Credible supervision and co-processing method and system for security and civil air defense

Abstract

The invention belongs to the technical field of intelligent security and protection, and particularly relates to a credible supervision and co-processing method and system for security and civil air defense. The method comprises the following steps: obtaining supervision rule configuration input by a user, generating a supervision task instruction according to the supervision rule configuration, and issuing the supervision task instruction to a residence point supervision terminal; receiving personnel behavior data collected by the terminal in response to the instruction, and generating an abnormal behavior signal based on space-time attention fusion analysis; generating a corresponding-level alarm instruction based on the abnormal signal, converting the alarm instruction into a multi-mode alarm message, and distributing the multi-mode alarm message to a management terminal; responding to the high-level alarm, calling the behavior data of the associated personnel, integrating the behavior data into a structured supervision evidence packet through a block chain smart contract, and carrying out distributed credible evidence storage; and generating a supervision report and persistently storing the supervision report in a database so as to automatically trigger a co-processing action. The problems that data islands are serious, evidence storage lacks a safety mechanism, and an automatic closed-loop processing mechanism does not exist in an existing scheme are solved.

Description

Technical Field

[0001] This disclosure belongs to the field of intelligent security technology, specifically relating to a reliable supervision and collaborative handling method and system for security personnel. Background Technology

[0002] Security personnel management refers to providing regional security for clients through the deployment and patrolling of security personnel, combined with basic security equipment, encompassing personnel management and on-site services. Supervision involves the collection and analysis of data on personnel behavior, work processes, and service quality within the security personnel management system to identify anomalies and achieve oversight and verification. Supervision is a crucial link in the management of security personnel management service quality. Through real-time monitoring and data analysis, supervision can effectively reduce operational losses for the company (such as false attendance reports and "ghost employees"), enhance customer loyalty (by strengthening trust through rapid response and transparent service), and address compliance and risk control challenges (such as legal risks arising from increased employee awareness of their rights).

[0003] The existing supervision technology in the security and manpower protection field relies on scattered independent equipment (such as time attendance machines and surveillance cameras) and manual operation: after data collection, it is stored on local devices or simple databases, and the project leader manually summarizes the attendance sheets, monitoring records and sign-in sheets every day, and notifies the handling via telephone or SMS; the software part only provides basic data recording functions and cannot realize in-depth analysis.

[0004] This model has significant drawbacks: data silos prevent the integration of multi-source data, resulting in delayed and error-prone anomaly detection; the lack of secure data storage mechanisms makes evidence easily tampered with or lost, hindering audit trails and reliable evidence preservation; and the absence of automated action triggers for simply recording data makes it difficult to form a closed-loop business process. These shortcomings exacerbate the industry's management difficulties, unstable service quality, and weak market competitiveness. Summary of the Invention

[0005] This disclosure proposes a reliable supervision and collaborative handling scheme for security personnel, which addresses the problems of existing schemes, such as severe data silos leading to difficulties in anomaly detection, lack of security mechanisms for evidence storage resulting in difficulty and unreliability, and lack of automatic closed-loop handling mechanisms leading to business link disruptions.

[0006] The first aspect of this disclosure provides a reliable supervision and collaborative handling method for security and human defense, including: The system obtains the supervision rule configuration input by the user, generates supervision task instructions based on the configuration, and sends the instructions to the supervision terminal at the station. The supervision rule configuration includes the supervision mode and the corresponding rule parameters. The system receives personnel behavior data uploaded from the supervision terminal in response to the supervision task instruction, and performs spatiotemporal attention-based fusion analysis on the personnel behavior data to generate abnormal behavior signals. The personnel behavior data includes video streams, attendance data, and patrol data from the duty station. Based on the abnormal behavior signal, an alarm command corresponding to the alarm level is generated, and the alarm command is converted into alarm messages of different modalities and distributed to the corresponding management terminal; In response to a high-level alarm command, the personnel behavior data associated with this alarm is retrieved and integrated into a structured supervisory evidence package based on a blockchain smart contract, and distributed trusted evidence storage is performed. The system generates inspection reports and persists them to the database to automatically trigger collaborative actions, including strengthening patrols, dispatching personnel, or notifying customer units. The inspection reports record the relevant data of this inspection event, including personnel behavior data, alarm instructions, and trusted evidence for high-level alarms.

[0007] In some embodiments of this disclosure, obtaining the user-inputted supervision rule configuration and generating supervision task instructions based on the configuration includes: The system receives the supervision mode and corresponding rule parameters selected by the user through the client interface. The supervision mode includes at least one of automatic supervision with uncertain rules, temporary supervision under special circumstances, and enhanced supervision of key personnel. The rule parameters include at least one of target duty point, time range, and personnel screening conditions. Parse the rule parameters; Based on the inspection mode and rule parameters, an executable inspection task instruction is generated, wherein the instruction includes the task type, execution time, and data collection requirements.

[0008] In some embodiments of this disclosure, the generation of executable supervision task instructions based on the supervision mode and rule parameters further includes: Trigger real-time screening of target personnel based on the industry blacklist personnel database and the personnel screening conditions, wherein the personnel screening conditions include at least one of the following: employee age range, insurance status, and criminal record; If individuals who are overage, underage, uninsured, or have violated the law are identified, an early warning signal is generated, and these individuals are marked as key targets for supervision. Based on the screening results, the generation logic of supervision task instructions is linked to adjust the frequency or priority of supervision. Adjusting the frequency or priority of supervision includes increasing the random sampling rate of marked individuals or prioritizing the supervision tasks at their respective duty stations. In response to the completion signal of the handling of the marked personnel, the personnel information is entered into the industry blacklist personnel database to complete the closed-loop risk management.

[0009] In some embodiments of this disclosure, the generation of inspection task instructions further includes: The geographical location and distribution density of target outposts were determined based on a GIS map visualization system. The inspection task instruction issuance path is optimized based on the distribution density to prioritize high-density or remote areas. The inspection task instruction includes the coordinates of the station point and the optimal path sequence.

[0010] In some embodiments of this disclosure, the step of performing spatiotemporal attention-based fusion analysis on the personnel behavior data to generate abnormal behavior signals includes: Obtaining spatiotemporal sequence data from the personnel behavior data includes: From the video stream data, spatial-temporal feature segments are extracted using an I3D network to generate a video temporal feature vector; Converting discrete events into continuous time series embedding vectors includes: first, using the time step of the video time series feature vector as a benchmark to preset a time interval, performing time-series completion on the discrete events; and then using an LSTM network to convert the discrete events into discrete event time series feature vectors with the same dimension as the video time series feature vector. The discrete events include at least check-in records and patrol point sign-in. Transforming trajectory data into a continuous time series embedding vector includes: interpolating patrol trajectory data with a preset time interval based on the time step of the video time series feature vector to generate a patrol trajectory time series matrix, then inputting it into an LSTM network to learn the changing pattern of patrol position over time, and outputting a patrol trajectory time series feature vector with the same dimension as the video time series feature vector. The spatiotemporal sequence data is fused and analyzed based on spatiotemporal attention, including: using a multi-head self-attention mechanism to perform cross-modal fusion of the video temporal feature vector, discrete event temporal feature vector, and patrol trajectory temporal feature vector; calculating the spatiotemporal dependency weights between different modal data; outputting an abnormal behavior score through a fully connected neural network based on the attention-weighted fusion features; and using the Sigmoid function to generate a binary abnormal signal, wherein the binary abnormal signal is an indication signal of normal or abnormal. An online gradient descent strategy is introduced to dynamically adjust the attention weight threshold based on historical alarm feedback data in order to reduce the false alarm rate.

[0011] In some embodiments of this disclosure, the step of performing spatiotemporal attention-based fusion analysis on the personnel behavior data to generate abnormal behavior signals further includes: The personnel behavior data also includes service evaluation data from customer units, wherein the service evaluation data is collected by scanning a QR code on the client side, including service quality ratings and text feedback; When performing spatiotemporal attention-based fusion analysis on the personnel behavior data, the service evaluation data is fused to adjust the abnormal behavior scores, wherein negative evaluations lower the score threshold and positive evaluations raise the score threshold.

[0012] In some embodiments of this disclosure, the process of integrating the personnel behavior data into a structured supervisory evidence package based on blockchain smart contracts and performing distributed trusted evidence storage includes: Retrieve personnel behavior data associated with the alarm, including video clips, attendance records, and patrol logs; The hash value of the data is generated through a blockchain smart contract, and the hash value and timestamp are stored in a distributed ledger. A trusted evidence storage certificate is generated based on a digital signature algorithm, wherein the certificate includes a transaction ID, storage block height, and data integrity verification information. The trusted evidence is linked to the inspection report to ensure that the data is tamper-proof and traceable.

[0013] In some embodiments of this disclosure, the generation of abnormal behavior signals further includes: Acquire employee training completion data, wherein the training completion data includes course completion status and completion rate indicators, and the courses include at least legal and regulatory courses and safety precautions courses; For key supervisory targets marked by the real-time screening, the weight threshold of their training completion is increased so that their abnormal scores are significantly improved when they fail to complete the preset key courses; Abnormal behavior signals are corrected based on the training completion rate, wherein the abnormal behavior signals are quantified by abnormal scores, and the abnormal scores of personnel who have not completed the key courses are increased, while the scores of personnel who have completed them are decreased.

[0014] In some embodiments of this disclosure, generating inspection reports and persistently storing them in a database to automatically trigger collaborative action includes: Related project contract information, including contract expiration warnings and overdue payment warnings; For high-level alarms involving specific customer sites, retrieve the project contract warning information of the customer and store the contract warning information together with the trusted evidence in the supervision report; Special inspection tasks are automatically triggered based on contract warning information, including enhanced on-site inspections of customers with overdue payments.

[0015] A second aspect of this disclosure provides a reliable supervision and collaborative response system for security and human defense, including a server communicatively connected to a supervision terminal and a management terminal, wherein the server is also connected to a database, wherein: The database is used to store supervision rule configurations, personnel behavior data, supervision reports, and credible evidence. The inspection terminal is used to collect video streams, attendance data and patrol data at the duty station, and upload personnel behavior data in response to inspection task instructions; The management terminal is used to receive alarm messages from the server and provides a user interface for inputting the supervision rule configuration and viewing the supervision report; The server is configured as follows: The system obtains the supervision rule configuration input by the user through the management terminal, generates supervision task instructions based on the configuration, and sends the instructions to the supervision terminal at the station. The supervision rule configuration includes the supervision mode and the corresponding rule parameters. The system receives personnel behavior data uploaded from the supervision terminal in response to the supervision task instruction, and performs spatiotemporal attention-based fusion analysis on the personnel behavior data to generate abnormal behavior signals. The personnel behavior data includes video streams, attendance data, and patrol data from the duty station. Based on the abnormal behavior signal, an alarm command corresponding to the alarm level is generated, and the alarm command is converted into alarm messages of different modalities and distributed to the corresponding management terminal; In response to a high-level alarm command, the personnel behavior data associated with this alarm is retrieved and integrated into a structured supervisory evidence package based on a blockchain smart contract, and distributed trusted evidence storage is performed. The system generates inspection reports and persists them to the database to automatically trigger collaborative actions, including strengthening patrols, dispatching personnel, or notifying customer units. The inspection reports record the relevant data of this inspection event, including personnel behavior data, alarm instructions, and trusted evidence for high-level alarms.

[0016] In summary, the trusted supervision and collaborative handling methods and systems for security personnel provided in the embodiments of this disclosure achieve fusion analysis of multi-source, multi-modal supervision data by integrating video streams, attendance, and patrol data based on a spatiotemporal attention mechanism. Furthermore, they generate abnormal behavior signals in real time using AI algorithms, improving the accuracy and speed of abnormal behavior detection. Simultaneously, they integrate supervision data into structured evidence packages based on blockchain smart contracts, performing distributed evidence storage to ensure data immutability and traceability. Finally, they automatically trigger supervision tasks based on the generated supervision reports, forming a closed-loop business process. Attached Figure Description

[0017] The features and advantages of this disclosure will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the scope of this disclosure in any way. Figure 1This is a schematic diagram of a reliable supervision and collaborative handling system for security and human defense, as shown in some embodiments of this disclosure; Figure 2 This is a flowchart illustrating a reliable supervision and collaborative handling method for security personnel and defense, based on some embodiments of this disclosure; Figure 3 This is an example of a supervision rule generated by the client on the client side in one embodiment of this disclosure; Figure 4 This is an example of a training management subsystem in one embodiment of this disclosure; Figure 5 This is an example of a contract management subsystem in one embodiment of this disclosure. Detailed Implementation

[0018] In the following detailed description, numerous specific details of this disclosure are illustrated by way of example in order to provide a thorough understanding of the relevant disclosure. However, it will be apparent to those skilled in the art that this disclosure can be practiced without these details. It should be understood that the terms “system,” “apparatus,” “unit,” and / or “module” used in this disclosure are a way of distinguishing different parts, components, sections, or parts at different levels in a sequential arrangement. However, these terms may be replaced with other expressions if they can achieve the same purpose.

[0019] It should be understood that when a device, unit, or module is referred to as being "on," "connected to," or "coupled to" another device, unit, or module, it may be directly connected to or coupled to, or communicate with, other devices, units, or modules, or there may be intermediate devices, units, or modules present, unless the context explicitly indicates otherwise. For example, the term "and / or" as used in this disclosure includes any one and all combinations of one or more of the associated listed items.

[0020] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to limit the scope of this disclosure. As shown in this specification and claims, unless the context clearly indicates otherwise, words such as "a," "an," "an," and / or "the" do not specifically refer to the singular and may include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified features, integrals, steps, operations, elements, and / or components, and such expressions do not constitute an exclusive list, in which other features, integrals, steps, operations, elements, and / or components may also be included.

[0021] Referring to the following description and accompanying drawings, these and other features and characteristics, operating methods, functions of related structural elements, combinations of parts, and economics of manufacture of this disclosure can be better understood, wherein the description and drawings form part of the specification. However, it is clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this disclosure. It is understood that the drawings are not drawn to scale.

[0022] Security personnel management refers to the core security service sector that provides regional security for clients such as enterprises, residential communities, commercial complexes, and government agencies through the on-site deployment, regular patrols, fixed-point duty, security inspections, and incident response of professional security personnel, combined with basic security equipment (such as time clocks and surveillance cameras). Its core business processes include: personnel management (recruitment and deployment, pre-job training, and scheduling), on-site services (stationed duty, patrol along designated lines, and initial handling of anomalies), data recording (attendance clocking, patrol sign-in, and incident registration), and client liaison (demand response and service feedback collection). Currently, the security and manpower protection industry still relies heavily on traditional models, with a generally low level of digitalization: personnel scheduling depends on manual shift schedules, service process records rely on paper ledgers or simple spreadsheets, and customer demand responses depend on telephone communication. This results in overall low service efficiency (such as delayed shift adjustments and easy loss of records) and weak risk management capabilities (such as the inability to monitor personnel on-duty status in real time and untimely handling of abnormal events), making it difficult to meet clients' modern demands for "traceable security and verifiable services." "Supervision," as a key component of security and manpower protection management, is precisely the core means to solve these problems and achieve service quality control.

[0023] "Supervision and inspection" refers to the management activities undertaken by security service companies or client units to monitor and verify the service quality and work standards of security personnel through processes such as formulating management rules, collecting service process data, analyzing personnel behavior compliance, and identifying abnormal events. It focuses on three core dimensions: "personnel behavior monitoring, service data collection, and abnormal event analysis," specifically including: Behavioral monitoring: Verify whether security personnel have engaged in any violations such as leaving their posts, neglecting their duties, deviating from patrol routes, or abandoning their posts without authorization; Data collection: Collect security personnel's attendance records, patrol check-in times and locations, abnormal event registration information (such as strangers breaking in, equipment malfunctions), and customer feedback data; Anomaly analysis: Through data comparison and behavior verification, we identify problems such as "false attendance reports", "patrol fraud", and "failure to report abnormal events" to ensure that services meet enterprise management standards and customer needs.

[0024] Supervision is a key link in the quality of security personnel management services. Through real-time monitoring and data analysis, supervision can effectively reduce business losses (such as false attendance reports and "ghost employees"), enhance customer loyalty (by strengthening trust through rapid response and transparent services), and address compliance and risk control challenges (such as legal risks arising from increased employee awareness of their rights).

[0025] Currently, the supervision technology in the security and manpower protection field has not yet formed a systematic solution, and mainly relies on a distributed model of "isolated equipment + manual processing". A typical solution is as follows: Data acquisition layer: Data acquisition is achieved using independent functional devices, including: Attendance equipment: Collects security personnel's attendance data by using fingerprint card readers, IC card card readers, etc. The data is stored only locally on the device or on a simple local area network server; Video surveillance tools: Analog cameras or basic network cameras are installed at key locations on duty points or patrol routes to capture video streams in real time. The video data is stored on a local hard disk recorder (NVR) and requires manual login to view. Patrol recording tools: Patrol routes are recorded via paper patrol sign-in sheets (with sign-in boxes set up at patrol points), a simple mobile APP (with manual signing-in button), or GPS or Bluetooth beacons. The data is archived only as "completion proof" and has no real-time analysis function.

[0026] Data processing layer: Data compilation: Every day, the project leader collects the Excel attendance sheets exported from the attendance devices, the surveillance video spot check records, the paper patrol sign-in sheets, etc., and manually organizes them into a "daily supervision log". Anomaly detection: The presence of anomalies is determined by "manual comparison" (e.g., if the attendance sheet shows "normal check-in" but no personnel appear in the monitoring, it is determined to be "proxy check-in"; if the patrol sign-in sheet shows "complete sign-in" but the route and time logic is contradictory, it is determined to be "patrol fraud"). Results feedback: The results of the anomaly assessment are communicated to the relevant personnel via telephone or WeChat, and the handling is recorded in a paper log. There is no automated feedback or early warning mechanism.

[0027] Application layer: Provides only basic data recording functions, with no systematic application output. Enterprise management needs to manually review the "daily supervision log" of each project to understand the overall service situation, and cannot obtain cross-project supervision data in real time; When providing feedback on service quality to clients, we can only provide basic data such as "monthly attendance summary" and "patrol count statistics", and cannot provide in-depth information such as the handling process of abnormal events and video evidence; Without data correlation analysis capabilities, it is impossible to comprehensively evaluate service quality through "attendance data + patrol data + customer feedback", and it is also impossible to trigger subsequent management actions (such as personnel reassignment or enhanced training) based on data.

[0028] The above plan has fundamental flaws, directly limiting the effectiveness of security personnel supervision: The problem of data silos is prominent, and multi-source data cannot be integrated and analyzed. In existing technologies, attendance data (clock-in machine), video data (NVR), and patrol trajectory data (sign-in sheets / APP / GPS trajectory data) are stored on separate devices or systems, lacking a unified data interaction interface and fusion analysis mechanism. Attendance data only records "whether clocked in," without linking it to "whether the person was on duty after clocking in" (manual video comparison is required). Patrol sign-in data only records "whether sign-in was performed," without linking it to "whether the check was completed according to regulations during sign-in" (no video or customer feedback to corroborate this). This "data isolation" leads to delayed anomaly detection (e.g., proxy clocking-in can only be detected by manual comparison the next day) and a high statistical error rate (e.g., missing a segment of surveillance footage leading to missed anomalies). Local storage lacks reliable guarantees, and evidence is easily tampered with or lost. Current technology employs a "local storage" model for supervisory data: attendance data is stored on the project's local server, video data is stored on the NVR hard drive, and paper patrol records are stored in the project office. This storage method presents two major risks: First, it is easily tampered with. Internal personnel can destroy or alter supervisory evidence by modifying server data (such as deleting proxy attendance records), formatting the NVR hard drive (destroying violation videos), or altering paper records (covering up patrol fraud). Second, it is easily lost. Local server failures, NVR hard drive damage, and lost paper records can all render evidence untraceable. In the event of customer complaints (such as questioning "security guards leaving their posts leading to theft") or employee rights protection (such as denying "violation judgments"), the company cannot provide tamper-proof and valid evidence. Therefore, "credible and traceable evidence" cannot be achieved.

[0029] Limited to data recording only, without automatic action triggering or business loop. The core function of existing technologies is limited to "data recording": attendance records are only used as a basis for payroll calculation, patrol records are only used as proof of service completion, and anomaly records are only used as a reference for penalties for violations, failing to form a closed business loop of "data-analysis-action-optimization." On the one hand, it cannot automatically trigger response actions (e.g., when a high-level anomaly is discovered, manual notification of personnel dispatch is required, and "reinforcement of personnel" instructions cannot be automatically generated); on the other hand, it cannot provide intelligent decision support (e.g., it cannot automatically suggest "increasing the patrol frequency in a certain area" based on "frequent anomalies in a certain area," requiring manual judgment based on experience). This "recording without action" model results in supervision remaining only at the "post-event verification" level, failing to achieve full-process management of "prevention before the event, handling during the event, and optimization after the event," and failing to meet the "full-process digital management" needs of the security and civil defense industry.

[0030] To address the aforementioned issues, this disclosure proposes a reliable supervision and collaborative response solution for security personnel. It integrates multi-source, multi-modal supervision data, including video, attendance, and patrol data, through a spatiotemporal attention mechanism. Based on AI analysis, it automatically and in real-time generates anomaly signals. Blockchain-based storage of structured evidence packages ensures the credibility and traceability of evidence. Supervision reports automatically trigger collaborative responses. This solution adapts to the supervision mode selected by the client, achieving accurate anomaly detection, traceable evidence, and a closed-loop business process, thus perfectly resolving the pain points of the security personnel industry.

[0031] Figure 1 This is a schematic diagram of a reliable supervision and collaborative response system for security and human defense, as shown in some embodiments of this disclosure. Figure 1 As shown, the system includes servers that are communicatively connected to both the inspection terminal and the management terminal, and the servers are also connected to a database. Wherein: The database is used to store supervision rule configurations, personnel behavior data, supervision reports, and trusted evidence. The database server can be any of a single machine, a cluster, or a distributed database.

[0032] The supervision terminal includes time clocks, video surveillance equipment, patrol recording equipment, etc., installed at the duty station. It is used to collect personnel behavior data at the duty station and upload the personnel behavior data in response to supervision task instructions as supervision data. The personnel behavior data includes at least the video stream, attendance data, and patrol data from the duty station.

[0033] The management terminal is used to receive alarm messages from the server and provides a user interface for inputting supervision rule configurations and viewing supervision reports. The supervision rule configuration includes supervision modes and corresponding rule parameters. The supervision modes include automatic supervision with undefined rules, temporary supervision for special circumstances, and enhanced supervision of key personnel. In some embodiments of this disclosure, the management terminal can be a smartphone, mobile terminal, or personal computer.

[0034] The server is configured as follows: The system obtains the supervision rule configuration input by the user through the management terminal, generates supervision task instructions based on the configuration, and sends the instructions to the supervision terminal at the station. The supervision rule configuration includes the supervision mode and the corresponding rule parameters. The system receives personnel behavior data uploaded from the supervision terminal in response to the supervision task instruction, and performs spatiotemporal attention-based fusion analysis on the personnel behavior data to generate abnormal behavior signals. The personnel behavior data includes video streams, attendance data, and patrol data from the duty station. Based on the abnormal behavior signal, an alarm command corresponding to the alarm level is generated, and the alarm command is converted into alarm messages of different modalities and distributed to the corresponding management terminal; In response to a high-level alarm command, the personnel behavior data associated with this alarm is retrieved and integrated into a structured supervisory evidence package based on a blockchain smart contract, and distributed trusted evidence storage is performed. The system generates inspection reports and persists them to the database to automatically trigger collaborative actions, including strengthening patrols, dispatching personnel, or notifying customer units. The inspection reports record the relevant data of this inspection event, including personnel behavior data, alarm instructions, and trusted evidence for high-level alarms.

[0035] The server can be a single machine, a cluster, or a distributed server.

[0036] Figure 2 This is a flowchart illustrating a reliable supervision and collaborative handling method for security and human defense, according to some embodiments of this disclosure. In some embodiments, the reliable supervision and collaborative handling method for security and human defense is comprised of... Figure 1 The server in the system shown executes the following steps in the trusted supervision and collaborative handling method for security and human defense: S210, obtain the supervision rule configuration input by the user, generate a supervision task instruction based on the configuration, and send the instruction to the supervision terminal at the station. The supervision rule configuration includes the supervision mode and the corresponding rule parameters.

[0037] Specifically, users configure supervision rules through a visual interface provided by the management terminal, including supervision modes and corresponding rule parameters. The supervision modes include: Unpredictable automatic supervision: The system intelligently generates random inspection tasks based on historical data to prevent personnel from knowing the patterns; Temporary inspections for special circumstances: Generate one-time inspection instructions in response to emergencies or temporary customer requests; Strengthen supervision of key personnel: Increase the frequency of supervision for high-risk personnel identified through screening (such as those who are overage, uninsured, or have criminal records).

[0038] The rule parameters include the target duty location, time range, and personnel screening criteria. The personnel screening criteria include the employee's age range, insurance status, and criminal record.

[0039] Figure 3 This is an example of a supervisory rule generated by a client in a management client in one embodiment of this disclosure.

[0040] After parsing the rule parameters, the system generates a supervision instruction based on the supervision mode, which includes the task type, execution time, and data collection requirements, and sends it to the supervision terminal.

[0041] In response to the pain points in the security and civil defense industry, such as low efficiency in issuing inspection tasks, unreasonable resource allocation, and delayed response to high-risk or remote areas due to the geographically dispersed and unevenly distributed project outposts, some embodiments of this disclosure integrate GIS (Geographic Information System) map visualization technology to intelligently optimize the issuance path of inspection task instructions, ensuring that inspection forces can prioritize and quickly cover key areas, thereby achieving accurate deployment and efficient utilization of inspection resources.

[0042] Specifically, it includes: First, the system automatically analyzes the spatial distribution density (e.g., the number of service points per unit area) and geographical attributes (e.g., whether they are located in remote suburbs or whether transportation is convenient) of these service points and automatically classifies them according to region, such as identifying "high-density clusters in downtown commercial areas", "medium-density strip distributions in development zones" and "low-density scattered locations in remote suburban counties".

[0043] Then, based on the built-in path planning algorithm (such as Dijkstra's algorithm, or an optimization algorithm for the multi-point traveling salesman problem), the sequence of issuing inspection task instructions is optimized according to the above density and geographical attribute analysis results.

[0044] For high-density areas: the algorithm generates the shortest patrol and inspection path covering all points in the area, ensuring that inspectors can complete the inspection tasks of multiple adjacent duty points in the area with the shortest travel time, greatly improving the inspection efficiency per unit time.

[0045] For remote and scattered areas: the algorithm will consider the isolation and accessibility of their geographical location, prioritize the generation of supervision instructions for them, and may combine them with several adjacent points into an optimal sequence to ensure that these easily overlooked areas can be covered in a timely manner and avoid management blind spots.

[0046] Ultimately, the optimized inspection task instructions not only include the specific inspection content, but also the precise coordinates of each target outpost and a recommended optimal path sequence, guiding inspectors to execute tasks sequentially and efficiently.

[0047] S220, receiving personnel behavior data uploaded from the supervision terminal in response to the supervision task instruction, and performing spatiotemporal attention-based fusion analysis on the personnel behavior data to generate abnormal behavior signals, wherein the personnel behavior data includes video streams, attendance data and patrol data from the duty station.

[0048] The core of this step lies in addressing the problems of delayed anomaly detection and high false alarm rates caused by "data silos" in traditional supervision. After receiving multi-source personnel behavior data uploaded from supervision terminals, the system does not perform independent or simple rule-based judgments. Instead, it employs a deep learning-based spatiotemporal attention fusion analysis model to achieve accurate cross-modal anomaly behavior perception. The specific implementation process is as follows: Multi-source data preprocessing and time series feature extraction The system first preprocesses the heterogeneous multimodal data, unifying it into temporal feature vectors suitable for deep learning models. This forms the basis for subsequent fusion analysis. Specifically, this includes: Video stream feature extraction: The system extracts spatial-temporal features from the received video stream data frame by frame. Specifically, the I3D (Inflated3D ConvNet) network is used as the core algorithm. This network can simultaneously capture spatial information within video frames (such as person posture and objects) and temporal information between consecutive frames (such as motion trajectory and action sequence), outputting a video temporal feature vector containing rich spatiotemporal semantics.

[0049] Discrete event serialization: Events such as attendance tracking and patrol point check-in are recorded as isolated discrete points in traditional systems. To address the timing mismatch between these events and the video stream, this system implements timing alignment and embedding. Temporal completion: Based on the time step of the aforementioned video temporal feature vector, the same time interval is preset. For time steps where no event occurs, the vector is filled with zero or a specific identifier vector.

[0050] Feature embedding: The completed discrete event sequence is input into a Long Short-Term Memory (LSTM) network. LSTM networks excel at learning long-term dependencies in time series. They can transform isolated check-in / sign-in events into discrete event time series feature vectors that reflect their temporal regularity and pattern, and whose dimensions are consistent with those of the video time series feature vectors.

[0051] Track data serialization: The patrol track data generated by GPS or Bluetooth beacons is also time-aligned. Trajectory interpolation: Based on the video stream time, the trajectory points are interpolated to generate a continuous patrol trajectory time series matrix with uniform time steps.

[0052] Trajectory feature learning: Input the matrix into the LSTM network to learn the changes in patrol position, speed and direction over time, and output a patrol trajectory temporal feature vector that can characterize whether the patrol behavior is normal or not.

[0053] Thus, the originally heterogeneous multi-source data is transformed into feature vectors that can be jointly analyzed within the same spatiotemporal dimension.

[0054] Cross-modal fusion and anomaly scoring based on spatiotemporal attention The system employs a multi-head self-attention mechanism to deeply fuse the three feature vectors mentioned above.

[0055] Cross-modal fusion: The model no longer treats video, discrete events, and trajectory data as independent components, but instead calculates the correlation weights between the three through a self-attention mechanism. For example, the model can learn cross-modal contradictions such as "at a certain point in time, the video shows a person located at point A, but the attendance record at this time comes from a distant point B," which cannot be detected by single-modal analysis.

[0056] Spatiotemporal dependency modeling: Attention mechanisms can capture the dependencies between different time steps and different features, thereby understanding the context of behavior. For example, it can identify whether "frequent clocking in within a short period of time" is a normal shift change behavior or an abnormal "proxy clocking in" behavior.

[0057] Anomaly score output: The attention-weighted fused features are fed into a fully connected neural network, ultimately outputting an anomaly behavior score. This score is mapped to a value between 0 and 1 using the Sigmoid function, thereby generating a binary anomaly signal (e.g., a score greater than 0.85 is considered "abnormal," otherwise it is considered "normal"), providing a precise and quantitative basis for subsequent alarm level determination.

[0058] Dynamic optimization and false alarm suppression To continuously improve system accuracy, the model incorporates an online gradient descent strategy. The system continuously collects feedback from administrators on historical alerts (such as "false alarm confirmed" or "confirmed true") and uses this feedback data to dynamically adjust the threshold parameters in the attention weights. This allows the model to learn continuously, becoming increasingly accurate in identifying genuine anomalies, effectively reducing the false alarm rate and minimizing unnecessary interference for administrators.

[0059] In one embodiment of this disclosure, the system dynamically adjusts the weight of anomaly scores based on employee training completion data, including: Before generating an abnormal signal, the system will obtain the employee's training completion data (such as the completion status and score of key courses such as laws and regulations and safety precautions) and use it as an important correction factor.

[0060] Adjustment of weight for key targets: For "key supervision targets" marked by real-time screening (such as over-age or uninsured personnel), the system will automatically increase the weight threshold of their training completion in the abnormal score.

[0061] Score Adjustment: Before fusion analysis, the model pre-adjusts the baseline scores based on training data. Individuals who have not completed the preset key courses will have significantly higher initial abnormal scores; while those who have completed them with excellent results will have their initial scores lowered accordingly. This allows the model to make prior judgments about compliance before analyzing behavioral data, making risk assessment more proactive.

[0062] The proposed implementation aims to enhance team skills and team atmosphere by incorporating training factors into the anomaly scoring model and thereby strengthening training. Figure 4 This is an example of the training management subsystem described in the embodiments.

[0063] To make anomaly detection more closely reflect actual customer perception, some other embodiments of this disclosure dynamically adjust the anomaly scoring threshold by integrating service evaluation data. Specifically, this includes: The system integrates service evaluation data from client organizations. This data, submitted by customers via QR code scanning, includes service quality ratings and text feedback.

[0064] Before the model finally uses the Sigmoid function to generate a binary anomaly signal, the scoring threshold will be dynamically adjusted based on the service evaluation data.

[0065] Negative feedback (such as low scores, complaint texts): The system will lower the scoring threshold, making it easier for the duty station or personnel to be judged as "abnormal", thereby triggering stricter supervision.

[0066] Positive feedback (such as high scores and praise texts): The system will raise the scoring threshold, reduce false alarms, and avoid unnecessary interference with outstanding individuals.

[0067] By adjusting the anomaly detection model based on "customer satisfaction," the anomaly detection standards are no longer mechanical but can flexibly fluctuate according to customer feedback, thus achieving a closed-loop linkage between management and service.

[0068] S230, generate an alarm command corresponding to the alarm level based on the abnormal behavior signal, and convert the alarm command into alarm messages of different modes and distribute them to the corresponding management terminal.

[0069] Specifically, the system does not simply throw out an "abnormal" signal directly. Instead, it first generates an alarm command with a corresponding alarm level based on the abnormal behavior score, the type of abnormality, and preset business rules. In one embodiment of this disclosure, the alarm levels are as follows: High-level alerts: These are for behaviors that may pose significant safety or operational risks, such as "confirmed absence from duty," "serious deviation from patrol route," or "suspected proxy clocking in." These alerts require immediate attention.

[0070] Medium-level alerts: These are for potential risk behaviors such as "minor delays at patrol points" or "persistently low service ratings." These alerts require attention and appropriate handling.

[0071] Low-level alarms: These are for minor compliance issues such as "untidy work clothes" or "brief stay in the area". These alarms are typically used for logging and reminders.

[0072] To address the issues of low information reach and slow feedback, the system automatically converts the same alarm command into multiple message modalities to adapt to different administrators' receiving habits and application scenarios, ensuring that critical information is delivered. These modalities include SMS notifications, app push notifications, voice broadcasts, and prominent display on visual dashboards.

[0073] S240, in response to a high-level alarm command, retrieves the personnel behavior data associated with this alarm and integrates the personnel behavior data into a structured supervisory evidence package based on a blockchain smart contract, and performs distributed trusted evidence storage.

[0074] Addressing the industry pain points of easily tampered or lost evidence, and the difficulty in achieving audit trails and credible evidence preservation, some embodiments of this disclosure utilize blockchain technology to achieve tamper-proof, traceable, and verifiable supervisory evidence. In the event of a labor dispute (such as an employee denying wrongdoing) or a customer complaint (such as questioning service quality), the company can present this certificate at any time. Any third-party verification agency can query the original evidence record on the blockchain using the information on the certificate and compare the hash value calculated at that time with the current data. This allows for efficient and low-cost self-proofing, greatly enhancing the authority and credibility of management.

[0075] The scheme includes: High-level alarms (such as "confirmed absence from duty" or "patrol fraud") automatically trigger the evidence association process. Based on the alarm's time, location (duty point), and associated personnel ID, the system automatically retrieves original personnel behavior data strongly correlated with the alarm from the database and storage system. This data constitutes the material for the evidence package, mainly including: Video clips: Raw video streams or key segments of surveillance video taken during a specific time period before and after an alarm.

[0076] Attendance records: detailed logs of the time and location (GPS coordinates or attendance machine ID) of each person.

[0077] Patrol log: Data such as personnel's patrol routes, GPS tracks, and patrol point check-in timestamps.

[0078] The retrieved raw data comes in various formats and is stored in a scattered manner. The system uses blockchain smart contracts to execute a standard evidence integration and solidification process. Data standardization: The smart contract encapsulates the received multi-source data according to a predefined structure (such as JSON format) to generate a structured supervisory evidence package. This evidence package contains the evidence data itself and its metadata (such as alarm ID, generation time, and associated objects).

[0079] Generating a digital fingerprint: The smart contract calls a hash algorithm (such as SHA-256) to calculate a unique, fixed-length hash value for the structured evidence package. This hash value acts as the "digital fingerprint" of the evidence package; any minor alteration to the original data will cause the hash value to change completely, thus verifying data integrity.

[0080] Distributed trusted evidence storage: The generated hash value and its key metadata (such as timestamps) are not stored on local servers, but are stored in a distributed ledger through a blockchain network.

[0081] The transaction is broadcast to multiple nodes in the blockchain network, and after being confirmed by the consensus mechanism, it is packaged into a new block.

[0082] Once the record is successfully made, the "fingerprint" of the evidence and the time of its preservation are permanently fixed and cannot be tampered with or deleted by any single entity (including the system operator).

[0083] Generate trusted evidence storage credentials: After successful notarization, the smart contract will generate a trusted notarization certificate based on a digital signature algorithm. This certificate serves as electronic proof of the legality and validity of the notarization activity and typically includes: Transaction ID (TxID): The unique transaction hash of this notarization operation on the blockchain, used to query the notarization record.

[0084] Storage block height: The block number where the evidence-based transaction is located, accurately pinpointing the evidence-based storage time.

[0085] Data integrity verification information: that is, the hash value of the original data, which is used for subsequent comparison and verification.

[0086] This certificate is usually generated in the form of a QR code or a short link for quick verification.

[0087] S250, generate an inspection report and persist it to the database to automatically trigger collaborative handling actions, including strengthening patrols, dispatching personnel, or notifying customer units. The inspection report records the relevant data of this inspection event, including personnel behavior data, alarm instructions, and trusted evidence for high-level alarms.

[0088] The core objective of this step is to transform the analysis results (anomaly signals, alarms, evidence) from previous steps into intelligent reports that can guide business actions. Specifically, this includes: The system automatically aggregates all data related to the inspection event, generates a structured inspection report, and persistently stores it in the database. The report must contain a complete chain of evidence. In some embodiments of this disclosure, this includes: core event data, analysis process data, credible evidence, and handling recommendations. Persistent storage is used to prevent the inspection report from being lost due to unforeseen circumstances or human error.

[0089] The generation and storage of reports automatically trigger preset collaborative actions, achieving a closed business loop. These actions intelligently vary depending on the alarm level and contract status, including: Strengthen patrols: Automatically generate new inspection tasks to increase the frequency of spot checks at the station in the following period.

[0090] Dispatcher: Notify the human resources department and suggest training, interviews, or reassignment of the personnel involved.

[0091] Notifying clients: For alarms related to service quality, the system can generate formatted notifications, which, after review, are sent to the client, reflecting transparent management.

[0092] In some embodiments of this disclosure, if a high-level alarm occurs during the inspection, the inspection report must include the high-level alarm, the corresponding trusted evidence, and the handling measures for the high-level alarm.

[0093] In some embodiments of this disclosure, the system automatically associates project contract information when generating reports, linking financial risks with security service quality for integrated management. This includes: For high-level alerts involving specific customer sites, the system will automatically retrieve the customer's project contract warning information through the customer unit ID, including: overdue payment warnings, contract expiration warnings, etc.

[0094] The system stores these contract warnings along with the credible evidence for this alert in the supervisory report. This allows a single report to simultaneously reflect both "service quality issues" and "customer financial risks," thus providing a new perspective for management decisions.

[0095] The system will automatically trigger special inspection tasks based on contract warning information. For example: When a late payment warning is detected, the system automatically generates an "enhanced patrol" task for all on-site locations of that customer. This is not only for security purposes, but also as a digital risk control and collection tool, maintaining customer relationships and urging payment by enhancing visible service value.

[0096] When a contract expiration warning is detected, the system can highlight it in the report and remind sales or management to prioritize the renewal of the customer's contract, while ensuring that service quality does not decline at the end of the contract period and guaranteeing the success rate of renewal.

[0097] One embodiment of this disclosure establishes an integrated security and human resources management system based on the linkage between supervision and contract management. Figure 5 This is an example of the contract management subsystem in the management system.

[0098] In summary, the trusted supervision and collaborative handling methods and systems for security personnel provided in the embodiments of this disclosure achieve fusion analysis of multi-source, multi-modal supervision data by integrating video streams, attendance, and patrol data based on a spatiotemporal attention mechanism. Furthermore, they generate abnormal behavior signals in real time using AI algorithms, improving the accuracy and speed of abnormal behavior detection. Simultaneously, they integrate supervision data into structured evidence packages based on blockchain smart contracts, performing distributed evidence storage to ensure data immutability and traceability. Finally, they automatically trigger supervision tasks based on the generated supervision reports, forming a closed-loop business process.

[0099] Although the subject matter described herein is provided in the general context of execution on a computer system in conjunction with an operating system and applications, those skilled in the art will recognize that other implementations can also be executed in conjunction with other types of program modules. Generally, program modules include routines, programs, components, data structures, and other types of structures that perform specific tasks or implement specific abstract data types. Those skilled in the art will understand that the subject matter described herein can be practiced using other computer system configurations, including handheld devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, minicomputers, mainframes, etc., and can also be used in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules may reside on both local and remote memory storage devices.

[0100] Those skilled in the art will recognize that the units and method 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 disclosure.

[0101] It should be understood that the specific embodiments described above are merely illustrative or explanatory of the principles of this disclosure and do not constitute a limitation thereof. Therefore, any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of this disclosure should be included within the protection scope of this disclosure. Furthermore, the appended claims are intended to cover all variations and modifications falling within the scope and boundaries of the appended claims, or equivalent forms of such scope and boundaries.

Claims

1. A security-oriented trusted supervision and collaborative handling method, characterized in that, The method comprises: obtaining user inputted supervision rule configuration, generating supervision task instruction based on the configuration, and issuing the instruction to the supervision terminal of the on-duty point, wherein the supervision rule configuration comprises supervision mode and corresponding rule parameters; receiving personnel behavior data uploaded from the supervision terminal in response to the supervision task instruction, and performing fusion analysis on the personnel behavior data based on space-time attention to generate abnormal behavior signal, wherein the personnel behavior data comprises video stream, attendance data and patrol data of the on-duty point; generating alarm instruction of corresponding alarm level according to the abnormal behavior signal, and converting the alarm instruction into alarm message of different modalities and distributing it to the corresponding management terminal; in response to high-level alarm instruction, calling the personnel behavior data associated with this alarm and integrating the personnel behavior data into structured supervision evidence package based on the blockchain smart contract to perform distributed trusted notarization; generating supervision report and persistently storing it to the database to automatically trigger collaborative disposal actions, including strengthening patrol, dispatching personnel or notifying customer units, wherein the supervision report records the associated data of this supervision event, including personnel behavior data, alarm instruction and trusted notarization certificate for high-level alarm.

2. The method of claim 1, wherein, The method of obtaining user inputted supervision rule configuration and generating supervision task instruction based on the configuration comprises: receiving supervision mode and corresponding rule parameters selected by the user through the client interface, the supervision mode comprising at least one of irregular rule automatic supervision, temporary supervision in special circumstances and intensive supervision of key personnel, and the rule parameters comprising at least one of target on-duty point, time range and personnel screening condition; analyzing the rule parameters; generating executable supervision task instruction based on the supervision mode and rule parameters, wherein the instruction contains task type, execution time and data acquisition requirement.

3. The method of claim 2, wherein, The method of generating executable supervision task instruction based on the supervision mode and rule parameters further comprises: triggering real-time screening of target personnel based on industry blacklist personnel database and the personnel screening condition, wherein the personnel screening condition comprises at least one of employee age range, insurance status and criminal record; if overage, underage, uninsured or illegal personnel are identified, generating a warning signal, marking the personnel as key supervision objects, and associating them to the generation logic of the supervision task instruction based on the screening result for adjusting supervision frequency or priority, wherein the adjusting supervision frequency or priority comprises increasing the random check rate of the marked personnel or prioritizing the supervision task of the on-duty point where they are located; in response to the disposal completion signal of the marked personnel, entering the personnel information into the industry blacklist personnel database to complete the risk closed-loop management.

4. The method of claim 3, wherein, The method of generating supervision task instruction further comprises: determining the geographical location and distribution density of the target on-duty point based on the GIS map visualization system; optimizing the supervision task instruction issuing path according to the distribution density to preferentially process high-density areas or remote areas, wherein the on-duty point coordinates and optimal path sequence are contained in the supervision task instruction.

5. The method of claim 1, wherein, The spatio-temporal attention-based fusion analysis on the personnel behavior data to generate the abnormal behavior signal comprises: obtaining spatio-temporal sequence data from the personnel behavior data, comprising: extracting a space-time feature fragment from the video stream data through an I3D network to generate a video time sequence feature vector; converting discrete events into continuous time sequence embedding vectors, comprising: first, taking the time step of the video time sequence feature vector as the basis to preset a time interval, and performing time sequence completion on the discrete events, and then converting the discrete events into a discrete event time sequence feature vector consistent with the dimension of the video time sequence feature vector through an LSTM network, wherein the discrete events at least include punch card records and patrol point check-ins; converting trajectory data into continuous time sequence embedding vectors, comprising: interpolating patrol trajectory data to generate a patrol trajectory time sequence matrix with the time step of the video time sequence feature vector as the basis to preset a time interval, and then inputting the patrol trajectory time sequence matrix into an LSTM network to learn the change rule of patrol positions over time, and outputting a patrol trajectory time sequence feature vector consistent with the dimension of the video time sequence feature vector; based on spatio-temporal attention, the spatio-temporal sequence data is fused and analyzed, comprising: adopting a multi-head self-attention mechanism to perform cross-modal fusion on the video time sequence feature vector, the discrete event time sequence feature vector and the patrol trajectory time sequence feature vector, calculating the spatio-temporal dependence weight between different modal data, based on the attention weighted fusion feature, outputting an abnormal behavior score through a fully connected neural network, and using a Sigmoid function to generate a binary abnormal signal, the binary abnormal signal is an indication signal of normal and abnormal; an online gradient descent strategy is introduced, and the attention weight threshold is dynamically adjusted according to the historical alarm feedback data to reduce the false alarm rate.

6. The method of claim 1, wherein, The spatio-temporal attention-based fusion analysis on the personnel behavior data to generate the abnormal behavior signal further comprises: The personnel behavior data further comprises service evaluation data from the client unit, wherein the service evaluation data is collected by the client through scanning the code, including service quality score and text feedback; When the personnel behavior data is fused and analyzed based on spatio-temporal attention, the service evaluation data is fused to adjust the abnormal behavior score, wherein negative evaluation reduces the score threshold and positive evaluation increases the score threshold.

7. The method of claim 1, wherein, The personnel behavior data is integrated into a structured supervision evidence package based on a blockchain smart contract, and the distributed trusted storage is executed, comprising: calling the personnel behavior data associated with the alarm, including video clips, attendance records and patrol logs; generating a hash value of the data through a blockchain smart contract, and storing the hash value and timestamp in a distributed ledger; generating a trusted storage certificate based on a digital signature algorithm, wherein the certificate includes transaction ID, storage block height and data integrity verification information; associate the trusted storage certificate to the supervision report to ensure that the data is tamper-proof and traceable.

8. The method of claim 1, wherein, The generation of the abnormal behavior signal further comprises: obtaining the training completion data of the employees, wherein the training completion data includes course completion status and completion degree index, and the courses at least include legal and regulatory courses and safety prevention courses; For the key supervision object marked by the real-time screening, the weight threshold of the training completion degree is increased so that the abnormal score is significantly increased when the preset key course is not completed; The abnormal behavior signal is corrected based on the training completion degree, wherein the abnormal behavior signal is quantified by an abnormal score, and the abnormal score of the personnel who have not completed the key course is increased, and the score of the personnel who have completed the key course is decreased.

9. The method of claim 1, wherein, The supervision report is generated and persistently stored in the database to automatically trigger the collaborative disposal action, including: Associating project contract information, including contract expiration warning and overdue payment warning; For high-level alarms involving specific customer sites, the project contract warning information of the customer is called, and the contract warning information is stored together with the trusted evidence certificate in the supervision report; Based on the contract warning information, special supervision tasks are automatically triggered, including strengthened on-site patrol for overdue payment customers.

10. A trusted supervision and collaborative disposal system for security and civil defense, comprising a server in communication connection with a supervision terminal and a management terminal respectively, and the server is also in data connection with a database, wherein: The database is used to store supervision rule configurations, personnel behavior data, supervision reports and trusted evidence certificates; The supervision terminal is used to collect video streams, attendance data and patrol data of the on-site point, and upload personnel behavior data in response to supervision task instructions; The management terminal is used to receive alarm messages from the server, and provide a user interface for inputting the supervision rule configurations and viewing the supervision reports; The server is configured to: Obtain the supervision rule configurations input by the user through the management terminal, generate supervision task instructions based on the configurations, and issue the instructions to the supervision terminal of the on-site point, wherein the supervision rule configurations include supervision modes and corresponding rule parameters; Receive personnel behavior data uploaded from the supervision terminal in response to the supervision task instructions, and perform fusion analysis on the personnel behavior data based on spatio-temporal attention to generate abnormal behavior signals, wherein the personnel behavior data includes video streams, attendance data and patrol data of the on-site point; Generate alarm instructions of corresponding alarm levels according to the abnormal behavior signals, and convert the alarm instructions into alarm messages of different modalities and distribute them to the corresponding management terminal; In response to high-level alarm instructions, the personnel behavior data associated with the current alarm is called and integrated into a structured supervision evidence package based on a blockchain smart contract to perform distributed trusted evidence storage; Generate a supervision report and persistently store it in the database to automatically trigger collaborative disposal actions, including strengthened patrol, dispatch personnel or notify the customer unit, wherein the supervision report records the associated data of the current supervision event, including personnel behavior data, alarm instructions and trusted evidence certificates for high-level alarms.

Images