Student school attendance guardian method and system based on secure encrypted network and cloud edge collaboration

Abstract

The application relates to the security technology field and provides a student going-to-and-from-school guardian method and system based on a secure encryption network and cloud edge cooperation, which comprises the following steps: acquiring video data collected by multi-source edge perception devices on a student going-to-and-from-school path; determining a current computing power mode and an execution position of identity recognition between a cloud center weak computing power mode and a cloud center strong computing power mode according to whether the edge perception devices have local reasoning capability; in the system running process, switching between the two computing power modes in response to a computing power mode switching instruction, and continuously executing identity recognition, travel state determination and safe leaving school determination in the switching process; wherein data interaction between the cloud edge and the end is encrypted by using a session dynamic password mechanism. While guaranteeing the real-time, continuity and security of recognition, the accuracy and reliability of student going-to-and-from-school guardianship are improved.

Description

Technical Field

[0001] This application relates to the field of security, specifically to a method and system for protecting students going to and from school based on a secure encrypted network and cloud-edge collaboration. Background Technology

[0002] In scenarios involving student safety during their commute to and from school, traditional solutions primarily rely on manual monitoring, fixed-point patrols, and parental supervision to control security at key locations such as school gates, residential area entrances, and intersections. While some areas have experimented with using video data collected by cameras to assist in security management as video surveillance equipment becomes more widespread, existing solutions still face the following challenges in practical application.

[0003] Existing video surveillance systems typically employ a single computing power architecture, either centralized cloud processing or fixed edge processing, which cannot adaptively adapt to the actual computing power capabilities of the front-end sensing devices. When edge devices possess local inference capabilities, sending the entire video stream back to the cloud for analysis results in wasted bandwidth and delayed recognition response. Conversely, when edge devices are ordinary cameras lacking computing power, forcing them to perform algorithm inference is impossible. This mismatch between computing power allocation and the actual capabilities of the devices makes it difficult for the system to balance real-time recognition with construction costs in complex deployment environments. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for protecting students going to and from school based on a secure encrypted network and cloud-edge collaboration, so as to alleviate the above-mentioned technical problems.

[0005] Firstly, this application provides a method for protecting students going to and from school based on a secure encrypted network and cloud-edge collaboration, including: Acquire video data collected by multi-source edge sensing devices along students' routes to and from school; Based on whether the edge sensing device has local reasoning capability, the current computing power mode is determined between the weak computing power mode and the strong computing power mode of the cloud center, and the execution location for identity recognition of the video data is determined, the execution location including the edge sensing device or the cloud center. The target in the video data is identified, and the travel status and safe departure from school of the target are determined based on the identification result and the location determination result, and the student's going to and from school protection status is output. During system operation, in response to the computing power mode switching command, the current computing power mode is switched between the weak computing power mode and the strong computing power mode of the cloud center, and the identity recognition, the trip status determination and the safe departure determination are continuously executed during the switching process; The data interaction between the cloud, edge, and device is encrypted using a session dynamic password mechanism.

[0006] In an optional implementation, determining the current computing power mode includes: When the edge sensing device has local reasoning capability, it determines that the current computing power mode is the cloud center weak computing power mode, and performs the identity recognition at the edge and only sends back the recognition result and feature data. When the edge sensing device does not have local reasoning capabilities, it determines that the current computing power mode is the cloud center strong computing power mode, and sends the video data back to the cloud to perform the identity recognition.

[0007] In an optional implementation, the response to the computing power mode switching command includes: In response to monitoring results of device status, network quality, and computing load, and / or administrator instructions, the computing mode switching instruction is generated; After completing the switch to the current computing power mode, reconfigure the algorithm deployment location and video transmission strategy.

[0008] In an optional implementation, the position determination result is determined in the following manner: The basic location of the target is determined by the coverage area of ​​the multi-source edge sensing device; After the target's identity is verified, the target's precise location is determined based on the visual calibration relationship between the image and geographic coordinates of the multi-source edge sensing device.

[0009] In an optional implementation, determining the precise location of the target based on the visual calibration relationship between the image and geographic coordinates of the multi-source edge sensing device includes: Obtain the pixel coordinates of the target in the captured image and the pixel coordinates of the preset calibration point; The pixel offset is determined based on the pixel coordinates of the target in the captured image and the pixel coordinates of the preset calibration point. Based on the visual calibration relationship between the image and geographic coordinates of the multi-source edge sensing device, the pixel offset is converted into latitude and longitude offset; The precise location of the target is obtained by superimposing the latitude and longitude offset onto the reference latitude and longitude of the multi-source edge sensing device.

[0010] In an optional implementation, the safe departure determination includes: Within the preset school dismissal time window, the identities of the target and the guardian are confirmed through facial recognition; Based on the location determination result, determine whether the target falls into the preset off-campus area; Detect the effective travel relationship between the target and the guard within the preset off-campus area; If no valid peer relationship is detected by the end of the preset school dismissal time window, an abnormal school departure status will be output.

[0011] In an optional implementation, the effective companion relationship is determined based on the companion matching degree, which is determined based on the spatial distance and the time difference of appearance between the target and the guardian; when the spatial distance is within a preset maximum companion distance and the time difference of appearance is within a preset maximum time difference, it is determined to be a valid companion relationship.

[0012] In an optional implementation, the output of the abnormal status includes: Based on the student's comprehensive confidence score, the peer matching score, and the off-campus area matching results, the confidence score for safe departure from campus is determined. When the confidence level for safe departure from school reaches the preset threshold for safe departure from school, it is determined that the departure from school is safe. When the confidence level for safe departure from school does not reach the preset safe departure threshold, the departure is determined to be abnormal.

[0013] In an optional implementation, the session dynamic password mechanism includes: During the communication session establishment phase, a dynamic password for the communication session is generated and bound to the communication session. During the duration of the communication session, the session dynamic password is periodically updated according to a preset iteration cycle; When an abnormal network event is detected, the current session dynamic password is forcibly invalidated and a new session dynamic password is generated. The session dynamic password is immediately destroyed when the communication session ends normally or is abnormally disconnected. The session dynamic cipher is generated based on the base key, terminal device identifier, role code, and session random number; communication sessions of different access roles use independent session dynamic ciphers to isolate the business loads of different roles; the session dynamic cipher of a single role can only decrypt data within the scope of that role's permissions.

[0014] Secondly, this application provides a student commuting monitoring system based on a secure encrypted network and cloud-edge collaboration, comprising: The video data acquisition module is used to acquire video data collected by multi-source edge sensing devices along the students' commuting route to and from school. The computing power configuration module is used to determine the current computing power mode between the weak computing power mode and the strong computing power mode of the cloud center based on whether the edge sensing device has local inference capability, and to determine the execution location for identity recognition of the video data, wherein the execution location includes the edge sensing device or the cloud center. The identification and determination module is used to identify the target in the video data, and to determine the travel status and safe departure from school of the target based on the identification result and the location determination result, and output the student's school arrival and departure protection status. The computing power switching module is used to switch the current computing power mode between the weak computing power mode and the strong computing power mode of the cloud center in response to the computing power mode switching command during system operation, and to continuously perform the identity recognition, the trip status determination and the safe departure determination during the switching process; The data interaction between the cloud, edge, and device is encrypted using a session dynamic password mechanism.

[0015] Beneficial effects: This application determines the current computing power mode based on whether the edge sensing device has local inference capabilities, enabling the system to adaptively adapt to deployment environments with different computing power conditions. When the edge device has local inference capabilities, lightweight algorithms such as identity recognition are executed at the edge, only transmitting the recognition results and feature data back, reducing the bandwidth consumption of video streaming and shortening the latency from video acquisition to recognition result output. When the edge device does not have local inference capabilities, video data is transmitted back to the cloud for centralized processing, ensuring compatible access for ordinary camera devices. Through this dual-mode adaptive mechanism, the system ensures real-time recognition while making full use of existing devices and reducing construction costs.

[0016] This application uses a hierarchical trigger-based positioning mechanism to first determine the basic location of the target based on the coverage area of ​​the sensing device, and then initiate precise positioning based on visual calibration after identity recognition is passed. This ensures that high-precision positioning calculations are performed only on registered personnel whose identities have been confirmed, avoiding the waste of computing resources caused by performing precise positioning on all targets appearing in the image, and improving the system's processing efficiency in multi-target scenarios during peak school hours.

[0017] This application uses a four-dimensional joint judgment mechanism of time window constraint, regional constraint, identity recognition and peer relationship to determine safe departure from school. It comprehensively determines the confidence level of safe departure from school by combining the student's face confidence level, peer matching degree and departure area matching results. It effectively distinguishes between two situations: students leaving school accompanied by a registered guardian and students leaving school alone. This reduces the probability of missed judgment and false judgment caused by single-dimensional judgment and improves the accuracy and robustness of departure safety judgment.

[0018] This application achieves hot switching of computing power mode by responding to computing power mode switching commands during system operation, switching the current computing power mode between the weak computing power mode and the strong computing power mode of the cloud center, and continuously performing identity recognition, travel status determination and safe departure determination during the switching process. This avoids the monitoring blind spots caused by the interruption of business and restart of equipment when switching computing power architecture in traditional solutions, and ensures continuous protection capabilities during peak student commuting hours.

[0019] This application encrypts data interaction between the cloud, edge, and endpoint through a session dynamic cryptography mechanism, constructing a three-level linked dynamic cryptographic transformation system of "session initialization—periodic iteration—forced update in case of anomalies." The session dynamic cryptography is generated based on a base key, terminal device identifier, role code, and session random numbers. Communication sessions for different access roles use independent session dynamic cryptography, isolating the business payloads of different roles. All keys are destroyed upon session termination, achieving full lifecycle controllability of keys. This overcomes the shortcomings of traditional static key encryption modes, such as long-term key immutability, risk diffusion after leakage, and lack of emergency key switching in case of anomalies, ensuring the security of sensitive information such as student travel data during transmission. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the student commuting protection method based on secure encrypted network and cloud-edge collaboration according to an embodiment of this application; Figure 2 This is a diagram of a 3D visualization monitoring interface for nursing and school scenarios according to an embodiment of this application; Figure 3 This is a diagram of the multi-angle face information collection interface according to an embodiment of this application; Figure 4 This is a schematic diagram of single-channel video point calibration according to an embodiment of this application; Figure 5 This is a schematic diagram of the integration of visual calibration points according to an embodiment of this application; Figure 6 This is an interface diagram of abnormal traffic behavior recognition and individual behavior profiling in an embodiment of this application; Figure 7 This is a schematic diagram of a student commuting protection system based on a secure encrypted network and cloud-edge collaboration, according to an embodiment of this application. Figure 8This is a schematic diagram of another student commuting protection system based on a secure encrypted network and cloud-edge collaboration, according to an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0024] like Figure 1 as well as Figure 8 As shown, this embodiment provides a method for protecting students going to and from school based on a secure encrypted network and cloud-edge collaboration. This method operates as follows: Figure 8 The three-layer architecture shown consists of a terminal perception layer, an intelligent central layer, and a smart application layer, which is used to protect the safety of students throughout their entire journey from home to school and back.

[0025] Step S110: Acquire video data collected by multi-source edge sensing devices along the student's route to and from school.

[0026] Multi-source edge sensing devices refer to various video acquisition devices deployed along students' routes to and from school, belonging to different management entities or having different functional roles. For example, cameras at school gates collect video data of students entering and leaving the school area, cameras at residential community entrances collect video data of students leaving or returning to their communities, cameras at intersections collect video data of students crossing urban road intersections, and gap-filling cameras cover the monitoring blind spots between the above devices. These devices can be intelligent cameras with local computing capabilities or ordinary cameras without local computing capabilities. A unified video access platform receives and aggregates the video data collected by various sensing devices and performs standardized processing on the video data, enabling video data of different formats and resolutions to be uniformly used in subsequent processes. The video data can be acquired as a real-time video stream or as a sequence of video frames extracted at preset time intervals, depending on the transmission capabilities of the sensing devices and network conditions.

[0027] In terms of dynamic planning of student journey sensing points, the spatial distribution of identity recognition points, event recognition points, and safe departure points is planned based on the student's "home-road-school" two-point trajectory. Identity recognition points are deployed at key travel nodes where student identity needs to be verified, such as community entrances and school gates; event recognition points are deployed on road sections where abnormal events such as running red lights, crowd gatherings, and falls need to be identified, such as intersections; safe departure points are deployed at and around the school gate to verify peer relationships. A single sensing point can simultaneously perform multiple recognition functions as needed, or an automatic algorithm rotation mechanism can be set up to activate different recognition algorithms at different times, achieving dynamic adaptation of point computing power. When the protected path changes or new sensing devices are added, the system supports expansion and adjustment of the point layout and algorithm deployment.

[0028] Step S120: Based on whether the edge sensing device has local inference capability, determine the current computing power mode between the weak computing power mode and the strong computing power mode of the cloud center, and determine the execution location for identity recognition of video data, which includes the edge sensing device or the cloud center.

[0029] Local inference capability refers to whether an edge sensing device has the ability to run algorithm models and complete inference calculations locally. Edge sensing devices can be smart cameras with integrated neural network inference chips or local servers deployed near the video capture site. When an edge sensing device has local inference capability, lightweight algorithms such as identity recognition can be deployed and executed locally at the edge. In this case, the cloud only needs to handle lightweight tasks such as data aggregation, status determination, and result push; this operating state corresponds to a cloud-centric, low-computing-power mode. When an edge sensing device lacks local inference capability, algorithm inference tasks such as identity recognition need to be centrally completed by the cloud; this operating state corresponds to a cloud-centric, high-computing-power mode.

[0030] The execution location of identity recognition corresponds to the current computing power mode: in the weak computing power mode of the cloud center, identity recognition is performed at the edge; in the strong computing power mode of the cloud center, identity recognition is performed in the cloud. Whether the edge sensing device has local inference capabilities can be automatically determined by the capability reporting information during device power-on registration, or it can be determined based on the configuration information configured by the administrator in the management backend.

[0031] Step S130: Identify the target in the video data, and determine the travel status and safe departure from school based on the identification result and the location determination result, and output the student's school arrival and departure protection status.

[0032] Identity verification refers to detecting targets from video data and confirming whether the target is a registered person in the system through methods such as facial recognition. Registered persons include the student themselves and the student's guardian or other authorized personnel. The execution location of identity verification is determined by the current computing power mode determined in step S120.

[0033] To improve the accuracy of facial recognition in scenarios involving children going to and from school, the system employs a facial recognition model optimized for children's facial features and complex outdoor environments. Children's facial features change more rapidly than adults', and outdoor environments present challenges such as varying lighting and diverse shooting angles. Therefore, the facial recognition model was adapted and optimized for children's samples and low-quality outdoor images during training. The system supports multi-angle photo input, including photos from the front, looking up, left side, and right side, thereby improving the recognition rate through multi-angle facial information. Figure 3 The system showcases a multi-angle facial information collection interface. It achieves accurate dual-identity recognition of both the protected individual and the guardian, simultaneously identifying the registered student and their guardian or other authorized personnel. To address the varying imaging conditions and crowd density at different locations such as school gates, residential area entrances, and intersections, the system supports setting separate facial recognition thresholds to strike a balance between accuracy and false recognition rates.

[0034] Trip status determination refers to identifying the stage of a target's journey to and from school based on identity verification and location determination results. Trip status can include three basic states: departure, transit, and arrival. Departure indicates the target has left their residential area or departure point; transit indicates the target is at an intermediate point on their journey to and from school; and arrival indicates the target has arrived at school or returned to their residential area. Trip status determination relies on the data correlation between identity verification and location determination results: identity verification confirms the target's identity, and location determination determines the target's location; only by combining both can the stage of a registered student's journey at a given time be determined.

[0035] Safe departure from school involves verifying the relationship between the individual and their guardian during dismissal time to determine if the individual is accompanied by a registered guardian. This process relies on both identity verification and location determination: identity verification confirms the identities of both the individual and their guardian, while location determination determines whether they are in the same spatial area.

[0036] The student commuting status is the result of the system's comprehensive assessment of the student's safety throughout their commuting process, and can include normal and abnormal states. A normal state indicates that the student completes their commuting journey according to the expected path and method, while an abnormal state indicates that the student deviates from the expected path, lacks accompanying personnel, or encounters other situations requiring attention during the journey.

[0037] Step S140: During system operation, in response to the computing power mode switching command, the current computing power mode is switched between the weak computing power mode and the strong computing power mode of the cloud center, and identity recognition, travel status determination and safe departure determination are continuously performed during the switching process; wherein, the data interaction between the cloud, edge and terminal is encrypted using a session dynamic password mechanism.

[0038] The computing power mode switching command can be triggered automatically by the system or issued by the administrator through the management backend. Automatic monitoring and triggering means that the system continuously monitors device status, network quality, and computing power load during operation, and automatically generates a computing power mode switching command when the monitoring results meet preset switching conditions. Administrator commands refer to the administrator manually initiating the computing power mode switch through the management backend based on actual operational needs.

[0039] During the switching process, the system does not interrupt the execution of identity verification, travel status determination, and safe departure from school determination. Specifically, when the system switches from the cloud-based weak computing power mode to the cloud-based strong computing power mode, the identity verification task originally executed at the edge is migrated to the cloud for continued execution, and the acquisition of video data and the detection and tracking of targets are uninterrupted during the migration process. When the system switches from the cloud-based strong computing power mode to the cloud-based weak computing power mode, the identity verification task is migrated from the cloud to the edge for execution, and the cloud continues to track the identified targets during the migration process until the edge takes over. Through the above hot-switching mechanism, the system will not have monitoring blind spots during the computing power mode switching, ensuring continuous protection capabilities during peak student arrival and departure times.

[0040] We will continue to elaborate on the process of determining the computing power mode and the specific implementation of hot switching in the running state.

[0041] When determining the current computing power mode, if the edge sensing device has local reasoning capabilities, the current computing power mode is determined to be the cloud center weak computing power mode, and identity recognition is performed at the edge and only the recognition results and feature data are returned.

[0042] In this process, the edge sensing device performs target detection, feature extraction, and identity recognition calculations on video frames locally, and then only transmits the recognition results and face recognition-related feature data back to the cloud. The recognition results may include structured data such as target ID, identity confirmation result, timestamp, and the target's location information in the frame. The feature data may include facial feature vectors, target appearance feature descriptors, and other data that can be used by the cloud for subsequent correlation analysis and status determination.

[0043] In a cloud-centric, low-computing-power mode, the entire video stream is not transmitted back to the cloud. Video data is processed at the edge and then discarded or cached locally. This reduces the computational load required for identity recognition, making it suitable for on-site processing at the edge. Transmitting the entire video stream back to the cloud would consume significant network bandwidth, and most of the video content is irrelevant to the identity recognition task. By transmitting only the recognition results and feature data, the system reduces the amount of data transmitted over the network by more than an order of magnitude while maintaining accuracy. It also shortens the latency from video capture to recognition result output, enabling millisecond-level recognition responses when the edge has the necessary computing power.

[0044] When edge-based sensing devices lack local inference capabilities, the current computing power mode is determined to be the cloud-centric high-power computing mode, and video data is transmitted back to the cloud for identity recognition. Situations where edge-based sensing devices lack local inference capabilities include: the device itself is a regular camera without integrated neural network inference chips or local computing units; or although the device has some computing hardware, it has not deployed the algorithm model required for identity recognition, or its computing power is insufficient to support real-time inference. In these cases, the edge-based sensing device is only responsible for video data acquisition, encoding, and transmission; all algorithm inference tasks, such as identity recognition, are completed by the cloud. Video data can be transmitted back to the cloud via a government dedicated network or a unified video network to ensure transmission stability and low latency. In the cloud-centric high-power computing mode, the cloud needs to allocate corresponding computing and bandwidth resources for the transmitted video stream to ensure the real-time nature of identity recognition. Through this dual-mode design, the system can simultaneously accommodate smart devices with edge computing capabilities and a large number of existing regular cameras, enabling devices with different computing power conditions to be included in a unified student commuting monitoring system.

[0045] Along the same student route to and from school, there may be both smart cameras with local reasoning capabilities and ordinary cameras without. For each edge sensing device, its corresponding computing power mode can be determined separately. For example, a smart camera deployed at the school gate has local reasoning capabilities, so a cloud-centric weak computing power mode is used, with identity recognition performed at the edge. Ordinary cameras at intersections, however, lack local reasoning capabilities, so a cloud-centric strong computing power mode is used, with video data transmitted back to the cloud for identity recognition. The recognition results from different computing power modes are then uniformly aggregated and correlated in the cloud, making the differences in computing power modes transparent to upper-layer applications.

[0046] During system operation, in response to monitoring results of device status, network quality, and computing load, and / or administrator commands, a computing mode switching command is generated. Device status monitoring includes the online, offline, and fault status of edge sensing devices. An online status indicates that the communication link between the device and the system is normal, and the device can report data normally; an offline status indicates that the communication link between the device and the system is interrupted, and the system cannot obtain video data from the device; a fault status indicates that there is an abnormality in the device's hardware or software, such as firmware upgrade failure, algorithm model loading anomalies, or sensor damage. Network quality monitoring includes indicators such as network bandwidth, latency, and packet loss rate.

[0047] Network bandwidth reflects the data transmission capacity of a link, latency reflects the time required for data to travel from the sender to the receiver, and packet loss rate reflects the proportion of data packets lost during data transmission.

[0048] Computing load monitoring includes the utilization of computing resources on edge devices and in the cloud, such as CPU utilization, GPU utilization, and memory utilization. When the monitoring results meet the preset switching conditions, the system automatically generates a computing mode switching command.

[0049] The preset switching conditions can be configured according to the actual deployment environment. For example, when the CPU utilization of the edge device continuously exceeds a preset threshold, it is considered that the device's local inference capability is insufficient to support real-time identity recognition, triggering a switch from the cloud center's weak computing power mode to the cloud center's strong computing power mode. When the network bandwidth continuously falls below a preset threshold or the packet loss rate continuously exceeds a preset threshold, it is considered that the video data backhaul link quality is insufficient to support the cloud center's strong computing power mode. If an alternative device with local inference capability exists, a switch from the cloud center's strong computing power mode to the cloud center's weak computing power mode is triggered. In addition to automatic monitoring triggers, administrators can also manually issue computing power mode switching commands through the management backend to handle planned equipment maintenance, firmware upgrades, or network adjustments.

[0050] After switching to the current computing power mode, the algorithm deployment location and video transmission strategy are reconfigured. Reconfiguration of the algorithm deployment location refers to redeploying identity recognition and other algorithm models between the edge and the cloud according to the switched computing power mode. When switching from a weak computing power mode to a strong computing power mode at the cloud center, the identity recognition algorithm originally deployed at the edge is migrated to the cloud for execution, and the edge no longer undertakes identity recognition tasks. When switching from a strong computing power mode to a weak computing power mode at the cloud center, the identity recognition algorithm is distributed to the edge devices, and the cloud releases the corresponding computing resources. Reconfiguration of the video transmission strategy refers to adjusting the content and method of data transmitted from the edge to the cloud according to the switched computing power mode. When switching to a weak computing power mode at the cloud center, the video transmission strategy is adjusted from transmitting the entire video stream to transmitting only the recognition results and feature data; when switching to a strong computing power mode at the cloud center, the video transmission strategy is adjusted from transmitting only the recognition results and feature data to transmitting the entire video stream. The reconfiguration of the algorithm deployment location and the reconfiguration of the video transmission strategy work together to complete the switching of the computing power mode.

[0051] The specific method for determining the location determination result in step S130 will be further elaborated. The location determination result is obtained through a hierarchical trigger-based positioning mechanism, which divides the location determination process into three stages: basic location determination, identity verification, and precise location calculation. There are clear trigger dependencies between the three stages.

[0052] First, the basic location of the target is determined by the coverage area of ​​the multi-source edge sensing device. The coverage area of ​​the multi-source edge sensing device refers to the fixed geographical area covered by the device's installation location. For example, the coverage area of ​​a camera at an intersection is the intersection and a certain surrounding road area; the coverage area of ​​a camera at a school gate is the school gate and an area extending outwards for a certain distance. When a target appears in the image captured by a multi-source edge sensing device, the system uses the fixed area covered by that device's installation location as the target's basic location. Basic location positioning does not rely on complex image calculations; it only needs to be determined based on which sensing device the target appears within its coverage area. Therefore, it has a fast response speed and low computational overhead. The role of the basic location is to quickly pinpoint the approximate range of the target, providing preliminary spatial constraints for subsequent identity recognition and precise positioning. For example, when a target appears within the coverage area of ​​an intersection camera, the system uses the fixed area covered by that camera's installation location as the basic location. At this point, the system can determine that the target is near the intersection, but it cannot yet determine the target's specific location within the intersection.

[0053] After determining the basic location, the target is identified to determine whether the target is a registered user of the system.

[0054] Only after successful identification of the target does the system determine its precise location based on the visual calibration relationship between the image from the multi-source edge sensing devices and its geographic coordinates. Successful identification means the system confirms the target is a person registered in the system, including the student, their guardian, or other authorized personnel. For targets whose identification fails, such as unregistered passersby, the system does not perform precise positioning calculations, retaining only basic location information or ignoring the target entirely. The rationale behind this design is that in student commuting scenarios, precise positioning is only truly needed for registered students and their guardians. Performing high-precision positioning calculations on all targets appearing in the frame would consume significant computing resources. By setting the trigger condition for precise positioning to successful identification, the system limits the computational load of high-precision positioning calculations to registered personnel, significantly reducing computing power consumption and improving overall system efficiency in multi-target scenarios during peak commuting hours.

[0055] Visual calibration relationship refers to the pre-established mapping relationship between image pixel coordinates and geographic coordinates of multi-source edge sensing devices. This mapping relationship defines the proportional relationship between pixel distance in the image and actual geographic distance, enabling the system to convert the pixel position of a target in the image into actual geographic latitude and longitude coordinates.

[0056] The process of establishing visual calibration relationships can be combined with Figure 4 and Figure 5 To understand. Figure 4 The diagram illustrates the calibration of single-channel video points. In the capture image of a certain sensing device, multiple reference points with known geographical coordinates are selected as preset calibration points, such as the endpoints of zebra crossings at intersections, corners of curbs, and tips of traffic islands, which are points with clear geographical location characteristics. Figure 5 The diagram illustrates the integration of visual calibration points, which integrates the visual calibration relationships established by multiple sensing devices to form a continuous positioning capability covering students' routes to and from school.

[0057] After the visual calibration relationship is established, the specific calculation process for the precise position is as follows.

[0058] The process involves acquiring the pixel coordinates of the target within the captured image and the pixel coordinates of preset calibration points. The pixel coordinates of the target within the captured image refer to the pixel position of the target in the video frame, which can be determined by the cloud center point or foot point of the target bounding box output by the target detection algorithm. The pixel coordinates of the preset calibration points refer to the pixel positions of pre-selected reference points with known geographic coordinates within the video frame. Preset calibration points are reference points with known geographic coordinates, such as the endpoints of zebra crossings at intersections or the corners of curbs. The geographic coordinates of these reference points can be obtained through on-site measurement or from a geographic information system (GIS).

[0059] The pixel offset is determined based on the pixel coordinates of the target in the captured image and the pixel coordinates of a preset calibration point. The pixel offset refers to the positional deviation of the target relative to the preset calibration point in the image, and can be calculated by the difference between the target's pixel coordinates and the preset calibration point's pixel coordinates. The pixel offset includes horizontal and vertical pixel offsets, corresponding to the horizontal and vertical axes of the image coordinate system, respectively.

[0060] Based on the visual calibration relationship between images and geographic coordinates from multi-source edge sensing devices, pixel offsets are converted into latitude and longitude offsets. The visual calibration relationship defines the proportional mapping between image pixel distances and actual geographic distances. This proportional mapping can be calculated using the known geographic coordinates and pixel coordinates of a preset calibration point. Multiplying the pixel offset by the corresponding scaling factor yields the actual geographic offset of the target relative to the preset calibration point, i.e., the latitude and longitude offset. The latitude and longitude offset includes offsets in the longitude direction and offsets in the latitude direction.

[0061] By superimposing latitude and longitude offsets onto the reference latitude and longitude of the multi-source edge sensing device, the precise location of the target can be obtained. The reference latitude and longitude of the multi-source edge sensing device refers to the fixed latitude and longitude coordinates of the device's installation location. By superimposing the latitude and longitude offsets onto the reference latitude and longitude, the precise latitude and longitude coordinates of the target in the actual geographic space can be obtained.

[0062] Furthermore, precise location calculation can be achieved in other ways. For example, after establishing the visual calibration relationship, the pixel coordinates of the target in the image can be directly mapped to geographic coordinates using a homography matrix, without explicitly calculating pixel offsets and latitude / longitude offsets. The homography matrix describes the perspective transformation relationship between the image plane and the geographic plane, and can be obtained by solving for the pixel coordinates and geographic coordinates of multiple preset calibration points. Regardless of the specific calculation method used, the core principle is to utilize the pre-established visual calibration relationship between the image and geographic coordinates to convert the target's position information in the image into actual geographic latitude / longitude coordinates.

[0063] This embodiment establishes a complete positioning chain from basic location determination to identity verification and then to precise location calculation through a hierarchical trigger-based positioning mechanism. Basic location determination provides rapid spatial constraints, identity verification provides the triggering conditions for precise positioning, and visual calibration provides the basis for converting pixel coordinates to geographic coordinates. The three stages have a clear sequential dependency, jointly achieving efficient and accurate positioning of registered personnel.

[0064] To quantify the reliability of the positioning results, a confidence level index is introduced into the positioning process. Base location confidence level. This indicates whether the target falls within the coverage area of ​​the sensing device. The value is 1 if the condition is met, and 0 otherwise. ; Identity recognition confidence The calculation method, which uses the average of facial similarity across multiple frames, is consistent with the comprehensive confidence score of student faces used in the safe departure criteria. ,in To continuously identify the number of frames, Let be the face similarity of the i-th frame.

[0065] Precisely locate the start-up coefficient satisfy Only when =1 and Reaching the preset identity recognition threshold T id Only then does it enter the precise calculation of latitude and longitude, T id For example, take 0.80. The precise latitude and longitude output satisfies: (longitude), (Latitude), among which , For precise longitude and latitude of the target, , To sense the longitude and latitude of the installation location of the sensing equipment, , These are the pixel offsets in the longitude and latitude directions, respectively. , This is the visual positioning ratio coefficient.

[0066] Visual calibration confidence level P calib The reliability of the visual calibration relationship and the accurate position calculation results can be determined by a combination of factors, including the number of calibration points, calibration residuals, and the target's position in the captured image. A higher number of calibration points and smaller calibration residuals result in higher visual calibration confidence. Personnel positioning overall confidence level P final Combining the confidence level of identity recognition and the confidence level of visual calibration, it satisfies P final =(R id +P calib ) / 2. The basic location confidence level and the precise positioning activating coefficient are necessary conditions for entering the precise positioning calculation, but they do not participate in the arithmetic synthesis of the comprehensive confidence level.

[0067] When P final Reaching the preset final threshold T final When P is used, the latitude and longitude obtained through precise positioning are used as the final location of the target; when P is used... final T not reached final When using the baseline position as the final target position, T final For example, we can take 0.85. Specifically, the final output target latitude and longitude G is:

[0068] By using the above confidence index classification, a balance is achieved between positioning accuracy and computational reliability, avoiding erroneous output of high-precision positions when identity recognition is unstable or visual calibration is insufficient.

[0069] The specific implementation of the safe departure determination in step S130 will be further elaborated. The safe departure determination adopts a four-dimensional joint determination mechanism of time window constraint, regional constraint, identity recognition and peer relationship. The four dimensions work together to form a complete basis for judging the safe departure status of the target.

[0070] The safe departure from school process includes the following steps: Within a preset dismissal time window, facial recognition is used to confirm the identity of the target and their guardian. The preset dismissal time window refers to a time period pre-set based on the school's dismissal time, such as 5 PM to 6 PM. The dismissal time window can be configured according to the school's schedule; dismissal times may vary between different schools and grades. The system supports setting corresponding dismissal time windows for different guardians. Guardians refer to the guardians or other authorized personnel registered in the system. Guardians are usually the student's parents or other legal guardians, while authorized personnel can be the student's grandparents, maternal grandparents, or other persons authorized by the guardian. The guardian's identity information and facial feature data are entered into the system during the initial information collection phase, establishing a connection with the corresponding student. Within the preset dismissal time window, the system performs facial recognition on targets appearing at the school gate and surrounding area to confirm whether the target is a registered student and to verify that other individuals appearing around the target are the registered guardians for that student. The execution location of facial recognition is determined by the current computing power mode.

[0071] The system determines whether the target has fallen into a pre-defined departure area based on the location determination result. The pre-defined departure area refers to the area around the school gate, such as an area extending outwards from the school gate as the cloud center, or specific areas connected to the school gate, such as pedestrian walkways or parent waiting areas. The specific range of the pre-defined departure area can be configured according to the school's actual layout and security management needs. The location determination result is obtained through a hierarchical trigger-based positioning mechanism, including basic location and precise location. The system determines whether the target is currently within the pre-defined departure area based on the location determination result. When the target is within the pre-defined departure area, it indicates that the target has left the campus and entered the departure passage area, at which point further verification of the companion relationship is required. When the target is not within the pre-defined departure area, it indicates that the target has not yet reached the departure area or has already left the departure area, and the system continues monitoring according to the specific situation.

[0072] Within a pre-defined area outside the school, the effective travel relationship between the target and the guardian is detected. An effective travel relationship refers to a situation where the target and the guardian meet pre-defined conditions both spatially and temporally. Spatial conditions refer to the spatial distance between the target and the guardian being within a pre-defined range, and temporal conditions refer to the time difference between the target and the guardian appearing in the same area being within a pre-defined range. The specific methods for determining an effective travel relationship will be further elaborated below.

[0073] If no valid companion relationship is detected by the end of the preset dismissal time window, an abnormal departure status will be output. This status indicates that the target did not leave school with a registered guardian within the dismissal time window, and relevant personnel need to monitor the target's safety after leaving school. After outputting the abnormal departure status, the system can trigger corresponding alarm procedures, such as sending a notification to the guardian or sending a notification to school administrators. It should be noted that the output of the abnormal departure status does not necessarily mean the target is in danger; rather, it indicates that the system cannot confirm that the target left school with a registered guardian and manual intervention is required.

[0074] In the above-mentioned judgment mechanism, the time window constraint limits the execution period of the safe departure judgment, avoiding unnecessary departure judgments on targets appearing at the school gate outside of school hours; the area constraint limits the spatial scope of the safe departure judgment, focusing the judgment on the departure passage area; the identity recognition confirms the identities of the target and the guardian, providing an object basis for the companion relationship detection; the companion relationship detection verifies the actual companionship status between the target and the guardian.

[0075] Effective peer-to-peer relationships are determined based on peer-to-peer matching degree, which is determined by the spatial distance and time difference between the target and the guard. Spatial distance refers to the Euclidean distance between the target and the guard in physical space, which can be calculated from the position determination results of both. When both the target and the guard are located within the preset off-campus area, the system obtains their precise locations and calculates the spatial distance between them. The time difference between appearance refers to the time difference between the target and the guard appearing within the coverage area of ​​the same sensing device. For example, if the target appears within the coverage area of ​​the camera at the school gate at a certain moment, and the guard appears within the coverage area of ​​the same camera at another moment, the difference between their appearance times is the time difference.

[0076] When the spatial distance is within the preset maximum travel distance and the time difference is within the preset maximum time difference, it is considered a valid travel relationship between the target and the guard. The preset maximum travel distance and preset maximum time difference can be configured according to the actual scenario. For example, the preset maximum travel distance can be set to 5 meters, and the preset maximum time difference can be set to 5 seconds. The above values ​​are only illustrative examples; in actual applications, they can be adjusted based on factors such as the population density at the school gate, the camera coverage area, and the age of the students. When the spatial distance is within the preset maximum travel distance, it indicates that the target and the guard are physically close enough to travel together; when the time difference is within the preset maximum time difference, it indicates that the target and the guard appear in the same area almost simultaneously, meeting the time condition for travel together. When both conditions are met, the system determines that a valid travel relationship exists between the target and the guard.

[0077] Peer matching score can be calculated in several ways. One possible implementation determines the peer matching score based on a combination of the ratio of spatial distance to a preset maximum peer distance and the ratio of the occurrence time difference to a preset maximum time difference. For example, peer matching score can be expressed as: ,in The matching degree is the same as the peer's score, with a value range of [0,1). The spatial distance between the target and the personnel guarding it; Preset maximum peer distance; To create a time difference; The preset maximum time difference; α is the spatial distance weight, and β is the time difference weight. α + β = 1, typically set to 0.5 and 0.5, or 0.6 and 0.4 depending on business focus. The smaller the spatial distance and the smaller the time difference, the closer the peer matching degree is to 1, indicating a closer peer relationship. The above formula is only an exemplary implementation of peer matching degree calculation. Those skilled in the art can use other calculation methods according to actual needs, as long as the peer matching degree can reflect the spatial and temporal proximity between the target and the guardian.

[0078] Building upon the established peer matching score, the safe departure from school assessment further incorporates a safe departure confidence score as a comprehensive evaluation indicator. The safe departure confidence score is determined based on the student's overall face confidence score, peer matching score, and departure area matching result. The student's overall face confidence score refers to the degree of credibility of identity verification calculated through a comprehensive analysis of multiple frames of face recognition results. Specifically, the system performs continuous multi-frame face recognition on the target within a preset dismissal time window, obtaining the face matching similarity for each frame. The average of these multi-frame face matching similarities is used as the student's overall face confidence score. The student's overall face confidence score reflects the reliability of target identity recognition; a higher confidence score indicates a greater likelihood that the target has been correctly identified as a registered student. The departure area matching result refers to the determination of whether the target is located within a preset departure area. When the target is within the preset departure area, the departure area matching result is set to 1; when the target is not within the preset departure area, the departure area matching result is set to 0.

[0079] When the confidence level for safe departure reaches the preset safe departure threshold, the departure is considered safe; when the confidence level does not reach the preset safe departure threshold, the departure is considered abnormal. The preset safe departure threshold can be configured according to the security management requirements of the actual scenario, for example, set to 0.75. The safe departure confidence level integrates information from three aspects: the reliability of identity recognition, the closeness of peer relationships, and the determinism of area matching, which can effectively reduce the impact of misjudgment by a single factor on the final judgment result.

[0080] There are several ways to calculate the confidence level for safe departure from school. One possible implementation is that the confidence level for safe departure from school can be expressed as... ,in The confidence level for safe departure from school is set to [0,1]. The overall confidence score for student faces is defined, with a value range of [0,1]. The matching degree is the same as the peer's score, with a value range of [0,1). This represents the matching result for areas outside the school, with a value of 0 or 1.

[0081] The above formula is only an exemplary implementation of the safe departure confidence score. Those skilled in the art can use weighted average, product or other comprehensive calculation methods according to actual needs, as long as the safe departure confidence score can comprehensively reflect the student's face confidence score, peer matching score and departure area matching result.

[0082] Furthermore, the time window and regional constraints in the joint judgment mechanism also contribute to the accuracy of the judgment. The time window constraint limits the execution period of the safe departure judgment to the preset dismissal time window, avoiding unnecessary departure judgments for students appearing at the school gate outside of dismissal time, thus reducing false alarms. The regional constraint limits the judgment scope to the preset departure area, avoiding misjudging situations where students are active on campus or passing by the school gate from a distance as departure behavior. Through the joint constraints of these four dimensions, the system reduces both the probability of missed judgments and the probability of false judgments, improving the overall accuracy and robustness of the safe departure judgment.

[0083] It should be noted that the specific values ​​given in this embodiment, such as the preset dismissal time window, preset maximum walking distance, preset maximum time difference, and preset safe departure threshold, are illustrative and do not constitute a limitation on the scope of protection. In practical applications, the above parameters can be flexibly configured according to factors such as the school's specific schedule, the spatial layout of the school gate, the camera coverage, the age characteristics of students, and security management requirements. For example, for lower-grade students, the preset maximum walking distance can be set smaller to require guards to maintain a closer distance to students; for school gate areas with high population density, the preset maximum time difference can be set shorter to reduce the possibility of misidentifying non-walking individuals as walking companions.

[0084] We will continue to elaborate on the encryption methods for data interaction between the cloud, edge, and device. Data interaction between the cloud, edge, and device employs a session dynamic password mechanism for encryption. This mechanism achieves information security during identity authentication, data transmission, and access control by generating a dynamic password when the session is established, periodically updating it during the session's duration, forcibly updating it under abnormal events, and destroying it when the session ends.

[0085] During the communication session establishment phase, the system generates a dynamic session password bound to that communication session. This dynamic session password corresponds only to the current business session and is invalidated upon session termination; it is not stored for long-term reuse. The dynamic session password is used as the data key for that session during data encryption and decryption. When the session initiator completes identity authentication, it uses the SM2 digital certificate to verify the entity's identity. The cloud-based password service generates the dynamic session password in real time based on the session context, terminal device identifier, and access role. Access roles include different entities such as edge devices, parents, and administrators. The dynamic session password is not transmitted in plaintext over the link; instead, key negotiation is completed via a cryptographic device. After negotiation, both ends of the session store the dynamic session password locally. When the session terminates normally or is abnormally disconnected, the dynamic session password is immediately destroyed and cannot be reused.

[0086] The session dynamic password is generated based on a base key, terminal device identifier, role code, and session random number. The base key is output by the cryptographic machine and serves as the root key source for generating the session dynamic password; the terminal device identifier is used to distinguish different edge devices or terminals; the role code is used to distinguish different access roles such as edge devices, parents, and administrators; and the session random number is used to ensure that the dynamic password generated for each session is unique.

[0087] Taking the initial dynamic password generated when a session is established as an example, its generation method can be represented as K init =H(K) base ⊕ID dev ⊕Prole⊕Nonce), where K init H is the initial dynamic password for the session, H is the SM3 cryptographic algorithm operation, and K is the dynamic password. base Basic key, ID dev For terminal device identification, Prole is the role code, Nonce is the random number for a single session, and ⊕ is the XOR operation.

[0088] It should be noted that the SM3 cryptographic algorithm performs a digest operation on all input factors as a whole, while the XOR operation is used to combine the input factors before inputting them into the SM3 algorithm. The SM3 operation and the XOR operation are at different stages of the key generation process; that is, the input factors are first XORed to form an intermediate input, and then the SM3 algorithm performs a digest operation on the intermediate input to obtain the session dynamic cipher. XOR combination is only one example of how to combine input factors; other combination methods such as concatenation and cascading can also be used, as long as all input factors can be uniformly input into the cryptographic algorithm for operation.

[0089] During the communication session, the system periodically updates the session's dynamic password according to a preset iteration cycle. When the session is not interrupted and the key reaches the iteration cycle, the system generates an iterated dynamic password. Upon receiving the new dynamic password, both parties use the new key and invalidate the old key.

[0090] Taking session-based iterative dynamic ciphers as an example, their generation method can be represented as K n+1 =H(K) n ⊕Counttim), where K n K is the session dynamic password being used in the nth round. n+1 The next round of dynamic cipher is generated iteratively, and Counttim is the session iteration counter, which increments by 1 with each iteration. Specifically, the session iteration counter is first XORed with the previous round's dynamic cipher to form an intermediate input, which is then digested using the SM3 algorithm to obtain the next round's dynamic cipher.

[0091] Upon detecting an abnormal network event, the current session's dynamic password is forcibly invalidated and a new session's dynamic password is generated. Abnormal network events include detected packet replay, packet tampering, abnormally high-frequency access, severe degradation of link quality, or suspicious access behavior. Once an abnormal update is triggered, the system no longer waits for the iteration cycle or for the session to end naturally; it immediately invalidates the current session's dynamic password, terminates the original session, re-performs authentication, and generates a completely new session's dynamic password.

[0092] The dynamic password for a communication session is immediately destroyed upon normal termination or abnormal disconnection. All dynamic passwords exist only in the memory of the cipher machine and both ends of the session, and are not persistently stored on disk. Different roles and devices use different session input factors, resulting in independent dynamic passwords.

[0093] Data interaction between the cloud, edge, and device operates on top of the underlying transmission channel, employing a business-layer cryptographic protection system driven by dynamic session ciphers to form a two-layer defense. Data reported by edge devices and business data sent to parents and administrators undergo business-layer encryption and integrity verification.

[0094] Communication sessions for different access roles use independent session dynamic ciphers, isolating the business payloads of different roles. Even when reusing the same underlying physical link, parent sessions, management sessions, and edge reporting sessions use their own session dynamic ciphers for encryption and decryption. Each role's business messages are encrypted and decrypted using its own dynamic cipher, ensuring isolation between the business payloads of different roles. A single role's session dynamic cipher can only decrypt data within the role's authorized scope and cannot decrypt encrypted data outside that scope. For example, a parent session's dynamic cipher can only decrypt the guardian information of the student corresponding to that parent and cannot decrypt guardian data for other students or the school as a whole.

[0095] This embodiment generates dynamic session passwords by combining device identifiers and role codes. All dynamic session passwords are destroyed when the session ends, achieving full lifecycle control of the key and isolation of multi-role communication. It overcomes the shortcomings of traditional static key encryption mode, such as long-term unchanging keys, risk diffusion after leakage, and lack of emergency key switching in case of anomalies, thus ensuring the security of data interaction between cloud, edge, and device.

[0096] This embodiment takes a specific scenario of protecting students throughout their commute to and from school as an example, combined with... Figure 1 and Figure 2 This illustrates the complete operation of the methods and systems described in the above embodiments in a real-world deployment environment. These embodiments are only used to demonstrate the feasibility and practicality of the technical solutions and do not constitute a limitation on the scope of protection.

[0097] Assume a primary school student, Zhang, lives in a residential community. His route to school is from the east gate of the community to the intersection of Xingfu Road, and then from the intersection of Xingfu Road to the south gate of the school. The system connects to three multi-source edge sensing devices along this route: a camera at the east gate of the community, a camera at the intersection of Xingfu Road, and a camera at the south gate of the school. The cameras at the east gate and the south gate are intelligent cameras with local inference capabilities, capable of executing lightweight algorithms such as identity recognition at the edge. The camera at the intersection of Xingfu Road is a regular camera without local inference capabilities, requiring video data to be transmitted back to the cloud for identity recognition. All three cameras are connected to the cloud through a unified video access platform. The cloud platform deploys modules for computing power configuration, recognition judgment, and computing power switching. Parents receive monitoring status messages pushed by the system via a WeChat mini-program.

[0098] At 7:30 AM, Zhang departed from the east gate of the residential compound. The camera at the east gate captured video data showing Zhang within its coverage area. Because this camera possesses local inference capabilities, and the system had already determined its edge computing power during the device access phase based on capability reporting information, a cloud-centric weak computing power mode was adopted for this device, with identity verification performed at the edge. The east gate camera completed target detection and facial recognition locally. After confirming Zhang's identity, it only transmitted the recognition results and feature data back to the cloud, without transmitting the full video stream. Upon receiving the recognition results, the cloud combined them with the location determination to confirm Zhang's location in the east gate area, classifying Zhang's journey status as "departure," and sent a departure reminder message to Zhang's parents via a WeChat mini-program.

[0099] At 7:40, Zhang passed through the intersection of Xingfu Road. The camera at the Xingfu Road intersection is a standard camera without local inference capabilities; the system uses a cloud-centric high-power computing mode for this device. The camera transmits the collected video data back to the cloud via the government intranet, where identity verification is performed. The cloud detects Zhang in the video data, completes facial recognition to confirm his identity, and combines this with location determination to determine that Zhang is located in the Xingfu Road intersection area, classifying his journey as "passing through." Because the Xingfu Road intersection camera lacks edge computing capabilities, the cloud needs to allocate corresponding computing and bandwidth resources to the device's video stream to ensure real-time identity verification.

[0100] At 7:50, Zhang arrived at the south gate of the school. The camera at the south gate has local inference capabilities, while the system uses a cloud-centric, low-power computing mode for this device. The camera at the south gate performs facial recognition at the edge, confirms Zhang's identity, and only sends the recognition result and feature data back to the cloud. The cloud, combined with the location determination result, determines that Zhang is located in the south gate area of ​​the school, classifies Zhang's travel status as "arrived," and pushes an arrival confirmation message to the parents via WeChat mini-program. Thus, Zhang's school journey monitoring is complete. The parents received two messages between 7:30 and 7:50: departure reminder and arrival confirmation, requiring no manual intervention throughout the process.

[0101] The preset dismissal time window is from 5:00 PM to 6:00 PM. During this time window, the camera at the south gate of the school continuously collects video data from the area around the school gate. The system detects Zhang's presence in the dismissal area, and simultaneously detects Zhang's registered guardian, namely his mother, in the same area. The south gate camera completes facial recognition of Zhang and his mother at the edge, confirming their identities. Based on the location determination result, the system calculates the spatial distance between the two to be 2 meters, and the time difference between their appearances within the same camera coverage area to be 2 seconds. The preset maximum walking distance is 5 meters, and the preset maximum time difference is 5 seconds. Both the spatial distance and the time difference are within the preset range, and the system calculates a walking match score of 0.6. At the same time, the system performs continuous multi-frame facial recognition on Zhang, calculating a student face comprehensive confidence score of 0.83. Zhang is located within the preset dismissal area, and the dismissal area match result is 1. The system takes into account the above three input factors and calculates a confidence level of 0.81 for safe departure from school, which exceeds the preset safe departure threshold of 0.75. Therefore, it is determined that the departure is safe and a confirmation message for safe departure from school is pushed to the parents via WeChat mini program.

[0102] During the aforementioned operation, the camera at the Xingfu Road intersection triggered a computing power mode switch command due to network quality degradation. Specifically, the system continuously monitored the camera's network quality indicators and detected that its link bandwidth was consistently below a preset threshold and its packet loss rate was consistently above a preset threshold. The video data transmission link quality was no longer sufficient to support real-time identity recognition under the high-power computing mode of the cloud center. The system automatically generated a computing power mode switch command, switching the recognition task associated with this device from the cloud to a nearby edge computing-enabled supplementary camera. During the switch, the system first deployed and loaded the identity recognition algorithm on the supplementary camera, and then migrated the recognition task associated with the Xingfu Road intersection camera to the supplementary camera. During the migration, the cloud maintained tracking of the identified target until the supplementary camera took over. Throughout the entire switch, Zhang's travel status was continuously determined, and no blind spots appeared in the monitoring.

[0103] exist Figure 2 The aforementioned monitoring process is presented in real-time graphically on the 3D visualization monitoring interface. Zhang's school route is dynamically drawn on a 3D twin map, with different icons marking the departure, transit, and arrival statuses at their respective locations. Safe departure events are displayed with a specific icon in the south gate area of ​​the school. Parents and administrators can both intuitively view Zhang's complete monitoring status through this interface. Figure 1 The process shown corresponds to the complete data flow in the above scenario, from edge device collection, capability detection, computing power mode judgment, algorithm distribution and deployment, data processing, result upload to the cloud, to multi-terminal applications.

[0104] From a system architecture perspective, the aforementioned multi-terminal presentation relies on a multi-terminal collaboration mechanism within the intelligent application layer. The system employs a microservice architecture, breaking down functions such as school safety monitoring, equipment management, event management, and user management into independent microservices. This enables modular adaptation and real-time data synchronization between the operational cloud center dashboard, the management backend, and the WeChat mini-program. The operational cloud center dashboard utilizes high-concurrency data visualization rendering technology based on WebGL to achieve real-time updates of operational metrics such as the number of school safety attendees, duration of school safety monitoring, and safety incidents. The management backend handles the review and configuration of protected objects and paths, as well as the full lifecycle management of sensing devices and safety incidents. The WeChat mini-program uses a lightweight login method with WeChat authorization, ensuring secure storage and anonymized display of user information. The system establishes a standardized data interaction protocol between the visualization engine and the frontend, enabling real-time synchronization of the sensing layer video stream, recognition algorithm results, children's school route data, and the 3D twin system. In the 3D twin map, the system dynamically draws the school safety trajectories, marks safety events in real time, and retrieves camera videos in conjunction with the system. It also supports customizing guardian paths within the 3D twin map, associating camera locations and recognition algorithms along these paths, and enabling coordinated configuration of guardian paths with the perception and algorithm layers. This adapts to the personalized home-road-school trajectories of different children. Message push notifications employ a mechanism combining identity recognition, location determination, and time triggering. Camera recognition results, location information, and preset time periods are used as push trigger factors, achieving precise delivery through the WeChat mini-program's message push interface. A message deduplication and delay compensation mechanism is implemented to prevent missed or duplicate messages. The system establishes a multi-role hierarchical permission system for the management committee, schools, and parents, dividing access permissions for school safety data. Parents only see their own child's school safety information, while the management committee and schools display global school safety data, ensuring secure data sharing and privacy protection.

[0105] It should be noted that the time points, distance values, time difference values, confidence levels, and thresholds given in this embodiment are all illustrative examples. In practical applications, they can be flexibly configured according to factors such as the school's specific schedule, the deployment location and computing power of each sensing device along the path, the network environment, and security management requirements. For example, more sensing devices can be connected along the school route, and the division of the travel status can be further refined according to the complexity of the actual path; the dismissal time window can be set according to the dismissal time of different grades; the preset maximum travel distance and preset maximum time difference can be adjusted according to the personnel density and spatial layout at the school gate. Regardless of how the above parameters change, the system follows the method described in the aforementioned embodiment, determines the computing power mode based on whether the edge sensing device has local inference capabilities, performs identity recognition on the target in the video data, and determines the travel status and safe departure from school based on the identity recognition result and the location determination result, and continues to execute the above recognition and determination tasks during the computing power mode switching process.

[0106] The above complete scenario demonstrates that the system continuously outputs student arrival and departure status during morning and afternoon school hours, allowing parents to monitor their children's departure, route, arrival, and safe departure from school in real time via a WeChat mini-program. Cameras with edge computing capabilities only transmit the recognition results and feature data after local identification, reducing bandwidth consumption for video streaming. Ordinary cameras without edge computing capabilities transmit video data to the cloud for centralized processing via a powerful cloud-based computing mode, ensuring compatible access for existing devices. When network quality deteriorates, rendering the original computing mode unusable for real-time requirements, the system performs a hot-swap of computing power modes to migrate the identification task to an alternative device with edge computing capabilities, ensuring uninterrupted service during the switch. This process fully demonstrates the closed-loop technical solution from data acquisition, computing power adaptation, identification and judgment to multi-terminal applications, proving the feasibility and practicality of this application in a real-world deployment environment.

[0107] like Figure 7 As shown, this embodiment provides a student commuting monitoring system based on a secure encrypted network and cloud-edge collaboration. The system operates within a three-layer architecture consisting of a terminal perception layer, an intelligent central layer, and a smart application layer, and is used to execute the student commuting monitoring method based on a secure encrypted network and cloud-edge collaboration described in the preceding embodiments. The system includes a video data acquisition module 701, a computing power configuration module 702, an identification and judgment module 703, and a computing power switching module 704. The functional division of these four modules corresponds one-to-one with the processing logic of steps S110 to S140 in the embodiments, and the data flow and control relationships between the modules constitute a complete processing link from video data acquisition to monitoring status output.

[0108] The video data acquisition module 701 is used to acquire video data collected by multi-source edge sensing devices along the student's commuting route. The video data acquisition module 701 is communicatively connected to the multi-source edge sensing devices, responsible for receiving and aggregating the video data collected by each sensing device along the student's commuting route, and performing standardized processing on the video data.

[0109] The communication connection method between the video data acquisition module 701 and the multi-source edge sensing device can be flexibly selected according to the actual deployment environment, such as data transmission via a government private network, a unified video network, or a wireless communication link. The video data acquisition module 701 performs standardization processing on the video data, enabling video data of different formats and resolutions to be uniformly used in subsequent processing flows. Standardization processing may include operations such as video encoding format conversion, resolution normalization, timestamp alignment, and device identifier association. Figure 7In the architecture shown, the video data acquisition module 701 is located at the junction between the terminal perception layer and the intelligent central layer. Its function is to aggregate the raw video data scattered in various sensing devices into a standardized data stream that can be uniformly processed by the upper-layer modules.

[0110] The computing power configuration module 702 is used to determine the current computing power mode between the weak computing power mode and the strong computing power mode of the cloud center based on whether the edge sensing device has local inference capabilities, and to determine the execution location for identity recognition of video data. The computing power configuration module 702 is connected to the video data acquisition module 701 and receives standardized video data output by the video data acquisition module 701 as well as capability information of each edge sensing device.

[0111] The computing power configuration module 702 transmits the determined computing power mode and identity recognition execution location information to the identification and judgment module 703, enabling the identification and judgment module 703 to execute the identity recognition task according to the corresponding computing power mode.

[0112] The identification and determination module 703 is used to identify targets in video data, and based on the identification and location determination results, to determine the target's travel status and safe departure from school, outputting the student's school arrival and departure monitoring status. The identification and determination module 703 is connected to the computing power configuration module 702, receives the computing power mode and identification execution location information determined by the computing power configuration module 702, and performs identification on targets in the video data according to the determined computing power mode.

[0113] The identification and judgment module 703 ultimately outputs the student's school attendance monitoring status, including normal and abnormal states. The monitoring status output by the identification and judgment module 703 can be further transmitted to various application terminals in the smart application layer, such as the large screen display terminal, the management backend, and the parent terminal mini-program, to realize multi-terminal presentation and message push of the monitoring status.

[0114] The computing power switching module 704 is used to switch the current computing power mode between the weak computing power mode and the strong computing power mode of the cloud center in response to the computing power mode switching command during system operation, and to continuously perform identity recognition, travel status determination, and safe departure determination during the switching process. The computing power switching module 704 is connected to the computing power configuration module 702 and the identification and determination module 703.

[0115] exist Figure 7In the overall system framework shown, the video data acquisition module 701 corresponds to the unified video access and management function of the terminal perception layer; the computing power configuration module 702 and the computing power switching module 704 correspond to the computing power scheduling function of the intelligent hub layer; and the identification and judgment module 703 corresponds to the algorithm identification function of the intelligent hub layer. The comprehensive scheduling, configuration management, mobile service terminal, message push, and data collaboration functions of the intelligent application layer are implemented by calling the guardian status data output by the identification and judgment module 703. A security system runs through the terminal perception layer, the intelligent hub layer, and the intelligent application layer, providing encrypted protection for data interaction between each layer.

[0116] This embodiment elaborates on the identification of abnormal traffic behavior and abnormal events in school safety incidents.

[0117] Abnormal traffic behavior recognition includes the dynamic detection of pedestrians and non-motorized vehicles running red lights. The system performs target detection on intersection video data, distinguishes between three types of targets: motor vehicles, non-motorized vehicles, and pedestrians, and identifies the time, location, and target type of each type of target running red lights. Figure 6 The system displays the detection results of pedestrians, motor vehicles, and non-motorized vehicles from the intersection surveillance footage, as well as the student travel protection statistics and individual behavior profiles generated based on these results.

[0118] Based on the identification of red-light running, the system performs group hazard analysis and individual habit analysis on the monitoring data. Group hazard analysis helps administrators determine the traffic risk level at intersections around schools. Specifically, it summarizes and statistically analyzes the frequency and type distribution of traffic violations at each intersection weekly and monthly, and also analyzes the frequency and type distribution of violations at each intersection from a spatiotemporal perspective to uncover the spatiotemporal patterns of group traffic hazards, providing a basis for schools and management departments to develop targeted safety education plans. Individual habit analysis analyzes the individual traffic habits of those being monitored. Specifically, it constructs a multi-dimensional behavioral characteristic indicator system based on children's red-light running and safe passage behavior data, generating a data profile of children's traffic habits. Then, based on the data profile, it generates an individual traffic habit profile, achieving quantitative analysis and feature extraction of individual behaviors. The individual traffic habit profile can be visually presented on a large screen and fed back to parents through the system. Parents can then use the profile information to provide targeted guidance and correct the undesirable traffic habits of their children.

[0119] Anomaly detection includes feature extraction and scene determination for both falls and crowd gatherings. The system extracts human posture features to identify falls and crowd density features to identify crowd gatherings, and sets trigger thresholds for anomalies. On roads around schools, the system uses detection data and algorithms for crowd gatherings and falls to identify suspected group fights. Upon detecting an anomaly, the system immediately pushes anomaly information, including the event type, location, and time, to the system dashboard, ensuring that relevant parties are promptly informed and can take appropriate action.

[0120] It should be noted that the abnormal traffic behavior recognition and abnormal event recognition described in this embodiment can be executed in parallel with the trip status determination and safe departure determination in the aforementioned embodiments, or they can be selectively deployed on different types of sensing points as needed. All of the above-mentioned recognition results and event information can be presented and pushed through multi-terminal applications.

[0121] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0122] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0123] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0124] It should be noted that if the function is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0125] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0126] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for protecting students going to and from school based on a secure encrypted network and cloud-edge collaboration, characterized in that, include: Acquire video data collected by multi-source edge sensing devices along students' routes to and from school; Based on whether the edge sensing device has local reasoning capability, the current computing power mode is determined between the weak computing power mode and the strong computing power mode of the cloud center, and the execution location for identity recognition of the video data is determined, the execution location including the edge sensing device or the cloud center. The target in the video data is identified, and the travel status and safe departure from school of the target are determined based on the identification result and the location determination result, and the student's going to and from school protection status is output. During system operation, in response to the computing power mode switching command, the current computing power mode is switched between the weak computing power mode and the strong computing power mode of the cloud center, and the identity recognition, the trip status determination and the safe departure determination are continuously executed during the switching process; The data interaction between the cloud, edge, and device is encrypted using a session dynamic password mechanism.

2. The method according to claim 1, characterized in that, Determining the current computing power mode includes: When the edge sensing device has local reasoning capability, it determines that the current computing power mode is the cloud center weak computing power mode, and performs the identity recognition on the edge sensing device and only returns the recognition result and feature data. When the edge sensing device does not have local reasoning capability, it determines that the current computing power mode is the cloud center's strong computing power mode, and sends the video data back to the cloud center to perform the identity recognition.

3. The method according to claim 2, characterized in that, The response to the computing power mode switching command includes: In response to monitoring results of device status, network quality, and computing load, and / or administrator instructions, the computing mode switching instruction is generated; After completing the switch to the current computing power mode, reconfigure the algorithm deployment location and video transmission strategy.

4. The method according to claim 1, characterized in that, The location determination result is determined in the following manner: The basic location of the target is determined by the coverage area of ​​the multi-source edge sensing device; After the target's identity is verified, the target's precise location is determined based on the visual calibration relationship between the image and geographic coordinates of the multi-source edge sensing device.

5. The method according to claim 4, characterized in that, Determining the precise location of the target based on the visual calibration relationship between the image and geographic coordinates of the multi-source edge sensing device includes: Obtain the pixel coordinates of the target in the captured image and the pixel coordinates of the preset calibration point; The pixel offset is determined based on the pixel coordinates of the target in the captured image and the pixel coordinates of the preset calibration point. Based on the visual calibration relationship between the image and geographic coordinates of the multi-source edge sensing device, the pixel offset is converted into latitude and longitude offset; The precise location of the target is obtained by superimposing the latitude and longitude offset onto the reference latitude and longitude of the multi-source edge sensing device.

6. The method according to claim 1, characterized in that, The safe departure criteria include: Within the preset school dismissal time window, the identities of the target and the guardian are confirmed through facial recognition; Based on the location determination result, determine whether the target falls into the preset off-campus area; Detect the effective travel relationship between the target and the guard within the preset off-campus area; If no valid peer relationship is detected by the end of the preset school dismissal time window, an abnormal school departure status will be output.

7. The method according to claim 6, characterized in that, The effective companion relationship is determined based on the companion matching degree, which is determined based on the spatial distance and the time difference of appearance between the target and the guardian; when the spatial distance is within the preset maximum companion distance and the time difference of appearance is within the preset maximum time difference, it is determined to be a valid companion relationship.

8. The method according to claim 7, characterized in that, The output of the abnormal status of leaving school includes: Based on the student's comprehensive confidence score, the peer matching score, and the off-campus area matching results, the confidence score for safe departure from campus is determined. When the confidence level for safe departure from school reaches the preset threshold for safe departure from school, it is determined that the departure from school is safe. When the confidence level for safe departure from school does not reach the preset safe departure threshold, the departure is determined to be abnormal.

9. The method according to claim 1, characterized in that, The session dynamic password mechanism includes: During the communication session establishment phase, a dynamic password for the communication session is generated and bound to the communication session. During the duration of the communication session, the session dynamic password is periodically updated according to a preset iteration cycle; When an abnormal network event is detected, the current session dynamic password is forcibly invalidated and a new session dynamic password is generated. The session dynamic password is immediately destroyed when the communication session ends normally or is abnormally disconnected. The session dynamic cipher is generated based on the base key, terminal device identifier, role code, and session random number; communication sessions of different access roles use independent session dynamic ciphers to isolate the business loads of different roles; the session dynamic cipher of a single role can only decrypt data within the scope of that role's permissions.

10. A student commuting monitoring system based on a secure encrypted network and cloud-edge collaboration, characterized in that: include: The video data acquisition module is used to acquire video data collected by multi-source edge sensing devices along the students' commuting route to and from school. The computing power configuration module is used to determine the current computing power mode between the weak computing power mode and the strong computing power mode of the cloud center based on whether the edge sensing device has local inference capability, and to determine the execution location for identity recognition of the video data, wherein the execution location includes the edge sensing device or the cloud center. The identification and determination module is used to identify the target in the video data, and to determine the travel status and safe departure from school of the target based on the identification result and the location determination result, and output the student's school arrival and departure protection status. The computing power switching module is used to switch the current computing power mode between the weak computing power mode and the strong computing power mode of the cloud center in response to the computing power mode switching command during system operation, and to continuously perform the identity recognition, the trip status determination and the safe departure determination during the switching process; The data interaction between the cloud, edge, and device is encrypted using a session dynamic password mechanism.

Images