Intelligent cabinet file signing and circulation method based on face recognition

Abstract

The application provides a kind of intelligent cabinet file signature flow method based on face recognition, comprising: responding to the delivery personnel identity recognition through result, read the delivery personnel identity, target signature personnel identity and unsigned cabinet state, generate delivery access record;After the paper file falls into unsigned cabinet and the cabinet door is closed, collect non-text appearance data and non-text bearing data, generate file non-content lifeline feature;The delivery access record, delivery fall cabinet record and file non-content lifeline feature are bound as signature flow task, and the task record to be signed is generated after integrity verification;After taking out paper file, generate off-cabinet task record to be signed, and generate back storage lifeline feature after its back storage paper file;The back storage lifeline feature is compared with the file non-content lifeline feature, and the signed task record to be taken is generated when comparison is passed;After the delivery personnel takes paper file, generate signature flow closed loop record.

Description

Technical Field

[0001] This invention relates to the field of intelligent cabinet document circulation management technology, and in particular to an intelligent cabinet document signing and circulation method based on facial recognition. Background Technology

[0002] Current paper document signing methods rely heavily on manual delivery, temporary office storage, or dedicated personnel for transfer, which can easily lead to long waiting times, unclear handover points, document delays, and difficulty in tracing responsibility. To improve efficiency, existing technologies have adopted smart cabinets for document delivery, signing, and retrieval. For example, after identification, the delivery person places the document in the unsigned compartment, the signatory retrieves and signs it before placing it in the signed compartment, and the delivery person then retrieves it.

[0003] However, existing smart lockers typically only record the person's identity, the time the locker compartment is opened, and the locker door's open / closed status. They struggle to prove whether the paper documents delivered, retrieved, stored, and collected are the same document or from the same batch. When multiple delivery personnel deliver documents to the same approver consecutively, the folders may have similar appearances, page numbers, or binding methods, easily leading to problems such as misplacement, concealment, substitution, omissions, or incorrect collection.

[0004] Using barcodes, QR codes, document number entry, photo recording, or full-text scanning to identify documents increases the workload for delivery personnel and may expose the document text, signatures, or seals. Conversely, relying solely on facial recognition and cabinet logs without entering document content fails to establish a reliable chain of evidence demonstrating the continuity of physical paper documents. Therefore, this invention proposes a facial recognition-based intelligent cabinet document signing and circulation method.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may contain prior art information that is not common knowledge to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing a smart cabinet document signing and circulation method based on facial recognition, thereby solving the technical problems mentioned in the background section.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for document signing and circulation in a smart cabinet based on facial recognition includes the following steps: S1. The multi-agent system responds to the delivery personnel's identity recognition result, reads the delivery personnel's identity, the target signatory's identity and the status of the unsigned cabinet, generates a delivery access record, generates a delivery cabinet record after the paper document falls into the unsigned cabinet and the cabinet door is closed, collects non-text appearance data and non-text carrying data, and generates non-content lifeline features of the document. S2. The task orchestration agent binds the delivery access record, delivery cabinet record, and non-content lifeline characteristics of the document into a signature and approval flow task. After the integrity verification is passed, a task record to be signed is generated and the unsigned cabinet retrieval permission is written. S3. When the identity of the target signatory is verified and the unsigned cabinet retrieval permission is effective, open the unsigned cabinet, update the pending task record to the off-cabinet pending task record according to the retrieval completion record, and write the signed cabinet save permission. S4. After the target signatory returns the paper document, a return lifeline feature is generated. The return lifeline feature is continuously compared with the non-content lifeline feature of the document. If the comparison is successful, a signed and pending task record is generated. If the comparison fails, a task lock status is generated. S5. When the delivery personnel's identity is verified, the signed cabinet retrieval permission is effective, and the task is not locked, open the signed cabinet and generate a closed-loop record of the signature and approval process based on the retrieval completion record, the non-content lifeline characteristics of the document, and the return lifeline characteristics.

[0008] S1 specifically includes: an identity recognition agent that collects facial images of delivery personnel and outputs their identity; a task scheduling agent that reads the delivery permission table, the signatory personnel table, and the cabinet mapping table, verifies the identity of the delivery personnel, the identity of the target signatory personnel, and the status of unsigned cabinets, and generates a delivery access record; a cabinet status agent that opens unsigned cabinets based on the delivery access record, and after paper documents are placed in and the cabinet door is closed, reads the cabinet door closing signal, door lock return signal, paper document occupancy signal, load weight signal, and cabinet static status signal, and generates a delivery and cabinet placement record; and a lifeline acquisition agent that collects non-text appearance data and non-text load data based on the delivery and cabinet placement record, deletes abnormal data, and generates non-content lifeline features of the document as the basis for task binding.

[0009] S2 specifically includes: The task orchestration agent reads the delivery access record, delivery locker record, and non-content lifeline features of the document; when the collection quality identifier is valid, it generates a task number and binds the delivery person's identity, the target signatory's identity, the unsigned locker number, the locker time, the document's placeholder result, and the document's non-content lifeline features into a signing and approval workflow task; The task orchestration agent performs integrity verification on the signing and approval workflow task, and generates a pending-signing task record when the personnel identity, locker ownership, document placeholder, feature validity, and historical conflicts all pass; The notification and authorization agent generates a reminder based on the pending-signing task record and writes the retrieval permission bound to the locker status agent with the task number, the target signatory's identity, the unsigned locker number, the pending-signing status, and the permission's expiration time.

[0010] S3 specifically includes: a task orchestration agent reads the pending task record and permission write record; an identity recognition agent collects the facial image of the target signatory and outputs the signatory's identity recognition result; after joint verification of identity, status, cabinet, permission time, permission write, and pending authorization anomalies, a retrieval access record is generated; a cabinet status agent reviews the unsigned cabinet's occupancy, weight, door lock, and abnormal status based on the retrieval access record; after review, the unsigned cabinet is opened, and a retrieval completion record is generated after the paper document is taken out and the cabinet door is closed; the task orchestration agent, based on the pending task record, retrieval access record, and retrieval completion record, migrates the pending sign status to the off-cabinet pending sign status, generates an off-cabinet pending sign task record, and writes it to the signed cabinet's back-to-access permission.

[0011] S4 specifically includes: a task orchestration agent that reads the records of tasks awaiting signature after leaving the cabinet and the record of write access permissions; an identity recognition agent that collects the facial images of the personnel returning to the cabinet and outputs the result of identity recognition; and a record of access permission is generated after the identity, status of waiting to sign after leaving the cabinet, status of waiting to return to the cabinet, number of the signed cabinet, access permission, and cabinet anomaly verification are passed; a cabinet status agent that opens the signed cabinet according to the record of access permission for returning to the cabinet, and a record of returning to the cabinet is generated after the paper document is placed in and the cabinet door is closed; a lifeline collection agent that generates the lifeline features of returning to the cabinet according to the same rules; and a lifeline comparison agent that compares the lifeline features of returning to the cabinet with the non-content lifeline features of the document for continuity. If the comparison is successful, a record of signed tasks awaiting collection is generated; if the comparison fails, a record of task lock status and continuity anomaly is generated.

[0012] S5 specifically includes: The notification and authorization agent reads the signed pending task record, task lock status, and continuity anomaly record; when the task is not locked and the continuity comparison passes, it writes the signed cabinet retrieval permission; the identity recognition agent collects the delivery person's facial image and outputs the recipient's identity; after joint verification by the task scheduling agent, a retrieval access record is generated; the cabinet status agent reviews the signed cabinet occupancy, weight, opening / closing records, and anomaly indicators based on the retrieval access record; after successful review, the signed cabinet is opened, and relevant status signals are collected after the paper document is removed and the cabinet door is closed; the task scheduling agent determines retrieval completion based on the above status signals and generates a retrieval completion record; subsequently, a closed-loop verification is performed based on the signed pending task record, retrieval access record, retrieval completion record, and two types of lifeline features; when the verification passes, a signed approval flow closed-loop record is generated, and the notification and authorization agent revokes the retrieval permission.

[0013] The beneficial effects of this invention are as follows: This invention establishes a signing and circulation task through identity recognition, verification of unsigned cabinet status, and delivery confirmation, binding the delivery of paper documents to the identity of the delivery person, the identity of the target signatory, and the unsigned cabinet number, thus reducing misdelivery, incorrect delivery, and invalid delivery. By collecting non-text appearance data and non-text-bearing data to generate non-content lifeline features of documents, the continuity of paper document entities can be characterized without reading the document's image or text, thus balancing document confidentiality and traceability.

[0014] This invention continuously compares the non-content lifeline characteristics of documents with the return lifeline characteristics, enabling the identification of anomalies such as replacement, concealment, misplacement, or omission during the return of paper documents, thereby improving the reliability of the return status of signed documents. Through status migration via pending signature task records, off-cabinet pending signature task records, signed pending collection task records, and closed-loop signature and approval flow records, the entire process of paper documents from delivery, collection, return, to retrieval has clear nodes and verifiable evidence.

[0015] This invention improves the security of document signing and circulation in smart lockers by combining identity verification results, locker access permissions, task status, and abnormal status during the retrieval, storage, and collection stages. This avoids opening lockers solely based on facial recognition results. When continuous comparison fails, access permission writing fails, locker status is abnormal, or collection is abnormal, a task lock status or abnormal handling record is generated to prevent the continued circulation of abnormal documents, facilitating subsequent review and accountability. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a smart cabinet document signing and circulation method based on facial recognition according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example: Figure 1 As shown, this embodiment provides a smart cabinet document signing and circulation method based on facial recognition, including the following steps: S1. The multi-agent system responds to the delivery personnel's identity recognition result, reads the delivery personnel's identity, the target signatory's identity and the status of the unsigned cabinet, generates a delivery access record, generates a delivery cabinet record after the paper document falls into the unsigned cabinet and the cabinet door is closed, collects non-text appearance data and non-text carrying data, and generates non-content lifeline features of the document. S2. The task orchestration agent binds the delivery access record, delivery cabinet record, and non-content lifeline characteristics of the document into a signature and approval flow task. After the integrity verification is passed, a task record to be signed is generated and the unsigned cabinet retrieval permission is written. S3. When the identity of the target signatory is verified and the unsigned cabinet retrieval permission is effective, open the unsigned cabinet, update the pending task record to the off-cabinet pending task record according to the retrieval completion record, and write the signed cabinet save permission. S4. After the target signatory returns the paper document, a return lifeline feature is generated. The return lifeline feature is continuously compared with the non-content lifeline feature of the document. If the comparison is successful, a signed and pending task record is generated. If the comparison fails, a task lock status is generated. S5. When the delivery personnel's identity is verified, the signed cabinet retrieval permission is effective, and the task is not locked, open the signed cabinet and generate a closed-loop record of the signature and approval process based on the retrieval completion record, the non-content lifeline characteristics of the document, and the return lifeline characteristics.

[0019] In this embodiment, the multi-agent system includes an identity recognition agent, a cabinet status agent, a task orchestration agent, a lifeline acquisition agent, a lifeline comparison agent, and a notification and authorization agent. The identity recognition agent is used to acquire facial images and output the identity recognition results. The cabinet status agent is used to read the cabinet door, door lock, occupancy, weight, and abnormal status and execute cabinet opening or blocking. The task orchestration agent is used to generate records, bind tasks, perform verification, and migrate task status. The lifeline acquisition agent is used to generate non-content lifeline features and lifeline features for file storage. The lifeline comparison agent is used to perform continuous comparison. The notification and authorization agent is used to generate reminders, write permissions, and revoke permissions. Abnormal records generated in each step are written to an abnormality handling queue. The abnormality handling queue includes task number, abnormality type, abnormal fields, abnormal time, blocking object, and review status, used for cabinet-side prompts, review prompts, and subsequent abnormality tracing.

[0020] S1 specifically includes the following sub-steps: S110. After receiving a delivery request initiated by a delivery person in front of the smart locker, the multi-agent system collects the delivery person's facial image and performs identity recognition. The identity recognition agent is a functional unit used to complete facial collection, feature matching, and identity output. The delivery person's identity output serves as the sole source of identity for this delivery permission verification. After the delivery person passes the identity recognition, they select the target signatory identity. The task scheduling agent reads the delivery permission table, the signatory table, and the locker mapping table to verify whether the delivery person's identity has valid delivery permission, whether the target signatory identity has valid signing permission, and whether the target signatory identity is bound to the unsigned locker number.

[0021] The cabinet status intelligent agent synchronously reads the unsigned cabinet number's corresponding unsigned cabinet occupancy status, unsigned cabinet door lock status, unsigned cabinet abnormality identifier, and historical non-closed loop task status. The historical non-closed loop task status is used to indicate whether there are any signing and approval flow tasks in the same unsigned cabinet that have not yet been transferred to the signed and pending status.

[0022] The task orchestration agent generates a delivery access record when the delivery personnel's identity is valid, the target signatory's identity is valid, the unsigned cabinet number matches, the unsigned cabinet door lock status is locked and waiting to be opened, the unsigned cabinet's abnormality flag is empty, and the historical non-closed-loop task status does not conflict with the current delivery. The delivery access record includes the delivery access record number, delivery personnel's identity, target signatory's identity, unsigned cabinet number, delivery access time, access verification result, and access failure reason fields.

[0023] When the access verification result is passed, the task orchestration agent sends the delivery access record to the cabinet status agent as the basis for opening the cabinet; when the access verification result is failed, the task orchestration agent generates a delivery rejection record, writes the delivery rejection record into the exception handling queue, and does not send an opening instruction to the cabinet status agent.

[0024] Through the above processing, S120 can only open unsigned cabinets based on delivery access records with passed access verification results, thus avoiding binding of legitimate personnel identities with incorrect cabinets, incorrect signature objects, or abnormal cabinet statuses.

[0025] S120. After the cabinet status intelligent agent reads the delivery access record, it opens the unsigned cabinet recorded in the delivery access record and records the opening time, unlocking current status and cabinet door opening result. After the delivery personnel put the paper documents into the unsigned cabinet and close the cabinet door, the cabinet status intelligent agent reads the cabinet door closing signal, door lock return signal, paper document occupancy signal, load weight signal and cabinet static state signal. The paper document occupancy signal is used to indicate whether there is an occupancy area in the unsigned cabinet that matches the shape of the paper document. The load weight signal is used to indicate the weight change of the unsigned cabinet's load-bearing surface before and after opening the cabinet.

[0026] The task scheduling agent generates a delivery record only when the cabinet door closure signal is valid, the door lock return signal is valid, the paper document occupancy signal is valid, the load weight signal changes positively relative to the baseline weight before opening the cabinet, and the static state signal inside the cabinet meets the acquisition conditions. The positive change in load weight can be set to no less than 10 grams to exclude the situation of accidentally closing an empty cabinet or not putting in any documents.

[0027] The delivery and placement record includes the delivery and placement record number, delivery access record number, delivery personnel identity, target signatory personnel identity, unsigned cabinet number, placement time, cabinet door closure result, document occupancy result, load-bearing weight change value, and placement verification result.

[0028] If the cabinet door fails to close within the preset waiting time, the lock fails to return to its original position, the file placeholder result is empty, or the load-bearing weight change value does not meet the set conditions, the task scheduling agent will not generate a delivery record, but will generate a delivery exception record and write the delivery exception record into the exception handling queue. At the same time, it will be sent to the notification authorization agent to remind the delivery personnel to close the cabinet again or re-deliver.

[0029] After the delivery record is generated, the task orchestration agent sends the delivery record to the lifeline collection agent as the trigger for S130 to start the collection of non-content lifeline features of documents, so that subsequent feature collection only occurs after the paper document has actually been placed in the unsigned cabinet and the cabinet door is in a valid closed state.

[0030] As a specific hardware implementation, an RGB-D depth camera is installed at the center of the top inner wall of the unsigned compartment, looking downwards, and a high-resolution array of thin-film pressure sensors is laid on the bottom surface. The paper document placeholder signal is extracted from the foreground mask of the top-view depth map output by the RGB-D depth camera; the load weight signal and the cabinet's static state signal are calculated from the dynamic pressure matrix continuously collected by the array of thin-film pressure sensors. When the variance of the sum of the dynamic pressure matrix within a preset time window is less than a set threshold, the cabinet's static state signal is activated.

[0031] S130, the lifeline data acquisition agent reads the delivery and placement records, and uses the unsigned cabinet number, placement time, and document occupancy results in the delivery and placement records as the basis for data acquisition and positioning. During the static detection period after the unsigned cabinet is closed, it collects non-text appearance data and non-text load data of paper documents. Non-text appearance data refers to the shape data that does not include the main text of the document, the text of the seal, and the text of the signature and approval, including the document edge outline, the stacking thickness outline, the binding corner shape, and the surface warping shape. Non-text load data is data used to reflect the load status of paper documents in the cabinet, including the load weight, the center of gravity distribution, and the occupancy area in the cabinet.

[0032] The static detection period begins from the moment the lock on the unsigned cabinet door is closed. The lifeline data acquisition agent continuously reads the changes in load weight, stacking thickness, and document edge contour, and determines the static detection completion time according to the following formula:

[0033] In the formula, t represents the time when the static detection is completed, and t represents the current time during the detection process. This indicates the change in the load at the current moment. This indicates the threshold for changes in load capacity. This indicates the change in layer thickness at the current moment. This indicates the threshold for changes in layer thickness. This indicates the amount of change in the file edge contour at the current moment. Indicates the threshold for changes in the file edge contour. This indicates the moment when the lock on the unsigned compartment door was closed. This indicates the minimum duration of stillness.

[0034] After the lifeline acquisition agent reaches the time of completion of static detection, it performs text region masking, edge processing, abnormal frame deletion, and normalization processing on the acquired data. Among them, text region masking is used to delete image areas that may contain text, edge processing is used to preserve the outer boundary of paper documents, abnormal frame deletion is used to delete data with missing acquisition time, cabinet door not closed, document not static, missing placeholder area, or abnormal sensor channel, and normalization processing is used to convert data from different sensor channels into comparable fields.

[0035] The lifeline acquisition agent organizes the processed valid data into file non-content lifeline features:

[0036] In the formula, E represents the non-content lifeline feature of the document, H represents the edge contour feature of the document, A represents the stacking thickness contour feature of the document, B represents the surface warping feature of the document, W represents the load-bearing weight feature, G represents the center of gravity distribution feature of the document, P represents the cabinet occupancy area feature of the document, and Q represents the acquisition quality indicator.

[0037] When the acquisition quality identifier is valid, the lifeline acquisition agent sends the non-content lifeline features of the file and the delivery record to the task orchestration agent as the basis for subsequent S210 binding and approval workflow tasks. When the acquisition quality identifier is invalid, the task orchestration agent generates a delivery exception record, writes the delivery exception record to the exception handling queue, and blocks the binding of pending tasks. The non-content lifeline features of the file are only used for subsequent binding of pending tasks and continuous comparison of data storage; they do not save the file's main text image or text.

[0038] As a preferred embodiment, the specific processing logic for text area masking is as follows: The color image output by the RGB-D depth camera is converted to grayscale and edge detected to extract all high-frequency connected components in the image. Regions whose connected component area is smaller than the overall file area by a preset ratio (e.g., 5%) and whose local gradient features match the text or stamp texture features are identified as potential text regions. By smoothly filling the pixel values ​​of potential text areas with surrounding background pixels or setting them to null values, the text content can be masked.

[0039] The surface warping morphology features are specifically extracted from the depth map (Z-axis data) output by the RGB-D depth camera, and the depth extrema points and depth change gradients within the edge contour of the paper document are used as the warping characterization vectors. The layer thickness profile feature is obtained by calculating the difference between the average depth of the area where the document is located and the reference depth of the bottom surface; the binding angle morphology feature is obtained by performing a Hough Transform on the edge profile to identify right-angle vertices and extracting the thickness bulge change rate of the vertices.

[0040] The center of gravity distribution characteristics are calculated based on the pressure matrix of the array-type thin-film pressure sensor. Specifically, the sum of the products of the position coordinates of each effective force point in the matrix and its corresponding pressure value is divided by the total pressure value to obtain the coordinates of the moment center point in the two-dimensional plane, which is used as the center of gravity distribution characteristics.

[0041] S2 specifically includes the following sub-steps: S210: The task orchestration agent reads the delivery access record generated by S110, the delivery locker record generated by S120, and the non-content lifeline characteristics of the file generated by S130, and uses the simultaneous existence of all three and the consistency of their field sources as a prerequisite for task binding. The task orchestration agent is a functional unit in the multi-agent system used to generate task numbers, bind task fields, and control task state transitions. It does not directly control the opening of the locker door, but only outputs the executable task status and permission basis to the locker status agent.

[0042] The task orchestration agent first reads the collection quality identifier in the non-content lifeline features of the file. If the collection quality identifier is valid, the task binding continues. If the collection quality identifier is invalid, the task orchestration agent generates an invalid lifeline feature record, writes the invalid lifeline feature record into the exception handling queue, and sends the record to the notification authorization agent to prompt the delivery personnel to re-deliver or re-close the cabinet. No approval and transfer task is generated.

[0043] When the quality identification is valid, the task scheduling agent uses the delivery personnel's identity, the target signatory's identity, the unsigned cabinet number, the cabinet placement time, and the non-content lifeline characteristics of the document as the unique sources of the task to generate a task number, and establishes a one-to-one association between the task number and the delivery access record number and the delivery cabinet placement record number; the task number is used to point to the same paper document circulation task in subsequent retrieval, storage, collection, and closed-loop records, and the task number will not be changed after the cabinet is opened, the document is stored, or the personnel collect it.

[0044] Approval workflow tasks can be organized by the following fields:

[0045] In the formula, This indicates the approval and transfer of tasks. Indicates the task number. Indicate the identity of the delivery person. Indicates the identity of the person who signed off on the target application. This indicates that the cabinet number was not signed. Indicates the time of placement in the cabinet. This indicates a non-content lifeline characteristic of the file. This indicates the file placeholder result. Indicates the task status.

[0046] In the above fields, the identity of the delivery person and the identity of the target signatory are obtained from the delivery access record; the unsigned cabinet number and cabinet placement time are obtained from the delivery and cabinet placement record; the non-content lifeline features of the document are obtained from the lifeline collection agent; the document placeholder result is obtained from the cabinet status agent; and the task status is written by the task scheduling agent. Through this binding method, the signing and transfer task can simultaneously point to the personnel, cabinet, time, and paper document entity features. Subsequently, S220 will not repeatedly identify the document content, but only perform integrity verification on the task field.

[0047] S220. The task orchestration agent performs integrity verification on the approval and transfer tasks. The integrity verification is used to determine whether the approval and transfer tasks meet the conditions for being converted into pending task records.

[0048] The task orchestration agent sequentially verifies whether the delivery personnel's identity matches the delivery personnel's identity in the delivery access record, whether the target signatory's identity matches the target signatory's identity in the delivery access record, whether the unsigned cabinet number matches the cabinet bound to the target signatory in the cabinet mapping table, whether the file placeholder result is valid, whether the collection quality identifier in the non-content lifeline characteristics of the file is valid, and whether the historical non-closed-loop task status does not conflict with the current signing and circulation task.

[0049] If multiple documents are allowed to be stacked in the same unsigned cabinet, the task orchestration agent also reads the cabinet stacking limit and the current number of occupied spaces. If the current number of occupied spaces is less than the cabinet stacking limit, verification can continue. If the current number of occupied spaces reaches the cabinet stacking limit, an abnormal record of pending signature binding is generated. For example, if the cabinet stacking limit is 5 documents and the current number of valid occupied spaces is 5 documents, the task orchestration agent will no longer bind newly delivered documents as pending signature task records.

[0050] The integrity verification result is determined according to the following logic:

[0051] In the formula, This indicates the result of the integrity verification of the approval and workflow task. This indicates the result of the delivery personnel's authorization verification. This indicates the result of the identity verification of the target approver. This indicates that the verification results for the ownership of the unsigned cabinets have not been completed. This indicates the file placeholder verification result. This indicates the validity verification result of the non-content lifeline feature of the document. This indicates the results of the verification of historical unclosed-loop mission conflicts. This indicates a logical AND relationship where all verification results must be true simultaneously.

[0052] Only when the integrity verification result is passed will the task orchestration agent write the approval flow task into the pending task record; if any verification result is failed, the task orchestration agent will generate a pending binding exception record and write the pending binding exception record into the exception handling queue. The pending binding exception record includes the task number, exception field, exception reason, exception generation time and blocking result, and blocks the subsequent writing of unsigned cabinet retrieval permissions.

[0053] Once verified, the generated pending task record includes the pending task record number, task number, delivery personnel identity, target signatory identity, unsigned cabinet number, non-content lifeline characteristics of the document, document placeholder result, pending signature status, task generation time, and permission expiration time.

[0054] The "Pending Signature" status indicates that the paper document has been delivered and is awaiting retrieval by the designated signatory. In this status, the system only allows the designated signatory matching the pending task record to request access to the corresponding unsigned locker. The delivery person is not allowed to retrieve the document, nor are other signatory personnel allowed to collect it, and the task cannot be directly moved to the "Signed and Awaiting Collection" status. The pending task record serves as the sole source of subsequent S230 generation of unsigned locker retrieval permissions.

[0055] S230. After reading the pending task record, the notification and authorization agent generates a pending signature reminder for the target approver and writes the unsigned cabinet retrieval permission corresponding to the pending task record to the cabinet status agent. The notification and authorization agent is a functional unit in the multi-agent system used to generate reminder records, write cabinet permissions, and record authorization results. The reminder records it generates do not replace the cabinet opening conditions. Cabinet opening is still determined by the identity recognition result subsequently output by the identity recognition agent and the unsigned cabinet retrieval permission.

[0056] The notification authorization agent generates a reminder sending record that includes the task number, the identity of the target signatory, the reminder time, the reminder method, and the sending result. If the reminder fails to send but the permission is successfully written, the pending task record remains valid, and the system retains the pending sign prompt on the cabinet screen or in the local task list. If the permission writing fails, the task orchestration agent marks the pending task record as an unauthorized abnormal state, writes the permission writing failure record into the exception handling queue, and blocks the subsequent generation of the pickup access record by S310.

[0057] Unsigned locker pickup permissions are linked to the task number, the identity of the target approver, the unsigned locker number, the pending signature status, and the validity expiration time of the permissions. The conditions for its activation are as follows:

[0058] In the formula, This indicates whether the unsigned locker pickup permission is active; a value of 1 indicates that it is active. This indicates that the person in front of the counter is being identified. Indicates the identity of the person who signed off on the target application. This indicates the cabinet number that is requested to be opened. This indicates that the cabinet number was not signed. Indicates the task status. Indicates a pending signature status. Indicates the current time. Indicates the expiration time of the permission. This indicates that the left and right conditions are equivalent.

[0059] The cabinet status agent will only allow S310 to open unsigned cabinets based on the identification results of the target signatory if all the above conditions are met. After the permission is written, the cabinet status agent returns the permission write record to the task orchestration agent. The permission write record includes the task number, the unsigned cabinet number, the permission write time, the permission expiration time, and the write result.

[0060] The task orchestration agent associates and saves the pending task records, reminder sending records, and permission writing records, enabling the S310 to read the pending status, pending authorization abnormal status, target signatory identity, unsigned cabinet number, and permission validity expiration time under the same task number, and complete the pickup access verification accordingly.

[0061] S3 specifically includes the following sub-steps: S310, the task orchestration agent reads the pending task record and permission write record generated by S230; when the target approver arrives at the smart cabinet and initiates a request to retrieve an unsigned document, the identity recognition agent collects the facial image of the person in front of the cabinet, performs identity recognition, and outputs the identity recognition result of the approver; the identity recognition result of the approver includes the identity of the person in front of the cabinet, the recognition confidence level, the liveness detection result, the recognition time, and the recognition window number. The recognition window number is used to distinguish the results of continuous or repeated recognition within the same time period.

[0062] Before outputting the identification results of the signatory, the intelligent identification agent deletes identification data that is missing face images, has an identification confidence level lower than a preset threshold, fails liveness detection, has multiple valid faces appearing in the same identification window, or has missing collection time. For example, when the identification confidence level threshold is set to 0.85, identification results below 0.85 will not be included in the document collection access verification.

[0063] The task orchestration agent reads the identification results of the signatory, the record of tasks to be signed, the permission writing record, the real-time status of unsigned cabinets, and the current time. It then performs joint verification on the identity of the target signatory, the task status, the number of unsigned cabinets, the abnormal identifier of unsigned cabinets, the expiration time of the permission, the permission writing result, and the abnormal status of pending authorization. The joint verification result is determined by the following formula:

[0064] In the formula, This indicates the result of the pickup access verification. This indicates the result of the identity verification of the target signatory in the package pickup access process. This indicates the task status verification result during the pickup approval process. This indicates the consistency verification result of the unsigned cabinet number in the package pickup access process. This indicates the verification result of the locked status of the unsigned locker during the package pickup access process. This indicates the verification result of the validity period of the access permission in the package pickup process. This indicates the verification result of the permission write result in the package pickup access process. This indicates the verification result of the pending authorization abnormal status in the pickup access process. This indicates a logical AND relationship where all verification results must be true simultaneously.

[0065] The task orchestration agent generates a pickup access record only when the pickup access verification result is passed. The pickup access record includes the task number, the identity of the target signatory, the unsigned cabinet number, the pickup access time, the identification window number, the pickup access verification result, and the validity expiration time of the permission. It is then sent to the cabinet status agent as the basis for subsequent cabinet opening.

[0066] If any verification item fails, the task orchestration agent generates a pickup rejection record and writes it to the exception handling queue. No cabinet opening instruction is sent to the cabinet status agent. The pickup rejection record includes the task number, rejection field, rejection reason, and rejection time. It is used for subsequent exception queries but does not trigger task status transition.

[0067] S320: After reading the pickup access record, the cabinet status intelligent agent first performs a pre-opening status review on the unsigned cabinet corresponding to the pickup access record, and then decides whether to open the unsigned cabinet.

[0068] The pre-opening status verification includes reading the file placeholder status, load-bearing weight status, the most recent opening and closing record of the cabinet door, the door lock return status, and the abnormality indicator of the unsigned cabinet. If the file placeholder status is empty, the load-bearing weight is abnormally reduced relative to the load-bearing weight feature in the pending task record, the most recent opening and closing record of the cabinet door shows unauthorized opening, or the abnormality indicator is not empty, the cabinet status agent will not execute the opening, the task orchestration agent will generate a pre-retrieval status abnormality record, write the pre-retrieval status abnormality record into the abnormality handling queue, and keep the pending task record in the pending signing state.

[0069] When the pre-opening status verification is passed, the cabinet status intelligent agent opens the corresponding unsigned cabinet for the access record and records the unlocking current status, cabinet door opening time and cabinet door opening result; after the target signatory takes out the paper document and closes the unsigned cabinet, the cabinet status intelligent agent reads the cabinet door closing result, door lock return result, document occupancy change result, load weight change result and item retrieval completion time.

[0070] The task orchestration agent determines whether the paper document has been retrieved based on the above results. The retrieval completion determination result is determined by the following formula:

[0071] In the formula, This indicates the completion of the pickup process. This indicates the cabinet door closing status during the pickup process. This indicates the door lock return status after the item retrieval is complete. This indicates the file placeholder changes during the pickup process. This indicates the change in the load-bearing weight during the pickup process. This indicates a logical relationship where all results must be true simultaneously.

[0072] If an unsigned cabinet corresponds to only one pending task, the document occupancy change result should change from effective occupancy to vacancy, and the load-bearing weight change result should show a change in the direction of retrieval relative to before the cabinet was opened. If an unsigned cabinet allows multiple documents from the same target signatory to be stacked, the task scheduling agent determines the target occupancy area corresponding to this task based on the cabinet occupancy area characteristics and load-bearing weight characteristics contained in the non-content lifeline characteristics of the documents in the pending task record. After the document is retrieved, the remaining occupancy area should still be able to match other pending task records.

[0073] When the item retrieval completion determination passes, the task orchestration agent generates an item retrieval completion record. This record includes the item retrieval completion record number, task number, target signatory's identity, unsigned cabinet number, retrieval completion time, pre-retrieval occupancy status, post-retrieval occupancy status, load weight change, cabinet door closure result, and door lock return result. If the item retrieval completion determination fails, the task orchestration agent generates an item retrieval incomplete record, writes it to the exception handling queue, and keeps the pending task record in a pending-signing state.

[0074] S330: The task orchestration agent reads the pending task record, the pickup access record, and the pickup completion record. When the task numbers of the three are consistent, the pickup access verification result is passed, the pickup completion judgment result is passed, the unsigned cabinet is effectively closed, and the door lock is effectively returned to its original position, the pending sign status in the pending task record is transferred to the off-cabinet pending sign status, and an off-cabinet pending sign task record is generated. The off-cabinet pending sign status indicates that the paper document has been taken out of the unsigned cabinet by the target signatory and is leaving the cabinet for signing and approval processing, and has not yet been returned to the signed cabinet.

[0075] The record of tasks pending signature after leaving the cabinet includes the task number, the identity of the delivery person, the identity of the target signatory, the number of the unsigned cabinet, the non-content lifeline features of the document, the pickup completion time, the start time of leaving the cabinet, the empty status of the unsigned cabinet, the status of waiting for return, and the status transition result. Among them, the non-content lifeline features of the document continue to use the features generated in S130 and bound in S210, which are used for continuous comparison by the lifeline comparison agent after the lifeline features are generated in S420.

[0076] The task orchestration agent must not use the cabinet door opening result as the basis for state transition; if the pickup access record exists but the pickup completion record is invalid, or the pickup completion record does not match the task number of the pending task record, the task orchestration agent shall retain the pending status and write the reason for the abnormality into the pickup incomplete record or the pickup pre-record abnormal status record.

[0077] After the task record for pending signatures after leaving the counter is generated, the authorized agent is notified to write the signed counter storage permission to the counter status agent based on the record. The signed counter storage permission is bound to the task number, the identity of the target signatory, the signed counter number, the pending signature status after leaving the counter, and the storage waiting status. If the task has not entered the pending signature status after leaving the counter, the authorized agent is notified not to write the signed counter storage permission.

[0078] The cabinet status agent returns a write record for the recall permission, which includes the task number, the signed cabinet number, the write time, the recall permission status, and the write result. The task orchestration agent associates and saves the records of tasks awaiting signature after leaving the cabinet with the write record for the recall permission, enabling S410 to subsequently read the identity of the target signatory under the same task number, the signed cabinet number, the status of waiting for signature after leaving the cabinet, the recall waiting status, and the non-content lifeline characteristics of the document, and to complete the recall identity verification and opening of the signed cabinet based on this information.

[0079] S4 specifically includes the following sub-steps: S410, the task orchestration agent reads the log of tasks awaiting signature after leaving the cabinet and the log of write access permissions generated by S330; when the person returning the document arrives at the smart cabinet and initiates a request to return the signed document, the identity recognition agent collects the person's facial image, performs identity recognition, and outputs the return identity recognition result; the return identity recognition result includes the person's identity, recognition confidence, liveness detection result, recognition time, and recognition window number. Recognition data with a recognition confidence lower than a preset threshold, liveness detection failure, multiple valid faces appearing in the same recognition window, or missing collection time are deleted by the identity recognition agent and do not enter the return access verification.

[0080] The task orchestration agent reads the stored identity verification results, the log of tasks awaiting signature after leaving the counter, the record of write access permissions, the real-time status of signed counters, and the current time. It then performs joint verification of the user's identity, the status of tasks awaiting signature after leaving the counter, the status of waiting for signature, the number of signed counters, the write access permission results, and the abnormal status of signed counters. The result of the signature access verification is determined by the following formula:

[0081] In the formula, This indicates that the access verification results have been saved. This indicates the result of the identity verification of the personnel returning to the deposit account during the deposit admission process. This indicates the verification result of the pending signature status during the data entry process. This indicates the verification result of the save-up waiting status in the save-up access process. This indicates the consistency verification result of the signed cabinet number in the access review process. This indicates the verification result of write access permission in the write access process. This indicates the verification result of the abnormal status of the signed cabinet in the access process. This indicates a logical AND relationship where all verification results must be true simultaneously.

[0082] Only when the readmission access verification result is passed will the task orchestration agent generate a readmission access record. This record includes the task number, the identity of the target approver, the signed cabinet number, the readmission access time, the identification window number, and the readmission access verification result. This record is then sent to the cabinet status agent as the basis for opening the signed cabinet. If any verification item fails, the task orchestration agent generates a readmission rejection record, writes it to the exception handling queue, keeps the pending tasks record unchanged, and prohibits the cabinet status agent from opening the signed cabinet based on this request.

[0083] S420: After the cabinet status intelligent agent reads and stores the access record, it opens the signed cabinet corresponding to the access record and records the unlocking current status, cabinet door opening time, and cabinet door opening result. After the target approver puts the signed paper document into the signed cabinet and closes the cabinet door, the cabinet status intelligent agent reads the cabinet door closing signal, door lock return signal, document occupancy signal, load weight signal, and cabinet static status signal.

[0084] The task orchestration agent generates a cabinet return record only when the cabinet door closure signal is valid, the door lock return signal is valid, the document placeholder signal is validly occupied, the load-bearing weight has changed positively relative to before the cabinet was opened, and the static state inside the cabinet meets the data collection conditions. The cabinet return record includes the cabinet return record number, task number, target signatory identity, signed cabinet number, return time, cabinet door closure result, door lock return result, document placeholder result, load-bearing weight change value, and cabinet return verification result.

[0085] If a signed cabinet allows multiple signed documents to be collected, the task orchestration agent also reads the existing and newly added spaced areas within the signed cabinet, confirming that the newly added spaced area does not cover the spaced areas corresponding to other signed and collected tasks. If it cannot confirm, it generates a spaced-out exception record, writes the spaced-out exception record to the exception handling queue, and does not start the spaced-out lifeline feature collection.

[0086] When the retrieval record is valid, the lifeline acquisition agent, during the static detection period after the signed cabinet is closed, uses the same text area masking rules, edge processing rules, abnormal frame deletion rules, and normalization rules as S130 to collect non-text appearance data and non-text data of the paper document, generating retrieval lifeline features:

[0087] In the formula, This indicates the characteristics of the lifeline. Indicates the edge contour features of the saved file. This indicates the profile characteristics of the stored stack thickness. Indicates the shape characteristics of the binding corner during storage. This indicates the warping morphology of the stored surface. This indicates the load-bearing capacity characteristic during storage. This indicates the distribution characteristics of the centroid of the stored data. This indicates the characteristics of the space occupied within the storage cabinet. This indicates the quality of the data collection and storage.

[0088] The lifeline acquisition agent deletes data that is not closed, is not static, has missing space, has abnormal sensor channels, or may contain text from the document. If valid data that meets the acquisition quality requirements still exists after deletion, the lifeline feature is sent back to the lifeline comparison agent. If there is insufficient valid data after deletion, the task orchestration agent generates a back-collection anomaly record, writes the back-collection anomaly record into the anomaly processing queue, and writes a back-collection pending review flag into the off-cabinet pending task record. No signed pending task record is generated.

[0089] S430: The lifeline comparison agent reads the non-content lifeline features of the document generated by S130 and bound by S210, the return lifeline features generated by S420, and the record of tasks awaiting signature after leaving the cabinet. It then compares the physical continuity of the paper document after it is delivered to the cabinet with that after it is returned to the cabinet. The continuity comparison does not require the two features to be completely identical. Slight changes caused by page turning, signing pressure, or short-term movement during the signing process are allowed, but abrupt changes in edge contours, stacking thickness, binding angle shape, load-bearing weight, center of gravity distribution, or the area occupied within the cabinet are not permitted.

[0090] The lifeline comparison agent calculates continuous difference values ​​based on the differences in each field:

[0091] In the formula, M represents the continuous difference value. Indicates the difference in file edge contours. Indicates the difference in layer thickness. This indicates the difference in binding angle shape. This indicates the difference in surface warping morphology. Indicates the difference in load-bearing weight. Indicates the difference in the distribution of the centroid. This indicates the difference in the area occupied within the cabinet. , , , , , , These represent the weighting coefficients corresponding to the aforementioned differences.

[0092] The lifeline comparison agent compares the continuity difference value with the preset continuity threshold and reads the back-store collection quality identifier at the same time. When the continuity difference value is not greater than the preset continuity threshold and the back-store collection quality identifier is valid, the task orchestration agent updates the out-of-cabinet pending task record to the signed pending task record. The signed pending task record includes the task number, the identity of the delivery person, the identity of the target signatory, the signed cabinet number, the non-content lifeline characteristics of the document, the back-store lifeline characteristics, the signed pending status, the back-store time, and the continuity comparison result, and serves as the input for the subsequent S510 to generate the retrieval permission.

[0093] When the continuity difference value exceeds the preset continuity threshold, or the data collection quality identifier is invalid, the task orchestration agent generates a task lock status and a data collection continuity anomaly record. The data collection continuity anomaly record includes the task number, anomaly field, continuity difference value, anomaly generation time, and review requirements. It notifies the authorized agent not to write the signed cabinet retrieval permission to the delivery personnel, and the cabinet status agent not to open the signed cabinet based on the task, thereby blocking the abnormal file from entering the retrieval process. The data collection continuity anomaly record is written to the anomaly processing queue.

[0094] To ensure the anti-counterfeiting capabilities of the comparison, the aforementioned weighting coefficients need to be configured differently based on the variability and forgery difficulty of each feature during the normal signing and circulation process.

[0095] In one specific embodiment, the range of each difference value is normalized to [0,1], and the preset continuity threshold is set to 0.3; the corresponding weight coefficients are assigned as follows: (Weight difference in load-bearing capacity) = 0.35 (Edge contour difference weight) = 0.20 (Weight of difference in layer thickness) = 0.15 (Weight of binding angle morphology difference) = 0.10, (Weight of surface warping morphology difference) = 0.10 (Weight of centroid distribution difference) = 0.05 (Weight of differences in the area occupied inside the cabinet) = 0.05.

[0096] The physical significance of using this weight allocation lies in: bearing weight ( ) and stack thickness ( This is a physical attribute, and it is difficult to maintain consistency when it is replaced or carried in, therefore it is given higher weight; however, the movement of paper documents during the signing process is unavoidable, resulting in the space occupied inside the cabinet ( ) and the distribution of the center of gravity ( It is prone to change, so it is given a very low weight to improve the fault tolerance.

[0097] S5 specifically includes the following sub-steps: S510: Notify the authorized agent to read the signed pending task record, task lock status and storage continuity anomaly record generated by S430, and first determine whether the task meets the authorization conditions; the authorization conditions include the existence of the signed pending task record and the task status being signed pending, the task lock status being unlocked, the continuity comparison result being passed, the signed cabinet number being valid and the signed cabinet not being in an abnormal lock status.

[0098] When all the above conditions are met, the authorization agent writes the signed cabinet retrieval permission to the cabinet status agent and generates a retrieval permission write record. The retrieval permission write record includes the task number, delivery personnel identity, signed cabinet number, permission write time, permission expiration time, and write result. If the permission write fails, the task orchestration agent generates a retrieval authorization exception record, writes the retrieval authorization exception record to the exception handling queue, keeps the signed pending task record unchanged, and does not allow the cabinet opening process to proceed.

[0099] After the delivery person arrives at the smart locker, the intelligent identity recognition agent collects the person's facial image and performs identity recognition, outputting the identity recognition result. The result includes the person's identity, recognition confidence level, liveness detection result, recognition time, and recognition window number. The intelligent identity recognition agent deletes recognition data that is missing from the facial image, has a recognition confidence level below a preset threshold, fails liveness detection, has multiple valid faces appearing in the same recognition window, or has missing data collection time.

[0100] The task orchestration agent reads the identity verification results of the person seeking the task, the record of signed and pending tasks, the record of writing the claiming permission, the task lock status, the real-time status of the signed cabinet, and the current time, and then performs the claiming access verification:

[0101] In the formula, This indicates that you are receiving the access verification results. This indicates the result of the identity verification of the recipient. This indicates the verification result of the "signed and awaiting collection" status. This indicates that the verification results have not been locked. This indicates that the consistency verification result of the signed cabinet number has been obtained. This indicates the verification result of writing the permission to receive data. This indicates the verification result of the validity period of the permission to claim. This indicates that the verification result for the abnormal status of the cabinet has been signed. This indicates a logical AND relationship where all verification results must be true simultaneously.

[0102] When the access verification result is passed, the task orchestration agent generates an access record; if any verification item fails, an access rejection record is generated, the access rejection record is written to the exception handling queue, and no cabinet opening command is sent to the cabinet status agent.

[0103] S520: After the cabinet status intelligent agent reads the retrieval access record, it performs a pre-retrieval status review on the signed cabinet before opening the cabinet. The pre-retrieval status review includes reading the document occupancy status, load-bearing weight status, the most recent cabinet door opening and closing record, abnormal indicators, and the cabinet occupancy area corresponding to the return lifeline feature in the signed cabinet; the target occupancy area is the return lifeline feature corresponding to the return cabinet occupancy area in the signed pending task record.

[0104] When the file placeholder status is empty, the load weight does not match the load weight characteristics stored in the signed pending task record, or the most recent cabinet door opening and closing record contains unauthorized opening or the abnormality identifier is not empty, the cabinet status agent does not open the cabinet, the task orchestration agent generates a pre-receipt status abnormality record, writes the pre-receipt status abnormality record into the abnormality handling queue, and keeps the signed pending task record unchanged.

[0105] When the pre-receipt status verification is passed, the cabinet status intelligent agent opens the signed cabinet and records the unlocking current status, cabinet door opening time, and cabinet door opening result. After the delivery personnel take out the paper documents and close the signed cabinet, the cabinet status intelligent agent reads the cabinet door closing result, door lock return result, target occupancy area change result, load-bearing weight change result, and retrieval completion time.

[0106] Based on the above results, the task orchestration agent determines whether the paper documents have been collected:

[0107] In the formula, This indicates that the judgment result has been received. This indicates that the cabinet door closure has been signed off. This indicates the result of the door lock returning to its original position. This indicates the result of changes in the target's occupied area. This indicates the result of the change in load-bearing weight. This indicates a logical AND relationship where all judgment results must be true simultaneously.

[0108] If a signed cabinet compartment corresponds to only one signed pending task, the target storage area should change from being effectively occupied to being vacant, and the load-bearing weight should change relative to the direction of removal before the cabinet is opened; if there are multiple signed pending documents in a signed cabinet compartment, the target storage area should correspond to the storage area in the signed pending task record, and the remaining storage area after removal should still be able to match other signed pending task records.

[0109] When the collection completion judgment is passed, the task orchestration agent generates a collection completion record; the collection completion record includes collection completion record number, task number, delivery personnel identity, signed cabinet number, collection completion time, occupancy status before collection, occupancy status after collection, load weight change value, cabinet door closing result, and door lock return result.

[0110] If the completion judgment fails, the task orchestration agent generates a record of incomplete collection, writes the record of incomplete collection to the exception handling queue, and keeps the task in the signed and awaiting collection state; if the target place area does not match the task, a record of suspected incorrect collection is generated, the record of suspected incorrect collection is written to the exception handling queue, and the task is locked.

[0111] S530: The task orchestration agent reads the signed and pending task records, the access record for receiving tasks, the record for receiving tasks completed, the task lock status, the non-content lifeline characteristics of the file, and the lifeline characteristics for saving back the file, and verifies whether the current signing and approval process can be closed-loop.

[0112] The closed-loop generation should not be based solely on whether the cabinet door is open or closed, but should be based on the following conditions: the task number matches, the retrieval access verification passes, the retrieval completion judgment passes, the retrieval personnel's identity matches the delivery personnel's identity, the task is not locked, and the target occupying area has been removed.

[0113] In the formula, This indicates that the closed-loop generation verification result is valid. This indicates the result of the task number consistency verification. This indicates that the access verification result has been passed. This indicates that the receipt process has been completed and approved. This indicates the result of the identity verification of the recipient. This indicates that the verification results have not been locked. This indicates the verification result of removing the target placeholder area. This indicates a logical AND relationship where all verification results must be true simultaneously.

[0114] When the closed-loop generation verification result is passed, the task orchestration agent generates a closed-loop record for the approval process. The closed-loop record for the approval process includes the task number, the identity of the delivery person, the identity of the target approver, the unsigned cabinet number, the signed cabinet number, the delivery access record number, the delivery and placement record number, the pending task record number, the retrieval completion record number, the return and placement record number, the continuity comparison result, the collection completion record number, the delivery access time, the placement time, the retrieval completion time, the return and placement time, the non-content lifeline characteristics of the document, the return and placement lifeline characteristics, the task lock status, the closed-loop generation time, and the closed-loop verification result.

[0115] After the closed-loop record of the approval process is generated, the task scheduling agent updates the signed and pending status to the closed-loop status, notifies the authorized agent to revoke the corresponding claiming permission for the task, and retains the claiming permission write record and the permission revocation time. The cabinet status agent releases the corresponding target occupancy area.

[0116] If the closed-loop generation verification result fails, the task orchestration agent does not terminate the approval flow task, but generates a closed-loop exception record based on the failure field. The closed-loop exception record is written to the exception handling queue as the basis for subsequent manual review and exception tracing.

[0117] All the above formulas are performed using dimensionless numerical calculations; the relevant formulas are based on empirical models that approximate the real situation, obtained through extensive data collection and software simulation fitting. The preset parameters and thresholds involved in the formulas can be conventionally set and adjusted by those skilled in the art according to the physical constraints of the actual application scenario.

[0118] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0119] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0120] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for document signing and circulation in a smart cabinet based on facial recognition, characterized in that, Includes the following steps: S1. The multi-agent system responds to the delivery personnel's identity recognition result, reads the delivery personnel's identity, the target signatory's identity and the status of the unsigned cabinet, generates a delivery access record, generates a delivery cabinet record after the paper document falls into the unsigned cabinet and the cabinet door is closed, collects non-text appearance data and non-text carrying data, and generates non-content lifeline features of the document. S2. The task orchestration agent binds the delivery access record, delivery cabinet record, and non-content lifeline characteristics of the document into a signature and approval flow task. After the integrity verification is passed, a task record to be signed is generated and the unsigned cabinet retrieval permission is written. S3. When the identity of the target signatory is verified and the unsigned cabinet retrieval permission is effective, open the unsigned cabinet, update the pending task record to the off-cabinet pending task record according to the retrieval completion record, and write the signed cabinet save permission. S4. After the target approver returns the paper document, a return lifeline feature is generated. The return lifeline feature is continuously compared with the non-content lifeline feature of the document. If the comparison is successful, a signed and pending task record is generated. If the comparison fails, a task lock status is generated.

2. The method for intelligent cabinet document signing and circulation based on facial recognition according to claim 1, characterized in that, Also includes: S5. When the delivery personnel's identity is verified, the signed cabinet retrieval permission is effective, and the task is not locked, open the signed cabinet and generate a closed-loop record of the signature and approval process based on the retrieval completion record, the non-content lifeline characteristics of the document, and the return lifeline characteristics.

3. The method for intelligent cabinet document signing and circulation based on facial recognition according to claim 1, characterized in that, S1 specifically includes: The identity recognition agent collects the facial image of the delivery personnel and outputs the identity of the delivery personnel. The task scheduling agent reads the delivery permission table, the signatory table and the cabinet mapping table, verifies the identity of the delivery personnel, the identity of the target signatory and the status of the unsigned cabinet, and generates a delivery access record. The cabinet status intelligent agent opens the unsigned cabinet according to the delivery access record. After the paper document is placed and the cabinet door is closed, it reads the cabinet door closing signal, door lock return signal, paper document occupancy signal, load weight signal and cabinet static status signal to generate a delivery record. The lifeline collection agent collects non-text appearance data and non-text content data based on the delivery and locker records. After deleting abnormal data, it generates non-content lifeline features of the file, which serve as the basis for task binding.

4. The method for intelligent cabinet document signing and circulation based on facial recognition according to claim 1, characterized in that, S2 specifically includes: The task orchestration agent reads the delivery access record, delivery cabinet record, and non-content lifeline features of the document. When the collection quality identifier is valid, it generates a task number and binds the delivery person's identity, the target signatory's identity, the unsigned cabinet number, the cabinet time, the document's occupancy result, and the document's non-content lifeline features into a signing and circulation task. The task orchestration agent performs integrity verification on the approval and transfer tasks, and generates a record of tasks to be signed when the personnel identity, cabinet ownership, document placement, feature validity and historical conflicts are all verified.

5. The method for intelligent cabinet document signing and circulation based on facial recognition according to claim 4, characterized in that, Also includes: The notification authorizes the intelligent agent to generate a reminder based on the pending task record, and writes the pickup permission to the cabinet status intelligent agent, which is bound to the task number, the identity of the target signatory, the unsigned cabinet number, the pending status, and the expiration time of the permission.

6. The method for intelligent cabinet document signing and circulation based on facial recognition according to claim 1, characterized in that, S3 specifically includes: The task orchestration agent reads the task record to be signed and the permission write record, and the identity recognition agent collects the facial image of the target signatory and outputs the identity recognition result of the signatory. After the identity, status, cabinet, permission time, permission write and unauthorized anomaly are jointly verified, the pick-up access record is generated. The cabinet status intelligent agent reviews the unsigned cabinet's occupancy, weight, door lock, and abnormal status based on the item retrieval access record. After the review is passed, the unsigned cabinet is opened, and an item retrieval completion record is generated after the paper document is removed and the cabinet door is closed. Based on the pending task record, pickup access record, and pickup completion record, the task orchestration agent transitions the pending signature status to the off-cabinet pending signature status, generates an off-cabinet pending signature task record, and writes it into the signed cabinet storage permission.

7. The method for intelligent cabinet document signing and circulation based on facial recognition according to claim 1, characterized in that, S4 specifically includes: The task orchestration agent reads the records of tasks awaiting signature after leaving the counter and the record of write access permission. The identity recognition agent collects the facial images of the person returning to the counter and outputs the identity recognition result. After the identity, waiting status after leaving the counter, waiting status for return, signed counter number, access permission, and counter anomaly verification are passed, a return access record is generated. The cabinet status agent opens the signed cabinet based on the stored access record. After the paper document is placed in and the cabinet door is closed, a stored cabinet record is generated. The lifeline collection agent generates stored lifeline features according to the same rules.

8. The method for intelligent cabinet document signing and circulation based on facial recognition according to claim 7, characterized in that, Also includes: The lifeline comparison agent will compare the continuity of the stored lifeline features with the non-content lifeline features of the file. If the comparison is successful, a signed and pending task record will be generated. If the comparison fails, a task lock status and a continuity anomaly record will be generated.

9. The method for intelligent cabinet document signing and circulation based on facial recognition according to claim 1, characterized in that, S5 specifically includes: The notification authorizes the intelligent agent to read the signed pending task records, task lock status, and storage continuity anomaly records. When the task is not locked and the continuity comparison passes, it writes the signed cabinet collection permission. The identity recognition intelligent agent collects the facial image of the delivery person and outputs the identity of the person to be collected. After the task scheduling intelligent agent passes the joint verification, it generates a collection access record. The smart agent for the cabinet status verifies the occupancy, weight, opening and closing records, and abnormal indicators of the signed cabinets based on the access record. After the verification is passed, the signed cabinet is opened, and relevant status signals are collected after the paper documents are taken out and the cabinet door is closed.

10. A method for intelligent cabinet document signing and circulation based on facial recognition according to claim 9, characterized in that, Also includes: The task orchestration agent determines that the task has been completed based on the above status signals and generates a completion record. Then, it performs a closed-loop verification based on the signed pending task record, the access record, the completion record, and the two types of lifeline features. When the verification passes, a closed-loop record of the approval flow is generated, and the authorized agent is notified to revoke the task completion permission.

Images