Point cloud model rendering method based on cloud edge coordination

Abstract

The invention relates to the technical field of point cloud model rendering, and discloses a point cloud model rendering method based on cloud edge coordination. A unique identifier is distributed for the electric power facility rendering task, and identifier generation, storage and verification are completed; secondly, converting the power point cloud data into a rendering format through a point cloud processing engine, checking geometric consistency in real time by using a data verification unit and a geometric analysis unit, and automatically correcting abnormal points; and finally, on the basis of cloud edge collaboration, power computing resource distribution is monitored through a load feature resource allocation unit, a task segmentation protocol is introduced, task priorities are dynamically sorted according to multi-dimensional features such as load intensity, equipment distance and power scene key levels, and rendering time periods are pre-allocated. The method realizes efficient scheduling of cloud edge resources in the power field, guarantees priority rendering of key areas such as transformer substation core equipment, and improves the processing precision and rendering efficiency of power point cloud data.

Description

Technical Field

[0001] This invention relates to the field of point cloud model rendering technology, specifically to a point cloud model rendering method based on cloud-edge coordination. Background Technology

[0002] In the power sector, with the continuous expansion of power grid scale and the improvement of its intelligence level, the demand for 3D modeling and rendering of power facilities is increasing. Point cloud data, as an important data source for 3D modeling, contains a large amount of information about the spatial coordinates, color, texture, etc. of power facilities, and can accurately reproduce the 3D shape of power facilities. However, in practical applications, point cloud data is characterized by large data volume and complex structure, which brings great challenges to the rendering of point cloud models.

[0003] Traditional point cloud model rendering methods mostly rely on a single cloud computing center or edge computing device, making it difficult to fully leverage the synergistic advantages of cloud computing and edge computing. When faced with large-scale power point cloud data rendering tasks, a single computing resource often becomes overloaded, leading to low rendering efficiency and even problems such as rendering stuttering and data loss. For example, when conducting 3D inspections of transmission lines, a large amount of point cloud data of power facilities such as towers and cables needs to be processed. Traditional methods may not be able to complete the rendering in a timely manner, affecting the efficiency and accuracy of the inspection.

[0004] Furthermore, traditional methods lack real-time checks and correction mechanisms for data geometric consistency during data processing. Point cloud data of power facilities may be affected by various factors during acquisition and transmission, such as sensor errors and environmental noise, leading to outliers or geometric deviations in the data. If these problems are not detected and addressed in a timely manner, they will affect the accuracy and reliability of the final rendered model, thus adversely impacting the planning, design, and maintenance of power facilities.

[0005] Meanwhile, traditional rendering methods also have shortcomings in task allocation and resource scheduling. They cannot dynamically adjust based on the load of computing resources and task priorities, resulting in critical areas not receiving priority rendering. In the power sector, the real-time performance and accuracy of rendering are crucial for certain critical areas, such as core equipment in substations and important transmission line sections, and traditional methods struggle to meet these requirements.

[0006] Furthermore, with the development of intelligent power systems, the demand for fusion processing of multi-sensor data is increasing. Traditional point cloud processing engines often cannot efficiently process point cloud data collected by multiple sensors, and cannot achieve full coverage and real-time processing of the collected area, thus affecting the quality and efficiency of point cloud model rendering. Summary of the Invention

[0007] The purpose of this invention is to provide a point cloud model rendering method based on cloud-edge coordination to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a point cloud model rendering method based on cloud-edge coordination, the method comprising: S1, Point Cloud Data Initialization: Assign a unique task identifier to each rendering task; S2, Point Cloud Data Processing: The system is equipped with a point cloud processing engine, which automatically converts point cloud data into a rendering format when data is input. The system's processing module integrates a data verification unit to check the geometric consistency of the data in real time during the conversion process. If anomalies are detected, the data is automatically corrected and marked as problems. S3, rendering optimization based on cloud-edge coordination: S31 embeds a resource allocation unit based on load characteristics into the point cloud processing engine to monitor the distribution and load of computing resources in real time. S32, To address the resource load issue, a task partitioning protocol is introduced. This protocol pre-allocates rendering segments through a fast segmentation mechanism and performs rendering according to task priority; the task priority ranking... By using a weighted model, the priority of tasks in rendering conflicts is dynamically sorted to ensure that critical areas are rendered first.

[0009] Preferably, S1 specifically includes: S11, Identifier Generation and Content Entry: Input identifier data, including task type, scene complexity, data source, task time and expected duration, through the rendering management platform. After the point cloud acquisition device is started, the device information is added to the identifier data to form a complete identifier record. S12, Identifier Storage and Association: Store the identifier data using the point cloud acquisition device, generate the corresponding task identifier, and associate the generated task identifier with the rendering task.

[0010] Preferably, step S1 further includes data verification: after the task identifier is associated, the task identifier content is read through the data verification module and compared with the identifier data stored in the rendering management platform. The verification content includes the completeness, accuracy and uniqueness of the task identifier; the successfully associated task identifier is uploaded to the rendering management platform to establish a long-term mapping relationship between the task identifier and the rendering task.

[0011] Preferably, S2 specifically includes: S21, A point cloud processing engine is installed at the system's input interface. The point cloud processing engine covers the acquisition area through a multi-sensor array. S22, when point cloud data input begins, the point cloud processing engine is triggered to detect point cloud signals in real time and automatically convert them into rendering data, including point cloud coordinates, color information, geometric structure, timestamps and key features. After successful conversion, the current status of the rendering task in the rendering management platform is updated.

[0012] Preferably, the system's processing module is configured with a multi-layered distributed geometric analysis unit to collect the overall structure and segmented geometric distribution of the data in real time. When data is transformed, the geometric analysis unit detects the geometric changes in the newly generated data, generates the actual structural data of the current data, and compares the data logic output by the point cloud processing engine with the actual structure collected by the geometric analysis unit. If a structural deviation is detected, the system will take action. If a set threshold is set, the current data is automatically marked as problematic, and the actual structural content is updated in the rendered data.

[0013] Preferably, S31 specifically includes: Resource awareness and load identification: The resource allocation unit senses the occupancy of computing resources in real time through the embedded load analysis chip. The load analysis chip uses a resource distribution analysis algorithm to detect the load distribution of resource transmission and mark computing activities with overlapping loads within the same resource type. Time-period detection and location: When resource load is detected, the specific time period during which the load occurs is located by calculating the intensity, distribution characteristics and time characteristics of the resource.

[0014] Preferably, the resource distribution analysis algorithm specifically includes: Resource occupancy status calculation: By analyzing and calculating the load energy distribution of resources, the resource occupancy status is determined. When the load energy exceeds the set energy threshold, it indicates that the resource is in an occupied state. The resource occupancy status detection condition is as follows: the resource occupancy status is determined by the energy threshold. If the energy is higher than the threshold, it indicates that the resource is occupied, and if it is lower than the threshold, it indicates that the resource is idle.

[0015] Preferably, the set of time periods for computing resource allocation is defined as a pre-allocated time period sequence. The time period sequence contains multiple consecutive time windows. In the time period sequence, each time window corresponds to a computing segment. The detection condition for overlapping load intensity is that the sum of the intensities of multiple computing resources within the time period exceeds a set conflict threshold. Load source location: If overlapping load intensity is detected, the set of load sources is located by calculating the contribution load intensity of each computing source. Computing sources whose contribution load intensity exceeds the proportional threshold are included in the load set.

[0016] Preferably, S32 specifically includes: S321, Fast Segmentation Mechanism for Pre-allocation of Time Periods: After load detection, the fast segmentation mechanism is used to dynamically adjust the time period allocation of computing resources, allocating specific processing time periods to each computing source to ensure that there is no overlap between time periods; S322, Dynamic prioritization of tasks: Based on the intensity of computing resources, the distance between the computing source and the processing device, and the critical level of the scene content, a multi-dimensional feature vector is generated, and a weight model is used to dynamically prioritize the multi-dimensional feature vectors of conflicting computing sources. S323, Priority-driven rendering execution: The processing time slots are allocated according to the sorting results.

[0017] Preferably, the multidimensional feature vector is represented as a feature combination of the computation source, and the feature combination includes load intensity, inverse distance, and content key level; The weighting model calculates the overall priority of the computing sources and sorts them based on the overall priority: the overall priority is balanced by the feature weight coefficients, and the reciprocal of the distance indicates that the computing source closer to the processing device has higher priority. The fast segmentation mechanism dynamically adjusts the allocation of computing resources by time period, which is represented as a time period start time sequence. The start time of the time period is determined by the reading order of the computing source, and the reading order is allocated by the priority sorting result.

[0018] Compared with the prior art, the beneficial effects of the present invention are: In the point cloud data initialization phase, a series of operations, including assigning a unique task identifier to each rendering task, generating the identifier, entering the content, storing and associating it, and verifying the data, ensure that each rendering task has a clear and unique identifier record. This allows for accurate tracking of relevant information for each task during the 3D modeling of power facilities, including task type, scene complexity, data source, task time, and expected duration. It also incorporates point cloud acquisition equipment information into the identifier data, providing a reliable foundation for subsequent rendering task management and data traceability. For example, when managing point cloud rendering tasks for different substations, relevant task information can be quickly queried and located using the task identifier, improving the efficiency and accuracy of task management.

[0019] During point cloud data processing, the system is equipped with a point cloud processing engine that automatically converts point cloud data into a rendering format upon input. It also integrates a data verification unit to check the geometric consistency of the data in real time, automatically correcting and marking any detected anomalies. This process ensures the accuracy and reliability of the input point cloud data, avoiding rendering model errors caused by data anomalies. In the power sector, the accuracy of point cloud data for power facilities directly impacts subsequent planning, design, and maintenance. Data processed in this way more accurately reflects the actual form of power facilities, providing power workers with a more accurate 3D model reference. For example, when processing point cloud data for transmission lines, timely correction of anomalies can prevent line model deviations caused by data issues, leading to more accurate judgments in subsequent line inspections and fault analysis.

[0020] The cloud-edge coordinated rendering optimization embeds a load-feature-based resource allocation unit into the point cloud processing engine. This unit monitors the distribution and load of computing resources in real time and dynamically adjusts resource allocation based on load status. When a resource load issue is detected, a task segmentation protocol is introduced. Rendering segments are pre-allocated through a rapid segmentation mechanism and rendered according to task priority. Task priority is determined using a weighted model, dynamically prioritizing tasks with rendering conflicts to ensure critical areas are rendered first. This approach fully leverages the advantages of cloud-edge coordination, improving computing resource utilization and avoiding resource waste and load imbalance. In power scenarios, for critical areas such as core equipment in substations and important transmission line sections, computing resources can be prioritized for rendering, ensuring the real-time performance and accuracy of rendering in these critical areas and providing strong support for key monitoring and management of power facilities. For example, in power emergency repair scenarios, point cloud models of fault areas can be rendered quickly, providing accurate 3D model data for repair plan development.

[0021] The resource allocation unit, through embedded load analysis chips and resource distribution analysis algorithms, can accurately perceive the occupancy of computing resources and pinpoint the specific time periods and sets of load sources. This enables the system to perform more precise resource scheduling and task allocation, improving system response speed and processing efficiency. In power systems, facing a large number of point cloud data rendering tasks, this precise resource management method ensures that the system can still operate stably under high load, avoiding problems such as rendering stuttering and task backlog.

[0022] The rapid segmentation mechanism enables pre-allocation of time slots, ensuring no overlap between slots. Dynamic task priority ranking, based on multi-dimensional feature vectors and weight models, rationally determines task priority according to factors such as computational resource intensity, distance between the computing source and processing equipment, and the criticality level of scene content. This dynamic adjustment and priority-driven rendering execution method allows the system to flexibly respond to different rendering tasks and scene requirements, further improving rendering efficiency and quality. In practical applications in the power sector, the priority of rendering tasks can be dynamically adjusted according to different power facility scenarios and task requirements, ensuring rendering effects for important tasks and critical areas, providing strong technical support for the intelligent development of the power industry. Attached Figure Description

[0023] Figure 1 This is a schematic diagram illustrating the working principle of the point cloud model rendering method based on cloud-edge coordination described in this invention. Figure 2 Design diagram for point cloud data processing and verification; Figure 3 Design diagram for cloud-edge resource load monitoring; Figure 4 Design diagram for resource allocation time period management; Figure 5 Render the design diagram for task priority execution. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figures 1-5 The present invention relates to a point cloud model rendering method based on cloud-edge coordination, the specific implementation steps of which are as follows: S1, Point Cloud Data Initialization: Assign a unique task identifier to each rendering task. In practical applications, the rendering management platform inputs identifier data containing task type, scene complexity, data source, task time, and expected duration. When the point cloud acquisition device starts, device information (such as device model, acquisition range, sensor parameters, etc.) is automatically added to the identifier data, forming a complete identifier record. The point cloud acquisition device stores the identifier data, generates the corresponding task identifier, and associates this task identifier with the rendering task. After association, the data verification module reads the task identifier content and verifies it against the identifier data stored in the rendering management platform. The verification covers the completeness, accuracy, and uniqueness of the task identifier. After successful verification, the successfully associated task identifier is uploaded to the rendering management platform, establishing a long-term mapping relationship between the task identifier and the rendering task for subsequent tracking and management of rendering tasks.

[0026] S2, Point Cloud Data Processing: The system is equipped with a point cloud processing engine that automatically converts point cloud data into rendering format upon input. The point cloud processing engine is installed at the system's input interface. This engine uses a multi-sensor array to cover the acquisition area, ensuring the comprehensiveness and accuracy of point cloud data acquisition. When point cloud data input begins, the point cloud processing engine is triggered to detect point cloud signals in real time and automatically convert them into rendering data, including point cloud coordinates, color information, geometric structure, timestamps, and key features. After successful conversion, the current status of the rendering task in the rendering management platform is updated synchronously, allowing administrators to monitor the data processing progress in real time. Furthermore, the system's processing module is equipped with a multi-layered distributed geometric analysis unit, which can acquire the overall structure and segmented geometric distribution of the data in real time. During data conversion, the geometric analysis unit detects the geometric changes in the newly generated data, generates the actual structural data of the current data, and compares the data logic output by the point cloud processing engine with the actual structure acquired by the geometric analysis unit. If a structural deviation is detected that exceeds a set threshold (which can be adjusted according to the actual power scenario requirements, for example, 0.5mm in substation cloud modeling), the current data will be automatically marked as problematic data, and the actual structural content will be updated in the rendered data to ensure the geometric consistency of the data.

[0027] S3, rendering optimization based on cloud-edge coordination: S31 embeds a load-characteristic-based resource allocation unit into the point cloud processing engine to monitor the distribution and load of computing resources in real time. The resource allocation unit uses an embedded load analysis chip to perceive the occupancy of computing resources in real time. This chip uses a resource distribution analysis algorithm to detect the load distribution of resource transmission and mark overlapping computing activities within the same resource type. The resource distribution analysis algorithm calculates the resource occupancy status by analyzing the load energy distribution of computing resources. When the load energy exceeds a set energy threshold, it indicates that the resource is occupied; otherwise, it is idle. When resource load is detected, the specific time period of the load occurrence is located by calculating the intensity, distribution characteristics, and temporal characteristics of the computing resources, providing a basis for subsequent resource scheduling.

[0028] S32 addresses resource load issues by introducing a task segmentation protocol. This protocol pre-allocates rendering segments through a fast segmentation mechanism and renders according to task priority. Task priority is determined using a weighted model, dynamically prioritizing tasks in rendering conflicts to ensure critical areas are rendered first. After load detection, the fast segmentation mechanism dynamically adjusts the allocation of computing resources across time slots, assigning specific processing time slots to each computing source to ensure no overlap between slots. The dynamic prioritization of tasks is based on the intensity of computing resources, the distance between the computing source and the processing device, and the criticality level of the scene content, generating multi-dimensional feature vectors. A weighted model is then used to dynamically prioritize these feature vectors of conflicting computing sources. Finally, processing time slots are allocated according to the ranking results, achieving priority-driven rendering execution.

[0029] Example 1: The specific implementation method during point cloud data initialization is as follows: Through the interactive interface of the rendering management platform, operators input the task type. For example, in the power industry, the task type could be a power line inspection point cloud rendering task, a substation equipment point cloud modeling task, etc. Different task types correspond to different processing flows and rendering requirements. Scene complexity needs to be assessed based on the actual situation of the power facilities. For example, when power lines pass through mountainous areas, the terrain is undulating and the vegetation is dense, so the scene complexity can be assessed as high; while in plains areas, the power facilities are relatively well-distributed, so the scene complexity can be assessed as medium or low. The data source needs to be clearly defined: whether the point cloud data is collected by UAV LiDAR, a ground 3D scanner, or other equipment, as this relates to the subsequent data processing methods and parameter settings. The task time must be accurate to the second, including the task start and end times. The estimated duration is estimated based on historical data or practical experience; for example, a substation point cloud rendering task is expected to last 2 hours.

[0030] Once the point cloud acquisition device is activated, its information is automatically added to the identification data. Taking a power line inspection drone equipped with LiDAR as an example, the device information includes the device's unique identifier, sensor model, sampling frequency, and scanning range. This information is crucial for subsequent data tracing and device performance evaluation. For instance, the sensor model determines the accuracy and resolution of the point cloud data acquisition, the sampling frequency affects the data density, and the scanning range limits the size of the acquisition area. Embedding this device information into the identification data creates a complete identification record, providing detailed background information for each rendering task.

[0031] The identification data is encrypted and stored using the built-in storage module of the point cloud acquisition device, and a unique task identifier is generated using a hash algorithm. The hash algorithm is one-way and unique, ensuring that the generated task identifier is unique and avoids duplication with other task identifiers. The task identifier can be a string composed of several bits of binary data, such as a 128-bit binary number. This length meets the needs of large-scale task management, ensuring that a large number of point cloud rendering tasks in the power system have a unique identifier.

[0032] After generating the task identifier, it needs to be associated with the rendering task. Specifically, the task identifier is transmitted to the rendering task management system via a wireless communication module, such as a 4G / 5G network. In power scenarios, multiple substations or transmission lines may be simultaneously collecting and rendering point clouds. The wireless communication module needs to ensure the stability and reliability of data transmission to prevent the task identifier from being lost or corrupted during transmission. After transmission, the system will associate the generated task identifier with the corresponding rendering task, establishing a preliminary correspondence.

[0033] After the association is completed, the system will automatically trigger the data verification process. The data verification module reads the task identifier content from the storage database and compares it field by field with the original identifier data in the rendering management platform. The verification content mainly includes three aspects: First, completeness verification, checking whether there is any missing data in the task identifier, such as whether it contains all the input identifier data fields, and whether any field values ​​are empty; second, accuracy verification, verifying whether the field values ​​in the task identifier are consistent with the actual input by the operator, such as whether the task type is correct, whether the scene complexity assessment is accurate, etc.; third, uniqueness verification, querying the database to ensure that the task identifier has not been used by other tasks, avoiding task management chaos due to duplicate identifiers.

[0034] In power systems, the accuracy and uniqueness of data are crucial, as incorrect or duplicate task identifiers can lead to mismatches between subsequent point cloud rendering data and the actual tasks, impacting the operation and maintenance of power facilities. For example, if the point cloud rendering task identifier for one transmission line is duplicated with that of another line, maintenance personnel may misanalyze the data and make incorrect decisions.

[0035] After successful verification, the associated task identifiers are uploaded to the central database of the rendering management platform, establishing a mapping table between task identifiers and rendering tasks. This mapping table contains multiple fields, such as task identifier, rendering task ID, association time, and device information. Through this mapping table, staff can quickly retrieve corresponding rendering task information by task identifier, understanding the task type, scene complexity, data source, etc.; they can also query associated task identifiers by rendering task ID, enabling comprehensive tracking and management of tasks.

[0036] In practical applications within the power sector, this mechanism ensures that point cloud rendering tasks for power facilities collected from different time periods and devices each have a unique identifier. For example, during regular inspections of a substation, each point cloud rendering task has a unique task identifier, facilitating comparison of point cloud models from different periods and analysis of changes in equipment operating status. Furthermore, in the event of sudden power outages, such as damage to a transmission line due to lightning strikes, emergency point cloud rendering tasks can be quickly located and managed using unique task identifiers, supporting rapid fault diagnosis and repair.

[0037] Furthermore, this method of initializing point cloud data facilitates data archiving and storage. In power systems, the accumulated amount of point cloud data is enormous. Through task identifiers and mapping tables, this data can be managed systematically, improving data retrieval and utilization efficiency. For example, when needing to view the point cloud model of a transmission line from a year ago, the corresponding rendering task data can be quickly found through the task identifier, eliminating the need for blind searching through massive amounts of data.

[0038] Example 2: In the specific implementation of point cloud data processing, the system deploys a point cloud processing engine and a multi-layered distributed geometric analysis unit to achieve format conversion and geometric consistency verification of point cloud data in power scenarios. First, a point cloud processing engine is deployed at the system input interface. This engine is designed based on edge computing nodes and can adapt to diverse acquisition devices in power scenarios. Taking substation point cloud acquisition as an example, the sensor array typically includes fixed LiDAR and mobile inspection equipment: fixed LiDAR is deployed at the four corners of the substation to cover the entire substation structure; the mobile inspection robot is equipped with portable scanning equipment to perform close-range, detailed scanning of key equipment such as transformers and insulators. Each sensor calibrates its time using the IEEE 1588 time synchronization protocol to ensure consistency in the time dimension of data collected by different devices, avoiding misalignment of point cloud stitching due to time deviations.

[0039] When point cloud data is input into the system from the sensor array, the point cloud processing engine is triggered and starts the real-time processing flow. The engine first filters and denoises the raw point cloud signal. This process, tailored to the electromagnetic interference characteristics of power equipment during operation, uses an adaptive filtering algorithm to identify and remove outliers. For example, point cloud data collected near a transformer may contain high-frequency noise due to electromagnetic interference. The filtering algorithm uses frequency domain analysis to distinguish such point cloud signals from normal equipment point clouds, retaining valid data points. After denoising, the engine converts the point cloud data into a rendering format. This includes converting the three-dimensional spatial coordinates to a coordinate system supported by the rendering engine (such as a right-handed coordinate system) with millimeter-level accuracy to meet the high-precision requirements of power equipment modeling; when extracting point cloud color information, RGB values ​​are mapped to equipment appearance features, such as the silver-gray of transformers and the color of steel-cored aluminum stranded wires in transmission lines; the geometric structure extraction module analyzes the normal vectors and curvature of the point cloud to generate basic geometric primitives such as planes and cylinders, providing structural support for subsequent model rendering.

[0040] During the conversion process, the engine adds a timestamp and key feature labels to each point cloud data point. The timestamps are accurate to the millisecond level, facilitating subsequent analysis of equipment status changes over different time periods. The key feature labels are pre-defined based on professional knowledge of power equipment, such as extracting sag features and insulator damage features for transmission lines, and bushing crack features for substation equipment. These features are automatically identified using machine learning algorithms, such as using convolutional neural networks to detect the damage edges of insulators, ensuring accurate extraction of key features. After successful conversion, the engine sends a status update command to the rendering management platform via a message queue. Upon receiving the command, the platform updates the rendering task status from "data acquisition" to "conversion complete" and records the conversion time and data volume, providing a reference for subsequent task scheduling.

[0041] The system processing module is configured with a multi-layered distributed geometric analysis unit that works collaboratively between the edge node layer and the cloud layer. The edge node layer is deployed on the local computing units of each acquisition device, collecting the geometric distribution of local point cloud data in real time, such as the local orientation of a transmission line. The cloud layer aggregates the data from each edge node to construct an overall structural model, such as the three-dimensional orientation of the entire transmission line. During data conversion, the geometric analysis unit continuously monitors the geometric changes in newly generated data: the edge node layer performs meshing processing on the real-time input point cloud data, generating a local triangular mesh model; the cloud layer then merges the local meshes to form a global three-dimensional model and performs differential analysis with the model from the previous moment to detect the amount of geometric change.

[0042] Taking power line point cloud as an example, the geometric analysis unit calculates parameters such as the line's sag value and tower verticality. When there is a deviation between the data logic output by the point cloud processing engine and the actual structure collected by the geometric analysis unit, the system initiates a verification mechanism: if the detected line sag deviation exceeds a preset threshold (this threshold can be set according to power industry standards, such as the allowable sag deviation for 110kV lines being ±2.5% of the design value), the system automatically marks this data segment as problematic data and triggers a correction process. During the correction process, the system prioritizes using the actual structural data collected by the geometric analysis unit to replace the deviation data output by the point cloud processing engine, such as updating the rendered data with measured sag values ​​to ensure the geometric accuracy of the point cloud model.

[0043] In substation equipment modeling scenarios, this mechanism effectively avoids model deviations caused by sensor errors or environmental interference. For example, when a fixed LiDAR is affected by strong light reflection and mistakenly captures the outline of the heat sink on top of the transformer as an abnormal protrusion, the geometric analysis unit can identify and correct this anomaly by comparing the geometric relationship between historical models and adjacent equipment, ensuring the outline accuracy of the transformer model. After correction, the system adds a correction marker to the rendered data, recording the correction time and content, facilitating subsequent data processing traceability.

[0044] Furthermore, the multi-layered distributed geometric analysis unit supports incremental updates. When local changes occur in equipment within a power scenario (such as the addition of insulators), the system only needs to update the point cloud data of the changed area, without reprocessing the entire dataset, thereby improving data processing efficiency. The edge node layer is responsible for handling the local geometric analysis of the changed area, while the cloud layer integrates and updates the global model, ensuring the model's real-time performance and accuracy.

[0045] Example 3: In cloud-edge coordinated rendering optimization, the specific implementation of resource awareness and load identification is as follows: A load analysis chip is embedded in the hardware architecture of the point cloud processing engine. This chip typically adopts an FPGA or ASIC architecture and is connected to the computing resources within the system, such as CPU, GPU, and memory, via a high-speed data bus to achieve real-time acquisition of the operating status of computing resources. Taking a power inspection point cloud rendering scenario as an example, when the system simultaneously processes point cloud model rendering tasks for multiple substations, the load analysis chip continuously monitors key operating parameters of each computing resource, such as CPU core utilization, GPU memory utilization, and memory bandwidth utilization. These parameters directly reflect the load status of computing resources.

[0046] The resource distribution analysis algorithm runs within the load analysis chip, first calculating the resource occupancy status. This algorithm converts the operating parameters of computing resources into load energy values. A standardized energy measurement system can be established, where 100% CPU utilization corresponds to an energy value of 100, 50% utilization to 50, and so on. Simultaneously, corresponding energy thresholds are set for different types of computing resources. For example, when the CPU's load energy value exceeds 80, the CPU resource is considered occupied; below 80, it is considered idle. This method of determining resource occupancy status through energy thresholds can adapt to the dynamic changes in computing resource demands during different times and for different tasks in power scenarios.

[0047] The load analysis chip continuously scans the load energy distribution of various computing resources within the system. When it detects that the load energy values ​​of multiple cores in a certain type of resource (such as a GPU) all exceed a set threshold, it marks these computing activities as a state of load overlap and generates a list of load activities. This list records information such as the time the load occurred, the type of computing resource involved, and the load energy value, providing basic data for subsequent resource scheduling. For example, during the daily morning power inspection data centralized processing period, the load analysis chip may detect that the load energy values ​​of multiple cores of GPU resources are continuously higher than 80, thus marking these computing activities as a high-load state.

[0048] After the time-period detection and positioning module obtains resource load information through the load analysis chip, it further analyzes the characteristics of the load, including load intensity, distribution characteristics, and temporal characteristics. Load intensity is reflected as the difference between the actual load energy value of the computing resources and the threshold; the larger the difference, the higher the load intensity. Distribution characteristics describe the distribution of the load across different computing resources, such as whether it is concentrated on a few GPU cores or CPU threads. Temporal characteristics determine the specific time period during which the load occurs, including the start time and duration. By analyzing these characteristics, the specific time period during which the load occurs can be accurately located. For example, in a power company's cloud-edge collaborative system, the load analysis chip detected that between 9:30 AM and 10:30 AM, the load energy value of GPU resources generally exceeded 80, and was mainly concentrated on the first four cores of the GPU, thus determining this period as a high-load period.

[0049] The set of time periods for allocating computing resources is defined as a pre-allocated time period sequence, which consists of multiple consecutive time windows, each corresponding to a computing segment. For example, each time window can be set to 10 minutes. The system sets the condition for detecting load intensity overlap as follows: within a certain time window, the sum of the load intensities of multiple computing resources exceeds a set conflict threshold. Taking GPU resources as an example, if the sum of the load intensities of all GPU cores exceeds 90, then it is determined that there is load intensity overlap within that time window. When such overlap is detected, the system calculates the contribution ratio of each computing source (i.e., each point cloud rendering task) to the load intensity. Specifically, it calculates the proportion of computing resources occupied by that task. For example, if a task occupies 20% of the GPU computing power, then its contribution ratio to the load intensity is 20%.

[0050] The system includes computational sources whose contribution to the load exceeds a set threshold into a load source set. For example, computational sources contributing more than 15% are identified as primary load sources. In power scenarios, this mechanism helps maintenance personnel accurately identify specific point cloud rendering tasks that cause high computational resource load. For instance, when performing 3D modeling of a substation, the high complexity of the substation's equipment and the large volume of point cloud data may cause its rendering tasks to become the primary load source. After the load source is located, the system transmits the load source set information to the subsequent task segmentation module, providing a basis for accurate resource scheduling and task prioritization.

[0051] In practical power applications, this resource awareness and load identification mechanism can effectively address the dynamic changes in point cloud data within power systems. For example, when power lines encounter severe weather (such as heavy rain or strong winds), urgent point cloud inspections of the lines are required. At this time, the system may simultaneously initiate multiple emergency rendering tasks, leading to a surge in computational resource load. Through real-time monitoring and load source location by the load analysis chip, the system can quickly identify the resource consumption of these emergency tasks, providing data support for subsequent resource allocation adjustments and prioritizing critical tasks. This ensures that in power emergency scenarios, the point cloud models of critical equipment can be rendered in a timely manner, providing a visual basis for fault diagnosis and repair decisions.

[0052] Furthermore, this mechanism supports the analysis of computing resource usage trends. By recording information such as the time periods and types of load sources over a long period, operations and maintenance personnel can summarize the load patterns of power point cloud rendering tasks. For example, computing resource load will experience periodic peaks on monthly equipment inspection days. Based on these patterns, operations and maintenance personnel can adjust resource configurations in advance, such as adding temporary computing nodes, to cope with upcoming high-load periods and improve the overall operating efficiency and stability of the system.

[0053] Example 4: In the task segmentation protocol, the specific implementation of the fast segmentation mechanism and task priority ranking needs to be tailored to the actual needs of the power scenario. Taking a power grid company in a certain region conducting point cloud inspections of transmission lines as an example, when the system simultaneously receives three rendering tasks—task A is the detailed modeling of the main transformer of a 220kV substation, task B is the status detection of insulators on a 110kV transmission line, and task C is the 3D reconstruction of the terrain surrounding the line—if GPU resource load overlap is detected, the fast segmentation mechanism will initiate dynamic time-segment allocation. This mechanism first analyzes the data volume and processing complexity of each task: main transformer modeling requires processing high-density point cloud data (approximately 1 billion points), insulator detection involves feature recognition (approximately 500 million points), and terrain reconstruction involves a relatively smaller data volume (approximately 300 million points). Based on processing requirements, the fast segmentation mechanism divides the available time slots into three non-overlapping time windows: 9:00-9:30 for task A, 9:30-10:00 for task B, and 10:00-10:30 for task C. Each time window corresponds to an independent slice of computing resources, ensuring that the GPU cores focus on processing a single task during different time periods and avoiding resource conflicts.

[0054] The dynamic prioritization of tasks is based on multi-dimensional feature vector generation and weight model calculation. Taking feature combinations in a power scenario as an example, the load intensity index reflects the proportion of resources occupied by a task: Task A, due to its large data volume, occupies 80% of GPU memory and 75% of computing power, corresponding to a load intensity feature value of 0.8; Task B occupies 60% of GPU memory and 60% of computing power, with a feature value of 0.6; Task C occupies 40% of GPU memory and 50% of computing power, with a feature value of 0.5. The reciprocal distance index is calculated based on the network latency between the computing source (edge ​​node) and the cloud server: The edge node of Task A is deployed in the local computer room of the substation, with a latency of 5ms, and the reciprocal distance is 1 / (5+1)≈0.17; the edge node of Task B is located at a base station in the middle of the transmission line, with a latency of 15ms, and the reciprocal distance is 1 / (15+1)≈0.06; the edge node of Task C is located in a remote monitoring center, with a latency of 30ms, and the reciprocal distance is 1 / (30+1)≈0.03. The criticality level of the content is set according to the importance of the power equipment: the main transformer, as the core equipment of the substation, has a high criticality level (characteristic value 1.0); the insulator affects the insulation performance of the line, so it has a medium criticality level (characteristic value 0.7); and the surrounding terrain has a low criticality level (characteristic value 0.3).

[0055] The weight coefficients of each feature in the weighted model can be adjusted according to power industry standards. For example, in a routine inspection scenario, the weights are set as follows: load intensity 0.4, distance reciprocal 0.3, and content criticality level 0.3. The comprehensive priority calculation for task A is: 0.8×0.4 + 0.17×0.3 + 1.0×0.3 = 0.32 + 0.051 + 0.3 = 0.671; for task B, it is: 0.6×0.4 + 0.06×0.3 + 0.7×0.3 = 0.24 + 0.018 + 0.21 = 0.468; and for task C, it is: 0.5×0.4 + 0.03×0.3 + 0.3×0.3 = 0.2 + 0.009 + 0.09 = 0.299. Thus, the task priority order is determined to be A > B > C, consistent with the time period allocation order of the rapid segmentation mechanism.

[0056] In power emergency scenarios, the weighted model can dynamically adjust parameters. For example, when a short-circuit fault occurs in a substation, a newly added task D is for emergency modeling of the faulty equipment. Its load intensity feature value is 0.7 (occupying 70% of GPU computing power), the latency between the edge node and the cloud server is 10ms (0.09 reciprocals of the distance), and the content criticality level is extremely high (feature value 1.0). At this time, the operation and maintenance personnel can manually increase the content criticality level weight to 0.5 and recalculate the priority: Task D becomes 0.7×0.4 + 0.09×0.3 + 1.0×0.5 = 0.28 + 0.027 + 0.5 = 0.807, which exceeds the priority of the original task A. The fast segmentation mechanism will immediately insert a new time window (such as 9:15-9:30) for task D and shift the time window of the original task A to the later period to ensure that the faulty equipment model is rendered first.

[0057] The time series generated by the rapid segmentation mechanism is strictly bound to the priority order. In a typical inspection scenario, task A, with higher priority, receives the earliest start point of the time series, and its data reading order takes precedence over other tasks. The system controls the data reading process through a scheduler: at 9:00, the scheduler sends a data reading command to the edge node of task A, and the edge node transmits the point cloud data of the main transformer to the cloud server via a low-latency link; at 9:30, the edge node of task B begins transmitting insulator data; and at 10:00, data transmission for task C begins. This priority-based reading order ensures that data from high-critical tasks enters the rendering process first, avoiding rendering delays caused by data transmission latency.

[0058] In power transmission line inspection scenarios, this mechanism can effectively handle concurrent multi-task situations. For example, a power transmission line crosses a mountainous area and simultaneously presents three rendering tasks: tree obstruction hazard (Task E requires analysis of the distance between trees and the line), insulator damage (Task F), and tower tilt (Task G). Task E has a load intensity of 0.6, a distance reciprocal of 0.1 (edge ​​node is at the foot of the mountain base station, delay 10ms), and a content criticality level of 0.6 (tree obstruction may cause short circuits); Task F has a load intensity of 0.5, a distance reciprocal of 0.08 (delay 12ms), and a criticality level of 0.8 (insulator damage directly affects safety); Task G has a load intensity of 0.4, a distance reciprocal of 0.15 (base station is close to the tower, delay 5ms), and a criticality level of 0.7 (tower tilt needs to be monitored). Based on the default weights: Task E is 0.6×0.4+0.1×0.3+0.6×0.3=0.24+0.03+0.18=0.45; Task F is 0.5×0.4+0.08×0.3+0.8×0.3=0.2+0.024+0.24=0.464; Task G is 0.4×0.4+0.15×0.3+0.7×0.3=0.16+0.045+0.21=0.415. Therefore, the priority order is F>E>G. The rapid segmentation mechanism allocates 8:00-8:20 for Task F, 8:20-8:40 for Task E, and 8:40-9:00 for Task G, ensuring that insulator damage detection tasks are prioritized, which conforms to the principle of "prevention first" in power safety.

[0059] During time slot allocation, the system monitors resource release in real time. If task F completes rendering ahead of schedule (e.g., actual processing time is 15 minutes), the remaining 5 minutes will be automatically allocated to task G, allowing it to start processing earlier. This dynamic adjustment mechanism improves resource utilization and avoids wasting idle time slots. In a substation expansion scenario, when the newly installed GIS equipment point cloud rendering task H is accessed, if the current time slot window is the processing time for task I (old equipment maintenance), the system will calculate the priority of task H: if the content criticality level of H is high (1.0), load intensity is 0.7, and distance reciprocal is 0.15, the priority calculated using the default weight is 0.7×0.4+0.15×0.3+1.0×0.3=0.28+0.045+0.3=0.625; if the priority of task I is 0.5, then task H will preempt the remaining time slot of task I, ensuring that the key equipment models of the expansion project are generated in a timely manner, providing a basis for construction acceptance.

[0060] Example 5: In the implementation of priority-driven rendering and multi-dimensional feature vectors, taking a power company's cloud-edge collaborative system as an example, when the system receives three point cloud rendering tasks with different priorities, the operation process of this mechanism is as follows: Task P is the emergency fault modeling of the main transformer of a 220kV substation, Task Q is the routine inspection modeling of a 110kV transmission line, and Task R is the terrain modeling of the substation's surrounding environment. The system first calculates the priority of each task based on the multi-dimensional feature vectors and allocates processing time slots according to the priority ranking results to ensure that critical tasks are executed first.

[0061] The construction of multidimensional feature vectors has clear physical meaning and practical application value in the power scenario. Taking task P as an example, its load intensity reflects the task's consumption of computing resources. Since the main transformer fault modeling requires processing high-density point cloud data to accurately present the fault details of the equipment, it occupies 80% of the GPU computing power and 75% of the video memory, with a corresponding load intensity feature value of 0.8. The reciprocal of distance reflects the network latency between the edge node and the cloud processing device. The edge node of task P is deployed in the local computer room of the substation, and the network latency with the cloud server is 5ms. According to the calculation method of the reciprocal of distance, its reciprocal of distance feature value is 1 / (5+1)≈0.17. In terms of content criticality level, the main transformer is the core equipment of the substation, and its fault directly affects the safe and stable operation of the power grid. Therefore, the content criticality level is set to the highest level of 1.0. The combination of these three features constitutes the multidimensional feature vector of task P [0.8, 0.17, 1.0].

[0062] Task Q, as a routine inspection modeling task, has a relatively small data volume, consuming 60% of the GPU computing power and 50% of the video memory, with a load intensity feature value of 0.6. The edge node of this task is located at a base station in the middle of the transmission line, with a network latency of 15ms to the cloud server, and a distance reciprocal feature value of 1 / (15+1)≈0.06. Regarding the content criticality level, although transmission line inspection is also important, it is a routine task, so it is set to 0.7, and its multidimensional feature vector is [0.6, 0.06, 0.7].

[0063] Task R models the surrounding terrain, requiring the least amount of data and consuming only 40% of the GPU computing power and 30% of the video memory. The load intensity feature value is 0.4. The edge nodes are located in a remote monitoring center with a network latency of 30ms and a distance reciprocal feature value of 1 / (30+1)≈0.03. The content criticality level is the lowest, set to 0.3, and the multidimensional feature vector is [0.4, 0.03, 0.3].

[0064] The weighted model assigns different weight coefficients to each feature based on the needs of the power scenario. In the emergency fault handling scenario, to ensure that the models of critical equipment are rendered first, the weight of the content criticality level is set to 0.5, the weight of load intensity is 0.3, and the weight of the inverse distance is 0.2. The comprehensive priority of each task is calculated by weighted summation: the comprehensive priority of task P is 0.8×0.3 + 0.17×0.2 + 1.0×0.5 = 0.24 + 0.034 + 0.5 = 0.774; the comprehensive priority of task Q is 0.6×0.3 + 0.06×0.2 + 0.7×0.5 = 0.18 + 0.012 + 0.35 = 0.542; and the comprehensive priority of task R is 0.4×0.3 + 0.03×0.2 + 0.3×0.5 = 0.12 + 0.006 + 0.15 = 0.276. Therefore, the priority order is determined as P > Q > R.

[0065] Based on the priority ranking, the system allocates processing time slots to each task. Task P, as the highest priority task, is assigned to the earliest time slot window, 9:00-9:30. During this time slot, the system prioritizes allocating major computing resources, such as 80% of GPU computing power and 90% of video memory, to task P to ensure it can complete rendering quickly. Task Q is assigned to the 9:30-10:30 time slot, and task R is assigned to the 10:30-11:00 time slot. At the beginning of each time slot, the system automatically switches the allocation of computing resources to ensure that the tasks in the current time slot receive sufficient resource support.

[0066] In power emergency scenarios, this priority-driven rendering execution mechanism plays a crucial role. For example, when a 220kV transmission line experiences a breakage due to lightning strikes, the system receives an emergency rendering task S for detailed modeling of the fault point. Task S has a load intensity feature value of 0.7, a network latency of 10ms between the edge node and the cloud, a distance reciprocal feature value of 0.09, and a content criticality level set to the highest value of 1.0. Due to the emergency fault handling, the operations and maintenance personnel temporarily adjust the weight of the content criticality level to 0.6, the load intensity weight to 0.3, and the distance reciprocal weight to 0.1. The calculated overall priority of task S is 0.7×0.3 + 0.09×0.1 + 1.0×0.6 = 0.21 + 0.009 + 0.6 = 0.819, which is higher than all currently executing tasks. At this point, the system will immediately activate the priority preemption mechanism, interrupt the currently executing low-priority task, and allocate a new time window for task S, such as the 10:00-10:30 time window that starts immediately, to ensure that the model of the fault point can be rendered as soon as possible, providing accurate visualization data for the repair personnel.

[0067] During rendering, the system monitors the progress and resource usage of each task in real time. If a high-priority task completes rendering ahead of schedule, the remaining time slot will be dynamically allocated to lower-priority tasks to improve resource utilization. For example, if task P completes rendering at 9:20, the remaining 10 minutes from 9:20 to 9:30 will be allocated to task Q, allowing it to start processing earlier. This dynamic adjustment mechanism is significant in power systems, especially when handling a large number of inspection tasks, as it effectively utilizes fragmented time resources and accelerates the overall task processing progress.

[0068] Taking the annual overhaul scenario of a substation as an example, the system needs to handle multiple point cloud rendering tasks, including main transformer maintenance modeling, GIS equipment maintenance modeling, insulator inspection modeling, and surrounding environment modeling. Through the calculation of multi-dimensional feature vectors and weight models, the priorities of each task are rationally ordered. Main transformer maintenance modeling, as the most critical task, is processed first, followed by GIS equipment maintenance modeling, then insulator inspection modeling, and finally, surrounding environment modeling. Throughout the overhaul process, the system allocates time slots according to priority, ensuring that the model of each critical piece of equipment can be rendered in a timely manner, providing maintenance personnel with accurate model references and guaranteeing the smooth progress of the overhaul work.

[0069] The reciprocal distance feature in multidimensional feature vectors also has practical significance in power scenarios. For example, in the inspection of transmission lines in remote areas, the network latency between edge nodes and cloud servers may be large, resulting in a small reciprocal distance feature value. In this case, the system will automatically adjust the task priority based on this feature, provide certain compensation in resource allocation, or adopt edge computing to perform partial rendering processing locally, thereby reducing data transmission volume and latency and improving rendering efficiency.

[0070] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0071] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A point cloud model rendering method based on cloud-edge coordination, characterized in that, Includes the following steps: S1, Point Cloud Data Initialization: Assign a unique task identifier to each rendering task; S2, Point Cloud Data Processing: The system is equipped with a point cloud processing engine, which automatically converts point cloud data into a rendering format when data is input. The system's processing module integrates a data verification unit, which checks the geometric consistency of the data in real time during the conversion process. If anomalies are detected, the data is automatically corrected and marked as problems. S3, rendering optimization based on cloud-edge coordination: S31 embeds a resource allocation unit based on load characteristics into the point cloud processing engine to monitor the distribution and load of computing resources in real time. S32, to address the resource load issue, a task segmentation protocol is introduced. The task segmentation protocol pre-allocates rendering segments through a fast segmentation mechanism and renders them according to task priority. The task priority is determined using a weight model, which dynamically sorts the priority of tasks in rendering conflicts to ensure that critical areas are rendered first.

2. The point cloud model rendering method based on cloud-edge coordination according to claim 1, characterized in that, S1 specifically includes: S11, Identifier Generation and Content Entry: Input identifier data, including task type, scene complexity, data source, task time and expected duration, through the rendering management platform. After the point cloud acquisition device is started, the device information is added to the identifier data to form a complete identifier record. S12, Identifier Storage and Association: Store the identifier data using the point cloud acquisition device, generate the corresponding task identifier, and associate the generated task identifier with the rendering task.

3. The point cloud model rendering method based on cloud-edge coordination according to claim 2, characterized in that, S1 also includes data verification: after the task identifier is associated, the data verification module reads... The task identifier content is retrieved and compared with the identifier data stored in the rendering management platform. The verification content includes the completeness, accuracy, and uniqueness of the task identifier. The successfully associated task identifiers are uploaded to the rendering management platform to establish a long-term mapping relationship between task identifiers and rendering tasks.

4. The point cloud model rendering method based on cloud-edge coordination according to claim 1, characterized in that, S2 specifically includes: S21, A point cloud processing engine is installed at the system's input interface. The point cloud processing engine covers the acquisition area through a multi-sensor array. S22, when point cloud data input begins, the point cloud processing engine is triggered to detect point cloud signals in real time and automatically convert them into rendering data, including point cloud coordinates, color information, geometric structure, timestamps and key features. After successful conversion, the current status of the rendering task in the rendering management platform is updated.

5. The point cloud model rendering method based on cloud-edge coordination according to claim 4, characterized in that, The system's processing module is equipped with a multi-layered distributed geometric analysis unit that collects the overall structure and segmented geometric distribution of data in real time. When data is transformed, the geometric analysis unit detects the geometric changes in the newly generated data and generates the actual structural data of the current data. The data logic output by the point cloud processing engine is compared with the actual structure collected by the geometric analysis unit. If the detected structural deviation exceeds the set threshold, the current data is automatically marked as problematic, and the actual structural content is updated in the rendered data.

6. The point cloud model rendering method based on cloud-edge coordination according to claim 1, characterized in that, S31 specifically includes: Resource awareness and load identification: The resource allocation unit senses the occupancy of computing resources in real time through the embedded load analysis chip. The load analysis chip uses a resource distribution analysis algorithm to detect the load distribution of resource transmission and mark computing activities with overlapping loads within the same resource type. Time-based detection and location: When resource load is detected, the intensity and distribution of resources are calculated. Characteristics and time characteristics are used to pinpoint the specific time period during which the load occurs.

7. The point cloud model rendering method based on cloud-edge coordination according to claim 6, characterized in that, The resource distribution analysis algorithm specifically includes: Resource occupancy status calculation: By analyzing and calculating the load energy distribution of resources, the resource occupancy status is determined. When the load energy exceeds the set energy threshold, it indicates that the resource is in an occupied state. The resource occupancy status detection condition is as follows: the resource occupancy status is determined by the energy threshold. If the energy is higher than the threshold, it indicates that the resource is occupied, and if it is lower than the threshold, it indicates that the resource is idle.

8. The point cloud model rendering method based on cloud-edge coordination according to claim 7, characterized in that, The set of time periods for computing resource allocation is defined as a pre-allocated time period sequence. The time period sequence contains multiple consecutive time windows. In the time period sequence, each time window corresponds to a computing segment. The detection condition for overlapping load intensity is that the sum of the intensities of multiple computing resources within the time period exceeds a set conflict threshold. Load source location: If overlapping load intensity is detected, the load source set is located by calculating the contribution load intensity of each computing source. Computing sources whose contribution load intensity exceeds the proportional threshold are included in the load set.

9. A point cloud model rendering method based on cloud-edge coordination according to claim 8, characterized in that, Specifically, S32 includes: S321, Fast Segmentation Mechanism for Pre-allocation of Time Periods: After load detection, the fast segmentation mechanism is used to dynamically adjust the time period allocation of computing resources, allocating specific processing time periods to each computing source to ensure that there is no overlap between time periods; S322, Dynamic prioritization of tasks: Based on the intensity of computing resources, the distance between the computing source and the processing device, and the critical level of the scene content, a multi-dimensional feature vector is generated, and a weight model is used to dynamically prioritize the multi-dimensional feature vectors of conflicting computing sources. S323, Priority-driven rendering execution: The processing time slots are allocated according to the sorting results.

10. A point cloud model rendering method based on cloud-edge coordination according to claim 9, characterized in that, The multidimensional feature vector is represented as a feature combination of the computation source, and the feature combination includes load intensity, inverse distance, and content key level; The weighting model calculates the overall priority of the computing sources and sorts them based on the overall priority: the overall priority is balanced by the feature weight coefficients, and the reciprocal of the distance indicates that the computing source closer to the processing device has higher priority. The fast segmentation mechanism dynamically adjusts the allocation of computing resources by time period, which is represented as a time period start time sequence. The start time of the time period is determined by the reading order of the computing source, and the reading order is allocated by the priority sorting result.

Images