A building digital reconstruction display method based on a high-fidelity voxel model

Abstract

This invention relates to the field of building digitization technology, and in particular to a method for building digitization reconstruction and display based on high-fidelity voxel models. The method includes the following steps: acquiring multi-source building site data; preprocessing the multi-source building site data to obtain high-fidelity voxel model data; extracting multi-level semantic information from the high-fidelity voxel model data to obtain structured semantic voxel data; constructing an interactive visualization scene based on the structured semantic voxel data to obtain three-dimensional visualization scene data; and publishing and driving the three-dimensional visualization scene data across multiple terminals to output the building digitization reconstruction and display results. This invention improves reconstruction accuracy, semantic expression capabilities, and multi-terminal display efficiency through building digitization reconstruction and display.

Description

Technical Field

[0001] This invention relates to the field of digital building technology, and in particular to a method for digital reconstruction and display of buildings based on high-fidelity voxel models. Background Technology

[0002] Existing digital reconstruction methods for buildings often rely on a single data source or low-precision models, typically generating 3D models from only 2D images or sparse point clouds. This results in deficiencies in geometric accuracy, detail representation, and spatial continuity, while also hindering component-level semantic analysis and attribute information association. In complex architectural environments, traditional methods struggle to effectively handle the registration problem of multi-source data. Data from different sources exhibits inconsistencies in coordinates, resolution differences, and noise interference, directly impacting model accuracy and integrity. Current visualizations are usually based on static 3D models, lacking interactive functions and multi-platform compatibility. Users cannot freely navigate the 3D scene, query components, or perform local sectioning observations, limiting the model's application. In scenarios such as building information management, construction monitoring, and historical building preservation, there is a lack of complete solutions that simultaneously achieve high-precision modeling, semantic representation, and multi-platform interactive display, failing to meet the needs of engineering decision-making and digital management. Summary of the Invention

[0003] Therefore, it is necessary to provide a method for digital reconstruction and display of buildings based on high-fidelity voxel models to solve at least one of the aforementioned technical problems.

[0004] To achieve the above objectives, a method for digital reconstruction and display of architecture based on high-fidelity voxel models includes the following steps:

[0005] Step S1: Acquire multi-source building site data, and preprocess the multi-source building site data to obtain high-fidelity voxel model data;

[0006] Step S2: Extract multi-level semantic information based on high-fidelity voxel model data to obtain structured semantic voxel data;

[0007] Step S3: Construct an interactive visualization scene based on structured semantic voxel data to obtain 3D visualization scene data;

[0008] Step S4: Based on the 3D visualization scene data, publish and drive the data across multiple terminals to output the results of the digital reconstruction of the building.

[0009] The beneficial effects of this invention are as follows: it achieves efficient, accurate, and interactive display of digital building reconstruction. By acquiring multi-source on-site building data and constructing high-fidelity voxel models, it can not only integrate multi-dimensional information from laser point clouds, images, and structural data, but also obtain a continuous and complete three-dimensional model while ensuring geometric accuracy. This provides an accurate data foundation for subsequent semantic analysis. Based on the high-fidelity voxel model, multi-level semantic information extraction and structured organization are performed, achieving fine identification, attribute binding, and spatial hierarchy management at the component level. This makes the building model not only have geometric form, but also complete semantic and attribute information, thereby improving the model's understandability and operability. By constructing an interactive three-dimensional visualization scene and integrating lighting, materials, and interactive logic, users can perform real-time roaming, component querying, and partial section observation, effectively enhancing the model's visualization experience and information presentation capabilities. Through multi-resolution slicing, streaming publishing, and multi-terminal adaptation, it achieves efficient loading and interactive display of the model on different platforms, improving data transmission efficiency and display smoothness, while ensuring cross-terminal consistency. Attached Figure Description

[0010] Figure 1 A schematic diagram illustrating the steps of a method for digital reconstruction and display of buildings based on high-fidelity voxel models;

[0011] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2.

[0012] Figure 3 A schematic diagram of the entire process of digital reconstruction of buildings;

[0013] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0014] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0015] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0016] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0017] To achieve the above objectives, please refer to Figures 1 to 3 A method for digital reconstruction and display of buildings based on high-fidelity voxel models includes the following steps:

[0018] Step S1: Acquire multi-source building site data, and preprocess the multi-source building site data to obtain high-fidelity voxel model data;

[0019] Step S2: Extract multi-level semantic information based on high-fidelity voxel model data to obtain structured semantic voxel data;

[0020] Step S3: Construct an interactive visualization scene based on structured semantic voxel data to obtain 3D visualization scene data;

[0021] Step S4: Based on the 3D visualization scene data, publish and drive the data across multiple terminals to output the results of the digital reconstruction of the building.

[0022] In one embodiment, multi-source data acquisition is performed on the target building. Specifically, point cloud data of the building's exterior facade and interior space can be acquired using a 3D laser scanning device, building exterior image data can be acquired using oblique photography by a drone, and local detail image data can be acquired using a handheld camera device. In some implementations, existing BIM models or CAD drawings can also be introduced as auxiliary data sources to improve the integrity of structural information.

[0023] After acquiring multi-source data, unified preprocessing is performed on the data from different sources. Specifically, this includes: denoising, outlier removal, and density equalization for point cloud data; and distortion correction, exposure equalization, and resolution unification for image data. Subsequently, cross-modal feature matching is performed based on point cloud features and image features, such as extracting geometric key points from the point cloud and feature points from the image, and establishing a correspondence through matching algorithms.

[0024] After feature matching is completed, spatial registration is performed on various types of data to unify data from different coordinate systems into the same world coordinate system. In one embodiment, a combination of initial coarse registration and fine registration can be used. First, coarse alignment is achieved through extrinsic parameter estimation, and then fine registration is performed using an iterative nearest-point algorithm to obtain a consistent data set. After registration is completed, 3D reconstruction is performed based on the fused data. Specifically, a dense point cloud is generated through multi-view stereo reconstruction, and a triangular mesh model is further generated through a surface reconstruction algorithm. For hole regions in the mesh model, they are repaired through neighborhood interpolation or surface fitting, and the surface is smoothed to obtain a continuous and complete initial 3D model.

[0025] After obtaining the initial 3D model, it is voxelized. Specifically, the voxel size is adaptively set according to the local geometric complexity of the model, with smaller voxel sizes used for structurally complex regions and larger voxel sizes used for flat regions, thereby controlling the data size while ensuring accuracy. Subsequently, the 3D space is regularly divided, and corresponding geometric information, color information, and normal vector information are sampled in each voxel unit, ultimately generating high-fidelity voxel model data.

[0026] Component-level semantic recognition is performed on high-fidelity voxel models. Specifically, geometric features (such as curvature and normal distribution), texture features, and spatial distribution features are extracted from the voxel data, and these features are input into a pre-trained component recognition model for inference processing. In one embodiment, the model can employ a deep learning model based on 3D point clouds or voxels to identify building component types such as walls, floors, doors, windows, beams, and columns, thereby obtaining component classification voxel data with category labels.

[0027] After obtaining the component classification results, attribute information is associated with each type of component. Specifically, based on the component category and its spatial range, corresponding dimensional parameters (such as length, thickness, and height), material properties (such as concrete, glass, and metal), and functional attributes (such as load-bearing or non-load-bearing components) are extracted, and the above attribute information is bound to the corresponding voxel set to form attributed voxel data.

[0028] Spatial relationship analysis and hierarchical organization are performed on attributed voxel data. Specifically, this includes analyzing the adjacency, contact, and containment relationships between components to construct the spatial topology of the components; simultaneously, based on architectural structural logic, the components are organized according to the hierarchy of "component—room—floor—overall building" to form hierarchical structured data. Finally, semantic tags, attribute information, and hierarchical relationships are uniformly integrated to obtain structured semantic voxel data.

[0029] A basic rendering scene is constructed based on structured semantic voxel data. Specifically, the voxel data is converted into a representation suitable for graphics rendering (such as a mesh or instantiated voxels), and corresponding material parameters, including color, texture mapping, and transparency, are assigned to different components. At the same time, camera parameters and viewing angles in the scene are set to form a basic 3D scene.

[0030] Building upon this foundation, interactive functionality logic is integrated. Specifically, interactive objects are created for each component within the scene, allowing users to select target components through clicking, box selection, or other methods, and display their attribute information in real time. Simultaneously, it supports component display and hiding controls, partial section browsing, and scene roaming operations, thereby enabling multi-angle observation of the building's internal structure.

[0031] After completing the interactive logic construction, the scene's visual effects are optimized. This includes: introducing a lighting model to enhance realism, such as using physically based rendering to simulate light reflection; setting shadow effects to improve spatial depth; and dynamically switching model precision at different viewing distances using Level of Detail (LOD) technology to improve rendering efficiency. The final result is complete 3D visualized scene data.

[0032] Multi-resolution slicing is performed on the 3D visualization scene data. Specifically, the overall scene is divided into multiple spatial blocks, and data versions with different precision levels are generated for each block, thereby forming published data suitable for streaming. In one embodiment, data blocks with different precision levels can be dynamically selected for loading based on viewing distance or screen resolution.

[0033] The generated data is adapted and encapsulated for multiple terminals. Specifically, the data format is converted according to different terminal types (such as browsers, mobile terminals, or immersive devices), and the rendering interface is adapted. For example, a WebGL-based rendering scheme is used for web terminals, and the model is simplified and compressed for mobile terminals to ensure smooth operation.

[0034] A visualization engine is used to load and drive the display of multi-platform adapted data. Specifically, during operation, corresponding data blocks are dynamically requested and loaded according to changes in the user's perspective, while the scene is rendered and updated in real time to achieve a smooth browsing experience. Finally, a unified digital reconstruction display of the building is output on various terminals, enabling an intuitive display of the building's structure and semantic information.

[0035] Please refer to [link / reference needed] for further information. Figure 3 First, multi-source field data is collected, and a basic model is obtained through registration, 3D reconstruction, and high-precision voxelization. Then, the semantics of the components are extracted, the associated attributes are identified, and the topology is sorted to form structured data. Next, an interactive visualization scene is built, streaming slices are made, and multi-terminal adaptation and release are performed.

[0036] Preferably, step S1 includes the following steps:

[0037] Step S11: Obtain multi-source building site data;

[0038] Step S12: Perform data registration and alignment based on multi-source building site data to obtain registered and aligned data;

[0039] Step S13: Reconstruct the three-dimensional space based on the registration and alignment data to obtain the initial three-dimensional model data;

[0040] Step S14: Perform high-precision voxelization processing based on the initial 3D model data to obtain high-fidelity voxel model data.

[0041] In one embodiment, multi-source data acquisition is performed on the target building to ensure the integrity and accuracy of subsequent modeling. Specifically, the building's exterior facade and interior space can be scanned using a ground-based 3D laser scanning device to obtain high-density point cloud data; simultaneously, a drone equipped with an oblique photography system can acquire multi-angle images of the building's exterior to obtain image data covering the roof and facade; for detailed interior areas, handheld camera devices or structured light scanning devices can be used for supplementary data acquisition.

[0042] Acquisition parameters for different data sources can be preset. For example, laser scanning resolution can be controlled within millimeter-level accuracy, and during image acquisition, a minimum overlap of 60% between adjacent images can be ensured, thereby improving the stability of subsequent feature matching. For buildings with historical data, existing BIM models or CAD drawings can be introduced as auxiliary input data to supplement information on unobservable areas.

[0043] This generates multi-source building site data, including point cloud data, image data, and auxiliary structural data.

[0044] Feature extraction is performed on different data sources. Specifically, geometric features, such as corner points, edge points, or curvature features, are extracted from point cloud data; scale-invariant feature points or texture feature points are extracted from image data, thus forming a feature set across data sources.

[0045] After obtaining the features, feature matching is performed to establish the correspondence between point clouds and images, as well as between different point cloud data. Initial matching can be performed using feature descriptor similarity, and mismatched point pairs can be eliminated through consistency constraints.

[0046] After feature matching is completed, coordinate system unification is performed. Specifically, the spatial transformation relationship between different data sources is estimated based on the matching results, including rotation matrices and translation vectors, and all data are mapped to the same coordinate system. In one embodiment, coarse registration can be performed first (e.g., based on extrinsic parameters or manual calibration), and then fine registration can be performed through iterative optimization algorithms to reduce accumulated errors.

[0047] Furthermore, the registered data undergoes fusion processing. Specifically, this includes weighted fusion of data in overlapping areas or selective retention of data with higher density and lower noise, thereby forming spatially consistent, continuous, and complete registered and aligned data.

[0048] Dense point clouds are generated based on the registered multi-source data. Specifically, multi-view geometric relationships can be used to perform stereo matching on image data to supplement point cloud density, making the original point cloud more complete in detail areas.

[0049] After obtaining the dense point cloud, surface reconstruction processing is performed. A triangular mesh generation algorithm converts the discrete point cloud into a continuous 3D surface model, resulting in triangular mesh model data. In one embodiment, mesh construction can be performed based on point cloud normal information to improve surface continuity.

[0050] After generating the triangular mesh, the model undergoes quality optimization. This includes: identifying voids in the mesh and filling them using neighborhood interpolation or surface fitting; smoothing noisy or uneven areas to eliminate local anomalies; and performing topology repair to ensure mesh connectivity. Through these processes, an initial 3D model with a complete structure and continuous surfaces is obtained.

[0051] Voxelization parameters are set based on the geometric features of the initial 3D model. Specifically, the voxel size can be adaptively determined according to the curvature variation, detail complexity, or point cloud density of different regions of the model: smaller voxel sizes are used for regions with rich detail to preserve structural details; larger voxel sizes are used for flat regions to reduce data size.

[0052] After determining the voxel dimensions, the 3D space is regularly divided to construct a 3D voxel mesh structure. Subsequently, the initial 3D model is mapped onto this voxel mesh, and attribute sampling is performed on each voxel unit. Specifically, it can be determined whether a voxel is occupied by the model, and the corresponding geometric information (such as occupancy status), color information (derived from texture or image mapping), and normal information are recorded. Further optimization processing of the voxel data is performed, such as subdividing boundary voxels to improve the accuracy of model edge representation; or filling empty regions to ensure the continuity of the voxel model. Finally, high-fidelity voxel model data with both spatial structure information and surface attribute information is obtained, providing basic data support for subsequent semantic extraction and visualization processing.

[0053] Preferably, step S12 includes the following steps:

[0054] Step S121: Perform point cloud and image feature matching based on multi-source building site data to obtain feature matching data;

[0055] Step S122: Perform a coordinate system transformation based on the feature matching data to obtain unified coordinate data;

[0056] Step S123: Perform data fusion based on the coordinate unified data to obtain registration and alignment data.

[0057] In one embodiment, point cloud and image feature matching processing is performed based on multi-source building site data. Specifically, representative geometric feature points, including edge points, corner points, and key points with significant curvature changes, are extracted from the point cloud data. Simultaneously, scale-invariant feature points or texture feature points are extracted from the image data, and corresponding feature descriptors are calculated. Subsequently, an initial matching relationship between point cloud features and image features is established through feature descriptor similarity measurement, and the matching results are filtered by combining spatial consistency constraints to eliminate mismatched point pairs, thereby obtaining stable and reliable feature matching data. After obtaining the feature matching data, coordinate system transformation processing is performed based on the matching relationship. Specifically, spatial transformation parameters between different data sources are estimated based on the matching point pairs, including rotation matrices and translation vectors, and a coordinate system transformation is constructed based on these parameters from each location. In one embodiment, the mapping relationship from the local coordinate system to the unified global coordinate system can be established by first performing a coarse pose estimation based on the initial matching results, and then finely adjusting the transformation parameters through iterative optimization to reduce alignment errors and improve overall consistency, thereby obtaining coordinate unified data. After completing coordinate unification, multi-source data fusion processing is performed based on the coordinate unified data. Specifically, when fusing point cloud data in spatially overlapping areas, weights can be set according to point density, acquisition accuracy, or noise level to perform weighted integration of multi-source data or to select and retain data with higher quality. At the same time, local smoothing and outlier removal processing are performed on the fused point cloud to eliminate boundary discontinuities and noise interference, ultimately obtaining registration and alignment data with consistent spatial position and continuous structure, providing a unified data foundation for subsequent 3D reconstruction.

[0058] Preferably, step S13 includes the following steps:

[0059] Step S131: Generate dense point cloud data based on the registration and alignment data;

[0060] Step S132: Reconstruct the surface mesh based on the dense point cloud data to obtain triangular mesh model data;

[0061] Step S133: Repair and smooth the holes in the triangular mesh model data to obtain the initial three-dimensional model data.

[0062] In one embodiment, dense point cloud generation is performed based on registration and alignment data. Specifically, using the geometric constraints between multi-view images, pixel-level matching calculations are performed on the spatially registered image data. The three-dimensional spatial coordinates corresponding to each pixel are recovered through disparity estimation, thereby supplementing the spatial point information of detailed regions on the basis of the original point cloud. At the same time, the generated result is constrained and corrected by the registered laser point cloud data to improve the spatial accuracy and density of the point cloud, ultimately obtaining dense point cloud data with uniform distribution and rich details. After obtaining the dense point cloud data, surface mesh reconstruction is performed based on the data. Specifically, the adjacency relationship between points is constructed according to the spatial distribution of the point cloud and its normal information, and generated by triangulation. A continuous triangular facet structure transforms a discrete set of points into triangular mesh model data with topological relationships. In one embodiment, the size of the triangular facets can be adaptively adjusted according to the point cloud density to balance detail representation and computational efficiency. After generating the triangular mesh model, hole repair and smoothing are performed. Specifically, for hole areas caused by missing data or occlusion, the missing parts are filled by neighborhood interpolation or surface fitting methods to keep the model surface continuous and complete. At the same time, areas with noise or irregular undulations are smoothed. Local abrupt changes are eliminated by iteratively adjusting the vertex positions, thereby improving the model surface quality and visual consistency, and finally obtaining initial three-dimensional model data with complete structure and continuous surface.

[0063] Preferably, step S14 includes the following steps:

[0064] Step S141: Set adaptive voxel size based on initial 3D model data to obtain voxelization parameter data;

[0065] Step S142: Based on the voxelization parameter data, perform spatial division and attribute sampling on the initial 3D model to obtain high-fidelity voxel model data.

[0066] In one embodiment, step S14 includes the following processing steps: First, an adaptive voxel size is set based on the initial 3D model data. Specifically, the voxel size is determined according to the geometric complexity, curvature variation, and local detail density of each region of the model. Smaller voxel sizes are used for regions with large curvature or rich details to ensure structural accuracy, while larger voxel sizes are used for flat or regular regions to reduce data volume. Simultaneously, the above size information and sampling strategy are used to form voxelized parameter data for subsequent spatial partitioning reference. Then, based on the voxelized parameter data, the initial 3D model is partitioned into 3D spaces. The entire model space is divided into equally spaced or adaptively sized voxel units according to voxel mesh rules. Attribute sampling is performed within each voxel unit, specifically including determining whether the voxel is occupied by the model, recording the occupancy status, surface color, normal vector, and texture information, and performing necessary subdivision processing on boundary voxels to preserve edge details. Simultaneously, interpolation filling is performed on empty or locally missing regions to ensure the continuity and integrity of the voxel model. Finally, high-fidelity voxel model data containing both spatial structural information and surface attribute information is formed, providing an accurate data foundation for subsequent semantic information extraction and visualization.

[0067] Preferably, step S2 includes the following steps:

[0068] Step S21: Identify building components based on high-fidelity voxel model data to obtain component classification voxel data;

[0069] Step S22: Associate component attribute information based on component classification voxel data to obtain attributed voxel data;

[0070] Step S23: Based on the attributed voxel data, perform spatial relationships and hierarchical organization to obtain structured semantic voxel data.

[0071] In one embodiment, step S2 includes the following processing steps: Building component identification based on high-fidelity voxel model data. Specifically, geometric features, texture features, and spatial distribution features of each voxel are extracted from the voxel model, and these features are input into a pre-trained component identification model for inference processing. The model can employ deep learning algorithms based on 3D convolutional networks or point cloud networks to identify different types of building components such as walls, floors, beams, columns, doors, and windows, thereby obtaining component classification voxel data with category labels. Subsequently, component attribute information is associated based on the component classification voxel data. Specifically, corresponding size parameters and material properties are extracted for each type of component. Information such as attributes and functional types is collected and bound to each voxel or component unit to form attributed voxel data. Further, spatial relationships and hierarchical organization are processed based on this attributed voxel data. Specifically, the spatial adjacency, contact, and containment relationships between components are analyzed to construct a spatial topology. Simultaneously, components are organized according to the hierarchical relationship of "component—room—floor—building as a whole" based on architectural logic. Semantic labels, attribute information, and hierarchical structure are then uniformly integrated to obtain structured semantic voxel data that contains both semantic classification and attribute and spatial hierarchical information, providing a complete data foundation for the subsequent construction of interactive visualization scenes.

[0072] Preferably, step S21 includes the following steps:

[0073] Step S211: Extract geometric and texture features based on high-fidelity voxel model data to obtain voxel feature data;

[0074] Step S212: Perform machine learning model inference based on voxel feature data to obtain component category prediction data;

[0075] Step S213: Classify and label voxels based on component category prediction data to obtain component classification voxel data.

[0076] In one embodiment, step S21 includes the following processing steps: Geometric and texture features are extracted from each voxel unit based on high-fidelity voxel model data. Specifically, local geometric features, such as curvature, normal vector distribution, edge strength, and spatial relationships within the voxel's neighborhood, are calculated from the voxel. Simultaneously, a comprehensive feature vector is formed by combining the corresponding color or texture information, thereby obtaining voxel feature data for each voxel. Subsequently, the voxel feature data is input into a pre-trained machine learning model for inference. The model can be based on a 3D convolutional network, PointNet, or a point cloud / voxel processing network, used to classify and predict voxels, outputting component category prediction data for each voxel, such as wall, floor slab, beam, column, door, and window types. Finally, the voxels are categorized and labeled based on the component category prediction data, uniformly identifying and grouping voxels of the same category, thereby obtaining component classification voxel data with category information, providing a semantic basis for subsequent component attribute association and spatial hierarchy organization.

[0077] Preferably, step S23 includes the following steps:

[0078] Step S231: Perform spatial adjacency analysis of components based on attributed voxel data to obtain spatial topological relationship data;

[0079] Step S232: Construct a hierarchical organizational structure based on spatial topological relationship data and component attributes to obtain hierarchical organizational structure data;

[0080] Step S233: Integrate structured semantic voxels based on hierarchical organizational structure data to obtain structured semantic voxel data.

[0081] In one embodiment, step S23 includes the following processing procedure: Based on attributed voxel data, perform spatial adjacency analysis of components. Specifically, by calculating the spatial distance, contact area, and relative positional relationship between each component voxel, identify the adjacency, support, and inclusion relationships between components, thereby obtaining spatial topological relationship data describing the spatial topology of each component. Subsequently, based on the spatial topological relationship data and component attribute information, construct a hierarchical organizational structure. Specifically, organize the components hierarchically according to architectural logic, for example, classifying basic components such as doors, windows, beams, and columns into room-level units. The process involves classifying room-level units into floor-level units, and then further classifying floor-level units into overall building units. Simultaneously, within each level, component attributes (such as function type, material, and load-bearing capacity) are combined to optimize the hierarchical division, thus forming a complete hierarchical organizational structure data. Finally, based on this hierarchical organizational structure data, attribute-bearing voxel data is integrated and processed, mapping semantic classification information, attribute information, and hierarchical structure uniformly into the voxel model. This achieves the fusion of semantic, attribute, and spatial hierarchical information at the voxel level, resulting in structured semantic voxel data, providing a complete data foundation for the subsequent construction of interactive visualization scenes.

[0082] Preferably, step S3 includes the following steps:

[0083] Step S31: Set up 3D scene rendering based on structured semantic voxel data to obtain basic rendering scene data;

[0084] Step S32: Integrate interactive function logic based on basic rendering scene data to obtain interactive 3D scene data;

[0085] Step S33: Optimize the lighting and effects of the interactive 3D scene data to obtain 3D visualized scene data.

[0086] In one embodiment, step S3 includes the following processing steps: Setting up 3D scene rendering based on structured semantic voxel data. Specifically, the voxel data is converted into a 3D representation suitable for rendering, including mesh representation or voxel instantiation. Material parameters, texture maps, and color information are assigned to different types of components. Simultaneously, the lighting direction, camera position, and viewing angle in the scene are set to generate basic rendering scene data. Subsequently, interactive function logic is integrated based on the basic rendering scene data. Specifically, interactive objects are created for each component, allowing users to perform interactive functions such as component selection, information display, hide / show control, partial section browsing, and scene roaming through mouse or touch operations, thereby obtaining 3D scene data with real-time interactive capabilities. Finally, the interactive 3D scene data undergoes lighting and effect optimization processing. Specifically, physically based lighting rendering models, shadow mapping, ambient occlusion, and Level of Detail (LOD) technology are applied to optimize material reflection, transparency, and anti-aliasing effects to improve rendering realism and visual coherence, ultimately obtaining complete 3D visualized scene data, providing a rendering foundation for multi-terminal publishing and display.

[0087] Preferably, step S4 includes the following steps:

[0088] Step S41: Construct multi-resolution model slices based on 3D visualization scene data to obtain streaming data;

[0089] Step S42: Perform multi-terminal adaptation and encapsulation based on streaming data to obtain multi-terminal adaptation data;

[0090] Step S43: Load and display the data based on the multi-terminal adapted data-driven visualization engine, and output the results of the digital reconstruction of the building.

[0091] In one embodiment, step S4 includes the following processing steps: First, constructing multi-resolution model slices based on 3D visualization scene data. Specifically, the overall 3D scene is divided into spatial blocks or voxel regions, and model data of different precision levels is generated for each spatial block. High-precision layers are used for close-up observation, and low-precision layers are used for long-distance display, thus forming published data suitable for streaming loading. Second, multi-terminal adaptation and encapsulation processing is performed based on the streaming published data. Specifically, according to the computing power and rendering interface of different terminal types, the data is converted into corresponding formats. For example, Web terminals use WebGL or WebGPU-parsable model formats, while mobile terminals perform model compression and simplification. Simultaneously, adapted loading scripts and resource indexes are generated, resulting in multi-terminal adapted data. Finally, the visualization engine is driven by the multi-terminal adapted data for loading and display. Specifically, during runtime on various terminals, data blocks of corresponding resolution are dynamically requested based on the user's perspective and operation, while the scene is rendered in real time and the interactive state is updated, achieving smooth roaming browsing and component information viewing. Ultimately, the results of digital reconstruction of the building are output, realizing an intuitive visualization of the building structure and semantic information on multiple terminals.

[0092] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A high-fidelity voxel model-based architectural digital reconstruction presentation method, characterized in that, Includes the following steps: Step S1: Acquire multi-source building site data, and preprocess the multi-source building site data to obtain high-fidelity voxel model data; Step S2: Extract multi-level semantic information based on high-fidelity voxel model data to obtain structured semantic voxel data; Step S3: Construct an interactive visualization scene based on structured semantic voxel data to obtain 3D visualization scene data; Step S4: Based on the 3D visualization scene data, publish and drive the data across multiple terminals to output the results of the digital reconstruction of the building.

2. The high-fidelity voxel model based architectural digitalization reconstruction presentation method according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain multi-source building site data; Step S12: Perform data registration and alignment based on multi-source building site data to obtain registered and aligned data; Step S13: Reconstruct the three-dimensional space based on the registration and alignment data to obtain the initial three-dimensional model data; Step S14: Perform high-precision voxelization processing based on the initial 3D model data to obtain high-fidelity voxel model data.

3. The high-fidelity voxel model based architectural digitalization reconstruction presentation method according to claim 2, characterized in that, Step S12 includes the following steps: Step S121: Perform point cloud and image feature matching based on multi-source building site data to obtain feature matching data; Step S122: Perform a coordinate system transformation based on the feature matching data to obtain unified coordinate data; Step S123: Perform data fusion based on the coordinate unified data to obtain registration and alignment data.

4. The high-fidelity voxel model based architectural digitalization reconstruction presentation method according to claim 2, characterized in that, Step S13 includes the following steps: Step S131: Generate dense point cloud data based on the registration and alignment data; Step S132: Reconstruct the surface mesh based on the dense point cloud data to obtain triangular mesh model data; Step S133: Repair and smooth the holes in the triangular mesh model data to obtain the initial three-dimensional model data.

5. The high-fidelity voxel model based architectural digitalization reconstruction presentation method according to claim 2, characterized in that, Step S14 includes the following steps: Step S141: Set adaptive voxel size based on initial 3D model data to obtain voxelization parameter data; Step S142: Based on the voxelization parameter data, perform spatial division and attribute sampling on the initial 3D model to obtain high-fidelity voxel model data.

6. The high-fidelity voxel model based architectural digitalization reconstruction presentation method according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Identify building components based on high-fidelity voxel model data to obtain component classification voxel data; Step S22: Associate component attribute information based on component classification voxel data to obtain attributed voxel data; Step S23: Based on the attributed voxel data, perform spatial relationships and hierarchical organization to obtain structured semantic voxel data.

7. The high-fidelity voxel model based architectural digitalization reconstruction presentation method according to claim 6, characterized in that, Step S21 includes the following steps: Step S211: Extract geometric and texture features based on high-fidelity voxel model data to obtain voxel feature data; Step S212: Perform machine learning model inference based on voxel feature data to obtain component category prediction data; Step S213: Classify and label voxels based on component category prediction data to obtain component classification voxel data.

8. The high-fidelity voxel model based architectural digitalization reconstruction presentation method according to claim 6, characterized in that, Step S23 includes the following steps: Step S231: Perform spatial adjacency analysis of components based on attributed voxel data to obtain spatial topological relationship data; Step S232: Construct a hierarchical organizational structure based on spatial topological relationship data and component attributes to obtain hierarchical organizational structure data; Step S233: Integrate structured semantic voxels based on hierarchical organizational structure data to obtain structured semantic voxel data.

9. The high-fidelity voxel model based architectural digitalization reconstruction presentation method according to claim 1, wherein, Step S3 includes the following steps: Step S31: Set up 3D scene rendering based on structured semantic voxel data to obtain basic rendering scene data; Step S32: Integrate interactive function logic based on basic rendering scene data to obtain interactive 3D scene data; Step S33: Optimize the lighting and effects of the interactive 3D scene data to obtain 3D visualized scene data.

10. The high-fidelity voxel model based architectural digitalization reconstruction presentation method according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Construct multi-resolution model slices based on 3D visualization scene data to obtain streaming data; Step S42: Perform multi-terminal adaptation and encapsulation based on streaming data to obtain multi-terminal adaptation data; Step S43: Load and display the data based on the multi-terminal adapted data-driven visualization engine, and output the results of the digital reconstruction of the building.

Images