Multi-mode attendance checking device system for key personnel

Abstract

The application discloses a key post personnel multi-mode verification attendance equipment system and belongs to the technical field of identity recognition and authentication. It comprises an information acquisition module, a face comparison module, a living body detection module, an attendance management module, an image storage module, a data encryption module and a platform management module. The face comparison module compares the identity after extracting the synchronous frame and the features of binocular video streams. The living body detection module detects the risk level of binocular living bodies. The image storage module generates a recording strategy according to the risk level, collects images and calculates hash values. The data encryption module divides the normal encryption area and the abnormal encryption area to perform partition encryption storage. The application drives the recording strategy and partition encryption by binocular living body detection and risk level, aligns and verifies the time stamp of attendance data and image evidence, realizes differential encryption, and simultaneously realizes multiple hash comparison and file integrity verification during file retrieval.

Description

Technical Field

[0001] This invention belongs to the field of identity recognition and authentication technology, specifically, it is a multimodal verification and attendance system for key personnel. Background Technology

[0002] With the advancement of real-name registration, facial recognition technology has become the mainstream method for personnel attendance tracking in key locations such as construction sites and data centers. Existing systems typically use a monocular camera to capture facial images, verify identity through facial comparison, and upload attendance records to a management platform for statistical analysis. Some systems also incorporate liveness detection technology to prevent attacks using photos or videos. Regarding data security, existing solutions mostly employ transport layer encryption or simple hash verification, and centrally store attendance records and image data. However, existing technologies still have the following shortcomings: First, facial images captured by monocular cameras lack depth information, making them susceptible to deception and resulting in insufficient reliability for identity verification. Second, there is a lack of temporal correlation and integrity verification between attendance records and video evidence, making the data easily tampered with or forged afterward, failing to meet the requirements for evidence in labor disputes and regulatory audits. Third, existing encryption schemes typically employ a unified key and a single storage strategy, failing to implement differentiated protection based on personnel risk levels, resulting in highly sensitive data being stored alongside routine data; once the key is leaked, all data is exposed. Fourth, attendance data and video materials are managed separately, requiring separate verification upon retrieval; the verification process is cumbersome and lacks a closed-loop verification mechanism, making it impossible to ensure the authenticity and integrity of the returned data. These problems result in significant shortcomings in existing attendance systems regarding security, data trustworthiness, and compliant evidence storage. Summary of the Invention

[0003] This invention aims to provide a multimodal verification and attendance system for key personnel to solve the above-mentioned technical problems. The technical solution adopted by this invention is as follows: A multimodal verification and attendance system for key personnel, including: The information acquisition module is used to acquire binocular video streams and on-duty information; The face comparison module is used to extract keyframes from the binocular video stream, extract features from the keyframes to obtain face features, and compare the face features with the registered face database to obtain the comparison result. The liveness detection module is used to extract synchronous frames from the binocular video stream to obtain binocular images, and to perform binocular liveness detection on the binocular images to obtain the risk level; The attendance management module is used to generate attendance records and stage identifiers based on on-duty information, determine anomalies in attendance records, and issue real-time alerts if there are abnormal attendance statuses. The image evidence module is used to perform hash calculation on keyframes to obtain the first hash value, generate basic recording based on stage identifiers, generate recording strategy based on risk level, acquire images from binocular video streams based on basic recording and recording strategy to obtain image evidence, perform hash calculation on image evidence to obtain the second hash value, and align keyframes, comparison results, image evidence and attendance records with timestamps to obtain verification files. The data encryption module is used to calculate the digest of the verification file to obtain the file verification value, write the file verification value, the first hash value and the second hash value into the attendance record, encrypt the written attendance record to obtain encrypted attendance, and store and encrypt the facial features, verification file, written attendance record and abnormal warning information to obtain encrypted storage data. The encrypted storage area is divided into a regular encryption area and an abnormal encryption area according to the risk level, and the verification file is partitioned and encrypted according to the risk level. The platform management module is used to decrypt encrypted attendance records and generate reports, receive file retrieval requests from the requesting end, retrieve and decrypt files from the regular encrypted area and the abnormal encrypted area according to the file retrieval request to obtain plaintext verification files, unpack and verify the plaintext verification files, compare the verification calculation results with the first hash value, the second hash value and the file verification value respectively, and send the plaintext verification files back to the requesting end when the comparison is successful. In the above scheme, on-duty information is obtained through binocular video stream, comparison results are obtained through face comparison, and risk level is obtained through liveness detection. Attendance management generates attendance records and stage identifiers and performs anomaly judgment. Image evidence risk level generates recording strategy and collects image evidence. First hash value and second hash value are calculated. Data encryption writes the file verification value and the two hash values ​​into the attendance record and divides it into regular encryption area and abnormal encryption area for partitioned encryption storage. Platform management decrypts the encrypted attendance to generate reports. After receiving file retrieval request, it unpacks and verifies the plaintext to verify the file, so that the attendance data and image evidence timestamps are aligned and integrity is verified. Encryption is differentiated according to risk level.

[0004] Furthermore, the liveness detection module includes: The frame extraction submodule is used to extract synchronous frames from the stereo video stream to obtain stereo images. The disparity analysis submodule is used to calculate the disparity of binocular images to obtain a depth map; The attack detection submodule is used to identify and detect cheating attacks on the depth map, obtain the attack confidence score, and generate a risk level based on the attack confidence score. In the above scheme, a stereo image is obtained by extracting synchronous frames from the stereo video stream, a depth map is obtained by calculating the disparity of the stereo image, and a cheating attack is detected and detected on the depth map to obtain the attack confidence score. A risk level is generated based on the attack confidence score, enabling liveness detection to identify cheating attacks based on depth information. This prevents non-real faces such as photos and videos from passing identity verification, reduces the possibility of non-real faces passing identity verification, and ensures the authenticity of identity verification for key personnel.

[0005] Furthermore, the image evidence storage module includes: The frame hashing submodule is used to perform frame preprocessing on keyframes to obtain frame features, and to perform hash calculation on the frame features to obtain the first hash value; The recording trigger submodule is used to perform stage parsing on the stage identifier to obtain the trigger signal, and generate basic recording based on the trigger signal; The strategy mapping submodule is used to quantify the risk level to obtain the recording duration, and generate a recording strategy based on the recording duration and the basic recording. The image acquisition submodule is used to acquire images from the binocular video stream based on the basic recording and recording strategy, and obtain image evidence through image buffering. The image hashing submodule is used to segment image evidence to obtain segment features, and then perform hash calculation on the segment features to obtain a second hash value. The document encapsulation submodule is used to extract and align timestamps from keyframes, comparison results, image evidence, and attendance records, and then encapsulate the data to obtain a verification document. In the above scheme, the first hash value is obtained by preprocessing and hashing keyframes; the trigger signal is obtained by parsing the stage identifier and generating basic recording; the recording duration is obtained by quantifying the risk level and generating a recording strategy; image evidence is obtained by acquiring images based on the basic recording and recording strategy and caching the images; the second hash value is obtained by segmenting and hashing the image evidence; and the timestamps are extracted, aligned, and encapsulated from keyframes, comparison results, image evidence, and attendance records to obtain a verification document. This enables image evidence storage to have both frame-level and segment-level dual hash verification, and binds attendance records to image evidence through timestamp alignment.

[0006] Furthermore, the policy mapping submodule includes: The risk mapping unit is used to extract and quantify the risk level and map it to multiple risk levels. The time period mapping unit is used to map the risk level to obtain the recording time period, decompose the basic recording to obtain the basic time period, align the recording time period with the basic time period, and extract the difference from the time period alignment result to obtain the recording duration. The parameter mapping unit is used to map the risk level to obtain frame rate parameters and resolution parameters, perform parameter verification on the frame rate parameters and resolution parameters, and perform parameter fusion on the parameter verification results to obtain a verification parameter group. The strategy synthesis unit is used to synthesize a recording strategy from the recording duration, basic recording, risk level, and verification parameter set. In the above scheme, the risk level is numerically extracted and quantified, then mapped to multiple risk levels. The risk levels are then mapped to time periods to obtain recording time periods, which are then aligned with the basic time periods, and the difference is extracted to obtain the recording duration. The risk levels are then mapped to parameters to obtain frame rate and resolution parameters, which are then verified and fused to obtain a verification parameter set. Finally, the recording duration, basic recording, risk level, and verification parameter set are synthesized to obtain the recording strategy, enabling the recording strategy to configure the recording duration, frame rate, and resolution differently based on the risk level.

[0007] Furthermore, the strategy synthesis unit includes: The basic analysis subunit is used to analyze the basic recording to obtain the basic time period, basic frame rate and basic resolution, and uses the basic time period, basic frame rate and basic resolution as the reference parameters; The duration overlay subunit is used to overlay the recording duration with the base time period to obtain the total time period; The level modulation subunit is used to obtain enhancement parameters by increasing the frame rate and resolution based on the risk level, the verification parameter group and the total time period; The strategy output subunit is used to encapsulate the baseline parameters, enhancement parameters, and total time period to obtain the recording strategy. In the above scheme, the baseline parameters are obtained by parsing the basic recording, the total time period is obtained by superimposing the recording duration with the baseline time period, the frame rate and resolution are adjusted according to the risk level, the verification parameter group, and the total time period to obtain the enhancement parameters, and the recording strategy is obtained by encapsulating the baseline parameters, enhancement parameters, and total time period. This allows the recording strategy to perform duration superposition and parameter adjustment based on the risk level on the basis of the baseline parameters, realizing the linkage configuration of recording duration and image acquisition parameters.

[0008] Furthermore, the data encryption module includes: The verification submodule is used to decompile the verification archive into a structured form to obtain the verification archive body, segment the verification archive body into archive segments and extract summary features, and perform summary calculation and summary encryption on the summary features to obtain the archive verification value. The verification and write submodule is used to perform verification aggregation on the file verification value, the first hash value, the second hash value and the attendance record. Through field mapping and write calculation, the file verification value, the first hash value and the second hash value are written to the attendance record to obtain a self-locking structure. The transmission encryption submodule is used to generate a session key, and to encapsulate and encrypt the self-locking structure according to the session key to obtain encrypted attendance. The storage encryption submodule is used to aggregate and group plaintext data such as facial features, verification files, written attendance records and abnormal warning information, and encrypt the data according to the storage algorithm to obtain encrypted storage data. The storage allocation submodule is used to divide the encrypted storage area into a regular encrypted area and an abnormal encrypted area according to the risk level, and to store the encrypted storage data in the regular encrypted area. The off-line storage submodule is used for risk level resolution and encryption selection. Verification files are encrypted according to the chosen encryption option to obtain either regular encryption or enhanced encryption. Regularly encrypted files are stored in the regular encryption area, while enhanced encrypted files are stored in the anomalous encryption area. In the above scheme, the verification file structure is unpacked and segmented, and digest features are extracted. Digest calculation and encryption yield the file verification value. The file verification value, two hash values, and attendance record verification are aggregated, and field mapping and writing calculations are performed to obtain a self-locking structure. A session key is generated to encrypt the transmission of the self-locking structure, resulting in encrypted attendance. Facial features, verification files, attendance records, and anomalous warning information are aggregated, grouped into plaintext, and stored encrypted according to the storage algorithm to obtain encrypted storage data. The encrypted storage area is divided into a regular encryption area and an anomalous encryption area according to the risk level, enabling attendance records to have a self-locking structure that embeds and binds the file verification value and hash value. Differentiated encryption areas are also defined according to the risk level.

[0009] Furthermore, the offloading storage submodule includes: The risk level analysis unit is used to determine the risk level and obtain a normal level or an abnormal level. The encryption selection unit is used to obtain a conventional algorithm by indexing a conventional algorithm according to the conventional level, to obtain a conventional key by deriving a conventional key from the device master key, and to encrypt the verification file using the conventional key to obtain conventional encryption; and to obtain an enhanced algorithm by indexing an enhanced algorithm according to the anomaly level, to obtain an enhanced key by deriving an enhanced key from the device master key, and to encrypt the verification file using the enhanced key to obtain enhanced encryption. The path selection unit is used to perform regular path indexing for regular encryption to obtain a regular storage path, and to perform abnormal path indexing for enhanced encryption to obtain an abnormal storage path. The archive unit is used to encapsulate regular encryption into ciphertext, generate integrity tags for the ciphertext, and store the ciphertext and integrity tags in the regular encryption area according to the regular storage path. Similarly, it encapsulates enhanced encryption into ciphertext, generates integrity tags for the enhanced ciphertext, and stores the enhanced ciphertext and integrity tags in the anomalous encryption area according to the anomalous storage path. In this scheme, a risk level is determined to obtain a regular or anomalous risk level. Based on the regular risk level, a regular key is derived from a regular algorithm index and a regular key, and then encrypted to obtain regular encryption. Based on the anomalous risk level, an enhanced key is derived from an enhanced algorithm index and an enhanced key, and then encrypted to obtain enhanced encryption. Path indexes are applied to both regular and enhanced encryption to obtain regular and anomalous storage paths, respectively. After ciphertext encapsulation and integrity tag generation, the data is stored in the regular encryption area and the anomalous encryption area, respectively. This ensures that regular and anomalous data are encrypted using different algorithms and keys and stored in different paths.

[0010] Furthermore, the encryption selection unit includes: The algorithm index subunit is used to index regular algorithms based on the regularity level and determine regular algorithms from the regular algorithm library; and to index enhanced algorithms based on the anomaly level and determine enhanced algorithms from the enhanced algorithm library. The key derivation subunit is used to generate a regular key seed based on the device master key, and to derive a regular key from the regular key seed; and to generate an enhanced key seed based on the device master key, and to derive an enhanced key from the enhanced key seed. The encryption operation subunit is used to perform conventional encryption operations on the verification file using a conventional key and a conventional algorithm to obtain conventional encryption; and to perform enhanced encryption operations on the verification file using an enhanced key and an enhanced algorithm to obtain enhanced encryption. In the above scheme, a conventional algorithm is determined from a conventional algorithm library based on the conventional level, and an enhanced algorithm is determined from an enhanced algorithm library based on the anomaly level. A conventional key seed is generated based on the device master key, and a conventional key is derived from it. An enhanced key seed is generated, and an enhanced key is derived from it. Conventional encryption operations are performed using the conventional algorithm and the conventional key, and enhanced encryption operations are performed using the enhanced algorithm and the enhanced key. This allows conventional encryption and enhanced encryption to achieve differentiated encryption operations based on independent algorithm libraries and derived keys.

[0011] Furthermore, the enhanced encryption operation employs a group round iterative structure, including: The verification file is divided into left and right file blocks. The enhanced key and risk level are fused to obtain the round key. The risk level is mapped to the number of rounds to obtain the number of iteration rounds. The left and right file blocks are iterated according to the iteration round number. In each iteration round, the right file block and the round key are XORed, the XOR result is cyclically shifted, the cyclically shifted result is XORed with the left file block, and the left and right file blocks are swapped. The left and right file blocks from the last iteration round are concatenated to obtain enhanced encryption. In the above scheme, the verification file is divided into left and right file blocks. The enhanced key and risk level are fused to obtain the round key. The risk level is mapped to the round number to obtain the iteration round number. The left and right file blocks are iterated according to the iteration round number. In each round, the right file block and the round key are XORed, cyclically shifted, and XORed, and the left and right file blocks are swapped. The last round is concatenated to obtain enhanced encryption, so that enhanced encryption achieves multi-round obfuscation and diffusion based on the grouped round iteration structure.

[0012] Furthermore, the unpacking and verification calculation of the plaintext verification file involves: unpacking the plaintext verification file to obtain keyframes, image evidence, comparison results, and attendance records; performing frame preprocessing and hash calculation on the keyframes to obtain the current frame hash value; performing image segmentation and hash calculation on the image evidence to obtain the current image hash value; and performing digest calculation and digest encryption on the keyframes, image evidence, comparison results, and attendance records to obtain the current file verification value. In the above scheme, by unpacking the plaintext verification file to obtain keyframes, image evidence, comparison results, and attendance records; performing frame preprocessing and hash calculation on the keyframes to obtain the current frame hash value; performing image segmentation and hash calculation on the image evidence to obtain the current image hash value; and performing digest calculation and digest encryption on the keyframes, image evidence, comparison results, and attendance records to obtain the current file verification value, the verification calculation after file retrieval has three verification bases: the current frame hash value, the current image hash value, and the current file verification value.

[0013] In summary, this invention prevents non-real faces such as photos and videos from passing identity verification through binocular liveness detection and cheating attack identification; through risk level-driven recording strategy generation and partitioned encrypted storage, attendance data and image evidence are matched with differentiated recording parameters and encryption strength according to risk level, and have timestamp alignment and integrity verification, realizing multiple hash comparisons and file integrity verification when retrieving files. Attached Figure Description

[0014] The present invention can be further illustrated by the non-limiting embodiments given in the accompanying drawings; Figure 1 This is an overall framework diagram of the multimodal verification and attendance equipment system for key personnel of the present invention; Figure 2 This is a framework diagram of the liveness detection module of the present invention; Figure 3 This is a framework diagram of the image evidence storage module of the present invention; Figure 4 This is a framework diagram of the data encryption module of the present invention. Detailed Implementation

[0015] To enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0016] This embodiment uses a housing construction project as an example. The tower crane operator, construction hoist operator, and site safety officer positions within the project are key positions. According to the real-name management requirements for construction workers, personnel in these positions must undergo identity verification before they can work. This system is deployed at attendance terminals at the verification points and on the project management platform. The attendance terminals are installed at the entrances to the tower crane operator's cab, the construction hoist car, and the safety officer's booth. In this embodiment, binocular video stream refers to the simultaneous video from the left and right eyes of a binocular camera; on-duty information refers to the job schedule and on-duty status information; and the registered face database refers to the facial template database collected during the real-name registration of personnel entering the site.

[0017] like Figures 1 to 4 As shown, the multimodal verification and attendance system for key personnel includes: The information acquisition module is used to acquire binocular video streams and on-duty information. In this embodiment, the attendance terminal is equipped with a binocular camera at the post verification point. The left and right eyes output two video streams, both sharing the same trigger clock. The trigger clock refers to the clock signal that simultaneously drives the left and right eyes to begin exposure. The time difference between a single frame is controlled within 10 milliseconds, and these are combined into a binocular video stream. The 10-millisecond time difference is based on the requirement that the left and right eye images must record at the same moment; exceeding this value will result in visible deviations in depth calculations. Furthermore, 10 milliseconds is less than the 40-millisecond interval between frames at 25 frames per second, and the time difference between the left and right eyes does not cross frame boundaries. On-duty information is read from the project's real-name management system, including the job number, shift time, and a list of personnel who should be present. The list of personnel who should be present is bound to the special operations personnel certificate information to prevent unlicensed personnel from entering the verification process.

[0018] The face comparison module is used to extract keyframes from the binocular video stream using synchronous frames, extract facial features from the keyframes, and compare the facial features with a registered face database to obtain the comparison result. Specifically, synchronous frame extraction refers to taking one frame every 15 frames from the left eye channel of the binocular video stream with a synchronization window of 10 milliseconds, calculating the facial clarity of each extracted frame, and taking the average of the grayscale change rate of the facial region edges as the keyframe. The frame with the highest clarity is selected. The 15-frame interval is based on the video stream frame rate of 25 frames per second, with 15 frames taking approximately 0.6 seconds. A single on-site verification action lasts 8 to 12 seconds, allowing for the extraction of 13 to 20 frames for selection. This interval balances real-time comparison with image quality. Given the large variations in lighting at construction sites, selecting the frame with the highest clarity avoids blurry frames caused by backlighting and dust from being included in the comparison. Facial features are extracted from the facial regions of the keyframes, and these facial features are multi-dimensional feature vectors. The facial features are compared one by one with the templates of the personnel required for the corresponding position in the registered face database. The one with the highest similarity is selected. The similarity is calculated as the cosine of the angle between the feature vectors of the facial features and the template. A similarity of 0.82 or higher is considered as the identity of the person, and the comparison result includes the conclusion of whether the person is the person or not, as well as the personnel number. The 0.82 is set based on the fact that the false recognition rate is less than one in ten thousand with a scale of 600 registered templates for the project. For example, there are 3 personnel required for the tower crane operator position, and there are 3 corresponding templates in the registered face database. Feature extraction is performed on the facial regions of keyframes to obtain 512-dimensional facial features. The cosine of the angle between the facial features and the 3 templates is calculated one by one, and the similarity results are 0.91, 0.31, and 0.28, respectively. The highest similarity of 0.91 is selected, which is greater than 0.82, and the person is identified as the person. If the highest similarity in a certain verification is 0.75, which is less than 0.82, the person is identified as not the person, and the comparison result includes the conclusion of not the person, and the personnel number is empty.

[0019] like Figure 2 As shown, the liveness detection module is used to extract synchronous frames from the binocular video stream to obtain binocular images, and then perform binocular liveness detection on the binocular images to determine the risk level. Binocular liveness detection refers to using the depth information of the binocular images to determine whether the verification object is a real person. Planar forgery methods cannot form the depth distribution of a real face, thus intercepting proxy attendance behavior. The risk level is used to indicate the degree to which the verification object is close to a cheating attack, and is divided into low risk, medium risk, and high risk. Specifically, the liveness detection module includes: The frame extraction submodule is used to extract synchronous frames from the stereo video stream to obtain a stereo image. Using the same synchronization window as the face comparison module, the frame extraction submodule extracts one frame from the left and one frame from the right eye at the same time from the stereo video stream, and the two frames are combined to form a stereo image.

[0020] The disparity analysis submodule is used to calculate the disparity of binocular images to obtain a depth map. Specifically, the disparity analysis submodule performs stereo matching on the binocular images, calculates the displacement of corresponding points in the left and right eyes pixel by pixel, i.e., the disparity, and converts the disparity into depth to obtain a depth map. The depth difference between the tip of the nose and the ear of a real human face is 3 to 4 centimeters, while the disparity between the photo and each point on the screen is close to zero. The depth map can intuitively present this difference.

[0021] The attack detection submodule is used to identify and detect cheating attacks on the depth map, obtain attack confidence, and generate a risk level based on the attack confidence. The attack detection submodule identifies three common cheating attacks found on construction sites, performing three types of detection on the depth map: First, it calculates the planar proportion by statistically analyzing the percentage of areas in the depth map with a depth difference of less than 1 cm. 1 cm is less than the minimum depth difference of a real face; the depth difference between a photo and the screen is usually less than this value, and a higher proportion is closer to a photo attack. Second, it detects the presence of regularly arranged periodic stripes in the depth map to obtain a texture score; a higher score is closer to a screen playback attack. Third, it compares the depth distribution of the depth map with a real face depth template area by area to obtain a mask deviation score; a higher score is closer to a mask attack. The real face depth template refers to the depth reference of each part of a real face obtained from statistical analysis of historical real-person samples from the project. The maximum value among the planar proportion, texture score, and mask deviation score is taken as the attack confidence, ranging from 0 to 1, with a larger value indicating a closer approximation of an attack. Risk levels are generated based on attack confidence: below 0.3 indicates low risk, 0.3 to 0.7 indicates medium risk, and above 0.7 indicates high risk. Below 0.3 is generally considered real-person verification, above 0.7 is generally considered an attack sample, and the middle range is treated as suspected. For example, when a verifier holds a printed photo to a camera, the photo is planar, and areas with a depth difference of less than 1 cm in the depth map account for 92% of the entire image, with a planar proportion of 0.92. When a verifier plays a pre-recorded video on a mobile phone, the screen pixel grid forms regularly arranged periodic stripes in the depth map, with a texture score of 0.85. When a verifier wears a silicone mask, the depth of the mask at the tip of the nose and cheekbone deviates from the depth template of a real face, with a mask deviation score of 0.78. For real-person verification, the planar proportion is 0.12, the texture score is 0.08, and the mask deviation score is 0.10. The maximum value of 0.10 is taken as the attack confidence; below 0.3, it is considered low risk. The attack confidence scores for the three scenarios—holding a printed photo, playing a video on a mobile phone, and wearing a silicone mask—were 0.92, 0.85, and 0.78, respectively, all higher than 0.7, indicating a high risk.

[0022] The attendance management module generates attendance records and stage identifiers based on on-duty information, performs anomaly detection on attendance records, and issues real-time alerts for abnormal attendance statuses. Specifically, attendance records include personnel number, job number, clock-in time, and shift. Stage identifiers mark the current stage of attendance, sequentially: on-duty verification stage, on-duty inspection stage, and off-duty settlement stage. Tower crane operators must complete on-duty verification and off-duty settlement once before and after each shift. Anomaly detection scenarios include: verification result indicating the person is not the registered individual, clock-in time outside the shift period, two consecutive verification failures for the same position, and absence of the safety officer during the on-duty inspection stage. When an abnormal attendance status is detected, a real-time alert is issued, and the alert information is pushed to the project management platform duty room, reminding dedicated safety management personnel to verify the situation.

[0023] like Figure 3 As shown, the image evidence storage module is used to perform hash calculation on keyframes to obtain a first hash value, generate basic recording based on stage identifiers, generate recording strategies based on risk levels, acquire images from the binocular video stream based on the basic recording and recording strategies to obtain image evidence, perform hash calculation on the image evidence to obtain a second hash value, and align keyframes, comparison results, image evidence, and attendance records with timestamps to obtain verification files; specifically, the image evidence storage module includes: The frame hashing submodule is used to perform frame preprocessing on keyframes to obtain frame features, and to perform hash calculation on the frame features to obtain the first hash value. Specifically, frame preprocessing includes size unification and grayscale conversion, and the frame features are obtained after processing. The frame features are then subjected to SM3 hash calculation to obtain the first hash value.

[0024] The recording trigger submodule is used to parse the stage identifier to obtain the trigger signal, and generate basic recording based on the trigger signal. Specifically, stage parsing refers to identifying the stage to which the stage identifier belongs: the on-duty verification stage generates a trigger signal to start recording, and the off-duty settlement stage generates a trigger signal to end recording, and basic recording is generated based on the trigger signal. The basic recording is the default recording benchmark, which is 15 seconds, 1080P, and 25 frames per second. According to project testing, the average time for one verification action is 8 to 12 seconds. 15 seconds can cover the complete action with a margin, and the storage size of a single piece of evidence is small, and the daily increase in storage of all attendance terminals in the project is controllable.

[0025] The strategy mapping submodule is used to quantify the risk level to obtain the recording duration, and generate a recording strategy based on the recording duration and the basic recording. Level quantization refers to comparing the attack confidence corresponding to the risk level with the tier boundary. The tier that the risk level falls into is quantified as the corresponding risk tier, and the recording duration is determined according to the risk tier. The higher the risk tier, the longer the recording duration. The specific tiering and determination process can be found in the risk mapping unit and the time period mapping unit. The relationship between the recording strategy and the basic recording is as follows: the basic recording ensures that there is basic recording under any circumstances, and the recording strategy adds recording duration and increases frame rate and resolution parameters when the risk increases.

[0026] The risk mapping unit is used to extract and quantify risk levels, and map them to multiple risk tiers. Numerical extraction refers to retrieving the attack confidence level used to generate the risk level. Tier quantification involves comparing the attack confidence level with two tier boundaries: 0.3 and 0.7. Levels below 0.3 are quantified as low-risk, 0.3 to 0.7 as medium-risk, and above 0.7 as high-risk. The tier boundaries are determined based on the same criteria as the attack detection submodule, i.e., the project's historical attack sample statistics. For ease of subsequent mapping, low-risk tiers are denoted as tier value 1, medium-risk tiers as tier value 2, and high-risk tiers as tier value 3. The time period mapping unit and parameter mapping unit look up the corresponding recording time period, frame rate parameter, and resolution parameter in a pre-stored mapping table based on the tier value. The mapping table is a pre-configured list of corresponding tier values ​​and recording parameters.

[0027] The time-segment mapping unit is used to map risk levels to obtain recording time segments, decompose basic recordings into basic time segments, align the recording time segments with the basic time segments, and extract the difference from the aligned time segments to obtain the recording duration. The time-segment mapping is as follows: 15 seconds for low-risk levels, 30 seconds for medium-risk levels, and 60 seconds for high-risk levels. The higher the risk, the more complete the recording process. High-risk cases often involve suspected proxy clocking in, requiring complete recording of the scene for future reference. The 15-second basic time segment can cover a complete verification action. The 30-second medium-risk time segment can capture a few seconds of environmental footage before and after the verification action to identify whether someone instructed someone to clock in for another person. The 60-second high-risk time segment can cover the entire process from the person entering the verification point to leaving, facilitating the tracing of accomplices. The basic recording is decomposed into time periods to obtain a basic time period of 15 seconds; time period alignment means aligning the recording time period with the basic time period at the same starting point; difference extraction means taking the portion of the recording time period that exceeds the basic time period as the recording duration, with 0 seconds for low-risk level, 15 seconds for medium-risk level, and 45 seconds for high-risk level.

[0028] The parameter mapping unit is used to map frame rate and resolution parameters to risk levels, perform parameter verification on these parameters, and fuse the verification results to obtain a set of verification parameters. Specifically, the parameter mapping is as follows: low-risk level corresponds to 25 frames per second and 1080P, medium-risk level corresponds to 30 frames per second and 2K, and high-risk level corresponds to 60 frames per second and 4K. The parameter verification content does not exceed the upper limit of the attendance terminal's recording capability; if it exceeds this limit, the upper limit will be applied. Parameter fusion refers to combining the verified frame rate and resolution parameters into a single set to obtain the set of verification parameters.

[0029] A strategy synthesis unit is used to synthesize a recording strategy from recording duration, basic recording, risk level, and verification parameter group. Specifically, the strategy synthesis unit includes: The basic analysis subunit is used to analyze the basic video recording to obtain the basic time period, basic frame rate, and basic resolution, which are then used as baseline parameters. Specifically, the basic analysis subunit analyzes the basic video recording to obtain a basic time period of 15 seconds, a basic frame rate of 25 frames per second, and a basic resolution of 1080P. These three parameters together serve as baseline parameters, representing the minimum recording standard implemented under any risk scenario. The significance of the baseline parameters is to provide a safety net; even if attack detection fails or the risk level is mistakenly judged as low risk, the attendance terminal still retains basic image evidence according to the baseline parameters, ensuring traceable video recordings under any circumstances. When the risk increases, the recording strategy adds recording duration and increases the frame rate and resolution parameters on top of the baseline parameters, forming a two-layer recording structure of safety net recording and enhanced recording. For example, the baseline parameters are a basic frame rate of 25 frames per second, a basic resolution of 1080P, and a basic time period of 15 seconds.

[0030] The duration overlay subunit is used to overlay the recording duration with the base time period to obtain the total time period. Specifically, the low-risk level corresponds to a recording duration of 0 seconds, the medium-risk level corresponds to 15 seconds, and the high-risk level corresponds to 45 seconds. After adding the recording duration to the base time period of 15 seconds, the total time period is obtained, with the total time periods corresponding to the three levels being 15 seconds, 30 seconds, and 60 seconds, respectively. The purpose of time period overlay is that the base time period is the minimum recording that is performed under any circumstances, and the recording duration is the additional recording when the risk increases. The two are connected end to end to form a single continuous video, rather than two independent video files, thus avoiding time gaps in the visual evidence.

[0031] The risk level modulation subunit is used to obtain enhancement parameters by increasing the frame rate and resolution based on the risk level, the verification parameter set, and the total time period. Specifically, the frame rate and resolution enhancement refer to replacing the base frame rate and base resolution in the baseline parameters with the frame rate and resolution parameters in the verification parameter set, respectively, to obtain the enhancement parameters. The total time period is used to determine the effective range of the enhancement parameters; that is, the base time period is executed according to the baseline parameters, and the portion of the total time period exceeding the base time period is executed according to the enhancement parameters. For example, under the high-risk level, the enhancement parameters are 60 frames per second and 4K, which can clearly restore the facial details of the person clocking in for someone else; under the low-risk level, the verification parameter set and the baseline parameters are the same, and the enhancement parameters are also the same as the baseline parameters.

[0032] The strategy output subunit is used to encapsulate the baseline parameters, enhancement parameters, and total time period into a recording strategy. Strategy encapsulation refers to organizing the baseline parameters, enhancement parameters, and total time period into a recording strategy according to the execution order: the first 15 seconds are recorded according to the baseline parameters, and subsequent time periods are recorded according to the enhancement parameters, until the total time period is reached.

[0033] The image acquisition submodule is used to acquire images from the stereo video stream according to the basic recording and recording strategy, and obtain image evidence through image caching. Specifically, basic recording is performed first, and the stereo video stream is enhanced according to the recording strategy; the image cache is a circular cache in the memory of the attendance terminal. Acquired segments are first written to the circular cache, and after it is full, they are written to local storage to obtain image evidence. The circular cache can avoid the loss of segments due to network fluctuations at the construction site.

[0034] The image hashing submodule is used to segment image evidence to obtain segment features, and then perform hash calculations on the segment features to obtain a second hash value. Specifically, image segmentation refers to dividing the image evidence into segments of 1 second each, extracting the segment length, segment order, and content bytes of each segment as segment features, and then performing SM3 hash calculations on the segment features to obtain a second hash value.

[0035] The document encapsulation submodule is used to extract and align timestamps from keyframes, comparison results, video evidence, and attendance records, and then encapsulate the data to obtain a verification document. The attendance terminal uses a unified clock with timestamps accurate to milliseconds. Timestamp extraction refers to retrieving the generation timestamps from keyframes, comparison results, video evidence, and attendance records. Timestamp alignment means placing all four on the same timeline. Data encapsulation involves packaging the aligned data into a single verification document and assigning each document a unique document number. This document number consists of the project number, terminal number, and generation timestamp, ensuring that every second in the video corresponds to the time in the attendance record.

[0036] like Figure 4As shown, the data encryption module is used to calculate the file verification value by performing digest calculation on the verification file, write the file verification value, the first hash value, and the second hash value into the attendance record, encrypt the written attendance record to obtain encrypted attendance, and store and encrypt facial features, verification files, written attendance records, and abnormal warning information to obtain encrypted storage data. The encrypted storage area is divided into a regular encryption area and an abnormal encryption area according to the risk level, and the verification files are partitioned and encrypted according to the risk level. Specifically, the data encryption module includes: The verification submodule is used to unpack the verification file to obtain the verification file body, segment the verification file body and extract digest features, and perform digest calculation and encryption on the digest features to obtain the file verification value. Specifically, unpacking refers to removing the encapsulation header to obtain the verification file body; segmenting the verification file body and extracting the segment length, segment order and content bytes of each segment as digest features; performing SM3 digest calculation on the digest features, and then using the device master key to encrypt the digest calculation result with SM4 digest to obtain the file verification value; the digest encryption is based on preventing attackers from recalculating the digest to forge the file verification value.

[0037] The verification and writing submodule is used to aggregate the file verification value, first hash value, second hash value, and attendance record. Through field mapping and write calculation, it writes the file verification value, first hash value, and second hash value into the attendance record, resulting in a self-locking structure. Specifically, verification aggregation refers to grouping the file verification value, first hash value, second hash value, and attendance record into a set to be written; field mapping means that the attendance record has a frame hash field, an image hash field, and a verification value field at the end, with the file verification value, first hash value, and second hash value corresponding one-to-one with these three fields; write calculation refers to writing the data sequentially according to the field order. After writing, a self-locking structure is obtained, meaning that the attendance record is mutually bound to the file verification value, first hash value, and second hash value. Any modification to any of these will cause a discrepancy in the platform's verification, which can be used to lock the original attendance facts in labor disputes.

[0038] The transmission encryption submodule is used to generate session keys, and then encapsulates and encrypts the self-locking structure according to the session keys to obtain encrypted attendance records. Specifically, the session key is randomly generated before each upload, with a length of 128 bits, based on the principle of one-time key, so that the leakage of a single record will not affect other records; the self-locking structure is encapsulated and encrypted according to the session key to obtain encrypted attendance records, which are then uploaded to the project management platform.

[0039] The storage encryption submodule is used to aggregate and group facial features, verification files, written attendance records, and anomaly warning information into plaintext groups, and then encrypt the data according to the storage algorithm to obtain encrypted storage data. Specifically, data aggregation refers to merging facial features, verification files, written attendance records, and anomaly warning information into a single set of data to be encrypted. During data aggregation, verification files with a high risk level are not included in the set of data to be encrypted and are encrypted and stored separately by the split storage submodule. Plaintext grouping refers to dividing the data into 16-byte groups; in this embodiment, the storage algorithm is SM4-CBC, and the encrypted storage data is obtained after encrypting each group.

[0040] The storage allocation submodule is used to divide the encrypted storage area into a regular encrypted area and an abnormal encrypted area based on the risk level, and to store the encrypted storage data in the regular encrypted area. Specifically, the regular encrypted area and the abnormal encrypted area are two logical partitions of the local storage of the attendance terminal. The regular encrypted area stores daily attendance data, and the abnormal encrypted area stores files of suspected cheating incidents and is mirrored and backed up to the project management platform daily. Data in the encrypted storage area that does not contain files of suspected cheating incidents is stored in the regular encrypted area.

[0041] The off-line storage submodule is used to perform risk level resolution and encryption selection. Based on the encryption selection, the verification file is encrypted to obtain either regular encryption or enhanced encryption. Regularly encrypted files are stored in the regular encryption area, and enhanced encrypted files are stored in the abnormal encryption area. Specifically, the off-line storage submodule includes: The risk level analysis unit is used to determine the risk level as either normal or abnormal. Specifically, low and medium risk are classified as normal, while high risk is classified as abnormal. High risk corresponds to suspected cheating incidents involving substitute workers, requiring enhanced protection and separate data retention.

[0042] An encryption selection unit is used to obtain a conventional algorithm by indexing a conventional algorithm according to the conventional level, to obtain a conventional key by deriving a conventional key from the device master key, and to encrypt the verification file using the conventional key to obtain conventional encryption; and to obtain an enhanced algorithm by indexing an enhanced algorithm according to the anomaly level, to obtain an enhanced key by deriving an enhanced key from the device master key, and to encrypt the verification file using the enhanced key to obtain enhanced encryption; specifically, the encryption selection unit includes: The algorithm index subunit is used to index conventional algorithms based on their level of abnormality, identifying conventional algorithms from the conventional algorithm library; and to index enhanced algorithms based on their level of abnormality, identifying enhanced algorithms from the enhanced algorithm library. Specifically, indexing refers to locating a unique algorithm in the corresponding algorithm library according to its level. The conventional algorithm library pre-stores the SM4-CBC algorithm, AES-CTR algorithm, and ZUC algorithm, while the enhanced algorithm library pre-stores the grouped round iterative algorithm and the multi-round AES algorithm. The significance of pre-stored multiple candidate algorithms in the algorithm library is that the determination of conventional and abnormal levels varies with project scenarios, and the chip capabilities and data compliance requirements of attendance terminals differ across projects. The algorithm index allows the same system to determine the algorithm matching the level from the library under different scenarios without modifying the system architecture. The basis for determining SM4-CBC as the conventional algorithm in this embodiment is that the cryptographic operation chip built into the attendance terminal supports SM4-CBC hardware, has low computational overhead, and is suitable for encrypting attendance records with a large amount of data under the conventional level. The basis for determining the group round iteration algorithm as the enhanced algorithm is that this algorithm improves the anti-cracking strength by increasing the number of iteration rounds. The computational overhead is greater than that of SM4-CBC, but the number of attendance records under the abnormal level is small and the security requirements are high, so the computational overhead is acceptable.

[0043] The key derivation subunit is used to generate a regular key seed based on the device master key, and to derive a regular key from the regular key seed; it also generates an enhanced key seed based on the device master key, and derives an enhanced key from the enhanced key seed. Specifically: the device master key and module identifier are concatenated and then SM3 calculated to generate a regular key seed. The module identifier refers to the fixed number of each functional unit within the attendance terminal. The regular key seed is then SM3 calculated again, and the first 128 bits are truncated to obtain the regular key. Similarly, an enhanced key seed is generated, and SM3 is performed again to obtain a 256-bit enhanced key. The device master key is stored in a secure chip and cannot be read. The secure chip refers to the cryptographic operation chip built into the attendance terminal. The derived keys are dedicated to specific purposes, and the leakage of a single key will not affect other data.

[0044] The encryption operation subunit is used to perform conventional encryption operations on the verification file using a conventional key and a conventional algorithm to obtain conventional encryption; and to perform enhanced encryption operations on the verification file using an enhanced key and an enhanced algorithm to obtain enhanced encryption. Specifically, the enhanced encryption operation adopts a grouped round-based iterative structure, including: The verification file is divided into left and right blocks. The enhanced key and risk level are fused to obtain the round key. The risk level is then mapped to the number of rounds to obtain the iteration rounds. Specifically, file division refers to splitting the verification file into 64-bit blocks, each block consisting of a 32-bit left block and a 32-bit right block. Key fusion involves concatenating the enhanced key with the risk level rank value and then calculating the round key using SM3. The round key varies with the risk level rank value. The round mapping is as follows: high risk is mapped to 32 rounds; more rounds mean stronger resistance to analysis. With 32 rounds, the attendance terminal encrypts a verification file in approximately 200 milliseconds, which is within an acceptable range.

[0045] The left and right file blocks are iterated through in rounds. In each round, the right file block is XORed with the round key. The XOR result is then cyclically shifted, and the shifted result is XORed with the left file block. The left and right files are then swapped. Finally, the left and right files from the last round are concatenated to obtain enhanced encryption. Each round of iteration proceeds as follows: the right file block is XORed with the round key; the XOR result is cyclically shifted left by 7 bits. The 7-bit shift is chosen because it's a fixed shift amount for easy implementation by the attendance terminal, and 7 bits ensures sufficient diffusion of changes within a single round. Furthermore, 7 bits are prime numbers, coprime to the 32-bit block length, guaranteeing sufficient bit rearrangement after each shift. The cyclically shifted result is XORed with the left file block, and the left and right files are swapped. After reaching the required number of rounds, the left and right files from the last round are concatenated sequentially to obtain enhanced encryption.

[0046] The path selection unit is used to obtain a regular storage path by performing a regular path index for regular encryption and an abnormal storage path by performing an abnormal path index for enhanced encryption. Specifically, the regular path index refers to locating the storage location according to the file number of the verified file, and the regular storage path points to the storage directory and fragmentation rules of the regular encryption area; similarly, the abnormal path index points to the storage directory and mirroring rules of the abnormal encryption area.

[0047] The archive unit is used to encapsulate conventional encryption with ciphertext to obtain conventionally encapsulated ciphertext, generate an integrity tag for the conventionally encapsulated ciphertext, and store the conventionally encapsulated ciphertext and integrity tag in the conventional encryption area according to the conventional storage path. Similarly, it performs ciphertext encapsulation on enhanced encryption to obtain enhancedly encapsulated ciphertext, generates an integrity tag for the enhancedly encapsulated ciphertext, and stores the enhancedly encapsulated ciphertext and integrity tag in the abnormal encryption area according to the abnormal storage path. Specifically, ciphertext encapsulation refers to adding the archive number and storage time to the ciphertext header; the integrity tag in this embodiment is taken from the SM3 calculation result of the encapsulated ciphertext. Conventional encryption, after encapsulation, yields conventionally encapsulated ciphertext, which, along with its integrity tag, is stored in the conventional encryption area according to the conventional storage path; enhanced encryption, after encapsulation, yields enhancedly encapsulated ciphertext, which, along with its integrity tag, is stored in the abnormal encryption area according to the abnormal storage path. During reading, the integrity tag is recalculated and compared with the stored value; if they do not match, it is determined to have been tampered with, and reading is refused.

[0048] The platform management module is used to decrypt encrypted attendance records and generate reports, receive file retrieval requests from the requesting end, retrieve and decrypt files from the regular encrypted area and the abnormal encrypted area according to the file retrieval request to obtain plaintext verification files, unpack and verify the plaintext verification files, compare the verification calculation results with the first hash value, the second hash value and the file verification value respectively, and send the plaintext verification files back to the requesting end when the comparison is successful. Specifically, the unpacking and verification calculation of the plaintext verification file involves: unpacking the plaintext verification file to obtain keyframes, image evidence, comparison results, and attendance records; performing frame preprocessing and hash calculation on the keyframes to obtain frame hash values; performing image segmentation and hash calculation on the image evidence to obtain image hash values; recombining the keyframes, image evidence, comparison results, and attendance records into the verification file body; segmenting the verification file body according to the same file segmentation method as the attendance terminal; extracting the segment length, segment order, and content bytes as digest features for each segment; performing SM3 digest calculation on the digest features; and then encrypting the digest calculation result with SM4 digest using the device master key to obtain the file verification value. After receiving the encrypted attendance, the project management platform decrypts it using the session key and generates a report, including team attendance sheets and a list of special operation personnel on duty, used for real-name management and labor wage calculation. The retrieval terminal is the project safety director's office terminal or the housing and construction authority's inspection terminal; the file retrieval request includes personnel number, date, and time period. Based on the file retrieval request, files are retrieved from both the regular and unusual encrypted areas and decrypted to obtain plaintext verification files. Unpacking yields keyframes, image evidence, comparison results, and attendance records. The keyframes undergo the same frame preprocessing and SM3 hash calculation as the attendance terminal to obtain the current frame hash value. Image evidence is segmented into 1-second segments and SM3 hashed to obtain the current image hash value. Digest calculation and encryption are performed on the keyframes, image evidence, comparison results, and attendance records to obtain the current file verification value. The algorithm and key used are consistent with those of the attendance terminal. The current frame hash value, current image hash value, and current file verification value are compared with the first hash value, second hash value, and file verification value written in the attendance record, respectively. If all match, the comparison passes, and the plaintext verification file is returned to the retrieving end. If any do not match, the file is considered modified, the return is rejected, and a verification log is recorded. The verification log records the retrieval person, retrieval time, file number, and verification result.

[0049] The present invention has been described in detail above. The specific embodiments are provided only to help understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make various improvements and modifications to the present invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A multimodal verification and attendance system for key personnel, characterized in that, include: The information acquisition module is used to acquire binocular video streams and on-duty information; The face comparison module is used to extract key frames from the binocular video stream, extract features from the key frames to obtain face features, and compare the face features with the registered face database to obtain the comparison result. The liveness detection module is used to extract synchronous frames from the binocular video stream to obtain binocular images, and to perform binocular liveness detection on the binocular images to obtain the risk level; The attendance management module is used to generate attendance records and stage identifiers based on the on-duty information, to determine the anomalies in the attendance records, and to issue real-time warnings if there are abnormal attendance statuses. The image evidence module is used to perform hash calculation on the key frame to obtain a first hash value, generate basic recording based on the stage identifier, generate a recording strategy based on the risk level, acquire images from the binocular video stream based on the basic recording and the recording strategy to obtain image evidence, perform hash calculation on the image evidence to obtain a second hash value, and align the key frame, comparison result, image evidence and attendance record with timestamps to obtain a verification file. The data encryption module is used to perform digest calculation on the verification file to obtain the file verification value, write the file verification value, the first hash value and the second hash value into the attendance record, encrypt the written attendance record to obtain encrypted attendance, and store and encrypt the facial features, verification file, written attendance record and abnormal warning information to obtain encrypted storage data. The encrypted storage area is divided into a regular encryption area and an abnormal encryption area according to the risk level, and the verification file is partitioned and encrypted according to the risk level. The platform management module is used to decrypt the encrypted attendance system and generate reports, receive file retrieval requests from the requesting end, retrieve and decrypt files from the regular encrypted area and the abnormal encrypted area according to the file retrieval request to obtain plaintext verification files, unpack and verify the plaintext verification files, compare the verification calculation results with the first hash value, the second hash value and the file verification value respectively, and send the plaintext verification files back to the requesting end when the comparison is successful.

2. The multimodal verification and attendance equipment system for key personnel according to claim 1, characterized in that, The liveness detection module includes: The frame extraction submodule is used to extract synchronous frames from the stereo video stream to obtain stereo images. The disparity analysis submodule is used to calculate the disparity of binocular images to obtain a depth map; The attack detection submodule is used to identify and detect cheating attacks on the depth map, obtain the attack confidence score, and generate a risk level based on the attack confidence score.

3. The multimodal verification and attendance equipment system for key personnel according to claim 1, characterized in that, The image storage module includes: The frame hashing submodule is used to perform frame preprocessing on keyframes to obtain frame features, and to perform hash calculation on the frame features to obtain the first hash value; The recording trigger submodule is used to perform stage parsing on the stage identifier to obtain the trigger signal, and generate basic recording based on the trigger signal; The strategy mapping submodule is used to quantify the risk level to obtain the recording duration, and generate a recording strategy based on the recording duration and the basic recording. The image acquisition submodule is used to acquire images from the binocular video stream based on the basic recording and recording strategy, and obtain image evidence through image buffering. The image hashing submodule is used to segment image evidence to obtain segment features, and then perform hash calculation on the segment features to obtain a second hash value. The document encapsulation submodule is used to extract and align timestamps from keyframes, comparison results, image evidence, and attendance records, and then encapsulate the data to obtain verification documents.

4. The multimodal verification and attendance equipment system for key personnel according to claim 3, characterized in that, The policy mapping submodule includes: The risk mapping unit is used to extract and quantify the risk level and map it to multiple risk levels. The time period mapping unit is used to map the risk level to obtain the recording time period, decompose the basic recording to obtain the basic time period, align the recording time period with the basic time period, and extract the difference from the time period alignment result to obtain the recording duration. The parameter mapping unit is used to map the risk level to obtain frame rate parameters and resolution parameters, perform parameter verification on the frame rate parameters and resolution parameters, and perform parameter fusion on the parameter verification results to obtain a verification parameter group. The strategy synthesis unit is used to synthesize the recording duration, basic recording, risk level and verification parameter group to obtain the recording strategy.

5. The multimodal verification and attendance equipment system for key personnel according to claim 4, characterized in that, The strategy synthesis unit includes: The basic analysis subunit is used to analyze the basic recording to obtain the basic time period, basic frame rate and basic resolution, and uses the basic time period, basic frame rate and basic resolution as the reference parameters; The duration overlay subunit is used to overlay the recording duration with the base time period to obtain the total time period; The level modulation subunit is used to obtain enhancement parameters by increasing the frame rate and resolution based on the risk level, the verification parameter group and the total time period; The strategy output subunit is used to encapsulate the baseline parameters, enhancement parameters, and total time period to obtain the recording strategy.

6. The multimodal verification and attendance equipment system for key personnel according to claim 1, characterized in that, The data encryption module includes: The verification submodule is used to decompile the verification archive into a structured form to obtain the verification archive body, segment the verification archive body into archive segments and extract summary features, and perform summary calculation and summary encryption on the summary features to obtain the archive verification value. The verification and write submodule is used to perform verification aggregation on the file verification value, the first hash value, the second hash value and the attendance record. Through field mapping and write calculation, the file verification value, the first hash value and the second hash value are written to the attendance record to obtain a self-locking structure. The transmission encryption submodule is used to generate a session key, and to encapsulate and encrypt the self-locking structure according to the session key to obtain encrypted attendance. The storage encryption submodule is used to aggregate and group plaintext data such as facial features, verification files, written attendance records and abnormal warning information, and encrypt the data according to the storage algorithm to obtain encrypted storage data. The storage allocation submodule is used to divide the encrypted storage area into a regular encrypted area and an abnormal encrypted area according to the risk level, and to store the encrypted storage data in the regular encrypted area. The off-line storage submodule is used to perform risk level analysis and encryption selection. The verification file is encrypted according to the encryption selection to obtain regular encryption or enhanced encryption. Regular encryption is stored in the regular encryption area, and enhanced encryption is stored in the abnormal encryption area.

7. The multimodal verification and attendance equipment system for key personnel according to claim 6, characterized in that, The off-flow storage submodule includes: The risk level analysis unit is used to determine the risk level and obtain a normal level or an abnormal level. The encryption selection unit is used to obtain a conventional algorithm by indexing a conventional algorithm according to the conventional level, to obtain a conventional key by deriving a conventional key from the device master key, and to encrypt the verification file using the conventional key to obtain conventional encryption; and to obtain an enhanced algorithm by indexing an enhanced algorithm according to the anomaly level, to obtain an enhanced key by deriving an enhanced key from the device master key, and to encrypt the verification file using the enhanced key to obtain enhanced encryption. The path selection unit is used to perform regular path indexing for regular encryption to obtain a regular storage path, and to perform abnormal path indexing for enhanced encryption to obtain an abnormal storage path. The archive unit is used to encapsulate regular encryption with ciphertext to obtain regular encapsulated ciphertext, generate an integrity tag for the regular encapsulated ciphertext, and store the regular encapsulated ciphertext and integrity tag in the regular encryption area according to the regular storage path. It also encapsulates enhanced encryption with ciphertext to obtain enhanced encapsulated ciphertext, generates an integrity tag for the enhanced encapsulated ciphertext, and stores the enhanced encapsulated ciphertext and integrity tag in the abnormal encryption area according to the abnormal storage path.

8. The multimodal verification and attendance equipment system for key personnel according to claim 7, characterized in that, The encryption selection unit includes: The algorithm index subunit is used to index regular algorithms based on the regularity level and determine regular algorithms from the regular algorithm library; and to index enhanced algorithms based on the anomaly level and determine enhanced algorithms from the enhanced algorithm library. The key derivation subunit is used to generate a regular key seed based on the device master key, and to derive a regular key from the regular key seed; and to generate an enhanced key seed based on the device master key, and to derive an enhanced key from the enhanced key seed. The encryption operation subunit is used to perform conventional encryption operations on the verification file using a conventional key and a conventional algorithm to obtain conventional encryption; and to perform enhanced encryption operations on the verification file using an enhanced key and an enhanced algorithm to obtain enhanced encryption.

9. The multimodal verification and attendance equipment system for key personnel according to claim 8, characterized in that, The enhanced encryption operation employs a group round iterative structure, including: The verification file is divided into left and right file blocks. The enhanced key and risk level are fused to obtain the round key. The risk level is mapped to the number of rounds to obtain the number of iteration rounds. The left and right file blocks are iterated in rounds according to the number of iterations. In each round of iteration, the right file block is XORed with the round key. The XOR result is then cyclically shifted. The cyclically shifted result is XORed with the left file block, and the left and right file blocks are swapped. The left and right file blocks from the last round of iteration are then concatenated to obtain enhanced encryption.

10. The multimodal verification and attendance equipment system for key personnel according to claim 1, characterized in that, The unpacking and verification calculation of the plaintext verification file is as follows: unpacking the plaintext verification file to obtain keyframes, image evidence, comparison results and attendance records; performing frame preprocessing and hash calculation on the keyframes to obtain the current frame hash value; performing image segmentation and hash calculation on the image evidence to obtain the current image hash value; and performing digest calculation and digest encryption on the keyframes, image evidence, comparison results and attendance records to obtain the current file verification value.

Images