Lightweight hierarchical digital twin modeling and rendering method for large-scale scenarios

Abstract

This application relates to the field of 3D modeling technology and provides a lightweight hierarchical digital twin modeling and rendering method for large-scale scenes. The method includes: analyzing the surface geometry of the original high-precision 3D model, calculating and extracting the curvature features of each local region corresponding to the high-precision 3D model, and constructing a curvature response distribution through the curvature features; based on the obtained curvature response distribution and dynamic curvature threshold, filtering and judging each model region corresponding to the original high-precision 3D model to obtain the second key detail region; generating independent multi-resolution detail enhancement data blocks for the second key detail region, and generating multiple data versions with decreasing detail levels for the same second key detail region to form a directly indexable multi-resolution detail library; in the real-time rendering stage, seamlessly merging the detail data blocks with the basic main model to achieve a balance between overall scene lightweighting and high-fidelity local details.

Description

Technical Field

[0001] This application relates to the field of 3D modeling technology, and in particular to a lightweight hierarchical digital twin modeling and rendering method for large-scale scenes. Background Technology

[0002] In the field of 3D digital twin modeling and real-time rendering, balancing lightweight large-scale scenes with high-fidelity local details is a core technical challenge. While existing technologies have achieved overall modeling efficiency improvements and rendering resource optimization, they still have significant shortcomings in handling critical details.

[0003] First, the lack of quantitative analysis methods for the geometric features of the model makes it impossible to automatically identify key areas with high curvature, resulting in the loss of details such as local complex structures and sharp edges during the lightweighting process, affecting the accuracy of virtual-real synchronization. Second, the absence of a hierarchical storage and dynamic scheduling mechanism makes it impossible to dynamically allocate computing resources according to the user's viewpoint, which can easily lead to loading delays and interactive stuttering in ultra-large-scale scenes. Third, the lack of lighting and texture consistency processing in the fusion of high-resolution details and lightweight basic models results in visual disjointedness.

[0004] Therefore, a method is urgently needed to solve at least one of the above problems. Summary of the Invention

[0005] This application provides a lightweight hierarchical digital twin modeling and rendering method for large-scale scenes, aiming to solve the core technical challenge of balancing lightweight large-scale scenes with high-fidelity local details in the field of 3D digital twin modeling and real-time rendering.

[0006] In a first aspect, embodiments of this application provide a lightweight hierarchical digital twin modeling and rendering method for large-scale scenarios, the method comprising:

[0007] The surface geometry of the original high-precision 3D model is analyzed, and the curvature features of each local region of the high-precision 3D model are calculated and extracted. The curvature features are used to distinguish the flat regions and the first critical detail regions of the original high-precision 3D model, and a curvature response distribution is constructed for subsequent region selection.

[0008] Based on the obtained curvature response distribution, a dynamic curvature threshold is generated by combining the global curvature statistics and detail fidelity adjustment of the original high-precision 3D model. Each model region corresponding to the original high-precision 3D model is screened and judged according to the dynamic curvature threshold. Regions with curvature response exceeding the dynamic curvature threshold are identified as second critical detail regions that cannot be simplified. At the same time, a unique semantic label is generated for each identified second critical detail region.

[0009] For the second critical detail region, a detail level technique is used to generate independent multi-resolution detail enhancement data blocks. Multiple data versions with decreasing detail levels are generated for the same second critical detail region. A multi-resolution detail library that can be directly indexed is formed based on the data versions.

[0010] During the real-time rendering phase, the distance between the viewing point and the spatial center of each key detail area is calculated. Based on the distance and the viewing distance sensitivity coefficient, detail data blocks of the corresponding resolution level in the multi-resolution detail library are dynamically scheduled. In the graphics rendering pipeline, the detail data blocks and the basic main model are seamlessly integrated for rendering, achieving a balance between lightweight overall scene and high-fidelity local details.

[0011] In some embodiments, the first critical detail region is a region containing complex structures, sharp edges, or intricate textures. The curvature features are used to quantify the surface curvature and detail richness of local regions of the model, and the curvature features of all first critical detail regions are normalized. The analysis of the surface geometry of the original high-precision 3D model, calculating and extracting the curvature features of each local region corresponding to the high-precision 3D model, includes: acquiring the triangular mesh data of the original high-precision 3D model, performing a sequential traversal analysis of each vertex and each facet in the triangular mesh data, and obtaining the set of directly adjacent vertices of each vertex. The unit normal vector corresponding to each vertex is used to calculate the discrete curvature value of each vertex by averaging the angle between the unit normal vectors in the neighborhood of each vertex. To suppress the interference of single-point noise on the subsequent judgment results, an area-weighted curvature aggregation mechanism is introduced at the facet level. Based on the geometric area weight of the facet to which each vertex belongs in the corresponding region, the discrete curvature values ​​of the vertex are aggregated and calculated to obtain the region-level geometric complexity measure. The geometric complexity measures of all regions are mapped to a unified normalized feature space to complete the normalization processing of the curvature features of all first key detail regions and eliminate the absolute scale difference between different models.

[0012] In some embodiments, constructing the curvature response distribution for subsequent region selection includes: after completing the normalization processing of the curvature features of all regions, calculating the global maximum and global minimum values ​​of the geometric complexity measure of all regions in the current original high-precision 3D model; based on the normalized curvature features, global maximum and global minimum values ​​of each region, constructing the distribution relationship of the detail response intensity to form the curvature response distribution for subsequent region selection.

[0013] In some embodiments, the step of generating a dynamic curvature threshold based on the obtained curvature response distribution, combined with the global curvature statistics corresponding to the original high-precision 3D model and the detail fidelity adjustment, includes: extracting and calculating the normalized mean curvature and the global maximum curvature response value of all regions of the original high-precision 3D model from the constructed curvature response distribution to obtain the global curvature statistics of the model; obtaining a preset detail fidelity adjustment coefficient, which is used to manually set or automatically adapt to the specific needs of different industries for detail fidelity to control the strictness of the key detail region identification standard; generating a dynamic curvature threshold for the current scene by weighting the normalized mean curvature, the global maximum curvature response value, and the detail fidelity adjustment coefficient; and adjusting the level of the dynamic curvature threshold by adjusting the detail fidelity adjustment coefficient to achieve strict or lenient control over the identification of key detail regions.

[0014] In some embodiments, the step of filtering and determining each model region corresponding to the original high-precision 3D model based on a dynamic curvature threshold, identifying regions with curvature responses exceeding the dynamic curvature threshold as second critical detail regions that cannot be simplified, and generating a unique semantic label for each identified second critical detail region includes: comparing the normalized curvature response value of each model region with the generated dynamic curvature threshold one by one; when the normalized curvature response value is greater than the dynamic curvature threshold, marking the corresponding model region as a second critical detail region; obtaining the spatial center coordinates, regional geometric feature vector, and local geometric descriptor of each marked region; and generating a unique and reproducible semantic label for each second critical detail region by weighted encoding of the spatial center coordinates, regional geometric feature vector, and local geometric descriptor. The semantic label is used to achieve precise positioning and management of critical detail regions in lightweighting, storage, and rendering.

[0015] In some embodiments, the step of generating independent multi-resolution detail enhancement data blocks for the second critical detail region using detail hierarchy technology, and generating multiple data versions with decreasing detail levels for the same second critical detail region, includes: acquiring all region data identified as second critical detail regions; for each second critical detail region, using detail hierarchy technology to preset multiple detail resolution levels, including the highest detail level and the most simplified detail level; acquiring a preset level decay factor to control the detail compression degree of different resolution levels, ensuring that high curvature regions can still retain key morphological features in low resolution levels; based on the original high-precision geometric features of each second critical detail region, combined with the level decay factor, generating multiple data versions with decreasing detail levels for each critical detail region; during the data version generation process, using a curvature-oriented local mesh resampling algorithm, adaptively adjusting the data vertex density of each resolution level according to the region curvature gradient, so that high curvature regions still retain geometric contours in low resolution levels, while low curvature regions undergo significant downsampling processing, achieving detail balance of data blocks at each resolution level, and generating independent multi-resolution detail enhancement data blocks corresponding to each second critical detail region.

[0016] In some embodiments, forming a directly indexable multi-resolution detail library based on data versions includes: classifying and organizing multiple data versions with decreasing levels of detail generated for each second key detail region; associating each data version with a corresponding key detail region identifier, resolution level identifier, and detail feature information corresponding to the data version; constructing a unified indexing mechanism to map each second key detail region to all data versions corresponding to that region, clarifying the correspondence between each key detail region and data versions at different resolution levels; and integrating all data versions of the second key detail regions based on the mapping relationship to form a directly indexable multi-resolution detail library, and retrieving and calling the corresponding data version through the key detail region identifier and resolution level identifier.

[0017] In some embodiments, all data versions are stored independently of the lightweight base model, in the form of data blocks, and the spatial location, curvature features, and semantic tags of the corresponding key detail regions are recorded in the data block metadata. The process of forming a directly indexable multi-resolution detail library based on data versions includes: storing multiple data versions corresponding to each second key detail region separately in the form of independent data blocks, with all data blocks independent of the lightweight base model; recording the spatial location information, curvature feature information, and generated semantic tags of the corresponding key detail region in the metadata of each data block, ensuring that each data block can be traced back to the corresponding key detail region; constructing a unified indexing system based on the metadata information of all data blocks, linking the key information in the data block metadata with the data blocks themselves to form a directly indexable multi-resolution detail library; and locating and extracting corresponding data blocks at different resolution levels by retrieving the spatial location, curvature features, or semantic tags of the key detail regions, thereby achieving efficient access to the data blocks.

[0018] In some embodiments, the step of calculating the distance between the viewing point and the spatial center of each key detail region during the real-time rendering stage, and dynamically scheduling detail data blocks of the corresponding resolution level in the multi-resolution detail library based on the distance and a viewing distance sensitivity coefficient, includes: during the real-time rendering process, acquiring the spatial coordinates of the current viewing point and the spatial center coordinates of each second key detail region in real time; obtaining the spatial proximity between the viewing point and each key detail region by calculating the straight-line distance between the spatial coordinates of the viewing point and the spatial center coordinates of each second key detail region; and acquiring a preset viewing distance sensitivity coefficient and a maximum viewing distance threshold, wherein the viewing distance sensitivity coefficient is used to control the sensitivity of detail level switching. The maximum viewing distance threshold is used to define the scheduling range of detail data blocks. Based on the calculated spatial distance, viewing distance sensitivity coefficient, and maximum viewing distance threshold, the detail level index corresponding to each second critical detail region is determined. When the spatial distance is less than the maximum viewing distance threshold, the closer the observation point is to the critical detail region, the higher the resolution level of the detail data block is scheduled for that region; the farther away it is, the lower the resolution level of the detail data block is scheduled. A prediction caching mechanism is introduced to obtain the movement vector and movement amplitude of the observation point in real time. Based on the movement vector and movement amplitude, the second critical detail region that may be accessed in the next frame is predicted, and the detail data blocks of the corresponding resolution level are loaded in advance to avoid loading delays.

[0019] In some embodiments, the seamless fusion rendering based on the detail data block and the base main model includes: performing real-time spatial alignment processing on the detail data block and the base main model to ensure that the detail data block and the base main model are accurately superimposed in three-dimensional space; then dynamically mixing the color information of the detail data block with the texture of the base main model through a texture fusion algorithm; and introducing a lighting correction mechanism to eliminate the lighting and shadow deviations between the detail data block and the base main model, thereby achieving seamless fusion rendering of the detail data block and the base main model and achieving a balance between overall scene lightweighting and high-fidelity local details.

[0020] This application achieves quantitative differentiation between flat areas and key details with high curvature through automated calculation and analysis of curvature features, avoiding the loss of structural details during lightweighting and improving the visual accuracy and functional verification effect of virtual-real synchronization. Based on a dynamic threshold generation mechanism using global curvature statistics and industry requirements, the identification standards for key areas can be flexibly adjusted to adapt to the detail fidelity requirements of various scenarios such as intelligent manufacturing and urban planning. A layered storage mode of independent multi-resolution detail data blocks avoids the full loading of high-poly model data, significantly reducing memory usage. A dynamic scheduling mechanism based on viewpoint distance prioritizes the allocation of computing resources to the visual focus area, improving rendering efficiency and interactive smoothness. Through a comprehensive strategy of spatial alignment, texture blending, and lighting correction, high-resolution details and lightweight base models are seamlessly superimposed in the rendered image, maintaining consistency in lighting, shadows, and material representation, and enhancing scene realism.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic flowchart illustrating the steps of a lightweight hierarchical digital twin modeling and rendering method for large-scale scenarios provided in an embodiment of this application;

[0024] Figure 2 This is a diagram illustrating the curvature feature extraction process provided in this application;

[0025] Figure 3 This is a key detail area filtering and marking diagram provided in this application;

[0026] Figure 4This is the multi-resolution detail-enhanced data block generation map provided in this application;

[0027] Figure 5 This is the dynamic detail scheduling graph based on viewpoint distance provided in this application;

[0028] Figure 6 This is a rendering of the details and basic model provided in this application;

[0029] Figure 7 This is a schematic block diagram of a lightweight hierarchical digital twin modeling and rendering system for large-scale scenarios provided in one embodiment of this application;

[0030] Figure 8 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application.

[0031] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0033] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0034] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0035] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0036] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0037] In the field of 3D digital twin modeling and real-time rendering, balancing lightweight large-scale scenes with high-fidelity local details is a core technical challenge. While existing technologies have achieved overall modeling efficiency improvements and rendering resource optimization, they still have significant shortcomings in handling critical details.

[0038] First, the lack of quantitative analysis methods for the geometric features of the model makes it impossible to automatically identify key areas with high curvature, resulting in the loss of details such as local complex structures and sharp edges during the lightweighting process, affecting the accuracy of virtual-real synchronization. Second, the absence of a hierarchical storage and dynamic scheduling mechanism makes it impossible to dynamically allocate computing resources according to the user's viewpoint, which can easily lead to loading delays and interactive stuttering in ultra-large-scale scenes. Third, the lack of lighting and texture consistency processing in the fusion of high-resolution details and lightweight basic models results in visual disjointedness.

[0039] Therefore, a method is urgently needed to solve at least one of the above problems.

[0040] To solve the above problem, please refer to Figure 1 This application provides a lightweight, hierarchical digital twin modeling and rendering method for large-scale scenarios, applicable to computer devices. These computer devices can be deployed on a single server or server cluster, or on handheld terminals, laptops, wearable devices, or robots. It should be noted that all information involved in the method provided in this application is extracted with the authorization of the relevant users and in accordance with relevant regulations, and will not infringe on user privacy.

[0041] The provided lightweight hierarchical digital twin modeling and rendering method for large-scale scenes includes steps S101 to S104. Details are as follows:

[0042] Step S101. Analyze the surface geometry of the original high-precision 3D model, calculate and extract the curvature features of each local region of the high-precision 3D model, distinguish the flat regions and the first key detail regions of the original high-precision 3D model through the curvature features, and construct the curvature response distribution for subsequent region selection.

[0043] Specifically, the core of this step is to address the lack of quantitative analysis methods for model geometric features in existing technologies. By systematically analyzing the surface geometry of the original high-precision 3D model, the curvature features of each local region are accurately calculated and extracted. This distinguishes flat areas from preliminary key detail areas, constructs curvature response distribution, and provides a quantitative basis for the accurate identification of subsequent key details, thus avoiding the loss of details due to the lack of quantitative standards during the lightweighting process.

[0044] By decomposing the surface geometry of the original high-precision 3D model, several local regions are divided using mesh patches, vertices, or voxels as basic units. Geometric calculation methods such as Gaussian curvature and mean curvature are used to numerically calculate the surface curvature of each local region and extract the corresponding curvature feature values. Based on the magnitude of the curvature feature values, the flat regions of the original model (curvature values ​​approaching 0) and the first critical detail region (curvature values ​​significantly higher than those of the flat regions, initially identified as high-curvature potential critical regions) are preliminarily distinguished. The curvature feature values ​​of each local region are integrated to construct a curvature response distribution covering the entire model, forming a quantitative map that can intuitively reflect the geometric complexity of each region, providing basic data for subsequent region selection.

[0045] Step S102. Based on the obtained curvature response distribution, a dynamic curvature threshold is generated by combining the global curvature statistics and detail fidelity of the original high-precision 3D model. Each model region corresponding to the original high-precision 3D model is screened and judged according to the dynamic curvature threshold. Regions whose curvature response exceeds the dynamic curvature threshold are identified as second key detail regions that cannot be simplified. At the same time, a unique semantic label is generated for each identified second key detail region.

[0046] Specifically, this step addresses the limitations of using fixed thresholds to filter key details based on curvature response distribution. By combining global curvature statistics with detail fidelity requirements to generate dynamic curvature thresholds, it accurately identifies the second key detail region that cannot be simplified and assigns it a unique semantic label. This achieves standardized and indexable identification of key details, laying the foundation for subsequent hierarchical storage and scheduling.

[0047] Based on the curvature response distribution obtained in step S101, the global curvature statistics of the original high-precision 3D model are calculated, including core statistical indicators such as curvature mean, variance, extrema, and quantiles. Combined with preset detail fidelity adjustment parameters (e.g., high, medium, and low fidelity levels, corresponding to different adjustment coefficients), the global curvature statistics and adjustment parameters are fused to generate a dynamic curvature threshold (e.g., threshold = curvature mean + fidelity coefficient × curvature variance). Each local region of the original model is traversed, and the curvature response value of each region is compared with the dynamic curvature threshold. Regions whose curvature response exceeds the threshold are identified as second critical detail regions that cannot be simplified. A unique semantic label is generated for each identified second critical detail region, containing information such as region number, spatial location, and geometric type (e.g., "pipeline interface_high curvature_region 005"), ensuring that each critical detail region can be accurately identified and indexed.

[0048] Step S103. Use the level of detail technique to generate independent multi-resolution detail enhancement data blocks for the second critical detail region, generate multiple data versions with decreasing levels of detail for the same second critical detail region, and form a directly indexable multi-resolution detail library based on the data versions.

[0049] Specifically, this step targets the second critical detail region, using Level of Detail (LOD) technology to generate independent multi-resolution detail enhancement data blocks, constructing a directly indexable multi-resolution detail library, solving the storage and fusion problem of high-resolution details and lightweight base models, realizing hierarchical storage of critical details, and providing data support for dynamic scheduling in the real-time rendering stage.

[0050] For each second critical detail region identified in step S102, a level-of-detail technique is used independently for processing without affecting the overall lightweighting process of the original model. Multiple data versions with decreasing levels of detail are generated for each second critical detail region, such as the highest fidelity version (LOD0, retaining complete geometry, texture, and lighting information), a moderately simplified version (LOD1, simplifying some non-critical faces), and a low-simplification version (LOD2, further simplifying the structure). The number of versions can be flexibly set according to the scene scale and rendering requirements. The detail data of each version is encapsulated into an independent detail enhancement data block, which contains complete information such as the geometric model, texture map, and lighting parameters for the corresponding resolution. Using the unique semantic tag of the second critical detail region as an index, all detail data blocks are integrated to build a multi-resolution detail library. Data in the library can be quickly retrieved for each resolution version of the corresponding region through tags.

[0051] Step S104. In the real-time rendering stage, calculate the distance between the observation point and the spatial center of each key detail area. Based on the distance and the viewing distance sensitivity coefficient, dynamically schedule detail data blocks of the corresponding resolution level in the multi-resolution detail library. In the graphics rendering pipeline, complete seamless fusion rendering based on the detail data blocks and the basic main model to achieve the unity of overall scene lightweighting and local detail high fidelity.

[0052] Specifically, in the real-time rendering stage, this step achieves dynamic scheduling of multi-resolution detail data through viewpoint distance and view distance sensitivity coefficient, and completes seamless fusion rendering in combination with the basic main model, solving the problems of loading delay, interaction stuttering and visual fragmentation in ultra-large-scale scenes, and ultimately achieving the unity of lightweight large-scale scenes and high-fidelity local details.

[0053] During real-time rendering, the spatial coordinates of the current viewpoint are continuously acquired, and the Euclidean distance between the viewpoint and the spatial center of each second critical detail area is calculated. A viewing distance sensitivity coefficient is preset, and a mapping rule between distance and resolution level is established based on the calculated viewpoint distance (e.g., LOD0 is scheduled for distance < 5 meters, LOD1 is scheduled for distance 5 meters ≤ distance < 15 meters, and LOD2 is scheduled for distance ≥ 15 meters). According to the mapping rule, detail data blocks of the corresponding resolution level are dynamically scheduled from the multi-resolution detail library to achieve on-demand allocation of computing resources. In the graphics rendering pipeline, the lightweight basic main model is spatially matched with the scheduled detail data blocks, and lighting parameters and texture mapping are uniformly processed to eliminate visual disjointness and achieve seamless rendering. This ensures both the lightweight rendering efficiency of the overall scene and the high-fidelity presentation of critical detail areas.

[0054] In some embodiments, the first critical detail region is a region containing complex structures, sharp edges, or intricate textures. The curvature features are used to quantify the surface curvature and detail richness of local regions of the model, and the curvature features of all first critical detail regions are normalized. The analysis of the surface geometry of the original high-precision 3D model, calculating and extracting the curvature features of each local region corresponding to the high-precision 3D model, includes: acquiring the triangular mesh data of the original high-precision 3D model, performing a sequential traversal analysis of each vertex and each facet in the triangular mesh data, and obtaining the set of directly adjacent vertices of each vertex. The unit normal vector corresponding to each vertex is used to calculate the discrete curvature value of each vertex by averaging the angle between the unit normal vectors in the neighborhood of each vertex. To suppress the interference of single-point noise on the subsequent judgment results, an area-weighted curvature aggregation mechanism is introduced at the facet level. Based on the geometric area weight of the facet to which each vertex belongs in the corresponding region, the discrete curvature values ​​of the vertex are aggregated and calculated to obtain the region-level geometric complexity measure. The geometric complexity measures of all regions are mapped to a unified normalized feature space to complete the normalization processing of the curvature features of all first key detail regions and eliminate the absolute scale difference between different models.

[0055] This embodiment refines the three core aspects of step S101: "extraction of curvature features from the original high-precision 3D model," "definition of the first critical detail region," and "curvature feature normalization." It addresses the problems of low curvature extraction accuracy, single-point noise interference in judgment, and inconsistent subsequent screening standards due to differences in model scales in existing technologies. It clarifies the specific range of the first critical detail region (complex structures, sharp edges, and areas with intricate textures), and through a standardized curvature extraction and normalization process, quantitatively characterizes the surface curvature and detail richness of local areas of the model, providing a unified quantitative basis for subsequent region screening.

[0056] Data Input and Analysis: Obtain the triangular mesh data of the original high-precision 3D model, which is the basis for the geometric analysis of the model surface; perform a traversal analysis on each vertex and each facet in the triangular mesh data to accurately obtain the set of direct adjacent vertices of each vertex and the unit normal vector corresponding to each vertex, providing basic data for subsequent curvature calculation.

[0057] Vertex discrete curvature calculation: Based on the set of directly adjacent vertices and their unit normal vectors of each vertex, the discrete curvature value of each vertex is obtained by averaging the angles between all unit normal vectors in the neighborhood of that vertex. This value directly reflects the degree of curvature of the local region where the vertex is located.

[0058] Region-level geometric complexity aggregation: To suppress the interference of single-point noise on the subsequent key region determination results, an area-weighted curvature aggregation mechanism is introduced at the facet level. Based on the geometric area weight of the facet to which each vertex belongs in the corresponding local region, the discrete curvature value of the vertex is aggregated and calculated to finally obtain the region-level geometric complexity measure, thereby achieving a stable representation of the geometric features of the local region.

[0059] Curvature feature normalization: The geometric complexity measures of all local regions are uniformly mapped to the preset normalized feature space to complete the normalization processing of the curvature features of all first key detail regions; this processing eliminates the absolute scale difference between different original models, ensuring that the subsequent curvature threshold screening can be applied to a unified judgment standard across models and scenes.

[0060] In some embodiments, constructing the curvature response distribution for subsequent region selection includes: after completing the normalization processing of the curvature features of all regions, calculating the global maximum and global minimum values ​​of the geometric complexity measure of all regions in the current original high-precision 3D model; based on the normalized curvature features, global maximum and global minimum values ​​of each region, constructing the distribution relationship of the detail response intensity to form the curvature response distribution for subsequent region selection.

[0061] This embodiment corresponds to step S101, "constructing a curvature response distribution for subsequent region screening." The core is to construct the distribution relationship of detailed response intensity based on the obtained normalized curvature features and the extreme values ​​of the global geometric complexity measurement of the statistical model, forming a standardized and reusable curvature response distribution, which provides intuitive and reliable data support for the generation of dynamic curvature threshold and region screening in step S102.

[0062] After normalizing the curvature features of all local regions, global extremum statistics perform global statistics on the geometric complexity of all regions in the current original high-precision 3D model, accurately extracting and recording the global maximum and global minimum values. These extrema are used to eliminate scale bias of curvature features in different regions.

[0063] The curvature response distribution is constructed based on the normalized curvature features of each local region. Combined with the statistically obtained global maximum and global minimum values, a corresponding distribution relationship between the intensity of the detail response and the geometric complexity of the region is constructed. This distribution relationship can intuitively reflect the richness of detail in each region of the model, forming a curvature response distribution for subsequent region selection, ensuring that subsequent threshold selection can accurately locate high-detail regions.

[0064] In some embodiments, the step of generating a dynamic curvature threshold based on the obtained curvature response distribution, combined with the global curvature statistics corresponding to the original high-precision 3D model and the detail fidelity adjustment, includes: extracting and calculating the normalized mean curvature and the global maximum curvature response value of all regions of the original high-precision 3D model from the constructed curvature response distribution to obtain the global curvature statistics of the model; obtaining a preset detail fidelity adjustment coefficient, which is used to manually set or automatically adapt to the specific needs of different industries for detail fidelity to control the strictness of the key detail region identification standard; generating a dynamic curvature threshold for the current scene by weighting the normalized mean curvature, the global maximum curvature response value, and the detail fidelity adjustment coefficient; and adjusting the level of the dynamic curvature threshold by adjusting the detail fidelity adjustment coefficient to achieve strict or lenient control over the identification of key detail regions.

[0065] This embodiment corresponds to step S102, "generating a dynamic curvature threshold based on curvature response distribution," which solves the problem in existing technologies where fixed thresholds cannot adapt to the detail fidelity requirements of different industries and scenarios. By fusing the model's global curvature statistics with the detail fidelity adjustment coefficient, a dynamically adjustable curvature threshold is generated, enabling precise control over the identification standard of key detail areas and adapting to the local accuracy requirements of different application scenarios.

[0066] Global curvature statistics extraction involves extracting and calculating the normalized mean curvature of all regions of the original high-precision 3D model from the constructed curvature response distribution, and simultaneously extracting the global maximum curvature response value. Both are used as global curvature statistics of the model to reflect the overall geometric detail distribution characteristics of the model.

[0067] The detail fidelity adjustment coefficient is obtained by acquiring a preset detail fidelity adjustment coefficient. This coefficient can be adjusted manually or automatically by the system according to the specific needs of different industries (such as intelligent manufacturing and urban planning) for detail fidelity. It is used to control the strictness of the identification standard for key detail areas (the larger the coefficient, the stricter the identification standard, the fewer key areas are selected but the higher the accuracy).

[0068] The dynamic curvature threshold calculation generates the dynamic curvature threshold for the current scene by weighting the normalized curvature mean, the global maximum curvature response value, and the detail fidelity adjustment coefficient. By adjusting the detail fidelity adjustment coefficient, the dynamic curvature threshold can be flexibly adjusted, thereby achieving strict or lenient control over the identification of key detail areas and meeting the personalized needs of different scenarios.

[0069] In some embodiments, the step of filtering and determining each model region corresponding to the original high-precision 3D model based on a dynamic curvature threshold, identifying regions with curvature responses exceeding the dynamic curvature threshold as second critical detail regions that cannot be simplified, and generating a unique semantic label for each identified second critical detail region includes: comparing the normalized curvature response value of each model region with the generated dynamic curvature threshold one by one; when the normalized curvature response value is greater than the dynamic curvature threshold, marking the corresponding model region as a second critical detail region; obtaining the spatial center coordinates, regional geometric feature vector, and local geometric descriptor of each marked region; and generating a unique and reproducible semantic label for each second critical detail region by weighted encoding of the spatial center coordinates, regional geometric feature vector, and local geometric descriptor. The semantic label is used to achieve precise positioning and management of critical detail regions in lightweighting, storage, and rendering.

[0070] This embodiment corresponds to step S102, "screening the second key detail region and generating semantic labels". The core is to accurately screen the key detail region based on the generated dynamic curvature threshold, and generate a unique semantic label through weighted encoding. This solves the problem that the key region cannot be accurately located, tracked and managed in the whole process, and provides clear region identification for subsequent multi-resolution data block generation and dynamic scheduling.

[0071] The second critical detail region screening involves comparing the normalized curvature response value of each model region with the dynamic curvature threshold generated in Example 3. When the normalized curvature response value is greater than the dynamic curvature threshold, the region is identified as the second critical detail region that cannot be simplified, thus clarifying the range of details that need to be retained in the future.

[0072] The regional key information collection involves acquiring core information for each region marked as the second key detail region, including the spatial center coordinates of the region, the regional geometric feature vector, and the local geometric descriptor. This information is used to generate unique semantic labels.

[0073] Semantic tag generation and application generates unique and reproducible semantic tags by weighting and encoding the spatial center coordinates, regional geometric feature vectors, and local geometric descriptors. These tags are used to achieve precise positioning and management of the second key detail region throughout the entire process of lightweight processing, data storage, and real-time rendering, ensuring that key details are not lost and are traceable.

[0074] In some embodiments, the step of generating independent multi-resolution detail enhancement data blocks for the second critical detail region using detail hierarchy technology, and generating multiple data versions with decreasing detail levels for the same second critical detail region, includes: acquiring all region data identified as second critical detail regions; for each second critical detail region, using detail hierarchy technology to preset multiple detail resolution levels, including the highest detail level and the most simplified detail level; acquiring a preset level decay factor to control the detail compression degree of different resolution levels, ensuring that high curvature regions can still retain key morphological features in low resolution levels; based on the original high-precision geometric features of each second critical detail region, combined with the level decay factor, generating multiple data versions with decreasing detail levels for each critical detail region; during the data version generation process, using a curvature-oriented local mesh resampling algorithm, adaptively adjusting the data vertex density of each resolution level according to the region curvature gradient, so that high curvature regions still retain geometric contours in low resolution levels, while low curvature regions undergo significant downsampling processing, achieving detail balance of data blocks at each resolution level, and generating independent multi-resolution detail enhancement data blocks corresponding to each second critical detail region.

[0075] This embodiment corresponds to step S103, "generating independent multi-resolution detail enhancement data blocks," which solves the problems of high coupling between high-resolution details and lightweight base models, large memory consumption, and unbalanced detail preservation in existing technologies. Through Level of Detail (LOD) technology, combined with a curvature-oriented local mesh resampling algorithm, it generates data versions with decreasing levels of detail for each second key detail region, optimizing memory usage and rendering efficiency while preserving key morphological features.

[0076] The resolution level preset acquires data from all regions identified as second critical detail regions. For each second critical detail region, multiple detail resolution levels are preset using Level of Detail (LOD) technology, including the highest level of detail (corresponding to the original high-precision detail) and the most simplified level of detail (corresponding to the lowest memory usage). The number of levels can be adjusted according to scene requirements.

[0077] The layer attenuation factor is obtained by acquiring a preset layer attenuation factor. This factor is used to control the degree of detail compression at different resolution layers. Its core function is to ensure that high curvature areas can still retain key morphological features in low resolution layers, while flat or minor details can be significantly simplified, achieving a balance between detail preservation and resource optimization.

[0078] Multi-resolution data version generation generates multiple data versions with decreasing levels of detail for each critical detail region by combining the original high-precision geometric features of each second critical detail region with a hierarchical decay factor. During the generation process, a curvature-oriented local mesh resampling algorithm is used to adaptively adjust the data vertex density of each resolution level according to the regional curvature gradient. High curvature regions retain more vertices to maintain the geometric contour, while low curvature regions are significantly downsampled to reduce memory usage, thus achieving a balance of detail in data blocks at each resolution level.

[0079] Data blocks are generated independently by generating independent detail-enhanced data blocks for each resolution level's data version, ensuring that they can be called independently during subsequent scheduling processes without depending on data from other levels.

[0080] In some embodiments, forming a directly indexable multi-resolution detail library based on data versions includes: classifying and organizing multiple data versions with decreasing levels of detail generated for each second key detail region; associating each data version with a corresponding key detail region identifier, resolution level identifier, and detail feature information corresponding to the data version; constructing a unified indexing mechanism to map each second key detail region to all data versions corresponding to that region, clarifying the correspondence between each key detail region and data versions at different resolution levels; and integrating all data versions of the second key detail regions based on the mapping relationship to form a directly indexable multi-resolution detail library, and retrieving and calling the corresponding data version through the key detail region identifier and resolution level identifier.

[0081] This embodiment corresponds to step S103, "forming a directly indexable multi-resolution detail library based on data version". The core is to build a standardized indexing mechanism, associate and map the second key detail area with the corresponding multi-resolution data version, and integrate them to form a multi-resolution detail library that can be quickly retrieved and efficiently called. This provides data support for the real-time dynamic scheduling in step S104 and solves the problems of low efficiency in calling detail data and chaotic indexing in the prior art.

[0082] Data version classification and organization involves classifying and organizing multiple data versions with decreasing levels of detail generated for each second key detail region. Each data version is associated with corresponding key information, including key detail region identifier, resolution level identifier, and detailed feature information (such as curvature range and number of vertices) to ensure data version traceability.

[0083] The indexing mechanism is constructed by creating a unified indexing mechanism that associates and maps each second key detail area with all the data versions corresponding to that area, clarifying the one-to-one correspondence between each key detail area and the data versions at different resolution levels, thus forming an index table.

[0084] The multi-resolution detail library integration integrates data versions of all second critical detail regions based on the above-mentioned association mapping relationship, forming a directly indexable multi-resolution detail library. This database supports quick retrieval and retrieval of the corresponding data version through critical detail region identifiers and resolution level identifiers, ensuring the efficiency of data scheduling during real-time rendering.

[0085] In some embodiments, all data versions are stored independently of the lightweight base model, in the form of data blocks, and the spatial location, curvature features, and semantic tags of the corresponding key detail regions are recorded in the data block metadata. The process of forming a directly indexable multi-resolution detail library based on data versions includes: storing multiple data versions corresponding to each second key detail region separately in the form of independent data blocks, with all data blocks independent of the lightweight base model; recording the spatial location information, curvature feature information, and generated semantic tags of the corresponding key detail region in the metadata of each data block, ensuring that each data block can be traced back to the corresponding key detail region; constructing a unified indexing system based on the metadata information of all data blocks, linking the key information in the data block metadata with the data blocks themselves to form a directly indexable multi-resolution detail library; and locating and extracting corresponding data blocks at different resolution levels by retrieving the spatial location, curvature features, or semantic tags of the key detail regions, thereby achieving efficient access to the data blocks.

[0086] This embodiment also corresponds to step S103, "forming a directly indexable multi-resolution detail library," focusing on solving the problems of high coupling between detail data blocks and the base model, untraceable data, and poor call flexibility. It explicitly states that all data versions are stored independently of the lightweight base model, records key information through metadata, and constructs a metadata-based indexing system to further improve the retrieval efficiency and maintainability of the detail library.

[0087] Independent data block storage stores multiple data versions corresponding to each second key detail area as independent data blocks. All data blocks are independent of the lightweight base model and can be loaded, called and maintained independently, reducing coupling and optimizing memory usage.

[0088] Metadata recording records the core information of the corresponding key detail area in detail in the metadata of each data block, including spatial location information, curvature feature information and semantic tags generated in Example 4, to ensure that each data block can be traced back to the corresponding second key detail area, which facilitates full-process management.

[0089] The index system is constructed by building a unified index system based on the metadata information of all data blocks. This index system associates key information (semantic tags, spatial location, curvature features) in the metadata of the data blocks with the data blocks themselves, forming a multi-resolution detail library that can be directly indexed.

[0090] Efficient retrieval and retrieval can quickly locate and extract corresponding data blocks at different resolution levels by retrieving the spatial location, curvature features, or semantic tags of key detail areas, enabling efficient retrieval of data blocks and adapting to the dynamic scheduling requirements of real-time rendering.

[0091] In some embodiments, the step of calculating the distance between the viewing point and the spatial center of each key detail region during the real-time rendering stage, and dynamically scheduling detail data blocks of the corresponding resolution level in the multi-resolution detail library based on the distance and a viewing distance sensitivity coefficient, includes: during the real-time rendering process, acquiring the spatial coordinates of the current viewing point and the spatial center coordinates of each second key detail region in real time; obtaining the spatial proximity between the viewing point and each key detail region by calculating the straight-line distance between the spatial coordinates of the viewing point and the spatial center coordinates of each second key detail region; and acquiring a preset viewing distance sensitivity coefficient and a maximum viewing distance threshold, wherein the viewing distance sensitivity coefficient is used to control the sensitivity of detail level switching. The maximum viewing distance threshold is used to define the scheduling range of detail data blocks. Based on the calculated spatial distance, viewing distance sensitivity coefficient, and maximum viewing distance threshold, the detail level index corresponding to each second critical detail region is determined. When the spatial distance is less than the maximum viewing distance threshold, the closer the observation point is to the critical detail region, the higher the resolution level of the detail data block is scheduled for that region; the farther away it is, the lower the resolution level of the detail data block is scheduled. A prediction caching mechanism is introduced to obtain the movement vector and movement amplitude of the observation point in real time. Based on the movement vector and movement amplitude, the second critical detail region that may be accessed in the next frame is predicted, and the detail data blocks of the corresponding resolution level are loaded in advance to avoid loading delays.

[0092] This embodiment corresponds to step S104, "dynamically scheduling detail data blocks during real-time rendering," and solves the problems of loading delays, interactive stuttering, and unreasonable allocation of computing power in existing technologies for ultra-large-scale scenes. By calculating the spatial distance between the observation point and key detail areas, and combining it with the viewing distance sensitivity coefficient, dynamic scheduling of detail data blocks is achieved. At the same time, a predictive caching mechanism is introduced to ensure rendering continuity and visual stability, thereby optimizing the user interaction experience.

[0093] Spatial coordinate acquisition provides basic data for distance calculation by acquiring the spatial coordinates of the current viewing point and the spatial center coordinates of each second key detail area in real time during the real-time rendering process.

[0094] Spatial distance calculation is performed by calculating the straight-line distance (Euclidean distance) between the spatial coordinates of the observation viewpoint and the spatial center coordinates of each second critical detail region. This distance directly determines the scheduling strategy for data block resolution.

[0095] The scheduling parameters are set by obtaining the preset line-of-sight sensitivity coefficient and the maximum line-of-sight threshold. The line-of-sight sensitivity coefficient is used to control the sensitivity of the switching of detail levels (the larger the coefficient, the more sensitive the level switching). The maximum line-of-sight threshold is used to define the scheduling range of detail data blocks (detail data blocks outside the threshold can be not scheduled to further optimize computing power).

[0096] The detail level determination and scheduling are based on the calculated spatial distance, viewing distance sensitivity coefficient, and maximum viewing distance threshold to determine the detail level index corresponding to each second critical detail region. When the spatial distance is less than the maximum viewing distance threshold, the closer the observation point is to the critical detail region, the higher the resolution level of the detail data block is scheduled for that region, and the farther the distance is, the lower the resolution level of the detail data block is scheduled, thus realizing the on-demand allocation of computing power.

[0097] Predictive caching optimization introduces a predictive caching mechanism to obtain the movement vector and amplitude of the observation viewpoint in real time. Based on the movement vector and amplitude, it predicts the second critical detail region that may be accessed in the next frame and preloads the detail data blocks of the corresponding resolution level to avoid loading delays and interactive stutters, ensuring rendering continuity and visual stability.

[0098] In some embodiments, the seamless fusion rendering based on the detail data block and the base main model includes: performing real-time spatial alignment processing on the detail data block and the base main model to ensure that the detail data block and the base main model are accurately superimposed in three-dimensional space; then dynamically mixing the color information of the detail data block with the texture of the base main model through a texture fusion algorithm; and introducing a lighting correction mechanism to eliminate the lighting and shadow deviations between the detail data block and the base main model, thereby achieving seamless fusion rendering of the detail data block and the base main model and achieving a balance between overall scene lightweighting and high-fidelity local details.

[0099] This embodiment corresponds to step S104, "performing seamless fusion rendering based on the detailed data blocks and the basic main model," which solves the problems of visual disjointness and lighting and shadow deviation that occur after fusion of high-resolution details and lightweight basic models in existing technologies. Through a comprehensive strategy of spatial alignment, texture blending, and lighting correction, seamless fusion of the two is achieved, ensuring visual consistency and ultimately achieving a unity between overall scene lightweighting and high-fidelity local details.

[0100] Spatial alignment processing involves real-time spatial alignment of the scheduled detail data blocks with the lightweight base model; by calculating the local affine transformation matrix, precise matching of the vertex coordinates and normals of the detail data blocks is achieved, ensuring that the detail data blocks and the base model are accurately superimposed in three-dimensional space without geometric offset.

[0101] Texture blending processing employs a texture blending algorithm to dynamically mix the color information of detail data blocks with the texture of the base model. During the blending process, the detail weight coefficient is calculated by combining the regional curvature response and the viewpoint distance to ensure that details with high curvature and close proximity are displayed first, while the colors of the base model are displayed first in distant or low curvature areas, thus achieving a natural texture transition.

[0102] Lighting correction processing eliminates the lighting and shadow deviations between the detail data blocks and the base model by introducing a lighting correction mechanism. By calculating the normal difference between the detail data blocks and the base model in real time and combining it with the lighting direction vector, the lighting parameters of the detail data blocks are corrected to ensure that the two are consistent in terms of lighting, shadows and material representation, thus eliminating the sense of visual disjointedness.

[0103] Seamless fusion rendering integrates the processing results of spatial alignment, texture blending, and lighting correction within the graphics rendering pipeline to achieve seamless fusion rendering of detailed data blocks and the basic main model. Ultimately, it achieves a balance between overall scene lightweighting and high-fidelity local details, enhancing visual realism and rendering effects.

[0104] In some embodiments, to resolve the contradiction between the loss of key details and the requirement for high fidelity during model lightweighting, this solution proposes a lightweight hierarchical digital twin modeling and rendering method for large-scale scenarios, such as... Figure 2 , Figure 3 , Figure 4 , Figure 5 and Figure 6 As shown, this method significantly improves the mechanism for identifying, storing, and dynamically rendering detailed features, specifically including the following steps:

[0105] Step 1: Automated analysis of the surface geometry of the original high-precision 3D model. The core of this process is to accurately calculate and extract the curvature features of each local area of ​​the model. Curvature, as a key mathematical indicator for measuring the degree of surface curvature and the richness of detail, can effectively distinguish between flat areas and key detail areas containing complex structures, sharp edges or fine textures, laying the data foundation for subsequent differential processing.

[0106] Step 2: The analysis results are intelligently filtered and labeled according to the preset curvature threshold. This threshold can be dynamically adjusted according to the specific requirements of different industries for detail fidelity. The system automatically identifies areas with curvature exceeding the threshold as key detail areas that cannot be simplified and loses them, and attaches exclusive semantic labels to these areas for tracking and management, thereby essentially distinguishing them from non-critical areas that can be greatly simplified in the lightweight process.

[0107] Step 3: Generate independent detail enhancement data blocks with multi-resolution levels for each marked key detail area. This process uses detail level technology to generate multiple data versions with decreasing levels of detail for the same key area. Each version is independent of the lightweight base model and is stored in the form of data blocks, thus avoiding the huge load caused by loading all high-precision model data into memory while retaining the original high-precision details.

[0108] Step 4: During the real-time rendering stage, the system dynamically schedules detail data blocks of the appropriate resolution based on the spatial distance between the viewing point and the key detail area. When the viewpoint is close, the system automatically loads and calls high-resolution detail data blocks to achieve a clear display. When the viewpoint is far away, it switches to low-resolution data blocks to save computing power. This dynamic scheduling mechanism ensures that limited computing resources are always prioritized for the details that the user is most likely to focus on.

[0109] Step 5: In the graphics rendering pipeline, the dynamically scheduled high-resolution detail data blocks are spatially aligned and textured in real time with the lightweight base model. Through specific shader programs and blending techniques, the superimposed details and base model are made to maintain visual consistency in terms of lighting, shadows and material representation, thereby achieving a unity and balance between overall scene lightweighting and local detail high fidelity in the final rendered image.

[0110] The five steps described above constitute a complete technical loop from data preprocessing to real-time rendering. Each step refines the output of the previous step, ensuring the lossless transfer and efficient utilization of key information throughout the process. This workflow boasts excellent configurability and industry adaptability. Users can flexibly adapt to the detail fidelity requirements of different scenarios, from industrial manufacturing to smart cities, by adjusting parameters such as curvature threshold, resolution layer number, and view distance sensitivity. Furthermore, this method supports integration with existing mainstream 3D engines and digital twin platforms, enabling smooth embedding of detail enhancement functions without reconstructing the original rendering pipeline, significantly improving the system's practicality and promotional value.

[0111] In step 1, the system takes the triangular mesh of the original high-precision 3D model as input and performs automated parsing processing on its surface geometry, both per face and per vertex. First, by continuously calculating the rate of change of the normal in the neighborhood of each vertex, a stable and scale-consistent local curvature expression is constructed. This curvature not only reflects the degree of geometric bending but also implicitly contains information on texture density and structural abrupt changes. Its mathematical expression is as follows:

[0112] ;

[0113] in Indicates the first Discrete curvature values ​​at each vertex This represents the set of vertices directly adjacent to the given vertex. and Let represent the unit normal vectors of the corresponding vertices. This formula, by averaging the angle between normals, can stably characterize the intensity of local geometric undulations even with uneven mesh density. The result approaches zero in flat regions but significantly increases in edge areas, etched textures, and microstructure regions. To avoid single-point noise interfering with subsequent judgments, the system further introduces an area-weighted curvature aggregation mechanism at the patch level, forming a region-level geometric complexity metric, defined as:

[0114] ;

[0115] in Indicates the area The comprehensive curvature complexity index, Represents vertices The geometric area weight of the corresponding patch within the region naturally suppresses isolated perturbations with high curvature but extremely small surface areas, while ensuring that true structural details are statistically amplified, thus providing a reliable basis for cross-scale consistency analysis. After completing the curvature and complexity calculations, the system assigns weights to all regions. Mapping to a unified normalized feature space and constructing a curvature response distribution function for subsequent discrimination:

[0116] ;

[0117] in This represents the normalized detailed response intensity. and These represent the global extrema of region complexity in the current model. The purpose of this function is to eliminate the absolute scale differences between different models, enabling curvature features to directly participate in a unified threshold decision-making mechanism across scenarios and industries. Through the above continuous and traceable geometric analysis process, the system has completed the accurate quantification of model detail density without introducing redundant semantic information, providing direct and stable data input for the next step of key detail region selection and semantic labeling based on curvature thresholds.

[0118] In step 2, the system is based on the normalized curvature response obtained in step 1. Dynamic thresholding is applied to each region to identify key detail areas. Thresholds Through industry demand parameters The weighted relationship between the model and the global curvature statistics is generated, and its calculation formula is as follows:

[0119] ;

[0120] in This represents the curvature threshold in the current scenario. This represents the normalized mean curvature across all regions of the model. This represents the global maximum curvature response value. The detail fidelity adjustment factor is adjusted by... The system can achieve strict or lenient control over the identification of key areas, thereby meeting the local accuracy requirements of different application scenarios. After obtaining... Then, for each region Perform threshold determination operation:

[0121] ;

[0122] in As a key marker, when Time, region Key detail regions identified as having non-simplifiable loss characteristics directly impact the data preservation strategy for subsequent lightweight processing. To ensure that key regions are traceable across multiple models and scenarios, the system further generates semantic labels for each labeled region. Its generation mechanism is accomplished through weighted encoding of region location, curvature features, and local geometric descriptors, and its expression is:

[0123] ;

[0124] in Indicates the area Spatial center coordinates, Represents the geometric feature vector of the region. This represents a vector concatenation operation. The system generates unique and reproducible identifiers to ensure that critical details can be accurately located and managed throughout the entire process, including lightweighting, storage, and rendering. Through the aforementioned dynamic threshold filtering and semantic tagging mechanism, the system achieves essential separation between critical details and non-critical regions while maintaining the overall lightweightness of the model. This provides clear regional ranges and identification information for the generation of multi-resolution detail enhancement data blocks in the next step.

[0125] In step 3, the system targets the areas marked as key details in step 2. Generate independent multi-resolution detail-enhancing data blocks for each region. Corresponding to a set of detail sequence ,in Indicates the maximum resolution level. Corresponding to the highest level of detail, Corresponding to the most simplified version, data block The generation depends on the local curvature response. Attenuation coefficient with preset resolution Its formula is:

[0126] ;

[0127] in Indicates the area The original high-precision geometric feature matrix, The attenuation factor controls the degree of detail compression at different resolution levels. This formula ensures that high-curvature regions retain key morphological features at lower resolution levels, while flat or minor details are significantly simplified, thereby optimizing memory usage and rendering efficiency. During data block generation, the system employs a curvature-guided local mesh resampling algorithm, defined as:

[0128] ;

[0129] in Indicates the area At the level The vertex coordinate matrix, The normal gradient matrix of the vertex. This represents element-wise multiplication. As the sampling step size control coefficient, this algorithm adaptively adjusts the vertex density based on the curvature gradient, ensuring that high-curvature regions retain their geometric contours in low-resolution layers, while low-curvature regions are significantly downsampled, thus achieving detail balance across multi-level data blocks. After generation, all... It is stored as an independent data block, and the corresponding spatial location, curvature features and semantic tags are recorded in the metadata. This creates a directly indexable multi-resolution detail library. Through the above processing, the system can access each resolution level on demand without loading the complete high-poly model, effectively decoupling high-precision detail preservation from the lightweight base model, and providing accurate and efficient data support for dynamic scheduling based on the observation viewpoint in the next step.

[0130] In step 4, during the real-time rendering phase, the system renders the data based on the viewpoint position. With each key detail area Space Center Calculate sight distance It is used for dynamically scheduling detailed data blocks at the corresponding level, and its distance calculation formula is:

[0131] ;

[0132] in Describing the Euclidean norm, Measuring the current viewpoint and region The spatial proximity of the data blocks directly determines the resolution of the data blocks that the system should load during rendering. Based on this, a detail-level scheduling function is defined. :

[0133] ;

[0134] in Indicates the area In distance The detailed level index to be invoked below, The line-of-sight sensitivity coefficient is used to control the sensitivity of layer switching. The maximum line-of-sight threshold set for the system. The maximum resolution level defined in step three is used, with the dot indicating a rounding down operation. This function ensures that nearby key areas are automatically mapped to high-resolution data blocks, while areas far from the viewpoint use lower-resolution versions, concentrating computing power on the user's current visual focus. During dynamic scheduling, the system introduces a prediction caching mechanism, using the viewpoint movement vector... The calculation expression for preloading data blocks for regions that may be accessed in the next frame is as follows:

[0135] ;

[0136] in Indicates the predicted data block to be loaded. The sensitivity factor is preloaded to balance data loading lead time with memory usage. This indicates the magnitude of viewpoint movement. Through this mechanism, the system can maintain detail continuity and visual stability under rapid viewpoint changes, while ensuring that memory and GPU computing resources are always prioritized for the most likely high-detail areas to be observed. Through the aforementioned dynamic resolution scheduling and predictive loading strategy based on spatial distance, the system achieves on-demand access to key detail data blocks, providing accurate and controllable data input for the real-time fusion of high-resolution details and the base model in the next step.

[0137] In step 5, when drawing the pipeline, the system first uses the detail data blocks obtained in step four. With lightweight basic model Spatial alignment is performed using a local affine transformation matrix. The formula for achieving an exact match between vertex coordinates and normals is as follows:

[0138] ;

[0139] in Represents data block The original vertex coordinate matrix in the middle, This represents the aligned vertex coordinate matrix. By region Space Center The system generates corresponding position vectors from the base model through a process that ensures precise superposition of detail blocks with the base model in 3D space without geometric offset. Subsequently, the system employs a texture fusion algorithm based on illumination consistency to integrate the color information of the detail blocks. With base model texture Dynamic mixing is performed, and the fusion formula is as follows:

[0140] ;

[0141] in The vertex or pixel color values ​​in the final rendered output. The detail weighting coefficients are derived from the region curvature response. Distance from viewpoint Joint computation ensures that details with high curvature and close proximity are displayed first, while the base model colors are displayed first in distant or low-curvature areas, thus maintaining visual consistency in overall lighting, shadows, and material representation. At the lighting interaction level, the system introduces a real-time differential lighting correction term. To eliminate shadow or reflection discrepancies between the detailed data block and the base model, the expression is:

[0142] ;

[0143] in This represents the normal of the aligned detail data block. Represents the normal of the base model. Represents the direction vector of illumination. For lighting calculation functions, The details are incorporated into the blend shader for correction, ensuring a high degree of consistency between the details and the base model in terms of lighting response. Through the aforementioned spatial alignment, texture blending, and lighting correction processes, the system achieves seamless overlay of dynamically scheduled high-resolution details and lightweight base models in the rendered image. This allows the overall scene to maintain its lightweight advantage while presenting high-fidelity effects for key local details, providing end users with a visual experience that combines realism and interactive efficiency.

[0144] Please see Figure 7 As shown, Figure 7 This is a schematic diagram of the structure of a lightweight hierarchical digital twin modeling and rendering system 200 for large-scale scenarios provided in this application embodiment. This lightweight hierarchical digital twin modeling and rendering system 200 for large-scale scenarios is used to execute the steps of the lightweight hierarchical digital twin modeling and rendering method for large-scale scenarios shown in the above embodiments. The lightweight hierarchical digital twin modeling and rendering system 200 for large-scale scenarios can be a single server or a server cluster, or it can be a terminal, such as a handheld terminal, laptop computer, wearable device, or robot.

[0145] like Figure 7 As shown, the lightweight hierarchical digital twin modeling and rendering system 200 for large-scale scenarios includes:

[0146] The geometric analysis unit 201 is used to analyze the surface geometry of the original high-precision 3D model, calculate and extract the curvature features of each local region of the high-precision 3D model, distinguish the flat regions and the first key detail regions of the original high-precision 3D model through curvature features, and construct the curvature response distribution for subsequent region selection.

[0147] The threshold generation unit 202 is used to generate a dynamic curvature threshold based on the obtained curvature response distribution, combined with the global curvature statistics and detail fidelity of the original high-precision 3D model. The dynamic curvature threshold is used to filter and judge each model region corresponding to the original high-precision 3D model. Regions whose curvature response exceeds the dynamic curvature threshold are identified as second key detail regions that cannot be simplified. At the same time, a unique semantic label is generated for each identified second key detail region.

[0148] The detail forming unit 203 is used to generate independent multi-resolution detail enhancement data blocks for the second key detail region using detail level technology, generate multiple data versions with decreasing detail levels for the same second key detail region, and form a directly indexable multi-resolution detail library based on the data versions.

[0149] The rendering completion unit 204 is used to calculate the distance between the viewing point and the spatial center of each key detail area during the real-time rendering stage. Based on the distance and the viewing distance sensitivity coefficient, it dynamically schedules the detail data blocks of the corresponding resolution level in the multi-resolution detail library. In the graphics rendering pipeline, it completes seamless fusion rendering based on the detail data blocks and the basic main model, achieving a balance between overall scene lightweighting and high-fidelity local details.

[0150] In some embodiments, the first critical detail region is a region containing complex structures, sharp edges, or intricate textures. The curvature features are used to quantify the surface curvature and detail richness of local regions of the model, and the curvature features of all first critical detail regions are normalized. The analysis of the surface geometry of the original high-precision 3D model, calculating and extracting the curvature features of each local region corresponding to the high-precision 3D model, includes: acquiring the triangular mesh data of the original high-precision 3D model, performing a sequential traversal analysis of each vertex and each facet in the triangular mesh data, and obtaining the set of directly adjacent vertices of each vertex. The unit normal vector corresponding to each vertex is used to calculate the discrete curvature value of each vertex by averaging the angle between the unit normal vectors in the neighborhood of each vertex. To suppress the interference of single-point noise on the subsequent judgment results, an area-weighted curvature aggregation mechanism is introduced at the facet level. Based on the geometric area weight of the facet to which each vertex belongs in the corresponding region, the discrete curvature values ​​of the vertex are aggregated and calculated to obtain the region-level geometric complexity measure. The geometric complexity measures of all regions are mapped to a unified normalized feature space to complete the normalization processing of the curvature features of all first key detail regions and eliminate the absolute scale difference between different models.

[0151] In some embodiments, constructing the curvature response distribution for subsequent region selection includes: after completing the normalization processing of the curvature features of all regions, calculating the global maximum and global minimum values ​​of the geometric complexity measure of all regions in the current original high-precision 3D model; based on the normalized curvature features, global maximum and global minimum values ​​of each region, constructing the distribution relationship of the detail response intensity to form the curvature response distribution for subsequent region selection.

[0152] In some embodiments, the step of generating a dynamic curvature threshold based on the obtained curvature response distribution, combined with the global curvature statistics corresponding to the original high-precision 3D model and the detail fidelity adjustment, includes: extracting and calculating the normalized mean curvature and the global maximum curvature response value of all regions of the original high-precision 3D model from the constructed curvature response distribution to obtain the global curvature statistics of the model; obtaining a preset detail fidelity adjustment coefficient, which is used to manually set or automatically adapt to the specific needs of different industries for detail fidelity to control the strictness of the key detail region identification standard; generating a dynamic curvature threshold for the current scene by weighting the normalized mean curvature, the global maximum curvature response value, and the detail fidelity adjustment coefficient; and adjusting the level of the dynamic curvature threshold by adjusting the detail fidelity adjustment coefficient to achieve strict or lenient control over the identification of key detail regions.

[0153] In some embodiments, the step of filtering and determining each model region corresponding to the original high-precision 3D model based on a dynamic curvature threshold, identifying regions with curvature responses exceeding the dynamic curvature threshold as second critical detail regions that cannot be simplified, and generating a unique semantic label for each identified second critical detail region includes: comparing the normalized curvature response value of each model region with the generated dynamic curvature threshold one by one; when the normalized curvature response value is greater than the dynamic curvature threshold, marking the corresponding model region as a second critical detail region; obtaining the spatial center coordinates, regional geometric feature vector, and local geometric descriptor of each marked region; and generating a unique and reproducible semantic label for each second critical detail region by weighted encoding of the spatial center coordinates, regional geometric feature vector, and local geometric descriptor. The semantic label is used to achieve precise positioning and management of critical detail regions in lightweighting, storage, and rendering.

[0154] In some embodiments, the step of generating independent multi-resolution detail enhancement data blocks for the second critical detail region using detail hierarchy technology, and generating multiple data versions with decreasing detail levels for the same second critical detail region, includes: acquiring all region data identified as second critical detail regions; for each second critical detail region, using detail hierarchy technology to preset multiple detail resolution levels, including the highest detail level and the most simplified detail level; acquiring a preset level decay factor to control the detail compression degree of different resolution levels, ensuring that high curvature regions can still retain key morphological features in low resolution levels; based on the original high-precision geometric features of each second critical detail region, combined with the level decay factor, generating multiple data versions with decreasing detail levels for each critical detail region; during the data version generation process, using a curvature-oriented local mesh resampling algorithm, adaptively adjusting the data vertex density of each resolution level according to the region curvature gradient, so that high curvature regions still retain geometric contours in low resolution levels, while low curvature regions undergo significant downsampling processing, achieving detail balance of data blocks at each resolution level, and generating independent multi-resolution detail enhancement data blocks corresponding to each second critical detail region.

[0155] In some embodiments, forming a directly indexable multi-resolution detail library based on data versions includes: classifying and organizing multiple data versions with decreasing levels of detail generated for each second key detail region; associating each data version with a corresponding key detail region identifier, resolution level identifier, and detail feature information corresponding to the data version; constructing a unified indexing mechanism to map each second key detail region to all data versions corresponding to that region, clarifying the correspondence between each key detail region and data versions at different resolution levels; and integrating all data versions of the second key detail regions based on the mapping relationship to form a directly indexable multi-resolution detail library, and retrieving and calling the corresponding data version through the key detail region identifier and resolution level identifier.

[0156] In some embodiments, all data versions are stored independently of the lightweight base model, in the form of data blocks, and the spatial location, curvature features, and semantic tags of the corresponding key detail regions are recorded in the data block metadata. The process of forming a directly indexable multi-resolution detail library based on data versions includes: storing multiple data versions corresponding to each second key detail region separately in the form of independent data blocks, with all data blocks independent of the lightweight base model; recording the spatial location information, curvature feature information, and generated semantic tags of the corresponding key detail region in the metadata of each data block, ensuring that each data block can be traced back to the corresponding key detail region; constructing a unified indexing system based on the metadata information of all data blocks, linking the key information in the data block metadata with the data blocks themselves to form a directly indexable multi-resolution detail library; and locating and extracting corresponding data blocks at different resolution levels by retrieving the spatial location, curvature features, or semantic tags of the key detail regions, thereby achieving efficient access to the data blocks.

[0157] In some embodiments, the step of calculating the distance between the viewing point and the spatial center of each key detail region during the real-time rendering stage, and dynamically scheduling detail data blocks of the corresponding resolution level in the multi-resolution detail library based on the distance and a viewing distance sensitivity coefficient, includes: during the real-time rendering process, acquiring the spatial coordinates of the current viewing point and the spatial center coordinates of each second key detail region in real time; obtaining the spatial proximity between the viewing point and each key detail region by calculating the straight-line distance between the spatial coordinates of the viewing point and the spatial center coordinates of each second key detail region; and acquiring a preset viewing distance sensitivity coefficient and a maximum viewing distance threshold, wherein the viewing distance sensitivity coefficient is used to control the sensitivity of detail level switching. The maximum viewing distance threshold is used to define the scheduling range of detail data blocks. Based on the calculated spatial distance, viewing distance sensitivity coefficient, and maximum viewing distance threshold, the detail level index corresponding to each second critical detail region is determined. When the spatial distance is less than the maximum viewing distance threshold, the closer the observation point is to the critical detail region, the higher the resolution level of the detail data block is scheduled for that region; the farther away it is, the lower the resolution level of the detail data block is scheduled. A prediction caching mechanism is introduced to obtain the movement vector and movement amplitude of the observation point in real time. Based on the movement vector and movement amplitude, the second critical detail region that may be accessed in the next frame is predicted, and the detail data blocks of the corresponding resolution level are loaded in advance to avoid loading delays.

[0158] In some embodiments, the seamless fusion rendering based on the detail data block and the base main model includes: performing real-time spatial alignment processing on the detail data block and the base main model to ensure that the detail data block and the base main model are accurately superimposed in three-dimensional space; then dynamically mixing the color information of the detail data block with the texture of the base main model through a texture fusion algorithm; and introducing a lighting correction mechanism to eliminate the lighting and shadow deviations between the detail data block and the base main model, thereby achieving seamless fusion rendering of the detail data block and the base main model and achieving a balance between overall scene lightweighting and high-fidelity local details.

[0159] It should be noted that, for the sake of convenience and brevity, the specific working processes of the lightweight hierarchical digital twin modeling and rendering system for large-scale scenarios and its modules described above can be found in the corresponding embodiments of the lightweight hierarchical digital twin modeling and rendering method for large-scale scenarios, and will not be repeated here.

[0160] The aforementioned lightweight hierarchical digital twin modeling and rendering method for large-scale scenarios can be implemented as a computer program, which can be used in scenarios such as... Figure 7 It runs on the device shown.

[0161] Please see Figure 8 , Figure 8 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application. The computer device includes a processor, a memory, and a network interface connected via a device bus, wherein the memory may include a storage medium and internal memory.

[0162] The storage medium can store operating devices and computer programs. The computer program includes program instructions that, when executed, cause the processor to perform any lightweight, hierarchical digital twin modeling and rendering method for large-scale scenes.

[0163] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0164] Internal memory provides an environment for the execution of computer programs in non-volatile storage media. When these computer programs are executed by the processor, the processor can execute any lightweight hierarchical digital twin modeling and rendering method for large-scale scenes.

[0165] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 8The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the terminal to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0166] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0167] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps:

[0168] The surface geometry of the original high-precision 3D model is analyzed, and the curvature features of each local region of the high-precision 3D model are calculated and extracted. The curvature features are used to distinguish the flat regions and the first critical detail regions of the original high-precision 3D model, and a curvature response distribution is constructed for subsequent region selection.

[0169] Based on the obtained curvature response distribution, a dynamic curvature threshold is generated by combining the global curvature statistics and detail fidelity adjustment of the original high-precision 3D model. Each model region corresponding to the original high-precision 3D model is screened and judged according to the dynamic curvature threshold. Regions with curvature response exceeding the dynamic curvature threshold are identified as second critical detail regions that cannot be simplified. At the same time, a unique semantic label is generated for each identified second critical detail region.

[0170] For the second critical detail region, a detail level technique is used to generate independent multi-resolution detail enhancement data blocks. Multiple data versions with decreasing detail levels are generated for the same second critical detail region. A multi-resolution detail library that can be directly indexed is formed based on the data versions.

[0171] During the real-time rendering phase, the distance between the viewing point and the spatial center of each key detail area is calculated. Based on the distance and the viewing distance sensitivity coefficient, detail data blocks of the corresponding resolution level in the multi-resolution detail library are dynamically scheduled. In the graphics rendering pipeline, the detail data blocks and the basic main model are seamlessly integrated for rendering, achieving a balance between lightweight overall scene and high-fidelity local details.

[0172] In some embodiments, the first critical detail region is a region containing complex structures, sharp edges, or intricate textures. The curvature features are used to quantify the surface curvature and detail richness of local regions of the model, and the curvature features of all first critical detail regions are normalized. The analysis of the surface geometry of the original high-precision 3D model, calculating and extracting the curvature features of each local region corresponding to the high-precision 3D model, includes: acquiring the triangular mesh data of the original high-precision 3D model, performing a sequential traversal analysis of each vertex and each facet in the triangular mesh data, and obtaining the set of directly adjacent vertices of each vertex. The unit normal vector corresponding to each vertex is used to calculate the discrete curvature value of each vertex by averaging the angle between the unit normal vectors in the neighborhood of each vertex. To suppress the interference of single-point noise on the subsequent judgment results, an area-weighted curvature aggregation mechanism is introduced at the facet level. Based on the geometric area weight of the facet to which each vertex belongs in the corresponding region, the discrete curvature values ​​of the vertex are aggregated and calculated to obtain the region-level geometric complexity measure. The geometric complexity measures of all regions are mapped to a unified normalized feature space to complete the normalization processing of the curvature features of all first key detail regions and eliminate the absolute scale difference between different models.

[0173] In some embodiments, constructing the curvature response distribution for subsequent region selection includes: after completing the normalization processing of the curvature features of all regions, calculating the global maximum and global minimum values ​​of the geometric complexity measure of all regions in the current original high-precision 3D model; based on the normalized curvature features, global maximum and global minimum values ​​of each region, constructing the distribution relationship of the detail response intensity to form the curvature response distribution for subsequent region selection.

[0174] In some embodiments, the step of generating a dynamic curvature threshold based on the obtained curvature response distribution, combined with the global curvature statistics corresponding to the original high-precision 3D model and the detail fidelity adjustment, includes: extracting and calculating the normalized mean curvature and the global maximum curvature response value of all regions of the original high-precision 3D model from the constructed curvature response distribution to obtain the global curvature statistics of the model; obtaining a preset detail fidelity adjustment coefficient, which is used to manually set or automatically adapt to the specific needs of different industries for detail fidelity to control the strictness of the key detail region identification standard; generating a dynamic curvature threshold for the current scene by weighting the normalized mean curvature, the global maximum curvature response value, and the detail fidelity adjustment coefficient; and adjusting the level of the dynamic curvature threshold by adjusting the detail fidelity adjustment coefficient to achieve strict or lenient control over the identification of key detail regions.

[0175] In some embodiments, the step of filtering and determining each model region corresponding to the original high-precision 3D model based on a dynamic curvature threshold, identifying regions with curvature responses exceeding the dynamic curvature threshold as second critical detail regions that cannot be simplified, and generating a unique semantic label for each identified second critical detail region includes: comparing the normalized curvature response value of each model region with the generated dynamic curvature threshold one by one; when the normalized curvature response value is greater than the dynamic curvature threshold, marking the corresponding model region as a second critical detail region; obtaining the spatial center coordinates, regional geometric feature vector, and local geometric descriptor of each marked region; and generating a unique and reproducible semantic label for each second critical detail region by weighted encoding of the spatial center coordinates, regional geometric feature vector, and local geometric descriptor. The semantic label is used to achieve precise positioning and management of critical detail regions in lightweighting, storage, and rendering.

[0176] In some embodiments, the step of generating independent multi-resolution detail enhancement data blocks for the second critical detail region using detail hierarchy technology, and generating multiple data versions with decreasing detail levels for the same second critical detail region, includes: acquiring all region data identified as second critical detail regions; for each second critical detail region, using detail hierarchy technology to preset multiple detail resolution levels, including the highest detail level and the most simplified detail level; acquiring a preset level decay factor to control the detail compression degree of different resolution levels, ensuring that high curvature regions can still retain key morphological features in low resolution levels; based on the original high-precision geometric features of each second critical detail region, combined with the level decay factor, generating multiple data versions with decreasing detail levels for each critical detail region; during the data version generation process, using a curvature-oriented local mesh resampling algorithm, adaptively adjusting the data vertex density of each resolution level according to the region curvature gradient, so that high curvature regions still retain geometric contours in low resolution levels, while low curvature regions undergo significant downsampling processing, achieving detail balance of data blocks at each resolution level, and generating independent multi-resolution detail enhancement data blocks corresponding to each second critical detail region.

[0177] In some embodiments, forming a directly indexable multi-resolution detail library based on data versions includes: classifying and organizing multiple data versions with decreasing levels of detail generated for each second key detail region; associating each data version with a corresponding key detail region identifier, resolution level identifier, and detail feature information corresponding to the data version; constructing a unified indexing mechanism to map each second key detail region to all data versions corresponding to that region, clarifying the correspondence between each key detail region and data versions at different resolution levels; and integrating all data versions of the second key detail regions based on the mapping relationship to form a directly indexable multi-resolution detail library, and retrieving and calling the corresponding data version through the key detail region identifier and resolution level identifier.

[0178] In some embodiments, all data versions are stored independently of the lightweight base model, in the form of data blocks, and the spatial location, curvature features, and semantic tags of the corresponding key detail regions are recorded in the data block metadata. The process of forming a directly indexable multi-resolution detail library based on data versions includes: storing multiple data versions corresponding to each second key detail region separately in the form of independent data blocks, with all data blocks independent of the lightweight base model; recording the spatial location information, curvature feature information, and generated semantic tags of the corresponding key detail region in the metadata of each data block, ensuring that each data block can be traced back to the corresponding key detail region; constructing a unified indexing system based on the metadata information of all data blocks, linking the key information in the data block metadata with the data blocks themselves to form a directly indexable multi-resolution detail library; and locating and extracting corresponding data blocks at different resolution levels by retrieving the spatial location, curvature features, or semantic tags of the key detail regions, thereby achieving efficient access to the data blocks.

[0179] In some embodiments, the step of calculating the distance between the viewing point and the spatial center of each key detail region during the real-time rendering stage, and dynamically scheduling detail data blocks of the corresponding resolution level in the multi-resolution detail library based on the distance and a viewing distance sensitivity coefficient, includes: during the real-time rendering process, acquiring the spatial coordinates of the current viewing point and the spatial center coordinates of each second key detail region in real time; obtaining the spatial proximity between the viewing point and each key detail region by calculating the straight-line distance between the spatial coordinates of the viewing point and the spatial center coordinates of each second key detail region; and acquiring a preset viewing distance sensitivity coefficient and a maximum viewing distance threshold, wherein the viewing distance sensitivity coefficient is used to control the sensitivity of detail level switching. The maximum viewing distance threshold is used to define the scheduling range of detail data blocks. Based on the calculated spatial distance, viewing distance sensitivity coefficient, and maximum viewing distance threshold, the detail level index corresponding to each second critical detail region is determined. When the spatial distance is less than the maximum viewing distance threshold, the closer the observation point is to the critical detail region, the higher the resolution level of the detail data block is scheduled for that region; the farther away it is, the lower the resolution level of the detail data block is scheduled. A prediction caching mechanism is introduced to obtain the movement vector and movement amplitude of the observation point in real time. Based on the movement vector and movement amplitude, the second critical detail region that may be accessed in the next frame is predicted, and the detail data blocks of the corresponding resolution level are loaded in advance to avoid loading delays.

[0180] In some embodiments, the seamless fusion rendering based on the detail data block and the base main model includes: performing real-time spatial alignment processing on the detail data block and the base main model to ensure that the detail data block and the base main model are accurately superimposed in three-dimensional space; then dynamically mixing the color information of the detail data block with the texture of the base main model through a texture fusion algorithm; and introducing a lighting correction mechanism to eliminate the lighting and shadow deviations between the detail data block and the base main model, thereby achieving seamless fusion rendering of the detail data block and the base main model and achieving a balance between overall scene lightweighting and high-fidelity local details.

[0181] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the steps of the lightweight hierarchical digital twin modeling and rendering method for large-scale scenes provided in any embodiment of this application.

[0182] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0183] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A lightweight hierarchical digital twin modeling and rendering method for large-scale scenarios, characterized in that, include: The surface geometry of the original high-precision 3D model is analyzed, and the curvature features of each local region of the high-precision 3D model are calculated and extracted. The curvature features are used to distinguish the flat regions and the first critical detail regions of the original high-precision 3D model, and a curvature response distribution is constructed for subsequent region selection. Based on the obtained curvature response distribution, dynamic curvature thresholds are generated by combining the global curvature statistics corresponding to the original high-precision 3D model with the detail fidelity adjustment. This includes: extracting and calculating the normalized mean curvature and the global maximum curvature response value of all regions of the original high-precision 3D model from the constructed curvature response distribution to obtain the global curvature statistics of the model. The system obtains a preset detail fidelity adjustment coefficient, which is used to control the strictness of the key detail region recognition standard by manually setting or automatically adapting it according to the specific needs of different industries for detail fidelity. A dynamic curvature threshold for the current scene is generated by weighting the normalized mean curvature, the global maximum curvature response value, and the detail fidelity adjustment coefficient. The dynamic curvature threshold is adjusted by changing the detail fidelity adjustment coefficient, thus achieving strict or lenient control over the recognition of key detail regions. Based on the dynamic curvature threshold, each model region corresponding to the original high-precision 3D model is screened and judged. Regions with curvature responses exceeding the dynamic curvature threshold are identified as second key detail regions that cannot be simplified. A unique semantic label is generated for each identified second key detail region. For the second critical detail region, a detail level technique is used to generate independent multi-resolution detail enhancement data blocks. Multiple data versions with decreasing detail levels are generated for the same second critical detail region. A multi-resolution detail library that can be directly indexed is formed based on the data versions. During the real-time rendering phase, the distance between the viewing point and the spatial center of each key detail area is calculated. Based on the distance and the viewing distance sensitivity coefficient, detail data blocks of the corresponding resolution level in the multi-resolution detail library are dynamically scheduled. In the graphics rendering pipeline, the detail data blocks and the basic main model are seamlessly integrated for rendering, achieving a balance between lightweight overall scene and high-fidelity local details.

2. The method according to claim 1, characterized in that, The first critical detail region is a region containing complex structures, sharp edges, or intricate textures. The curvature features are used to quantify the surface curvature and detail richness of local regions of the model. The curvature features of all first critical detail regions are normalized. The analysis of the surface geometry of the original high-precision 3D model, calculating and extracting the curvature features of each local region corresponding to the high-precision 3D model, includes: Obtain the triangular mesh data of the original high-precision 3D model, and traverse and analyze each vertex and each face in the triangular mesh data one by one to obtain the set of direct adjacent vertices of each vertex and the unit normal vector corresponding to each vertex. The discrete curvature value of each vertex is obtained by averaging the angle between the unit normal vectors in the neighborhood of each vertex. To suppress the interference of single-point noise on subsequent judgment results, an area-weighted curvature aggregation mechanism is introduced at the face layer. Based on the geometric area weight of the face to which each vertex belongs in the corresponding region, the discrete curvature values ​​of the vertex are aggregated and calculated to obtain the region-level geometric complexity measure. The geometric complexity measures of all regions are mapped to a unified normalized feature space, and the curvature features of all first critical detail regions are normalized to eliminate the absolute scale differences between different models.

3. The method according to claim 2, characterized in that, The construction of the curvature response distribution for subsequent region selection includes: After normalizing the curvature features of all regions, the global maximum and global minimum values ​​of the geometric complexity measure of all regions in the current original high-precision 3D model are calculated. Based on the normalized curvature features, global maximum and global minimum values ​​of each region, the distribution relationship of detailed response intensity is constructed to form a curvature response distribution for subsequent region selection.

4. The method according to claim 1, characterized in that, The process involves filtering and determining each model region corresponding to the original high-precision 3D model based on a dynamic curvature threshold. Regions with curvature responses exceeding the dynamic curvature threshold are identified as second critical detail regions that cannot be simplified. Simultaneously, a unique semantic label is generated for each identified second critical detail region, including: The normalized curvature response value of each model region is compared with the generated dynamic curvature threshold one by one. When the normalized curvature response value is greater than the dynamic curvature threshold, the corresponding model region is marked as the second critical detail region. Obtain the spatial center coordinates, regional geometric feature vector, and local geometric descriptor for each marked region; By weighted encoding of the spatial center coordinates, regional geometric feature vectors, and local geometric descriptors, a unique and reproducible semantic label is generated for each second critical detail region. The semantic label is used to achieve precise positioning and management of critical detail regions in lightweighting, storage, and rendering.

5. The method according to claim 1, characterized in that, The process of generating independent multi-resolution detail-enhanced data blocks for the second critical detail region using detail-level techniques, and generating multiple data versions with decreasing levels of detail for the same second critical detail region, includes: Acquire all region data identified as second critical detail regions, and for each second critical detail region, use detail level technology to preset multiple detail resolution levels, including the highest detail level and the most simplified detail level; Obtain a preset layer attenuation factor to control the degree of detail compression at different resolution layers, ensuring that high curvature areas can still retain key morphological features in low resolution layers; Based on the original high-precision geometric features of each second critical detail region, and combined with the hierarchical decay factor, multiple data versions with decreasing levels of detail are generated for each critical detail region. During the data version generation process, a curvature-oriented local mesh resampling algorithm is adopted to adaptively adjust the data vertex density of each resolution level according to the regional curvature gradient. This ensures that high curvature regions retain their geometric contours in low resolution levels, while low curvature regions undergo significant downsampling. This achieves detail balance in data blocks at each resolution level and generates independent multi-resolution detail enhancement data blocks corresponding to each second critical detail region.

6. The method according to claim 5, characterized in that, The process of creating a directly indexable multi-resolution detail library based on data versions includes: For each second key detail region, multiple data versions with decreasing levels of detail are categorized and organized, and each data version is associated with the corresponding key detail region identifier, resolution level identifier, and detail feature information corresponding to the data version. A unified indexing mechanism is constructed to associate and map each second key detail area with all data versions corresponding to that area, clarifying the correspondence between each key detail area and data versions at different resolution levels; Based on the association mapping relationship, the data versions of all second key detail areas are integrated to form a multi-resolution detail library that can be directly indexed. The corresponding data version can be retrieved and called through the key detail area identifier and resolution level identifier.

7. The method according to claim 1, characterized in that, All data versions are independent of the lightweight base model, stored in the form of data blocks, and the spatial location, curvature features, and semantic tags of the corresponding key detail regions are recorded in the data block metadata. The formation of a directly indexable multi-resolution detail library based on the data versions includes: Each of the multiple data versions corresponding to the second key detail area is stored separately as an independent data block, and all data blocks do not depend on the lightweight base model. The metadata of each data block records the spatial location information, curvature feature information, and generated semantic tags of the corresponding key detail area, ensuring that each data block can be traced back to the corresponding key detail area; Based on the metadata information of all data blocks, a unified index system is constructed. The index system associates the key information in the metadata of the data blocks with the data blocks themselves, forming a multi-resolution detail library that can be directly indexed. By retrieving the spatial location, curvature features, or semantic tags of key detail regions, the system can locate and extract corresponding data blocks at different resolution levels, enabling efficient access to these data blocks.

8. The method according to claim 1, characterized in that, During the real-time rendering phase, the distance between the viewing point and the spatial center of each key detail region is calculated. Based on the distance and the viewing distance sensitivity coefficient, detail data blocks of the corresponding resolution level in the multi-resolution detail library are dynamically scheduled, including: During real-time rendering, the spatial coordinates of the current viewpoint and the spatial center coordinates of each second key detail area are obtained in real time. The spatial proximity between the observation viewpoint and each key detail area is obtained by calculating the straight-line distance between the spatial coordinates of the observation viewpoint and the spatial center coordinates of each second key detail area. Obtain the preset line-of-sight sensitivity coefficient and the maximum line-of-sight threshold, where the line-of-sight sensitivity coefficient is used to control the sensitivity of detail level switching, and the maximum line-of-sight threshold is used to define the scheduling range of detail data blocks; Based on the calculated spatial distance, viewing distance sensitivity coefficient, and maximum viewing distance threshold, the detail level index corresponding to each second key detail region is determined. When the spatial distance is less than the maximum viewing distance threshold, the closer the observation point is to the key detail region, the higher the resolution level of the detail data block in that region is scheduled, and the farther the distance is, the lower the resolution level of the detail data block is scheduled. A predictive caching mechanism is introduced to obtain the movement vector and movement amplitude of the observation point in real time. Based on the movement vector and movement amplitude, the second key detail region that may be accessed in the next frame is predicted, and the detail data blocks of the corresponding resolution level are loaded in advance to avoid loading delays.

9. The method according to claim 1, characterized in that, The seamless fusion rendering based on detailed data blocks and the basic main model includes: The detailed data blocks and the basic main model are spatially aligned in real time to ensure that they are accurately superimposed in 3D space. Then, the color information of the detailed data blocks and the texture of the basic main model are dynamically mixed through a texture fusion algorithm. A lighting correction mechanism is introduced to eliminate the lighting and shadow deviations between the detailed data blocks and the basic main model, so as to achieve seamless fusion rendering of the detailed data blocks and the basic main model and achieve the unity of lightweight overall scene and high-fidelity local details.

Images