A three-dimensional modeling task distribution and merging method, device and medium

Abstract

The application discloses a three-dimensional modeling task distribution and merging method, equipment and medium, the method comprises the following steps: based on the complexity evaluation result of the to-be-modeled scene, the modeling task is distributed to a plurality of edge computing nodes to form a distributed modeling task set; the edge computing nodes process the distributed local modeling task in parallel, generate the corresponding local three-dimensional model, and upload the local three-dimensional model to a cloud server; after the cloud server receives the local three-dimensional model from a plurality of edge computing nodes, model coordinate alignment, mesh fusion and hierarchical optimization processing are performed to generate a complete three-dimensional scene model. The application constructs an edge end distributed modeling task and cloud end model merging mechanism, improves the efficiency and scalability of three-dimensional modeling, and ensures that the model precision meets the actual application requirements.

Description

Technical Field

[0001] This invention relates to the field of 3D modeling technology, specifically to a method, device, and medium for distributing and merging 3D modeling tasks. Background Technology

[0002] With the widespread application of 3D modeling technology in fields such as virtual reality, augmented reality, smart cities, and industrial design, the demand for large-scale scene modeling is increasing. Traditional 3D modeling methods typically rely on centralized processing, uploading all data to the cloud or local servers for unified processing. This approach faces many challenges when dealing with massive amounts of data: First, centralized processing requires high-performance computing resources, which are costly and have poor scalability. The computing power of a single server is limited, and when the scene data volume reaches tens to hundreds of gigabytes, the modeling time often lasts for hours to days, making it difficult to meet the needs of real-time modeling. Second, network bandwidth limitations lead to data transmission delays. In the transmission of point cloud data and texture images in large-scale scenes, stuttering and packet loss are prone to occur, which not only affect modeling efficiency but may also lead to a decrease in modeling accuracy. In addition, existing methods lack effective task distribution and merging mechanisms, making it difficult to support multi-user collaborative modeling. Poor data interaction between modeling nodes can easily lead to model conflicts and duplicate modeling.

[0003] For example, in smart city applications, 3D modeling of entire city blocks is required, with data volumes typically reaching hundreds of gigabytes. Centralized processing often leads to excessively long processing times and resource bottlenecks, making it difficult to respond promptly to real-time needs such as urban planning and emergency rescue. In the field of nuclear power facility operation and maintenance, millimeter-level precision 3D modeling of nuclear power plant buildings is required. Traditional centralized methods struggle to balance modeling accuracy and efficiency, and the on-site transmission of large amounts of data poses security risks. In the digitalization of industrial plants, when multiple workstations simultaneously collect data for modeling, existing methods cannot achieve reasonable task allocation and seamless model merging, resulting in long modeling cycles and poor model consistency.

[0004] The above-mentioned defects urgently need to be addressed. Summary of the Invention

[0005] To overcome the problems of low efficiency, poor scalability, large data transmission latency, and weak collaboration capabilities in existing centralized 3D modeling methods, this invention provides a 3D modeling task distribution and merging method, device, and medium. By constructing a mechanism for merging edge-distributed modeling tasks with cloud-based models, the efficiency and scalability of 3D modeling are improved, while ensuring that the model accuracy meets the needs of practical applications.

[0006] The technical solution of this invention is as follows:

[0007] In a first aspect, the present invention provides a method for distributing and merging three-dimensional modeling tasks, including:

[0008] Based on the complexity assessment results of the scenario to be modeled, the modeling task is distributed to multiple edge computing nodes to form a distributed modeling task set;

[0009] The edge computing nodes process the assigned local modeling tasks in parallel, generate corresponding local 3D models, and upload the local 3D models to the cloud server.

[0010] After receiving the local 3D model from multiple edge computing nodes, the cloud server performs model coordinate alignment, mesh fusion, and hierarchical optimization to generate a complete 3D scene model.

[0011] As a preferred embodiment of the present invention, the step of distributing the modeling task to multiple edge computing nodes based on the complexity assessment result of the scene to be modeled includes:

[0012] A multi-dimensional evaluation system is constructed to quantitatively evaluate the scene to be modeled from four dimensions: scene data volume, geometric topological complexity, texture accuracy requirements, and real-time indicators. The complexity level is output through a weighted summation algorithm.

[0013] Assess the data volume, geometric topology complexity, and computational resource requirements of the scenario to determine the task distribution strategy; the task distribution strategy includes static pre-segmentation of tasks and dynamic adjustment of tasks, and the task distribution strategy is dynamically adapted according to the complexity level;

[0014] Based on the computing power and network status of the edge computing nodes, modeling tasks are dynamically distributed, and the task allocation priority of each edge computing node is determined by the analytic hierarchy process.

[0015] A task queue scheduling and distribution process is adopted. The task queue monitors the load rate of each edge computing node in real time. When the load rate of an edge computing node exceeds a third preset percentage, a task migration mechanism is triggered to migrate a portion of the unfinished tasks of the edge computing node with a load rate exceeding the third preset percentage to an edge computing node with a load rate below a fourth preset percentage, where the third preset percentage is greater than the fourth preset percentage. When the load rate of an edge computing node is lower than a fifth preset percentage, a new task is allocated from the task queue to an edge computing node with a load rate below the fifth preset percentage, where the fifth preset percentage is less than the fourth preset percentage.

[0016] As a preferred embodiment of the present invention, the task queue splits tasks with a complexity level exceeding a preset level into multiple sub-tasks and assigns them to different edge computing nodes for parallel processing, and the sub-tasks synchronize their progress through a data interface.

[0017] As a preferred embodiment of the present invention, the task static pre-segmentation is based on a spatial clustering algorithm to divide the scene to be modeled into multiple non-overlapping sub-regions, each sub-region corresponding to a basic modeling task, and the size of the sub-region is set according to the average computing power of the edge computing nodes; the task dynamic adjustment is based on the real-time running status of the edge computing nodes, and a greedy algorithm is used to optimize task allocation so that the load rate difference of each edge computing node does not exceed a first preset percentage, and the task completion time deviation does not exceed a second preset percentage.

[0018] As a preferred embodiment of the present invention, the edge computing node processes the assigned local modeling task in parallel, generates the corresponding local 3D model, and uploads the local 3D model to the cloud server, including:

[0019] The edge computing node acquires local scene data and performs data preprocessing and feature extraction to extract multiple key feature points of the scene;

[0020] Based on the extracted key feature points of the scene, a 3D reconstruction is performed by combining a vision-based structure-of-motion reconstruction algorithm and a multi-view stereo matching algorithm to generate the local 3D model. A simplification algorithm is then applied to simplify the local 3D model.

[0021] The simplified local 3D model is compressed while retaining key geometric information;

[0022] The compressed local 3D model is uploaded to the cloud server.

[0023] As a preferred embodiment of the present invention, after the cloud server receives the local 3D model from multiple edge computing nodes, it performs model coordinate alignment, mesh fusion, and hierarchical optimization processing, including:

[0024] The coordinate system of the local 3D model is aligned using a feature matching algorithm;

[0025] The local 3D model after coordinate alignment is merged using a mesh fusion algorithm;

[0026] The merged local 3D model hierarchy is optimized using the LOD algorithm, and the merged local 3D model is divided into a far-distance rendering layer, a mid-distance rendering layer, a close-up rendering layer, and a extreme close-up rendering layer.

[0027] As a preferred embodiment of the present invention, after the cloud server receives the local 3D models from multiple edge computing nodes, it also performs texture optimization and lighting consistency adjustment; the texture optimization adopts a texture remapping algorithm to eliminate the texture color difference of each local 3D model; the lighting consistency adjustment is based on a global lighting model to uniformly verify the lighting parameters of each local 3D model.

[0028] As a preferred embodiment of the present invention, after the cloud server generates a complete 3D scene model, it performs quality inspection on the 3D scene model. The quality inspection is quantitatively evaluated from three dimensions: model accuracy, completeness, and visualization effect.

[0029] In a second aspect, the present invention provides an electronic device, comprising:

[0030] At least one processor; and,

[0031] A memory communicatively connected to the at least one processor; wherein,

[0032] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the above-described 3D modeling task distribution and merging method.

[0033] Thirdly, the present invention provides a computer-readable storage medium, including a data storage area and a program storage area, wherein the data storage area stores created data and the program storage area stores a computer program; wherein, when the computer program is executed by a processor, it implements the three-dimensional modeling task distribution and merging method described above.

[0034] According to the above-described solution, the beneficial effects of this invention are as follows:

[0035] 1. Modeling efficiency is significantly improved. Through parallel computing processing of edge computing nodes, the modeling efficiency is improved by more than 60% compared with traditional methods, and the time spent on modeling large-scale scenes is reduced to 1 / 3 to 1 / 2 of the original time.

[0036] 2. Highly scalable, supporting collaborative parallel modeling of ≥50 edge computing nodes, and the number of edge computing nodes can be dynamically increased or decreased according to scenario requirements to adapt to modeling tasks of different scales;

[0037] 3. The model has excellent accuracy. Through optimization processing such as improved ICP algorithm and Poisson fusion algorithm, the model coordinate system deviation does not exceed 0.1mm and the surface roughness does not exceed 0.05mm, which meets the requirements of millimeter-level applications.

[0038] 4. Low bandwidth consumption: By adopting a model compression strategy, data transmission volume is reduced by more than 80%, significantly alleviating network bandwidth pressure.

[0039] 5. Strong fault tolerance: Through task migration and multi-replica backup mechanisms, it ensures seamless continuation of modeling tasks and zero data loss when edge computing nodes fail. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating a three-dimensional modeling task distribution and merging method according to an embodiment of the present invention;

[0041] Figure 2 for Figure 1 A detailed flowchart of step S1 is shown below;

[0042] Figure 3 for Figure 1 A detailed flowchart of step S2 is shown below;

[0043] Figure 4 for Figure 1 A detailed flowchart of step S3 is shown below;

[0044] Figure 5 This is a schematic diagram of the structure of a 3D modeling task distribution and merging device in one embodiment of the present invention;

[0045] Figure 6 This is a schematic diagram of the structure of a 3D modeling task distribution and merging device in another embodiment of the present invention;

[0046] Figure 7 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0047] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0048] It should be noted that the terms "comprising" and "having," and any variations thereof, in the specification and claims of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0049] like Figures 1 to 4 As shown, one embodiment of the present invention provides a method for distributing and merging 3D modeling tasks, including:

[0050] Step S1: Based on the complexity assessment results of the scene to be modeled, the modeling task is distributed to multiple edge computing nodes to form a distributed modeling task set. Complexity assessment is the core of achieving reasonable task distribution. A multi-dimensional assessment system is adopted to quantitatively assess the scene data volume, geometric topological complexity, texture accuracy requirements, and real-time indicators, and the complexity level is output through a weighted summation algorithm.

[0051] This step aims to comprehensively improve modeling efficiency by rationally allocating modeling tasks through scientific complexity assessment and task scheduling strategies, thereby achieving load balancing across edge computing nodes. The specific implementation process is as follows:

[0052] Step S11, Scene Data Acquisition and Preprocessing: First, the basic structural information of the nuclear power plant is retrieved through the cloud platform, including the land area, number of building floors, number of in-service equipment, and modeling accuracy requirements (millimeter level). Then, edge computing nodes deployed on-site are used to conduct a preliminary scan of the plant, acquiring point cloud data and texture images of the overall scene. The acquired data undergoes preprocessing, including data denoising, data format conversion, and data fragmentation. Data denoising uses a statistical filtering algorithm, setting the number of neighboring points to 50 and the standard deviation to 3, to remove outliers from the point cloud. Data format conversion converts the point cloud data to PLY format and the texture images to PNG format. Data fragmentation divides the overall data into multiple data blocks according to floors and areas, with each data block corresponding to a potential modeling task.

[0053] Step S12, Scene Complexity Assessment: Construct a multi-dimensional assessment system to quantitatively evaluate the scene to be modeled for each data block from four dimensions: scene data volume, geometric topological complexity, texture accuracy requirements, and real-time performance indicators. Output the complexity level (e.g., 0-10) using a weighted summation algorithm. The weight coefficients of the weighted summation algorithm are dynamically adapted based on the scene type. For example, in industrial scenes (such as nuclear power plants and industrial workshops), the data volume weight accounts for 40%, geometric topological complexity for 35%, texture accuracy for 12.5%, and real-time performance for 12.5%. In smart city scenes, the weights are adjusted to 35% for scene data volume, 30% for geometric topological complexity, 20% for texture accuracy, and 15% for real-time performance, ensuring that the weights adapt to the core requirements of different scenes. Taking the equipment area on the first floor of the factory as an example, the data volume is approximately 10GB (8GB point cloud, 2GB texture), scoring 8 points in the data volume dimension. This area has dense equipment, with an estimated 2 million triangular faces, dramatic changes in surface curvature, and complex topological connections, resulting in a geometric topological complexity score of 9 points. Clear display of equipment interfaces, welds, and other details is required, necessitating high texture resolution and mapping accuracy, resulting in a texture accuracy requirement score of 8 points. As a core area of ​​the factory, real-time performance is crucial, and modeling must be prioritized, resulting in a real-time performance score of 8 points. A weighted summation (weights: scene data volume 40%, geometric topological complexity 35%, texture accuracy requirement 12.5%, real-time performance 12.5%) yields a complexity level of 8×0.4+9×0.35+8×0.125+8×0.125=8.35, classifying it as a high-complexity sub-scene.

[0054] Step S13: Task Distribution Strategy Confirmation: Assess the scene data volume, geometric topological complexity, and computational resource requirements to determine the task distribution strategy. Scene data volume is quantified by point cloud data volume (unit: millions of points) and texture image resolution (unit: pixels). For example, the point cloud data volume of a smart city scene is typically 500-1000 million points, with a texture image resolution of 4K (3840×2160). The point cloud data volume of an industrial plant scene is 100-500 million points, with a texture image resolution of 2K (2560×1440). Geometric topological complexity is assessed by the number of triangular faces, the rate of change of surface curvature, and topological connectivity. The more triangular faces, the more drastic the curvature change, and the more complex the topological connectivity, the higher the geometric complexity level. Computational resource requirements are predicted based on the single-frame modeling time and memory usage. If the single-frame modeling time exceeds 500ms and the memory usage exceeds 8GB, it is determined to be a high-computational-requirement scenario. The task distribution strategy includes static pre-segmentation and dynamic task adjustment. Static pre-segmentation uses a spatial clustering algorithm to divide the scene to be modeled into multiple non-overlapping sub-regions. Each sub-region corresponds to a basic modeling task. The size of the sub-region is set according to the average computing power of the edge computing nodes, generally controlled so that a single edge computing node can complete the basic modeling within 30 minutes. Dynamic task adjustment is based on the real-time running status of the edge computing nodes and uses a greedy algorithm to optimize task allocation, ensuring that the load rate difference between edge computing nodes does not exceed a first preset percentage (e.g., 15%) and the task completion time deviation does not exceed a second preset percentage (e.g., 10%), avoiding situations where some nodes are overloaded while others are idle. The task distribution strategy is dynamically adapted according to the complexity level. For example, low-complexity scenes with a complexity level of 0-3 use static pre-segmentation, while medium-to-high-complexity scenes with a complexity level of 4-10 use dynamic task adjustment, optimizing task scheduling in real time to ensure efficiency and load balance.

[0055] Step S14: Task Allocation Priority Confirmation: Based on the computing power and network status of the edge computing nodes, modeling tasks are dynamically distributed, and the task allocation priority of each edge computing node is determined using the analytic hierarchy process (AHP). Computing power is quantitatively evaluated using CPU clock speed (GHz), GPU memory capacity (GB), and parallel computing cores. Nodes with a CPU clock speed of at least 3.0 GHz, GPU memory of at least 16 GB, and at least 2048 parallel computing cores are classified as high-performance nodes and assigned high-complexity sub-tasks; nodes with a CPU clock speed of 2.0-3.0 GHz, GPU memory of 8-16 GB, and 1024-2048 parallel computing cores are classified as medium-performance nodes and assigned medium-complexity sub-tasks; nodes with a CPU clock speed below 2.0 GHz, GPU memory below 8 GB, and fewer than 1024 parallel computing cores are classified as low-performance nodes and assigned low-complexity sub-tasks. Network status includes uplink bandwidth (Mbps), transmission latency (ms), and packet loss rate (%). Nodes with uplink bandwidth of at least 100Mbps, transmission latency of less than 50ms, and packet loss rate of less than 1% are prioritized for tasks with larger data volumes; nodes with uplink bandwidth of 50-100Mbps, transmission latency of 50-100ms, and packet loss rate of 1%-3% are assigned medium-sized data volume tasks; and nodes with uplink bandwidth of less than 50Mbps, transmission latency of more than 100ms, and packet loss rate of more than 3% are assigned small-sized data volume tasks. The task allocation priority calculation formula for edge computing nodes is: P=0.4×C+0.3×B+0.2×L+0.1×R, where C is the computing power score (0-10 points), B is the uplink bandwidth score (0-10 points), L is the transmission latency score (0-10 points), and R is the packet loss rate score (0-10 points). Edge computing nodes with higher priority are assigned tasks first.

[0056] Step S15, Task Queue Management: A task queue scheduling and distribution process is adopted to ensure load balancing. The task queue adopts a priority queue structure, which combines task complexity and node priority to ensure that high-complexity tasks are preferentially assigned to high-priority nodes. It also supports dynamic task insertion and migration. The task queue monitors the load rate of each edge computing node in real time. When the load rate of an edge computing node exceeds a third preset percentage (e.g., 75%), a task migration mechanism is triggered, migrating a portion (e.g., 30%) of unfinished tasks from the edge computing node with a load rate exceeding the third preset percentage to an edge computing node with a load rate below the fourth preset percentage (e.g., 50%). The third preset percentage is greater than the fourth preset percentage. When the load rate of an edge computing node is below the fifth preset percentage (e.g., 30%), new tasks are allocated from the task queue to edge computing nodes with a load rate below the fifth preset percentage. The fifth preset percentage is less than the fourth preset percentage, ensuring that the load rate of each node is maintained within a reasonable range (e.g., 30%-75%) between the fifth and third preset percentages. The task queue also supports task splitting, allowing tasks with complexity levels exceeding a preset level (e.g., level 8) to be divided into multiple subtasks and assigned to different edge computing nodes for parallel processing. The subtasks synchronize their progress via data interfaces, avoiding efficiency degradation caused by task dependencies. The migration process uses incremental data transmission, transferring only incomplete intermediate modeling results, with a migration latency of no more than 200ms.

[0057] Step S2: Edge computing nodes process the assigned local modeling tasks in parallel, generate corresponding local 3D models, and upload the local 3D models to the cloud server. The edge computing nodes adopt a heterogeneous computing architecture, integrating multi-core CPUs (no less than 16 cores) and high-performance GPUs (no less than 16GB of video memory), supporting multi-threaded parallel modeling, which greatly improves the efficiency of local modeling. The uploaded local 3D models carry node identifiers, modeling timestamps, and scene positioning information.

[0058] This step is a crucial link in the local 3D model. Relying on the heterogeneous computing capabilities of edge computing nodes and multi-source data fusion technology, a high-precision local 3D model is constructed. The specific implementation process is as follows:

[0059] Step S21, Multi-source Data Acquisition and Fusion: Edge computing nodes acquire local scene data. These nodes acquire local scene data in real time through sensors (LiDAR, camera, IMU) or read pre-processed historical data from local storage devices. LiDAR acquires point cloud data, industrial cameras acquire frame texture images, and IMUs acquire inertial data. The acquired local scene data is integrated using a data fusion algorithm. A combination of Kalman filtering and Bayesian estimation is employed, combining the geometric accuracy of the LiDAR point cloud with the texture information of the visual image. Simultaneously, IMU data is introduced to correct camera pose, effectively suppressing motion blur. The state equation for the Kalman filter is set as X=[x, y, z, rx, ry, rz, vx, vy, vz], and the observation equation is set as Z=[xl, yl, zl, xv, yv, zv, rxIMU, ryIMU, rzIMU], where (x, y, z) are the three-dimensional coordinates, (rx, ry, rz) are the attitude angles, (vx, vy, vz) are the velocities, xl, yl, zl are the lidar observations, xv, yv, zv are the camera observations, and rxIMU, ryIMU, rzIMU are the IMU observations. The position accuracy error of the filtered data does not exceed 1 mm, and the attitude accuracy error does not exceed 0.01°.

[0060] Step S22, Data Preprocessing and Feature Extraction: Data preprocessing and feature extraction are performed on the fused local scene data to extract multiple key feature points. Data preprocessing includes data denoising, data registration, data format conversion, and data augmentation. For data denoising, a statistical filtering algorithm is used to remove outliers from the LiDAR point cloud data, and a Gaussian filtering algorithm is used for the visual image data, with a window size of 3×3 and a standard deviation of 0.8, effectively eliminating image noise and improving the quality of the original data. Data registration is based on the SIFT feature point matching algorithm to achieve spatiotemporal alignment of multi-source data (point cloud, RGB image, IMU), with registration accuracy error controlled within 0.5 pixels. Data format conversion unifies the data format to PLY point cloud format and PNG texture format to meet the standardized input requirements for subsequent 3D modeling and rendering. Data augmentation expands the training samples through geometric transformations such as random rotation, translation, and scaling, improving the robustness and generalization ability of the model. Feature extraction employs the ORB algorithm, combined with FAST corner detection and BRIEF descriptors, extracting no fewer than 500 key scene feature points. The scene key feature point descriptors have a dimension of 256. Simultaneously, a feature point clustering algorithm is used to remove redundant feature points, reducing computational load. The feature extraction accuracy is no less than 95%.

[0061] Step S23, 3D Reconstruction and Model Optimization: Based on the extracted key feature points of multiple scenes, a 3D reconstruction is performed by combining the vision-based Structure of Motion (SfM) algorithm and the Multi-View Stereo Matching (MVS) algorithm to generate a local 3D model. A simplification algorithm is then applied to simplify the local 3D model. The SfM algorithm recovers the camera pose and scene structure from the multi-view images, generating a sparse point cloud. The camera pose estimation error is better than 0.01°, and the sparse point cloud density is controlled at 2-3 points / cm². 2 The MVS algorithm generates high-density point clouds from sparse point clouds, and constructs a 3D mesh model through a patch growing algorithm. The density of the high-density point cloud is controlled at 5-10 points / cm. 2 The number of triangles in the mesh model is adjusted according to the complexity of different scenarios: no less than 1 million triangles for industrial scenarios and ≥5 million triangles for smart city scenarios. Model simplification employs a quadrilateral folding algorithm, reducing the number of triangles by merging adjacent quadrilaterals. While maintaining the model's topology, this reduces the number of triangles by 30%-50%, resulting in a Hausdorff distance of ≤2mm between the simplified and original models, ensuring that model accuracy is not significantly affected. Furthermore, a texture mapping algorithm is used to apply 2D texture images to the surface of the 3D mesh model, achieving a texture mapping accuracy of ≥98% and effectively suppressing texture stretching and misalignment.

[0062] Step S24, Model Compression: The simplified local 3D model is compressed using an improved Draco compression algorithm. The improved Draco compression algorithm optimizes compression parameters for point cloud data and mesh data separately. Point cloud data compression focuses on preserving key geometric information, while mesh data compression focuses on preserving topological structure. The compression ratio can be configured according to scene requirements, adjustable from 10:1 to 50:1. In industrial scenes, the compression ratio for geometrically critical areas (such as equipment interfaces and welds) should not exceed 20:1, while the compression ratio for non-critical areas (such as walls and floors) can be increased to 50:1. In smart city scenes, the compression ratio for distant areas can be 50:1, and the compression ratio for near-field areas can be 20:1. After compression, the model data size is reduced by over 80%, significantly reducing transmission bandwidth usage.

[0063] Step S25, Model Transmission: The compressed partial 3D model is encrypted and uploaded to the cloud server via HTTPS protocol. A CRC check mechanism is embedded during transmission, and the cloud server verifies data integrity in real time. In case of an anomaly, a retransmission is requested to ensure reliability. After verifying the data's integrity, the cloud server sends confirmation information to the edge computing node, which then deletes the compressed partial 3D model, releasing storage resources. Simultaneously, breakpoint resumption technology is employed, transmitting only the unfinished portion after network recovery, improving transmission efficiency.

[0064] Step S3: After receiving the local 3D model from multiple edge computing nodes, the cloud server performs model coordinate alignment, mesh fusion and hierarchical optimization to generate a complete 3D scene model.

[0065] This step is the core processing stage on the cloud server. It employs a pipelined architecture to accurately stitch and optimize multiple local 3D models, ultimately generating a complete 3D scene model that meets the needs of practical applications. The specific implementation process is as follows:

[0066] Step S31, Model Reception and Verification: The cloud server receives local 3D models uploaded by each edge computing node in parallel through a distributed data receiving module. Employing a block storage and redundancy backup mechanism, the model data is written to an SSD array in blocks according to node identifiers and sub-scene types. Each block of data corresponds to three copies (two on the primary server and one on the backup server), comprehensively ensuring data security. After reception, the cloud server automatically performs dual verification: First, integrity verification, comparing the CRC-32 checksum uploaded by the edge computing node (e.g., node 7) with the locally calculated value. The verification pass rate must reach 100%. If verification fails (e.g., the model data uploaded by node 7 has a verification deviation of 0.02%), the cloud server immediately sends a retransmission command to that edge computing node, limiting the retransmission of abnormal data blocks to 30 seconds. If the number of retransmissions exceeds three, a node failure warning is triggered. Second, format standardization verification, uniformly converting all fragmented models to the standardized PLY format, eliminating invalid topology structures and redundant texture data to ensure compatibility in subsequent processing. After successful verification, the cloud server generates a model receiving receipt, which includes information such as the receiving timestamp, data size, and verification result. This receipt is then sent back to the corresponding edge computing node, and the task queue status is updated to pending merging.

[0067] Step S32, Model Coordinate Alignment: Align the coordinate systems of local 3D models using a feature matching algorithm. Specifically, after the cloud server receives the local 3D models uploaded by each edge computing node, it first extracts the key feature points of each model, generates feature point descriptors based on the ORB algorithm, and establishes the correspondence between different models through feature point matching. The coordinate alignment uses an improved ICP algorithm, introducing RANSAC robust estimation to remove outliers and avoid the impact of abnormal feature points on alignment accuracy. The number of iterations is set to 20-50, and the convergence threshold is set to 0.01mm. Iteration stops when the model deviation between two iterations is less than the convergence threshold. During the alignment process, a global coordinate system is established. Using the coordinate system of the first uploaded model as a reference, other models are transformed into the global coordinate system. After alignment, the model coordinate system deviation does not exceed 0.1mm, ensuring accurate stitching of each local 3D model. For large-scale scenes, a layered alignment strategy is adopted: first, adjacent local 3D models are aligned pairwise, and then the aligned models are aligned as a whole to the global coordinate system, improving alignment efficiency and accuracy.

[0068] Step S33: Mesh Blending and Seam Optimization: The mesh fusion algorithm merges the coordinate-aligned local 3D models, eliminating seams and overlaps. The mesh fusion algorithm uses the Poisson fusion algorithm, based on the implicit surface reconstruction principle, to construct a globally consistent mesh topology. First, the normal vectors and vertex coordinates of each local 3D model are calculated. For vertices in overlapping areas, a weighted average method is used, with weights determined based on the distance from the vertex to the model center; the closer the distance, the greater the weight. The normal vectors of overlapping areas are smoothed to ensure continuous change. A 3×3 window-sized smoothing filter algorithm is used to optimize the seams. By adjusting the coordinates and normal vectors of the vertices on both sides of the seam, visible seam traces are eliminated, and the surface roughness of the merged model is controlled within 0.05mm. For mesh conflict areas, a mesh cutting algorithm is used to remove redundant meshes, preserving accurate geometry, thereby ensuring the integrity and consistency of the merged model.

[0069] Step S34, Texture Optimization and Illumination Consistency Adjustment: Texture optimization and illumination consistency adjustment are performed on the merged local 3D models. Specifically, to alleviate the texture inconsistency problem among multiple edge computing nodes, a texture remapping algorithm based on feature point matching is selected to eliminate texture color differences between local 3D models: First, texture feature points of each sub-model are extracted to construct a globally unified texture coordinate mapping; for regions in the CIE Lab color space with a color difference ΔE > 2.5, histogram equalization is used to correct their brightness and contrast, reducing the texture color difference between adjacent models to ΔE ≤ 1.2. Illumination consistency adjustment is based on a global illumination model, uniformly verifying the illumination parameters (including ambient light intensity, diffuse coefficient, and specular coefficient) of each local 3D model to ensure uniform and consistent illumination effects in the merged local 3D models.

[0070] Step S35, LOD Level Optimization and Rendering Adaptation: The LOD (Level of Detail) algorithm optimizes the merged local 3D model hierarchy to improve rendering efficiency. The LOD algorithm divides the merged local 3D model into four levels: a far-range rendering layer (Level 0, 80% reduction in polygon count), a mid-range rendering layer (Level 1, 50% reduction in polygon count), a close-range rendering layer (Level 2, 20% reduction in polygon count), and a close-up rendering layer (Level 3, original precision). Each level automatically switches based on view distance and perspective, with a switching latency of no more than 50ms. Simultaneously, an LOD caching mechanism is employed to preload model data from adjacent levels, preventing stuttering during switching and improving the real-time rendering experience.

[0071] Step S36, Model Quality Inspection: The generated complete 3D scene model undergoes quality inspection, which is quantitatively evaluated from three dimensions: model accuracy, integrity, and visualization effect. Accuracy inspection selects multiple key measurement points, comparing the measured values ​​with the theoretical values ​​of the model using a laser tracker; the deviation is ≤0.1mm, meeting millimeter-level requirements. Integrity inspection compares each equipment component in the model with a pre-set list of multiple factory equipment items. Visualization effect inspection combines automatic inspection and manual verification to confirm the absence of mesh gaps, texture misalignment, abnormal lighting, and other issues.

[0072] The 3D modeling task distribution and merging method in this embodiment achieves efficient and scalable 3D modeling, and is particularly suitable for collaborative modeling applications involving massive scenes. Compared with traditional centralized modeling methods, the advantages of this invention are: First, significantly improved modeling efficiency. Through parallel computing processing of edge computing nodes, modeling efficiency is improved by more than 60% compared with traditional methods, and the time consumption for large-scale scene modeling is reduced to 1 / 3 to 1 / 2 of the original. Second, strong scalability. It supports collaborative parallel modeling of ≥50 edge computing nodes, and the number of edge computing nodes can be dynamically increased or decreased according to scene requirements to adapt to modeling tasks of different scales. Third, excellent model accuracy. Through optimization processing such as improved ICP algorithm and Poisson fusion algorithm, the model coordinate system deviation does not exceed 0.1mm, and the surface roughness does not exceed 0.05mm, meeting the requirements of millimeter-level applications. Fourth, low bandwidth consumption. By adopting a model compression strategy, the amount of data transmission is reduced by more than 80%, greatly alleviating network bandwidth pressure. Fifth, strong fault tolerance. Through task migration and multi-copy backup mechanisms, it ensures seamless continuation of modeling tasks and zero data loss when edge computing nodes fail. This invention can be widely used in fields such as industrial plants, smart cities, nuclear power facility operation and maintenance, and virtual reality, and has both outstanding technological advancement and broad market prospects.

[0073] like Figure 5 As shown, one embodiment of the present invention provides a 3D modeling task distribution and merging device, comprising:

[0074] Task distribution module 1 is used to distribute modeling tasks to multiple edge computing nodes based on the complexity assessment results of the scenario to be modeled, forming a distributed modeling task set. Task distribution module 1 integrates a complexity assessment unit, a node status monitoring unit, and a task queue management unit. The complexity assessment unit outputs the complexity level using a multi-level assessment system, the node status monitoring unit collects the running parameters of the edge computing nodes in real time, and the task queue management unit uses a priority queue structure to realize dynamic task allocation.

[0075] Edge processing module 2 is used to process the assigned local modeling tasks in parallel on the edge computing nodes, generate the corresponding local 3D models, and upload the local 3D models to the cloud server. Edge processing module 2 includes a data preprocessing unit, a 3D reconstruction unit, a model optimization unit, and a data compression unit. The data preprocessing unit performs noise reduction and registration of multi-source data. The 3D reconstruction unit uses the SfM algorithm and the MVS algorithm to generate dense point clouds. The model optimization unit simplifies the model through the quadrilateral folding algorithm. The data compression unit uses the Draco algorithm to compress the model data.

[0076] The cloud merging module 3 receives local 3D models from multiple edge computing nodes, performs model coordinate alignment, mesh fusion, and hierarchical optimization to generate a complete 3D scene model. The cloud merging module 3 includes a coordinate alignment unit, a mesh fusion unit, a texture optimization unit, a lighting adjustment unit, a LOD optimization unit, and a quality inspection unit. The coordinate alignment unit uses an improved ICP algorithm for coordinate system calibration; the mesh fusion unit eliminates model seams through Poisson fusion; the texture optimization unit uses a texture remapping algorithm to eliminate texture color differences; the lighting adjustment unit uses a global lighting model to unify lighting parameters, ensuring consistent visual effects after merging; the LOD optimization unit constructs a multi-level-of-detail model; and the quality inspection unit quantitatively evaluates the model's accuracy and integrity.

[0077] The 3D modeling task distribution and merging device provided in this application embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0078] like Figure 6 As shown, in one embodiment, the 3D modeling task distribution and merging device may further include an incremental update module 4 and a fault-tolerant processing module 5. The incremental update module 4 identifies the changed areas through a scene change detection algorithm and only updates the modeling of the affected parts. The fault-tolerant processing module 5 monitors the running status of the edge computing nodes and automatically migrates the tasks and restores the intermediate modeling results when the nodes fail, ensuring that the tasks continue.

[0079] like Figure 7 The diagram shown is a structural schematic of the electronic device that implements the three-dimensional modeling task distribution and merging method of the present invention.

[0080] The electronic device may include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and may also include a computer program stored in the memory 11 and capable of running on the processor 10, such as a 3D modeling task distribution and merging program.

[0081] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (e.g., executing 3D modeling task distribution and merging programs) and calls data stored in the memory 11 to perform various functions of the electronic device and process data.

[0082] The memory 11 includes at least one type of readable storage medium, including flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of an electronic device, such as a portable hard drive. In other embodiments, the memory 11 can be an external storage device of the electronic device, such as a plug-in portable hard drive, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. Furthermore, the memory 11 can include both internal and external storage units of the electronic device. The memory 11 can be used not only to store application software and various types of data installed on the electronic device, such as the code of a 3D modeling task distribution and merging program, but also to temporarily store data that has been output or will be output.

[0083] The communication bus 12 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable communication between the memory 11 and at least one processor 10, etc.

[0084] Communication interface 13 is used for communication between the aforementioned electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, Bluetooth interface, etc.), typically used to establish communication connections between the electronic device and other electronic devices. The user interface may be a display, an input unit (such as a keyboard), and optionally, a standard wired or wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device and to display a visual user interface.

[0085] Figure 7 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 7 The structure shown does not constitute a limitation on the electronic device and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0086] For example, a power supply, although not shown, may also include a power source (such as a battery) to power various components. Preferably, the power supply can be logically connected to at least one processor 10 via a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power sources, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be elaborated further here.

[0087] It should be understood that the embodiments are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0088] The memory 11 in the electronic device stores a 3D modeling task distribution and merging program, which is a combination of multiple computer programs.

[0089] For the specific implementation method of the above computer program by processor 10, please refer to Figure 1 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0090] Furthermore, if the modules / units integrated into an electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, a computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0091] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following functions: Figure 1 The flowchart illustrates the method for distributing and merging 3D modeling tasks.

[0092] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0093] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0094] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0095] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0096] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0097] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0098] Furthermore, although the operations of the method of the present invention are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0099] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

[0100] The present invention has been described above with reference to the accompanying drawings. Obviously, the implementation of the present invention is not limited to the above-described manner. Any improvements made using the inventive concept and technical solution of the present invention, or the direct application of the inventive concept and technical solution of the present invention to other situations without modification, are all within the protection scope of the present invention.

Claims

1. A method for distributing and merging 3D modeling tasks, characterized in that, include: Based on the complexity assessment results of the scenario to be modeled, the modeling task is distributed to multiple edge computing nodes to form a distributed modeling task set; The edge computing nodes process the assigned local modeling tasks in parallel, generate corresponding local 3D models, and upload the local 3D models to the cloud server. After receiving the local 3D model from multiple edge computing nodes, the cloud server performs model coordinate alignment, mesh fusion, and hierarchical optimization to generate a complete 3D scene model.

2. The method for distributing and merging 3D modeling tasks according to claim 1, characterized in that, Based on the complexity assessment results of the scenario to be modeled, the modeling task is distributed to multiple edge computing nodes, including: A multi-dimensional evaluation system is constructed to quantitatively evaluate the scene to be modeled from four dimensions: scene data volume, geometric topological complexity, texture accuracy requirements, and real-time indicators. The complexity level is output through a weighted summation algorithm. Assess the data volume, geometric topology complexity, and computational resource requirements of the scenario to determine the task distribution strategy; the task distribution strategy includes static pre-segmentation of tasks and dynamic adjustment of tasks, and the task distribution strategy is dynamically adapted according to the complexity level; Based on the computing power and network status of the edge computing nodes, modeling tasks are dynamically distributed, and the task allocation priority of each edge computing node is determined by the analytic hierarchy process. A task queue scheduling and distribution process is adopted. The task queue monitors the load rate of each edge computing node in real time. When the load rate of an edge computing node exceeds a third preset percentage, a task migration mechanism is triggered to migrate a portion of the unfinished tasks of the edge computing node with a load rate exceeding the third preset percentage to an edge computing node with a load rate below a fourth preset percentage, where the third preset percentage is greater than the fourth preset percentage. When the load rate of an edge computing node is lower than a fifth preset percentage, a new task is allocated from the task queue to an edge computing node with a load rate below the fifth preset percentage, where the fifth preset percentage is less than the fourth preset percentage.

3. The method for distributing and merging 3D modeling tasks according to claim 2, characterized in that, The task queue splits tasks with a complexity level exceeding a preset level into multiple subtasks and assigns them to different edge computing nodes for parallel processing, with the subtasks synchronizing their progress through a data interface.

4. The method for distributing and merging 3D modeling tasks according to claim 2, characterized in that, The static pre-segmentation of the task is based on a spatial clustering algorithm to divide the scene to be modeled into multiple non-overlapping sub-regions. Each sub-region corresponds to a basic modeling task. The size of the sub-region is set according to the average computing power of the edge computing nodes. The dynamic adjustment of the task is based on the real-time running status of the edge computing nodes. A greedy algorithm is used to optimize the task allocation so that the load rate difference of each edge computing node does not exceed a first preset percentage and the task completion time deviation does not exceed a second preset percentage.

5. The method for distributing and merging 3D modeling tasks according to claim 1, characterized in that, The edge computing nodes process the assigned local modeling tasks in parallel, generate corresponding local 3D models, and upload the local 3D models to the cloud server, including: The edge computing node acquires local scene data and performs data preprocessing and feature extraction to extract multiple key feature points of the scene; Based on the extracted key feature points of the scene, a 3D reconstruction is performed by combining a vision-based motion recovery structure algorithm and a multi-view stereo matching algorithm to generate the local 3D model. A simplification algorithm is then applied to simplify the local 3D model. The simplified local 3D model is compressed while retaining key geometric information; The compressed local 3D model is uploaded to the cloud server.

6. The method for distributing and merging 3D modeling tasks according to claim 1, characterized in that, After receiving the local 3D model from multiple edge computing nodes, the cloud server performs model coordinate alignment, mesh fusion, and hierarchical optimization processing, including: The coordinate system of the local 3D model is aligned using a feature matching algorithm; The local 3D model after coordinate alignment is merged using a mesh fusion algorithm; The merged local 3D model hierarchy is optimized using the LOD algorithm, and the merged local 3D model is divided into a far-distance rendering layer, a mid-distance rendering layer, a close-up rendering layer, and a extreme close-up rendering layer.

7. The method for distributing and merging 3D modeling tasks according to claim 1, characterized in that, After receiving the local 3D models from multiple edge computing nodes, the cloud server performs texture optimization and lighting consistency adjustment. The texture optimization uses a texture remapping algorithm to eliminate the texture color difference of each local 3D model. The lighting consistency adjustment is based on a global lighting model to uniformly verify the lighting parameters of each local 3D model.

8. The method for distributing and merging 3D modeling tasks according to claim 1, characterized in that, After the cloud server generates a complete 3D scene model, it performs quality inspection on the 3D scene model. The quality inspection is quantitatively evaluated from three dimensions: model accuracy, completeness, and visualization effect.

9. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the 3D modeling task distribution and merging method as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing computer-executable instructions, characterized in that, It includes a data storage area and a program storage area. The data storage area stores the created data, and the program storage area stores the computer program. When the computer program is executed by the processor, it implements the three-dimensional modeling task distribution and merging method as described in any one of claims 1 to 8.

Images