breaking determination module is: , in, is the gate-breaking risk value, The peak acceleration detected by the gate acceleration sensor, in units of , is the acceleration threshold, is the number of consecutive face recognition failures, The crossing time of infrared monitoring, in seconds, =2S is the safe passage time, =0.5, =0.3, =0.2 is the risk factor; when The sound and light alarm is triggered and the gate is locked.

[0040] This solution can achieve the fusion of multi-dimensional parameters, such as It can quantify the severity of the external impact on the gate. Set the acceleration threshold Based on the statistical value of the swing acceleration of the gate when the human body passes normally, the violent gate-breaking acceleration can reach 8-10m / s 2 .

[0041] and It can reflect the possibility of intentional identity fraud. A normal user may fail occasionally, but if the failure occurs more than 3 times in a row, , which significantly increases the risk value. Ability to measure abnormal traffic behavior. Safety time =2S based on normal passage time (1.5-2 seconds), when people are stranded or rush in quickly 1S triggers risks at any time.

[0042] In the above formula =0.5 gives the highest weight to physical impact, as violent gate-breaking directly threatens equipment safety; =0.3 can correlate identity fraud risks and prevent multiple attack attempts; =0.2 can supplement the time dimension anomalies and avoid underreporting.

[0043] when When the sound and light alarm is triggered and the gate is locked. For example: That is, the impact is significant, , failed to break through the gate twice, =1.5S, then R=0.5×(8 / 5)+0.3×2+0.2×(1.5 / 2)=0.8+0.6+0.15=1.55, exceeding the threshold and triggering an alarm.

[0044] This scheme designs a risk value formula for gate-breaking Through multi-dimensional data fusion, acceleration threshold The design can effectively distinguish between normal passage and violent gate-breaking; the cumulative number of recognition failures Identify intentional identity fraud; and this design can save the video streams before and after the event and add acceleration and face matching metadata to support post-event tracing and responsibility determination.

[0045] For example, traditional attendance systems rely on single recognition results and have the following defects: ‌Occasional equipment failures, temporary camera damage, or network outages can lead to single recognition failures, resulting in false positives for absences; ‌ Historical data is not utilized, and the reliability of current records is not corrected through time series analysis; Manual review is costly and requires administrators to check abnormal records one by one, which is inefficient. The confidence correction formula of the attendance data optimization module is: , in, The confidence level of the corrected attendance ranges from 0 to 1. For the The success flag of the historical recognition is 1 for success and 0 for failure. is the current time, =0,1 is the time attenuation coefficient; when It is marked as abnormal attendance and manual review is initiated.

[0046] The above time decay model, exponential weight distribution Assign decay weights to historical data, where =0,1 controls the decay rate, for example: the weight of data 24 hours ago is , 1 hour ago , ensuring that recent data dominates the confidence calculation. The above formula is normalized, and the denominator sums all weights to avoid bias caused by differences in data volume.

[0047] In the above formula, anomaly detection mechanism is designed. It is marked as abnormal attendance and manual review is initiated.

[0048] Assume that a worker has successfully completed 9 of his 10 attendance checks recently. , the most recent failure ; The time interval is 1 hour, , calculate the molecule

[0049] Denominator

[0050] If it fails three times in a row, Trigger review.

[0051] It's worth noting that the confidence correction formula in the above solution can optimize attendance data in the following ways: attenuating weight distribution so that recent successful recognition records contribute more to the current confidence level, reducing interference from earlier data; automatically marking anomalies, triggering manual review when C < 0.7, and reducing misjudgments due to occasional equipment failures; and combining time series analysis to prevent a single failed record from excessively influencing attendance results.

[0052] For example, traditional construction site management systems have data silos: Attendance data needs to be manually exported and uploaded to the housing and construction platform, which cannot meet the real-time supervision requirements; gate-breaking incidents only have local sound and light alarms and are not synchronized to mobile terminals, resulting in delayed emergency responses; attendance data is independent of salary accounting and safety training, resulting in untrained personnel entering the site illegally. The data linkage management module includes a data interface protocol that connects to the Ministry of Housing and Urban-Rural Development platform, which uploads personnel names, types of work, attendance times, and gate-breaking incidents in real time; dynamically displays the number of people present, high-risk behavior statistics, and abnormal temperature alarm information on the LED display screen; pushes alarm notifications containing the location of malicious gate-breaking incidents, on-site video clips, and handling suggestions to security personnel through the mobile terminal APP; at the same time, the abnormal attendance data is synchronized to the enterprise ERP system for salary accounting and contract management, and is linked to the construction site real-name training platform to automatically prohibit personnel who have not completed safety training from passing. This solution uses the data linkage module to achieve full-process management through the following mechanisms.

[0053] It can realize real-time data reporting and automatically encrypt and upload personnel access data to the housing and construction platform to meet government regulatory requirements; multi-terminal collaboration, LED screens display on-site headcount, mobile apps push alarm details, and improve emergency response speed; business closed loop, synchronizing abnormal attendance to the ERP system to block problematic salary payments, and linking with the training platform to prohibit untrained personnel from passing.

[0054] For example, traditional construction site attendance methods have fragmented processes: Single verification steps: Relying solely on facial recognition or card swiping, they are unable to handle complex scenarios; Delayed gate-breaking detection: Video footage is retrieved only after the incident occurs, making real-time interception impossible; Inefficient data management: Attendance records need to be manually exported and cleaned, which is prone to errors.

[0055] Based on this, the present application provides a management method, which is applied to a smart construction site face recognition supervision and management system as described in any one of the above, including: S1: Capture facial images through binocular cameras and simultaneously collect ambient light intensity. It also reads ID card or IC card information and matches it with a pre-stored whitelist, and monitors gate acceleration and infrared cross-border signals in real time. S2: Calculates the liveness detection score and the person-document comparison similarity, adjusts the weight parameters according to the ambient light intensity, and uses a dynamic weight formula to determine the legitimacy of the identity. If the verification passes, access is authorized and attendance is recorded. Otherwise, IC card-assisted verification is initiated. S3: When it is detected that the gate acceleration exceeds the threshold or the infrared crossing time is lower than the safety value, the facial area image is extracted and the gate-breaking risk value is calculated. If the risk value exceeds the threshold, an alarm is triggered and the gate-breaking video evidence is saved; S4: Based on historical recognition records and the time decay model, the confidence level of attendance data is corrected. Low-confidence data is marked as abnormal and the administrator is notified for review. The corrected attendance data is encrypted and uploaded to the Ministry of Housing and Urban-Rural Development platform and the enterprise management system. S5: Display the real-time number of people present and gate-breaking event statistics on the LED screen, push event details and processing priorities to the security personnel's mobile terminal, and lock the access rights of the team to which the malicious gate-breaking person belongs.

[0056] It is worth mentioning that this solution achieves the following improvements through process integration: Multimodal verification: simultaneous collection of facial, ID, and ambient lighting data, and dynamic adjustment of verification strategies; Real-time risk interception: triggering alarms and preserving evidence the moment a gate-breaking behavior occurs, avoiding difficulties in subsequent accountability; Automated management: Automating the entire process of confidence level correction, encrypted upload, and multi-terminal alarms to reduce manual intervention.

[0057] For example, traditional IC card authentication has vulnerabilities: ‌Risk of impersonation‌: using someone else's IC card to gain access without being able to link it to the actual cardholder's identity; ‌Log loss‌: abnormal card number usage records are not recorded, making it difficult to trace responsibility; ‌Verification Disjunction‌: IC card and facial recognition are performed independently, unable to complement each other to enhance security. Therefore, IC card-assisted verification involves comparing the IC card number with facial features associated with a pre-stored whitelist. If the card number is valid but the facial features do not match, the number of consecutive recognition failures is accumulated and the gate-breaking risk assessment process is triggered. The anomalous card number is also flagged as suspicious and an audit log is generated.

[0058] It is worth mentioning that this solution improves security through the IC card auxiliary authentication mechanism in the following ways: ‌Person-card binding‌: Compare IC card numbers with facial features in a pre-stored whitelist to prevent theft; ‌Audit trace‌: Mark abnormal card numbers and generate logs to support subsequent tracing and permission revocation; ‌Risk linkage‌: The cumulative number of recognition failures triggers a gate-breaking judgment to block continuous attacks.

[0059] Traditional video evidence preservation methods suffer from the following shortcomings: "Information isolation": Video clips are not linked to data such as acceleration and identification records, making it difficult to fully reconstruct the entire incident; "Storage redundancy": Saving all-day video footage consumes significant storage space, making it inefficient to retrieve key events; and "Data tampering vulnerability": Unencrypted video files can be maliciously modified, compromising the effectiveness of judicial evidence. To address these issues, the proposed gate-breaking video evidence preservation method captures a high-definition video stream from 10 seconds before to 5 seconds after the incident. The video file is then supplemented with metadata including gate acceleration data, infrared boundary crossing time, and facial match calculation results. The video clips are then linked to associated attendance records and personnel identity information and stored in a secure audit database. Notably, this video evidence preservation mechanism ensures evidence integrity through the following improvements: "Critical segment capture": Only the 15 seconds of video before and after the incident are saved, reducing storage costs; "Metadata binding": Acceleration and facial match data are appended to form a complete chain of evidence; and "Secure storage": Encrypted data is stored in an audit database to prevent tampering and support judicial access.

[0060] For example, traditional data uploading methods have the following risks: plaintext transmission: attendance data contains sensitive information and can be easily intercepted and leaked; network dependence: data is lost and cannot be recovered when the network is disconnected; transmission congestion: batch uploading during peak hours causes network delays, affecting real-time performance.

[0061] The encrypted upload uses the national secret SM4 algorithm to encrypt the attendance time, work number and facial feature vector, and performs batch upload of data every 30 minutes. When the network is interrupted, local storage is automatically enabled and the retransmission flag is marked. After the network is restored, the encrypted data packets that have not been uploaded are transmitted first.

[0062] This solution uses an encrypted upload mechanism to ensure data security and reliability in the following ways: ‌National Secret Algorithm Encryption‌: Uses the SM4 algorithm to encrypt sensitive information such as facial features and work numbers; ‌Breakpoint Resume‌: Locally stores encrypted data packets when the network is interrupted, and prioritizes retransmission after recovery; ‌Batch Upload‌: Batch uploads are performed every 30 minutes to balance real-time performance and network load.

[0063] For example, the traditional incident handling mechanism is inefficient: ‌Priority confusion‌: All alerts are handled uniformly, resulting in delayed responses to high-risk incidents; ‌Responsibilities are unclear‌: Work orders are not automatically dispatched, relying on manual task assignment; ‌ Lack of reporting‌: Security reports are not generated regularly, making it difficult to evaluate the effectiveness of system operation. Based on this, the method for determining the processing priority is to divide the risk value of the gate into high, medium and low risk levels. High-risk incidents must be handled on-site by security personnel within 10 minutes and reported to the management department. Medium-risk incidents generate standardized work orders and are automatically distributed to the relevant responsible persons for processing within a time limit. Low-risk incidents are recorded in the system log and a weekly security report is generated. It is worth mentioning that the hierarchical processing mechanism of the above scheme optimizes the management process in the following ways: ‌ Risk grading‌: Divide the gate risk value (RR) into high, medium and low risk levels, and give priority to high-risk incidents; ‌ Work order automation‌: Medium-risk incidents automatically generate work orders and distribute them to the responsible persons, shortening the processing cycle; ‌ Periodic reporting‌: Record low-risk incidents and generate weekly reports to assist managers in optimizing security strategies.

[0064] Unless otherwise specified, the device components involved in the above embodiments are all conventional device components, and the connection methods and control methods involved are all conventional connection methods and control methods unless otherwise specified.

[0065] The present invention has been described in detail above with reference to the embodiments. However, those skilled in the art will appreciate that, without departing from the spirit of the present invention, the specific parameters in the above embodiments may be modified to form multiple specific embodiments, which are all within the common variation range of the present invention and will not be described in detail here.

Claims

1. A smart construction site face recognition attendance supervision and management system, characterized by: include: Multimodal data acquisition module: including binocular liveness detection camera, ID card reader, IC / ID card reader, infrared monitoring sensor and gate acceleration sensor; Dynamic weight calculation module: used to dynamically adjust the weight parameters of face matching based on liveness detection results, person-document comparison data, and ambient lighting conditions; Malicious gate-breaking determination module: Generates a gate-breaking risk value based on gate acceleration sensor data, infrared monitoring of cross-border behavior, and the number of consecutive facial recognition failures; Attendance data optimization module: Modify the confidence level of attendance records through historical recognition success rate and time series analysis; Data linkage management module: used to upload attendance data to the Ministry of Housing and Urban-Rural Development platform in real time, and to link with the construction site LED display screen, mobile terminal and alarm device.

2. A smart construction site face recognition attendance supervision and management system according to claim 1, characterized in that: The face matching weight adjustment formula in the dynamic weight calculation module is: , in, is the final matching weight value, is the confidence score of binocular liveness detection, To compare the similarity of human evidence, is the ambient light intensity and the unit is lux, =500 is the illumination reference value, =0.6, =0.3, =0.1 is the weight coefficient, =0.01 is the illumination compensation coefficient; when If the authentication is successful, it is considered as legal. Otherwise, secondary verification is triggered.

3. A smart construction site face recognition attendance supervision and management system according to claim 1, characterized in that: The calculation formula of the gate-breaking risk of the malicious gate-breaking determination module is: ， in, is the gate-breaking risk value, The peak acceleration detected by the gate acceleration sensor, in units of , is the acceleration threshold, is the number of consecutive face recognition failures, The crossing time of infrared monitoring, in seconds, =2S is the safe passage time, =0.5, =0.3, =0.2 is the risk factor; when The sound and light alarm is triggered and the gate is locked.

4. A smart construction site face recognition attendance supervision and management system according to claim 1, characterized in that: The confidence correction formula of the attendance data optimization module is: , in, The confidence level of the corrected attendance ranges from 0 to 1. For the The success flag of the historical recognition is 1 for success and 0 for failure. is the current time, =0,1 is the time attenuation coefficient; when It is marked as abnormal attendance and manual review is initiated.

5. The smart construction site face recognition attendance supervision and management system according to claim 1 is characterized in that: The data linkage management module includes a data interface protocol connected to the Ministry of Housing and Urban-Rural Development platform, which uploads personnel names, types of work, attendance times and gate-breaking incidents in real time; dynamically displays the number of people present, high-risk behavior statistics and temperature abnormality alarm information on the LED display screen; pushes alarm notifications containing the location of malicious gate-breaking incidents, on-site video clips and handling suggestions to security personnel through the mobile terminal APP; at the same time, the abnormal attendance data is synchronized to the enterprise ERP system for salary accounting and contract management, and is linked to the construction site real-name training platform to automatically prohibit personnel who have not completed safety training from passing.

6. A management method, applied to a smart construction site face recognition supervision and management system as described in any one of claims 1 to 5, characterized in that: include: S1: Capture facial images through binocular cameras and simultaneously collect ambient light intensity. It also reads ID card or IC card information and matches it with a pre-stored whitelist, and monitors gate acceleration and infrared cross-border signals in real time. S2: Calculates the liveness detection score and the person-document comparison similarity, adjusts the weight parameters according to the ambient light intensity, and uses a dynamic weight formula to determine the legitimacy of the identity. If the verification passes, access is authorized and attendance is recorded. Otherwise, IC card-assisted verification is initiated. S3: When it is detected that the gate acceleration exceeds the threshold or the infrared crossing time is lower than the safety value, the facial area image is extracted and the gate-breaking risk value is calculated. If the risk value exceeds the threshold, an alarm is triggered and the gate-breaking video evidence is saved; S4: Based on historical recognition records and the time decay model, the confidence level of attendance data is corrected. Low-confidence data is marked as abnormal and the administrator is notified for review. The corrected attendance data is encrypted and uploaded to the Ministry of Housing and Urban-Rural Development platform and the enterprise management system. S5: Display the real-time number of people present and gate-breaking event statistics on the LED screen, push event details and processing priorities to the security personnel's mobile terminal, and lock the access rights of the team to which the malicious gate-breaking person belongs.

7. The management method according to claim 6, characterized in that: The IC card auxiliary verification includes comparing the IC card number with the facial features associated with the pre-stored whitelist. If the card number is valid but the facial features do not match, the number of consecutive recognition failures is accumulated and the gate-breaking risk assessment process is triggered. At the same time, the abnormal card number is marked as a suspicious document and an audit log is generated.

8. The management method according to claim 6, characterized in that: The method of preserving the gate-breaking video evidence is to intercept the high-definition video stream from 10 seconds before to 5 seconds after the incident, and attach the gate acceleration data, infrared crossing time and face matching calculation results as metadata to the video file. At the same time, the video clips are bound to the associated attendance records and personnel identity information and stored in the security audit database.

9. The management method according to claim 6, characterized in that: The encrypted upload uses the national secret SM4 algorithm to encrypt the attendance time, work number and facial feature vector, and performs batch upload of data every 30 minutes. When the network is interrupted, local storage is automatically enabled and the retransmission flag is marked. After the network is restored, the encrypted data packets that have not been uploaded are transmitted first.

10. The management method according to claim 6, characterized in that: The method for determining the processing priority is to divide the risk value of breaking through the gate into high, medium and low risk levels. High-risk incidents must be handled on-site by security personnel within 10 minutes and reported to the management department. Medium-risk incidents generate standardized work orders and are automatically distributed to relevant responsible persons for processing within a time limit. Low-risk incidents are recorded in the system log and a weekly safety report is generated.

Images