A method and system for dynamic aggregation and sharing of end environmental interaction information for a closed park

Abstract

The application provides a closed park-oriented end environment interaction information dynamic aggregation and sharing method and system. The method is applied to a cloud server and comprises the following steps: receiving initial environment interaction data sets uploaded by a plurality of first terminals; clustering rich media segments according to scene attribute tags, and respectively classifying the rich media segments into an entrance management interaction resource pool, a pre-stored point guidance resource pool and a high-density vertical building route finding resource pool; calling a fusion scoring model based on machine vision objective features and user crowdsourcing feedback weights to perform dynamic quality scoring, and constructing a multi-modal end environment interaction information library; receiving an environment interaction request of a second terminal, extracting a rich media segment with the highest dynamic quality score as target scene guidance data according to real-time node state parameters of a delivery task, and issuing the target scene guidance data. The application dynamically maintains data timeliness by using a natural index-based negative feedback index decay model, overcomes blind area drift through a multi-source fusion positioning and anti-fraud verification mechanism, and introduces an end-side sandbox hosting and random noise physical overwriting mechanism to realize an end-to-end privacy closed loop, thereby significantly improving the performance and data security of closed park end delivery.

Description

Technical Field

[0001] The present invention relates to the fields of computer network, distributed data processing and computer vision, in particular to a method and system for dynamic aggregation and sharing of terminal environment interaction information for closed parks, and particularly covers data privacy protection, distributed authorized cooperative architecture, real-time image privacy feature hiding processing based on artificial intelligence and geofence-based multi-modal data dynamic aggregation technology. Background Art

[0002] In the existing logistics distribution and public navigation technologies, map navigation and route planning technologies based on public roads have been developed very maturely. However, for the last-mile terminal delivery link in non-standard closed parks (such as closed residential quarters, large commercial complexes and high-density high-rise office buildings), operators still face great technical bottlenecks of information interaction and positioning failure. The specific technical defects are as follows:

[0003] The physical difference of the terminal environment hinders the effectiveness of prior data: different closed parks have great differences in access rules, property management release procedures, internal temporary storage facility configuration and elevator control logic. Traditional standard map navigation belongs to flat road network data, which cannot cover long-tail micro-environment features such as security room access structural parameters, vertical elevator split-drive rules, accurate positions of freight elevators and property collection counters, resulting in serious delay of operators due to the lack of accurate multi-modal guidance.

[0004] Defects of redundancy and timeliness lag of long-tail guidance data: even if some users upload local guidance for specific closed spaces, due to the lack of standard technical classification and spatiotemporal clustering algorithms, a large number of similar or expired photos and video fragments are disorderedly accumulated in the cloud, causing serious waste of storage and computing resources. Meanwhile, when closed parks change traffic rules due to sudden control, property management change or physical facility maintenance, static guidance data cannot achieve real-time update and route triggering interruption.

[0005] Signal blind areas in closed spaces cause failure of traditional navigation technologies: closed parks often have high-density high-rise building groups or underground garages. Such physical structures seriously block satellite navigation signals, and easily cause multi-path effect leading to abrupt positioning coordinate changes or signal loss. Existing navigation solutions rely on a single global positioning system, which is prone to fail to trigger data ejection of corresponding nodes due to positioning drift in signal blind areas, and lack an anti-cheating verification mechanism for virtual positioning to steal private data. Summary of the Invention

[0006] The present invention provides a method and system for dynamic aggregation and sharing of terminal environment interaction information for closed parks, which is applied to a cloud server.

[0007] To address the aforementioned technical problems, this invention provides a method for dynamic aggregation and sharing of terminal environment interaction information in closed campuses, comprising the following steps:

[0008] Receive initial environment interaction datasets uploaded by multiple first terminals. The initial environment interaction datasets include the logical boundaries of the target closed campus, rich media segments for specific spatial delivery nodes, and synchronously associated scene attribute tags.

[0009] Based on the scene attribute tags, multiple rich media segments belonging to the same closed park are clustered in spatial business flow, and divided and assigned to the corresponding specific scene resource pools. The specific scene resource pools include at least the entrance property management interaction scene resource pool, the pre-storage point guidance scene resource pool, and the high-density vertical building wayfinding scene resource pool.

[0010] A fusion scoring model based on machine vision objective features and user crowdsourced feedback weights is invoked to dynamically score the quality of multiple rich media segments in the same scene resource pool, and a multimodal terminal environment interaction information database for the target closed park is aggregated and constructed in the cloud server.

[0011] Receive an environmental interaction request sent by a second terminal, the environmental interaction request carrying the current target closed park identifier and the real-time node status parameters of the current delivery task;

[0012] Based on the real-time node status parameters, the rich media segment with the highest dynamic quality score is extracted from the multimodal terminal environment interaction information database and sent to the second terminal as target scene guidance data.

[0013] As a further improvement of the present invention, the correspondence between the scene attribute tags and the scene resource pool is specifically as follows:

[0014] When the scene attribute label is the gate access control label, the corresponding entrance property management interaction scene resource pool includes property management access structured control parameter data and access gate control multimodal media segments.

[0015] When the scene attribute tag is the item collection and temporary storage tag, the corresponding front temporary storage point guidance scene resource pool includes real-scene image data of the public temporary storage area and location guidance fragments of non-standard collection counters.

[0016] When the scene attribute label is a building interior space guidance label, the corresponding high-density vertical building wayfinding scene resource pool includes vertical elevator group distribution images and multimodal media clips of internal freight passage entrances and exits.

[0017] As a further improvement of the present invention, the fusion scoring model for constructing the multimodal terminal environment interaction information database specifically adopts the following calculation logic:

[0018] The cloud server calculates an objective machine vision score based on the image clarity and landmark exposure in the rich media clip, and collects crowdsourcing positive and negative feedback rates and actual delivery delay fluctuation data after the second terminal calls the clip in history, and calculates the crowdsourcing interaction score.

[0019] The weighted sum is obtained by multiplying the objective machine vision score by the first preset weight coefficient and then adding the product of the crowdsourcing interaction score and the second preset weight coefficient.

[0020] The weighted summation result is multiplied by a real-time time decay coefficient calculated based on the timestamp generated by the rich media segment minus the current system timestamp to generate the dynamic quality score;

[0021] In the multimodal terminal environment interaction information database, the rich media segment with the highest dynamic quality score in the same scenario is configured as the default preferred pop-up node.

[0022] As a further improvement to the present invention, after the target scene guidance data is distributed, a dynamic circuit breaker and path replacement mechanism based on real-time crowdsourcing feedback is also included:

[0023] Real-time monitoring of the error reporting rate submitted by users on the second terminal side for the currently issued rich media segment;

[0024] When the number of reports of temporary blocking or physical rule changes for a certain default preferred pop-up node exceeds the risk control threshold within a preset time window, the rich media segment is marked as invalid and blocked in the cloud server.

[0025] The default preferred pop-up node in the multimodal terminal environment interaction information database is automatically and dynamically switched to the backup rich media segment with the second highest dynamic quality score in the current scene resource pool.

[0026] As a further improvement of the present invention, the step of extracting rich media segments based on the real-time node state parameters specifically includes an adaptive control flow triggering mechanism:

[0027] When the real-time trajectory of the second terminal is detected to be close to the logical boundary of the target closed park, the real-time node status parameter is automatically triggered to the entrance preparation node status, and the optimal rich media segment in the entrance property management interaction scene resource pool is pushed to it.

[0028] When the second terminal reports a user-side rejection or transfer of service signal, the real-time node status parameter is automatically changed to the pre-storage node status, and the optimal rich media segment in the pre-storage point guidance scene resource pool is pushed to it.

[0029] As a further improvement of the present invention, the step of receiving the environmental interaction request sent by the second terminal also includes a multi-source fusion positioning and anti-spoofing verification mechanism for weak signal blind zones in enclosed spaces:

[0030] The system collects the second terminal's current global satellite navigation system basic positioning data, surrounding cellular base station triangulation positioning data, and wireless local area network probe scan fingerprint information in parallel to perform fusion positioning and generate the current fusion positioning coordinates;

[0031] The spatial deviation between the base station triangulation data and the global satellite navigation system basic positioning data is calculated in the cloud server. When the fingerprint information scanned by the wireless local area network probe does not match the pre-stored fingerprint spatial distribution feature database, a positioning spoofing interception signal is triggered to block subsequent data transmission.

[0032] As a further improvement of the present invention, in the calculation process of the dynamic quality score, the real-time time decay coefficient adopts a negative feedback exponential decay model based on the natural exponent:

[0033] The specific calculation method is to use the base of the natural logarithm as the base and the product of the negative preset decay constant and the time difference in days as the exponent for power operation, wherein the time difference in days is the difference in days between the current system timestamp and the timestamp generated by the rich media segment.

[0034] As the number of days of the time difference increases, the dynamic quality score is subjected to non-linear penalty weight reduction.

[0035] As a further improvement to the present invention, after the data is sent to the second terminal, a terminal-side sandbox hosting and forced cleanup mechanism is also included:

[0036] As a further improvement of the present invention, in the calculation process of the dynamic quality score, the real-time time decay coefficient adopts a negative feedback exponential decay model based on the natural exponent: the specific calculation method is to use the base of the natural logarithm as the base, and the product of the negative preset decay constant and the time difference in days as the exponent for power operation, wherein the time difference in days is the difference in days between the current system timestamp and the timestamp of the rich media segment generation; as the time difference in days increases, the dynamic quality score is subjected to nonlinear penalty weight reduction.

[0037] The above solution can ensure the timeliness of the cloud database and completely eliminate outdated and inflexible data through exponential negative feedback penalties.

[0038] As a further improvement of the present invention, after the data is sent to the second terminal, an end-side sandbox hosting and forced cleanup mechanism is also included: the target scene guidance data is forcibly hosted in an isolated sandbox cache opened locally on the second terminal, and its access control command is configured to a read-only state for a single task cycle.

[0039] When the second terminal is continuously monitored to leave the logical boundary of the target closed park, or when a business completion signal is received indicating that the current delivery task has been completed, the cloud server remotely ejects a write-overwrite command to the second terminal, forcibly clearing and rewriting the isolation sandbox cache with random noise data.

[0040] The above scheme achieves the read-and-burn functionality of fine-grained micro-addresses and multimodal images at the execution end through the hardware layer, thus achieving an end-to-end closed loop for absolute data privacy protection. Attached Figure Description

[0041] Figure 1 is a system architecture diagram of dynamic aggregation and sharing of terminal environment interaction information for closed campuses provided by an embodiment of the present invention.

[0042] Figure 2 is a classification diagram of the multimodal terminal environment interaction information database scenario resource pooling provided in an embodiment of the present invention.

[0043] Figure 3 is a timing diagram of the crowdsourcing aggregation and scoring process of end-to-end environmental interaction information provided in an embodiment of the present invention.

[0044] Figure 4 is a flow diagram of the end-point adaptive control flow based on real-time trajectory and service signals provided in an embodiment of the present invention. Detailed Implementation

[0045] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0046] System basic hardware, software modules, and underlying dictionary architecture description:

[0047] As shown in Figure 1, this system includes multiple first terminals 510, a distributed cloud server 520, and a second terminal 530 for performing end-to-end fulfillment and pathfinding.

[0048] The first terminal 510 is typically a worker terminal with crowdsourced shooting or data reporting capabilities, used to collect and upload environmental interaction datasets consisting of detailed latitude and longitude coordinates, route guidance videos, or images. The internal software architecture of the cloud server 520 is divided into a dataset receiving module 521, a spatial business flow clustering module 522, an information database construction module 523, a multimodal terminal environmental interaction information database 524, a node status matching module 525, and a directional guidance distribution module 526.

[0049] To completely overcome the limitations of conventional map navigation technology in areas without satellite navigation signals, such as indoor spaces and underground parking garages, and to protect users' long-tail address privacy, this system's cloud-based multimodal terminal environment interaction information database 524 adopts a dictionary architecture based on scene attribute pooling classification.

[0050] As shown in Figure 2, for each specific target closed campus, its multimodal terminal environment interaction information database 524 is internally decoupled vertically into three completely isolated long-tail business scenario resource pools:

[0051] Entrance Property Management Interaction Scenario Resource Pool 602: This pool stores data on how personnel at the second terminal 530 pass through the physical barrier of the closed park gate. Among them, the property management access structured control parameter data 605 includes plain text control fields such as the specific restricted hours for the gate, whether ID cards are required as collateral, and registration rules for food delivery or express delivery; the access gate control multimodal media clip 606 is a real-scene image historically collected by the first terminal 510, clearly indicating the spatial arrangement of the hidden barcode reader or facial recognition system of the access control.

[0052] Pre-delivery point guidance scenario resource pool 603: The data stored in this pool is for overflow scenarios of non-face-to-face delivery. Public temporary storage area real-scene image data 607 includes real-scene images of the storage shelves and self-pickup lockers behind the security room, with machine vision annotation boxes; non-standard collection counter location guidance segment 608 includes multimodal real-scene trajectory guidance from the main gate to specific property self-built temporary storage points.

[0053] High-density vertical building wayfinding scenario resource pool 604: The data stored in this pool targets the vertical high-dimensional space inside large office buildings or high-rise residential buildings. Vertical elevator group distribution image 609 clearly identifies the high and low floor separation rules and asymmetrical operation floor restrictions of specific elevator groups; multimodal media clips 610 at the entrances and exits of internal freight passages guide operators on how to avoid the main entrance of passenger elevators and quickly enter lower-level freight elevators or fire underground passages.

[0054] Deduplication and cleaning of rich media data streams at the end based on a multi-dimensional fusion weighting algorithm (Figures 1 and 2) Figure 3 This paper analyzes the clustering and processing flow of multi-source crowdsourced data in the cloud, specifically including the following steps:

[0055] Step 701, Dataset Upload: Multiple first terminals 510 representing crowdsourced collection ends collect video clips of specific closed park gates or interiors, add scene attribute tags, and upload them to the cloud server 520, where the dataset receiving module 521 performs unified aggregation.

[0056] Step 702, Spatial Service Flow Clustering and Cleaning: After identifying the aforementioned scene attribute labels, the spatial service flow clustering module 522 first calls the machine vision optical flow tracing and feature point matching algorithm to identify that multiple video segments all have high-frequency exposure of the same physical landmark within a specific time period. Based on this, the system determines that these videos belong to the same physical spatial node and uniformly classifies them into the corresponding specific scene resource pool for homogeneous deduplication comparison.

[0057] Step 703, Scoring and Database Construction: The database construction module 523 calls a preset mathematical model to automatically assess the quality of multiple video segments. The specific quality scoring and evaluation logic is described below:

[0058] The final dynamic quality score for each rich media segment is generated by a cascaded calculation of machine vision objective score, crowdsourced interaction score, first and second preset weighting coefficients, and real-time time decay coefficient. Specifically, the cloud server 520 first automatically extracts the Laplacian variance of the video to calculate the image sharpness, calculates the smoothness of lens shake through pixel displacement frame difference, and identifies the total exposure time of valid architectural landmarks in the image. The three factors are weighted to generate a machine vision objective score with a value ranging from zero to one hundred.

[0059] Secondly, based on historical data on the positive and negative feedback rates and actual delivery delay fluctuations of the second terminal after actually calling the rich media segment, the system calculates a crowdsourcing interaction score ranging from zero to one hundred. Specifically, the calculation is as follows: subtract the number of historical negative reviews and error reports from the number of historical positive reviews, divide the difference by the total number of historical calls to the segment, multiply the calculated ratio by a coefficient of one hundred, and then add a time-consuming correction factor. If the actual delivery delay of the delivery person looking at the video to find the target location is lower than the historical average route-finding time, the time-consuming correction factor is positive; otherwise, a deduction penalty is applied.

[0060] Subsequently, the cloud server 520 multiplies the machine vision objective score by the first preset weight coefficient, and then performs a weighted summation of the product of the crowdsourcing interaction score and the second preset weight coefficient to obtain a weighted total score; wherein, the sum of the first preset weight coefficient and the second preset weight coefficient is constantly restricted to one point zero.

[0061] Finally, the system multiplies the weighted total score by a real-time time decay coefficient calculated using a negative feedback exponential decay model based on the natural index, thus obtaining the dynamic quality score of the rich media segment. The specific calculation method for the real-time time decay coefficient is as follows: It is operated on by exponentiation using the base of the natural logarithm as the base and the product of a negative preset decay constant and the number of days of time difference as the exponent; the number of days of time difference is the difference between the current system timestamp and the timestamp at which the rich media segment was generated. As the segment ages and the difference increases, the time decay coefficient exhibits a non-linear penalty weighting.

[0062] Internet server 520 automatically calculates the rich media clip with the highest dynamic quality score in similar scenarios and configures it as the default preferred pop-up node in the multimodal terminal environment interaction information database 524. The remaining low-scoring videos are either cold-stored for later use or periodically physically deleted, achieving data deduplication at the plain text level without relying on any mathematical formulas or symbols.

[0063] Example 2: Cascaded control process of signal blind zone fusion positioning, dynamic fuse breaking and path replacement:

[0064] Referring to Figures 1, 2, and 4, when an operator using the second terminal 530 is performing a last-mile delivery task in a specific office building or residential community, the system adaptively triggers control flow guidance based on real-time trajectory and special business signals, and executes closed-loop data destruction on the terminal side. The specific steps include the following:

[0065] Step 704, request reception, and step 801, monitoring the second terminal: The second terminal 530 sends an environmental interaction request to the cloud server 520. To combat satellite positioning signal drift, loss, or virtual positioning spoofing caused by high-density buildings in enclosed parks and indoor environments, the node state matching module 525 activates a multi-source fusion positioning and anti-spoofing verification mechanism. It collects the second terminal 530's current global satellite navigation system basic positioning data, surrounding cellular base station triangulation data, and wireless LAN probe scan fingerprint information in parallel. The cloud server 520 calculates the spatial deviation between the base station triangulation data and the GNSS positioning data. When the Wi-Fi probe scan fingerprint information does not match the pre-stored fingerprint spatial distribution feature library, it determines that the current approximate coordinates contain virtual positioning spoofing, triggers a positioning spoofing interception signal, and blocks subsequent data transmission. After successful verification, the current fused positioning coordinates are generated.

[0066] Step 802, Logical Boundary Determination and Step 803, Entrance Preparation State Trigger: When Step 802 determines that the current fused positioning coordinates of the second terminal 530 have entered the logical boundary of the target closed park, the system automatically triggers the real-time node status parameter to the entrance preparation node state. The directional guidance module 526 executes step 705, extracting the highest-rated multimodal media segment 606 of the access gate control from the entrance property management interaction scenario resource pool 602 in the information database and displaying it on the second terminal 530 to guide the staff to quickly pass through the gate.

[0067] Step 804, Business Signal Judgment and Step 805, Temporary Storage Node Status Trigger: In subsequent delivery, if the second terminal 530 receives special business signals such as user rejection or transfer to storage, or if the operator manually triggers transfer to storage, after capturing the signal in step 804, the real-time node status parameters are automatically changed to the previous temporary storage node status. The directional guidance delivery module 526 accurately extracts rich media data containing non-standard collection counter location guidance segments 608 from the previous temporary storage point guidance scenario resource pool 603 and pushes it to the second terminal 530, guiding it to quickly find the hidden goods storage location.

[0068] Step 806, Error Rate Monitoring and Step 807, Threshold Comparison: When the second terminal 530 performs pathfinding, the system activates a dynamic circuit breaker and path substitution mechanism based on real-time crowdsourced feedback. Step 806 continuously monitors the error reporting rate submitted by users on the second terminal 530 for the currently issued rich media segment. When the number of reports of temporary blocking or physical rule changes for a certain default preferred pop-up node exceeds the risk control threshold in step 807 within a preset time window, the process flows to step 808.

[0069] Step 808, Failure Marking and Step 809, Backup Dynamic Switching: The cloud server 520 marks the rich media segment as an failed interception state in the cloud and initiates step 809, automatically and dynamically switching the default preferred pop-up node in the multimodal terminal environment interaction information database 524 to the backup rich media segment with the second highest dynamic quality score in the current scene resource pool for path replacement, ensuring the continuity of the performance process.

[0070] End-side sandbox hosting and forced cleanup mechanism: To prevent long-tail data leakage caused by the long-term plaintext retention of highly sensitive micro-level privacy data within the closed campus on the terminal side, the system implements end-side sandbox hosting control logic. All target scene guidance data sent to the second terminal 530 is forcibly hosted in a locally established isolated sandbox cache area. Its access control instructions are configured to be read-only for a single task cycle, blocking any form of local album saving, screenshot export, or external read handle. When the system continuously monitors that the current location of the second terminal 530 has left the logical boundary of the target closed campus, or receives a business completion signal indicating that the current delivery task has been completed, the cloud server 520 immediately revokes the data access authorization and remotely ejects and overwrites the destruction instruction to the second terminal 530. After receiving the instruction, the end-side control software module of the second terminal 530 seamlessly activates the end-side cleanup program in the background of the system, releases the memory handle, and calls the disk sector erasure interface of the underlying file system. It uses random noise data to perform multiple rounds of seamless overwriting and erasing of the physical sectors of the isolated sandbox cache that temporarily stores the privacy data. In this way, all precise spatial coordinates and multimodal boot data in the local cache are completely cleared and destroyed at the physical hardware level, realizing end-to-end absolute closed-loop protection of location privacy.

Claims

1. A method for dynamic aggregation and sharing of terminal environment interaction information in closed campuses, characterized in that, The method is applied to a cloud server and includes the following steps: Receive initial environment interaction datasets uploaded by multiple first terminals. The initial environment interaction datasets include the logical boundaries of the target closed campus, rich media segments for specific spatial delivery nodes, and synchronously associated scene attribute tags. Based on the scene attribute tags, multiple rich media segments belonging to the same closed park are clustered in spatial business flow, and divided and assigned to the corresponding specific scene resource pools. The specific scene resource pools include at least the entrance property management interaction scene resource pool, the pre-storage point guidance scene resource pool, and the high-density vertical building wayfinding scene resource pool. A fusion scoring model based on machine vision objective features and user crowdsourced feedback weights is invoked to dynamically score the quality of multiple rich media segments in the same scene resource pool, and a multimodal terminal environment interaction information database for the target closed park is aggregated and constructed in the cloud server. Receive an environmental interaction request sent by a second terminal, the environmental interaction request carrying the current target closed park identifier and the real-time node status parameters of the current delivery task; Based on the real-time node status parameters, the rich media segment with the highest dynamic quality score is extracted from the multimodal terminal environment interaction information database and sent to the second terminal as target scene guidance data.

2. A dynamic aggregation and sharing system for terminal environmental interaction information in closed campuses, characterized in that: include: Data set receiving module: used to receive initial environment interaction datasets uploaded by multiple first terminals. The initial environment interaction datasets include the logical boundaries of the target closed park, rich media segments for specific spatial delivery nodes, and synchronously associated scene attribute tags. Spatial business flow clustering module: Based on the scene attribute tags, it is used to divide and classify multiple rich media segments belonging to the same closed park into the corresponding entrance property management interaction scene resource pool, the pre-storage point guidance scene resource pool and the high-density vertical building wayfinding scene resource pool. Information database construction module: used to call the fusion scoring model based on machine vision objective features and user crowdsourced feedback weights to dynamically score the quality of multiple rich media segments in the same scene resource pool, and aggregate and construct a multimodal terminal environment interaction information database for the target closed park; Node status matching module: used to receive the environmental interaction request sent by the second terminal, and perform target segment matching based on the target closed campus identifier and the real-time node status parameters of the current delivery task carried therein; Oriented guidance delivery module: used to extract the rich media segment with the highest dynamic quality score from the multimodal terminal environment interaction information database according to the real-time node status parameters and deliver it as target scene guidance data to the second terminal.

3. The method according to claim 1, characterized in that, The specific correspondence between the scene attribute tags and the scene resource pool is as follows: When the scene attribute label is the gate access control label, the corresponding entrance property management interaction scene resource pool includes property management access structured control parameter data and access gate control multimodal media segments. When the scene attribute label is the item collection and temporary storage label, the corresponding front temporary storage point guidance scene resource pool includes real-scene image data of the public temporary storage area and location guidance fragments of non-standard collection counters. When the scene attribute label is a building interior space guidance label, the corresponding high-density vertical building wayfinding scene resource pool includes images of vertical elevator group distribution and multimodal media clips of internal freight passage entrances and exits.

4. The method according to claim 1, characterized in that, The fusion scoring model used in constructing the multimodal terminal environment interaction information database specifically employs the following calculation logic: The cloud server calculates an objective machine vision score based on the image clarity and landmark exposure in the rich media clip, and collects crowdsourcing positive and negative feedback rates and actual delivery delay fluctuation data after the second terminal calls the clip in history, and calculates the crowdsourcing interaction score. The weighted sum is obtained by multiplying the objective machine vision score by the first preset weight coefficient and then adding the product of the crowdsourcing interaction score and the second preset weight coefficient. The weighted summation result is multiplied by a real-time time decay coefficient calculated based on the timestamp generated by the rich media segment minus the current system timestamp to generate the dynamic quality score; In the multimodal terminal environment interaction information database, the rich media segment with the highest dynamic quality score in the same scenario is configured as the default preferred pop-up node.

5. The method according to claim 1, characterized in that, After the target scenario guidance data is distributed, a dynamic circuit breaker and path replacement mechanism based on real-time crowdsourcing feedback is also included: Real-time monitoring of the error reporting rate submitted by users on the second terminal side for the currently issued rich media segment; When the number of reports of temporary blocking or physical rule changes for a certain default preferred pop-up node exceeds the risk control threshold within a preset time window, the rich media segment is marked as invalid and blocked in the cloud server. The default preferred pop-up node in the multimodal terminal environment interaction information database is automatically and dynamically switched to the backup rich media segment with the second highest dynamic quality score in the current scene resource pool.

6. The method according to claim 1, characterized in that, The step of extracting rich media segments based on the real-time node status parameters specifically includes an adaptive control flow triggering mechanism: When the real-time trajectory of the second terminal is detected to be close to the logical boundary of the target closed park, the real-time node status parameter is automatically triggered to the entrance preparation node status, and the optimal rich media segment in the entrance property management interaction scene resource pool is pushed to it. When the second terminal reports a user-side rejection or transfer of service signal, the real-time node status parameter is automatically changed to the pre-temporary storage node status, and the optimal rich media segment in the pre-temporary storage point guidance scene resource pool is pushed to it.

7. The method according to claim 1, characterized in that, The step of receiving the environmental interaction request sent by the second terminal also includes a multi-source fusion positioning and anti-spoofing verification mechanism for weak signal blind spots in enclosed spaces: The system collects the second terminal's current global satellite navigation system basic positioning data, surrounding cellular base station triangulation positioning data, and wireless local area network probe scan fingerprint information in parallel for fusion positioning to generate the current fusion positioning coordinates; The spatial deviation between the base station triangulation data and the global satellite navigation system basic positioning data is calculated in the cloud server. When the fingerprint information scanned by the wireless local area network probe does not match the pre-stored fingerprint spatial distribution feature database, a positioning spoofing interception signal is triggered to block subsequent data transmission.

8. The method according to claim 4, characterized in that, The real-time decay coefficient adopts a negative feedback exponential decay model based on the natural exponential: The specific calculation method is to use the base of the natural logarithm as the base and the product of the negative preset decay constant and the time difference in days as the exponent for power operation, wherein the time difference in days is the difference in days between the current system timestamp and the timestamp generated by the rich media segment. As the number of days of the time difference increases, the dynamic quality score is subjected to non-linear penalty weight reduction.

9. The method according to claim 1, characterized in that, The process of sending the data to the second terminal also includes end-side sandbox hosting and forced cleanup mechanisms: The target scene guidance data is forcibly hosted in an isolated sandbox cache opened locally on the second terminal, and its access control command is configured to be read-only for a single task cycle. When the second terminal is continuously monitored to leave the logical boundary of the target closed park, or when a business completion signal is received indicating that the current delivery task has been completed, the cloud server remotely ejects a write-overwrite command to the second terminal, forcibly clearing and rewriting the isolation sandbox cache with random noise data.

Images