A historical building BIM fine modeling method based on multi-source data fusion

Abstract

This application belongs to the field of Building Information Modeling (BIM) and digital preservation of cultural heritage, specifically involving a refined BIM modeling method for historical buildings based on multi-source data fusion. This method includes: collecting multi-source heterogeneous data such as 3D laser point clouds, UAV oblique photography, and survey drawings; performing spatial calibration and constructing a structured metadata database under a unified geographic coordinate and time benchmark; performing parametric BIM modeling based on building component classification and precision grading standards to realistically restore details such as the platform, brackets, and roof; simultaneously carrying out parametric reconstruction of the vegetation system and environmental integration; and finally importing the data into Unreal Engine, combining physical rendering, dynamic weather, day-night cycle, and AI interaction logic to construct a high-fidelity, highly immersive digital cultural scene. This application achieves a full-element, high-precision, and interactive digital twin of historical buildings, significantly improving the level of protection, display, and revitalization of cultural heritage.

Description

Technical Field

[0001] This application belongs to the field of building information modeling and digital protection of cultural heritage, specifically involving a refined BIM modeling method for historical buildings based on multi-source data fusion. Background Technology

[0002] With the development of digital technology, Building Information Modeling (BIM) technology, as a core means of constructing digital heritage, can transform physical building entities into high-precision digital models, providing data support for the restoration, management, revitalization and utilization of cultural relics and the construction of smart cities.

[0003] Historical buildings typically possess unique construction techniques and complex decorative components (such as brackets, beams, etc.). Their detailed modeling requires the integration of multi-source heterogeneous data, including laser point clouds, UAV oblique photography, and survey drawings, to achieve a high-fidelity digital reconstruction of the building itself and its surrounding environment.

[0004] However, existing modeling methods still have significant shortcomings when dealing with such objects: on the one hand, traditional 3D modeling is difficult to effectively identify and express the geometric and semantic features of non-standard components, resulting in distortion of detailed structures; on the other hand, there is a lack of a unified spatial benchmark and attribute association mechanism in the process of multi-source data fusion, making it difficult to coordinate point clouds, images and structured information; in addition, existing methods are mostly based on regular geometric assumptions, which makes it difficult to accurately reflect the common damage, deformation and irregular forms of historical buildings, and it is difficult to meet the requirements of model accuracy for their full life cycle management. Summary of the Invention

[0005] The purpose of this invention is to provide a method for refined BIM modeling of historical buildings based on multi-source data fusion, which can effectively solve the problems mentioned in the background technology.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for refined BIM modeling of historical buildings based on multi-source data fusion includes the following specific steps:

[0008] Step 1: Collect multi-source heterogeneous basic data of historical buildings: Use 3D laser scanning equipment to obtain point cloud data with millimeter-level precision, and simultaneously use drones equipped with industrial cameras to conduct oblique photography to obtain macro-topographic images. Combine this with on-site survey drawings to collect building structure size information, surface texture mapping information, and reference photos to construct a full-element original dataset covering the building itself and its surrounding environment.

[0009] Step 2: Construct a unified spatial-temporal reference system and attribute association logic: Spatially calibrate and align the collected multi-source heterogeneous data within a preset geographic information coordinate framework, and establish a structured meta-database including basic building attributes, component characteristics, protection requirements, and drawing archives to achieve accurate positioning and detailed retrieval of each historical building data in the 3D scene;

[0010] Step 3: Perform parametric fine modeling of the historical building body: Based on the extracted outline and feature points, in the building information modeling software, the components are divided into categories such as platform, column, enclosure, beam, decoration, roof and bracket according to the structural system characteristics of the historical building. Then, according to the fineness grading standard, parametric modeling and assembly of family files are performed to form a building entity model containing three-dimensional information of geometry, appearance and attributes.

[0011] Step 4: Conduct parametric restoration and environmental integration of vegetation system: Use vegetation modeling software to perform three-dimensional geometric reconstruction based on the morphological characteristics of tree branches and trunks and real-world images, restore the hierarchical relationship of branches and trunks and the canopy structure, and set physical wind field parameters to achieve high-fidelity simulation of ancient and famous trees and general plant landscapes, and complete the deployment of full-element digital twin scene assets.

[0012] Step 5: Implement high-fidelity visualization applications and multi-dimensional interactive integration: Import the constructed building entity model and vegetation model into the Unreal Engine environment, apply physically realistic rendering technology to perform real-time lighting processing and light and shadow baking, and integrate a dynamic weather system, day and night cycle switching logic, and a high-quality dynamic character model library to build a highly immersive digital humanistic interactive scene of historical buildings.

[0013] Preferably, the process of collecting multi-source heterogeneous basic data in step 1 includes acquiring complete information on the surrounding environment, roof, facade, and interior. The surrounding environment should include information on the main entrance of the building, surrounding structures, roads, squares, water bodies, mountains, and greenery, with a focus on collecting historical environmental elements such as ancient wells, ancient trees, courtyard walls, traditional streets and alleys, and gardens. The roof section should include the roof form, roof structure, and additions and alterations. The facade section should cover all visible facades and collect information on materials, decorations, structural details, doors and windows, and damage and deformation. The interior space includes information on accessible spaces, interior layouts with artistic value, and structural details that reflect the style and characteristics.

[0014] Preferably, when collecting texture photo data in step 1, a multi-angle shooting method is used to take pictures of historical buildings and surrounding facilities along the route. When shooting, ensure that the camera's line of sight is facing the subject and that the angle between the camera's angle and the subject is greater than or equal to 70 degrees. For special areas where the road is narrow and it is impossible to take a full picture, adopt the logic of first taking a panoramic picture and then taking a scanning picture of each store face to perform post-processing photo compositing.

[0015] Preferably, the establishment of the spatial reference system in step 2 follows these specifications: the plane coordinate system adopts the local coordinate system of a certain region, the elevation datum adopts the national elevation datum and is associated with the spherical coordinate system, and the date in the time reference system adopts the Gregorian calendar and the time adopts Beijing time.

[0016] Preferably, the point cloud data acquisition and stitching in step 2 executes the following logic: the point cloud data acquisition accuracy is set to no less than 10 mm, the point cloud data includes three-dimensional coordinates, color values, classification values, intensity values ​​and time features, the stitching error of the point cloud data acquired in batches is less than 5 mm, and it is ensured that there is no point cloud layering phenomenon caused by registration error exceeding the limit.

[0017] Preferably, the parametric fine modeling process of the historical building body in step 3 is based on the integrated design data feature of the building information modeling software. The parametric function is used to realize the coordination and change management between graphic elements. The geometric expression of the model unit includes spatial positioning, spatial occupancy and geometric accuracy elements. The model coordinates of the city-level model unit are consistent with the platform coordinates. The component-level model unit marks the positioning base point and is quantitatively characterized in the attribute information table with parameters such as base point coordinates and spatial occupancy size.

[0018] Preferably, the precision grading standard in step 3 follows the following grading logic: Level 1 reflects the building outline, and the model is made based on the geometric shape and height; Level 2 represents the main structure based on the survey drawings, ignoring auxiliary components such as steps, chimneys, and eaves, and the dimensions of doors and windows are consistent with the drawings; Level 3 reflects the outlines of balconies, doors, windows, eaves, etc. on the facade based on the survey drawings and referring to on-site pictures, and distinguishes different materials; Level 4 realistically reflects the appearance details, including roof tiles, complex decorations, and chimney details, so that the model's appearance is consistent with the real thing, and the materials used for each component of the model completely correspond to the actual appearance features of the building.

[0019] Preferably, in step 3, the detailed modeling of historical buildings, while satisfying the visual effect, adopts texture mapping to represent complex views to reduce geometric complexity. For components that reflect traditional construction techniques, the process is recorded in detail in the model. The roof model represents the tile layer and structural layer, the wall model represents the decorative layer and structural layer, and the ground model represents the surface layer and structural layer. The overall thickness of each part is strictly consistent with the survey data.

[0020] Preferably, the following judgment rules are applied to handle data discrepancies in step 3: For the interior of historical buildings, if the part is not reflected in the survey information but is reflected in the image information and does not conform to the historical characteristics, the model is built according to the survey information; if the structure is inconsistent, the model is built according to the on-site situation; for the exterior of historical buildings, the model is built according to the actual on-site situation; if the on-site reconstruction has destroyed the original appearance, the original historical appearance is restored according to the survey information; for auxiliary components, if the survey information does not reflect them and they are not part of the building structure, the model will not represent them.

[0021] Preferably, in step 3, the naming convention for the model follows a combination of Latin characters, numbers, and underscores, with no spaces between characters and symbols. In the floor code, the basement is represented by the character B, and the above-ground part is represented by the character F. The file name consists of the building name, project code, zoning system, professional code, type, elevation, and description information in sequence, separated by hyphens.

[0022] Preferably, the vegetation system modeling in step 4 involves the following specific operations: using vegetation modeling tools to reconstruct the structure of the tree trunks, branches, and leaves; drawing the shape of the tree trunk by hand and correcting the shape using adjustment points; adding at least two different levels of branches and leaves using geometric components; importing color maps, normal maps, and transparency maps for material filling; and setting wind field parameters by adjusting sliders to control wind intensity and gust frequency.

[0023] Preferably, the rendering process of the visualization application in step 5 includes: applying physically based rendering technology to achieve advanced dynamic shadows, screen space reflections and lighting channel processing, using sequence editor tools to create non-linear real-time animations, and using Unreal Engine's built-in scene query system and behavior tree to execute advanced artificial intelligence logic.

[0024] Preferably, the dynamic environment simulation in step 5 includes: applying high-quality global illumination technology to simulate the propagation path of light and the indirect light reflected by objects; controlling the light intensity and color by adjusting curves and gradients; using dynamic physical material effect parameters to adjust presets to enrich scene functions; and applying a dynamic weather system to execute the changing seasons and the switching of day and night cycles, wherein the day and night cycle supports the calculation of the sun and moon positions based on geographical latitude and longitude.

[0025] Preferably, the method also involves the result quality control logic: the difference between the model plane positioning and the survey drawing is less than 0.05 meters, the error between the model component size and the actual size on site is limited to within ±50 millimeters, the modeling accuracy requirement for important protected parts is to reach level 4, and the sampling inspection ratio is 100%, and the modeling accuracy for the main structure is to reach level 3, with a sampling inspection ratio of not less than 50%.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] This invention constructs a multi-source heterogeneous dataset with millimeter-level precision by comprehensively applying 3D laser scanning, oblique photogrammetry, and on-site survey drawings, effectively solving the problem that traditional methods struggle to reproduce the complex geometric features of historical buildings. By establishing a unified spatial-temporal reference system, it achieves precise logical association between discrete point clouds and structured attribute information. For non-standard components unique to historical buildings, such as brackets and beams, parametric modeling is adopted, which not only improves the geometric fidelity of the components but also comprehensively records the building's age, materials, craftsmanship, and other cultural attributes through a metadata database, providing solid data support for the full lifecycle management of historical buildings.

[0028] This invention employs a strict classification of detail levels, particularly for the fourth level of detailed modeling of critical protected areas. This achieves a realistic reproduction of tile textures, decorative component details, and damage and deformation characteristics, enabling the digital model to simulate physical entities with high fidelity. Simultaneously, this invention introduces parametric vegetation modeling and surrounding scene integration technology, comprehensively integrating historical buildings with their natural environment and geographical information to construct a complete digital twin asset, significantly enhancing the systematic nature and authenticity of the digital cultural scene.

[0029] This invention utilizes a visualization platform built on Unreal Engine, employing physically based rendering technology and high-quality global illumination algorithms to significantly enhance the realism and real-time performance of lighting and shadow effects. By integrating a dynamic weather system, day-night switching logic, and a dynamic character model library, this invention achieves a leap from static display to dynamic interactive scenes. It not only provides a high level of visual effects but also offers users a deeply interactive participation method through artificial intelligence behavior trees and physically based material effects. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0031] Figure 2 This is a schematic diagram of the core principle framework of the multi-source heterogeneous data fusion and space-time reference system construction of the present invention;

[0032] Figure 3 This is a logical flowchart of the parametric fine modeling of the historical building body and vegetation system in this invention.

[0033] Figure 4 This is a schematic diagram of the multi-level interaction relationships and data flow of the high-fidelity visualization application and multi-dimensional interactive integration of the present invention. Detailed Implementation

[0034] Example 1

[0035] Please refer to Figures 1 to 4To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0036] Currently, in the field of historical building preservation and digital restoration, traditional technical solutions often face core pain points such as single data sources, insufficient model accuracy, and lack of cultural attribute information. This results in digital assets that fail to accurately reproduce the complex geometric features and craftsmanship details of historical buildings, and lack sufficient interactive depth and visual realism in subsequent revitalization and utilization. To address these technical problems, this invention proposes a refined BIM modeling method for historical buildings based on multi-source data fusion. Through deep coupling of millimeter-level point cloud acquisition, parametric component assembly, vegetation physical simulation, and high-fidelity realistic rendering, a complete, high-fidelity digital twin scene with multi-dimensional attribute associations is constructed, providing precise underlying data support for cultural relic preservation and smart city construction.

[0037] In the aforementioned method for refined BIM modeling of historical buildings based on multi-source data fusion, step 1 involves collecting multi-source heterogeneous basic data of the historical building. This includes acquiring millimeter-level precision point cloud data using a 3D laser scanning device, simultaneously using a drone equipped with an industrial camera for oblique photography to obtain macroscopic terrain images, and combining this with on-site survey drawings to collect building structural dimensions, surface texture mapping information, and reference photographs, thus constructing a comprehensive original dataset covering the building itself and its surrounding environment.

[0038] In this phase, the 3D laser scanning process is performed as a full-coverage scan of the building body, with a scanning accuracy set to 1 mm and a point cloud density configured such that the point spacing is less than 2 mm within a 10-meter range. Given the complex and varied surface features of historical buildings, the scanning stations are deployed in a surrounding pattern to ensure that each facade, corner, and detailed decorative component is covered by scan lines at at least three different angles, eliminating blind spots and enhancing the redundancy verification capability of the point cloud data.

[0039] Among these, high-resolution optical imagery boasts a spatial resolution better than 0.05 meters, synthetic aperture radar imagery is used to assist in penetrating sparse vegetation to obtain accurate surface elevation, and hyperspectral imagery is used to identify the chemical composition characteristics of building materials. The acquisition of multi-source heterogeneous basic data also includes obtaining complete information on the surrounding environment, roofs, facades, and interiors. The surrounding environment includes information on the main building entrance, surrounding structures, roads, plazas, water bodies, mountains, and greenery, with a focus on historical environmental elements such as ancient wells, ancient trees, courtyard walls, traditional streets and alleys, and gardens. The roof section includes roof type, roof structures, and details of additions and alterations.

[0040] The facade section covers all visible facades and collects information on materials, decorations, structural details, doors, windows, and damage / deformation. The interior space includes information on all accessible spaces, artistically valuable interior layouts, and structural details reflecting the architectural style. During texture mapping and photographic data collection, a high-definition industrial camera was used to perform panoramic photography along the route, capturing images of historical buildings and surrounding facilities from multiple angles. The shooting process strictly adhered to the principle of orthogonal optical axes, meaning the camera's line of sight was directly facing the photographed surface, and the angle between the camera's viewpoint and the photographed surface was limited to 70 degrees or greater. For areas with narrow traditional streets and limited visibility, a combination of wide-angle panoramic shooting and local scanning was employed. A multi-scale feature matching algorithm was used to perform high-precision post-processing image synthesis, ensuring that texture maps did not exhibit ghosting or misalignment at edge seams.

[0041] In the aforementioned method for refined BIM modeling of historical buildings based on multi-source data fusion, step 2 involves constructing a unified spatial-temporal reference system and attribute association logic. The collected multi-source heterogeneous data is spatially calibrated and aligned within a pre-defined geographic information coordinate framework. A structured metadata database is established, including basic building attributes, component characteristics, protection requirements, and drawing archives, enabling precise positioning and detailed retrieval of each piece of historical building data in the 3D scene. The establishment of the spatial reference system follows these specifications: the planar coordinate system uses a local coordinate system for a given region, the elevation datum uses the national elevation datum, and a semantic association is established with the spherical coordinate system. The temporal reference system follows a globally unified standard, using the Gregorian calendar for dates and Beijing time for time. Strict accuracy checks are performed during the acquisition and stitching process of point cloud data. Point cloud data includes 3D coordinates, color values, classification values, intensity values, and temporal characteristics. Point cloud data acquired in batches is registered using an iterative nearest-point algorithm, with stitching errors controlled to less than 5 millimeters. The root mean square error of point cloud registration is quantitatively evaluated using the following formula:

[0042]

[0043] in, Characterizing the root mean square error, The number of feature point pairs participating in registration, For the target point cloud Coordinates of feature points The coordinates of the corresponding feature points in the source point cloud. The transformation matrix includes rotation and translation. By minimizing this error value, it is ensured that there is no point cloud layering caused by exceeding the registration error limit. In the attribute association logic construction stage, semantic association between entities is realized by establishing a graph neural network based on graph attention. The node feature vector dimension is set to 128, the edge weight is calculated by multiplying the spatial distance decay function with the functional similarity, the decay coefficient is set to 0.8, and the similarity threshold is set to 0.7. In addition, spatial autocorrelation analysis is performed, and the spatial clustering characteristics of building clusters are identified by Moran's index. The calculation window radius is set to 5000 meters, and the significance level is set to 0.05. Each data point of historical buildings is associated with real estate registration data through spatial coding. The real estate data covers property rights information such as land, forest land, and water area, and the matching error is configured to be less than 0.5 meters.

[0044] In the aforementioned method for refined BIM modeling of historical buildings based on multi-source data fusion, step 3 involves performing parametric refined modeling of the historical building entity. Based on the extracted contour lines and feature points, the components are categorized into platforms, columns, enclosures, beams, decorations, roofs, and brackets according to the structural system characteristics of the historical building in the BIM software. Parametric modeling and assembly of family files are then performed according to the refinement grading standards to form a building entity model containing three-dimensional information on geometry, appearance, and attributes. The parametric modeling process leverages the integrated design data feature of the BIM software, utilizing its parameter-driven mechanism to achieve automatic coordination between graphic elements.

[0045] The geometric representation depth of the model units follows a strict hierarchical logic: Level 1 reflects the building outline, based on simplified geometric blocks and total height; Level 2 represents the main load-bearing structure based on the survey drawings, ignoring non-structural auxiliary components, and the dimensions of doors and windows are strictly consistent with the drawings; Level 3 adds the outlines of balconies, eaves, and key decorative elements on the facade based on Level 2, and applies texture differentiation to different materials such as stone, brick walls, and wood structures; Level 4 requires the realistic reproduction of every appearance detail, including the arrangement of roof tiles, the number of steps in the brackets, and the damage and cracks in the chimney, so that the digital model is completely consistent with the physical entity in terms of perception.

[0046] During component assembly, a collision detection algorithm is executed to ensure that the connection error between components is limited to within 1 mm. For complex components that reflect traditional architectural features, such as dougong (bracket sets), parametric family nesting technology is adopted, with control variables set according to the number of steps, the length of the bracket set, and the slope of the bracket set angle. While meeting the visual restoration depth requirements, for minor decorative patterns that do not affect structural analysis, a combination of normal mapping and displacement mapping is used to replace high-poly geometry, optimizing the topological complexity of the model in the real-time rendering scene. For data discrepancies, preset judgment rules are executed: if there is a conflict between the survey drawings and the real-time images collected on-site in terms of interior decoration, priority is given to restoring the original historical appearance based on the survey information; if on-site reconstruction has severely damaged the original appearance, historical imagery is retrieved, and the model is restored according to the original features of the survey information. All model naming follows a strict combination of letters, numbers, and underscores, with no spaces between characters. The file naming structure is defined as: full building name, unique project identifier, professional code, floor code, and software version number. For example, the basement floor code is represented by the character B, and the above-ground floor code is represented by the character F.

[0047] In the aforementioned method for refined BIM modeling of historical buildings based on multi-source data fusion, step 4 involves parametric restoration and environmental integration of the vegetation system. Using vegetation modeling software, 3D geometric reconstruction is performed based on the morphological characteristics of tree branches and trunks and real-world images. This restores the hierarchical relationships of branches and trunks and the canopy structure. Physical wind field parameters are set to achieve high-fidelity simulation of ancient and famous trees and general plant landscapes, completing the deployment of full-element digital twin scene assets. Vegetation modeling executes parametric growth logic. The growth trajectory of the main trunk is determined in virtual space through hand-drawing, and control points are used to adjust the diameter and curvature of the trunk. A branching system is generated by adding multi-level geometric components, including at least three levels: main branches, secondary branches, and twigs. The leaf system is generated using instantiation rendering technology, with each leaf containing independent deflection angle parameters. To achieve dynamic realism of the vegetation in the digital twin environment, a physical wind field model is introduced to perform stress analysis. The wind load on the vegetation is calculated using the following simplified formula:

[0048]

[0049] in, For wind load, The density of air is 1.225 kg per cubic meter. For real-time wind speed, The drag coefficient of the vegetation canopy. This represents the windward area of ​​the blade system. The wind force and gust frequency can be controlled in real-time by adjusting sliders within the software. For material representation, a multi-layered material including diffuse, normal, and alpha mask maps is imported, and the light transmission characteristics of the blades are reproduced using a subsurface scattering shader. A reinforcement learning agent is introduced to optimize vegetation distribution. The agent employs a deep deterministic policy gradient algorithm. The state space includes terrain slope, light intensity, and historical vegetation distribution records; the action space is the coordinate vector of vegetation placement; and the reward function is set as a weighted product of the rationality of vegetation distribution and landscape harmony.

[0050] In the aforementioned method for refined BIM modeling of historical buildings based on multi-source data fusion, step 5 involves integrating high-fidelity visualization applications with multi-dimensional interaction. The constructed building entity model and vegetation model are imported into the Unreal Engine environment. Physically realistic rendering technology is applied to perform real-time lighting processing and shadow baking. A dynamic weather system, day / night cycle switching logic, and a high-quality dynamic character model library are integrated to construct a highly immersive digital humanistic interactive scene for historical buildings. The rendering process for the visualization application adopts a deferred rendering path, combined with screen space reflection and global illumination algorithms. High-quality global illumination technology is used to simulate multiple reflections of light between complex ancient building components. By adjusting brightness weights and color overflow parameters, the sense of layering in the shadow areas under wooden beams and rafters is enhanced.

[0051] The dynamic weather system supports real-time switching from sunny to heavy rain and blizzard, with rainwater flowing on roof tiles using dynamic material offset. The day-night cycle switching logic is based on real latitude and longitude coordinates, automatically calculating the sun and moon's trajectories and altitude angles to achieve long shadow effects during the transition from dawn to dusk. For interactive experience, a high-quality dynamic character model library is integrated, with each character entity possessing independent behavior tree logic and navigation grid path planning capabilities. The visualization decision support function supports dynamic generation of assessment reports by annual and five-year planning cycles, and the spatial distribution heatmap uses a 7-level hierarchical color scheme for visualization. Furthermore, the system interfaces with a land spatial planning map system through a standard geographic information service interface to perform spatial consistency checks, ensuring that the control boundary error within the digital twin scenario is less than one pixel.

[0052] To verify the reliability and accuracy of the aforementioned BIM-based refined modeling method for historical buildings using multi-source data fusion in a real-world project, this embodiment constructs a restoration application example for a core building complex in a historical district. In this example, the target buildings are a group of traditional courtyards featuring brick-and-wood structures, complex bracket sets, and intricately carved brick facades.

[0053] First, during the data acquisition phase, the team deployed 32 laser scanning stations, covering all visual spaces inside and outside the courtyard. The 3D laser scanner was configured to emit 1 million laser pulses per second, with a resolution of 3 millimeters per scan at a distance of 10 meters. A drone equipped with a five-lens oblique photography system performed two flights at a height of 60 meters, acquiring a total of 1500 high-resolution images. During the acquisition process, the team used handheld industrial cameras to perform 360-degree panoramic photography, focusing on multi-angle macro photography of the decorative roof ornaments and wood carvings in the porch, ensuring that the image resolution for every detail was better than 1 millimeter.

[0054] Secondly, in the data preprocessing and coordinate alignment stage, all point cloud data were converted to a local coordinate system using four benchmark control points deployed on-site. The root mean square error of point cloud registration was calculated to be 3.2 mm, far better than the standard requirement of 5 mm. The established metadata database contains 28 attribute dimensions, including the building's construction year, main building materials, and protection level. A graph neural network was used to automatically semantically associate the geometric entities of each door and window with historical survey records, and the association confidence level was assessed at 98%.

[0055] Subsequently, in the parametric modeling phase, Revit software was used to perform detailed modeling at LOD 400 level. For the 12 main load-bearing columns within the building, parametric family files were created, including column diameter scales and column base shapes. The roof tile modeling employed array-associative logic, with the spacing between each row of small blue tiles precisely configured to 120 mm. For the damaged cracks on the facade, instead of using complex geometric meshes, the team extracted crack trajectories from high-precision point clouds, combined with normal maps and roughness maps, to achieve a highly realistic visual reproduction in the renderer. Collision checking during the modeling process identified and corrected three gaps in component connections, ensuring the integrity of the model structure.

[0056] Next, in the environmental integration phase, SpeedTree software was used to recreate the two century-old locust trees in the courtyard. The texture of the main trunk was generated by seamlessly smoothing photographs of the tree bark taken on-site. The wind field parameters were set to simulate a level 3 gust environment common in a certain region during summer, with the wind speed set at 4.5 meters per second. By adjusting the drag coefficient, the swaying frequency of the ancient locust tree branches and leaves was made to closely match the on-site observation results.

[0057] Finally, in the visualization integration phase, the model was imported into the Unreal Engine editor. Ray tracing technology was enabled to handle the complex self-occluding shadows generated by the bracket components. A dynamic weather system simulated a typical Jiangnan plum rain scene, showcasing the visual effect of rainwater gathering along the roof ridges and dripping from the eaves through the collaboration of a raindrop particle system and a roof fluid shader. The interactive interface allows users to click on specific architectural components to bring up a property panel containing the component's survey drawing and current photograph in real time. Testing showed that the real-time rendering frame rate of this digital twin scene on a high-performance workstation remained stable at 90 frames per second, meeting the requirements for immersive digital display.

[0058] The quality control logic in this embodiment is consistent throughout: the difference between the model's planar positioning and the original survey drawings is strictly limited to 0.035 meters, conforming to the quality specification of less than 0.05 meters. The sampling inspection rate for critical protected areas is set at 100%, and all components pass the Level 4 precision acceptance. The sampling inspection rate for the main load-bearing structure is 60%, ensuring the overall structural robustness of the model.

[0059] Example 2

[0060] Building upon Example 1, this example further proposes a deep learning-assisted modeling technique for ultra-complex bracket arch components, aiming to address the low efficiency of manually creating complex parametric family files in Example 1. In the above method, step 3, performing parametric fine modeling of the historical building entity, also includes automatically classifying and extracting parameters from irregularly shaped components using a deep learning model.

[0061] Specifically, this embodiment introduces a component recognition framework based on a voxel convolutional neural network. The discrete point set extracted from point cloud data is transformed into a fixed-resolution 3D mesh through voxelization. The node feature vector includes the point cloud density within the voxel, the mean of the surface normal vector, and color moment features. A deep learning agent is configured to perform multi-class recognition of typical components in historical buildings. During the model training phase, 100,000 configuration cases stored in the knowledge base are used as prior knowledge support, and a semantic retrieval engine extracts similar component geometric templates.

[0062] The classification accuracy of the identification process is optimized using the cross-entropy loss function, and the convergence criterion is set as the rate of change of the Pareto front. Once the neural network identifies a specific component type, it automatically calls a preset parameterized family template and performs adaptive matching based on the boundary features of the point cloud data. For example, for a five-tiered bracket set in an ancient style, the algorithm automatically extracts the length of the ang (a type of bracket arm), the inclination of the ang angle, and the relative height of each dou (a type of bracket arm), and maps them to the control variables in the family file. This automated extraction logic significantly shortens the modeling cycle of complex components, reducing the modeling time for a single complex bracket set from 3 hours in Example 1 to 15 minutes.

[0063] Furthermore, this embodiment adds an interactive module based on a virtual reality headset in the visualization application of step 5. Users can disassemble and assemble building components using the controllers, and the system calculates the physical constraints between components in real time. If a logical error occurs during assembly, such as component collision or size mismatch, the system issues a visual warning through a particle system and retrieves the correct process rules from the knowledge base to provide repair suggestions. In this way, this embodiment not only achieves visual restoration but also realizes digital evidence preservation and teaching of historical building construction techniques at the process logic level.

[0064] In terms of deliverables management, this embodiment establishes a BIM data archiving system based on distributed storage. Digital deliverables are backed up using different storage media and an off-site backup strategy is implemented. The archiving path for BIM data is strictly configured to create independent subdirectories by profession, ensuring data independence and traceability. Version control technology ensures that during multiple update iterations, each original surveying data and final model deliverable can be logically traced back using a unique timestamp.

[0065] Example 3

[0066] Based on the above embodiments, this embodiment proposes a technical solution for the deep integration of BIM and GIS for the full-element scene integration of large-scale historical blocks. In the above method, step 2, constructing a unified spatial-temporal reference system and attribute association logic, also includes establishing a spatial data indexing mechanism with multi-scale detail levels.

[0067] This embodiment integrates high-resolution optical imagery, synthetic aperture radar imagery, and 3D point cloud data to construct a continuously scaled scene from the macroscopic city scale to the microscopic component scale. At the macroscopic scale, the tile rendering technology of the Geographic Information System (GIS) is applied, and geometric correction based on control points is used to achieve accurate alignment of multi-temporal remote sensing images. Spatial autocorrelation analysis is used to identify the landscape harmony deviation within blocks. The deviation index is calculated by weighting spatial mismatch rate, target deviation degree, and execution lag coefficient, with weights of 0.4, 0.4, and 0.2, respectively.

[0068] In the visualization interaction of step 5, a multi-dimensional indicator display function of radar charts was introduced. The radar charts constitute a five-dimensional evaluation space. The resource allocation scheme is continuously optimized through a deep deterministic policy gradient algorithm. The reward function is set as the product of the decrease in the configuration deviation index and the scene stability, aiming to find the optimal historical building function replacement scheme.

[0069] To ensure strict consistency between resource allocation plans and higher-level plans, the system links in real-time with the land spatial planning map system through a standard geographic information service interface. The spatial distribution heat map uses a 7-level color-coding method to mark the distribution of conservation value within the block. When a user attempts to undertake construction activities that do not comply with planning control requirements within the scene, the system automatically triggers a risk warning and classifies and labels the risk using red, yellow, and blue color levels in the visualized decision report. This embodiment, through this comprehensive, cross-scale integration approach, achieves a technological leap from individual building restoration to holistic block governance.

[0070] In summary, the present invention proposes a refined BIM modeling method for historical buildings based on multi-source data fusion. Through end-to-end optimization of millimeter-level acquisition accuracy, parametric family library construction, vegetation physical feature simulation, and the Unreal Engine rendering pipeline, it successfully solves the challenges of fidelity and interactivity in the digital restoration of historical buildings. This solution not only provides a high level of real-time visual effects but also ensures the scientific validity and authority of digital assets through deep attribute association and intelligent quality control mechanisms.

[0071] The quality control process involved in this invention is strictly implemented in three stages: mutual inspection and self-inspection. The self-inspection stage focuses on the preliminary verification of the model's planar positioning and elevation; the mutual inspection stage introduces a high-intensity quality inspection standard of 100% coverage for LOD400 level and 50% sampling inspection for LOD300 level. A weighted evaluation system consisting of mathematical accuracy, completeness, logical consistency, and the quality of attachments ensures that the delivered BIM model data, texture maps, attribute data, and quality inspection reports fully comply with current standards and data specifications.

[0072] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for refined BIM modeling of historical buildings based on multi-source data fusion, characterized in that, include: Collect multi-source heterogeneous basic data of historical buildings. The multi-source heterogeneous basic data includes point cloud data obtained by 3D laser scanning, macroscopic terrain images obtained by UAV oblique photography, structural dimension information, surface texture mapping information and reference photo records from on-site survey drawings; A unified spatial-temporal reference system and attribute association logic are constructed. The multi-source heterogeneous basic data are spatially calibrated and aligned within a preset geographic information coordinate framework. A structured meta-database containing basic building attributes, component characteristics, protection requirements, and drawing archive information is established to achieve accurate positioning and detailed retrieval of each historical building data in a three-dimensional scene. Parametric fine modeling of the historical building body is performed. Based on the extracted contour lines and feature points, the components are divided into categories such as platform, column, enclosure, beam, decoration, roof and bracket in the building information modeling software according to the historical building structure system. Parametric family files are generated according to the fineness grading standard and the model is assembled. Parametric restoration and environmental integration of vegetation systems were carried out. Vegetation modeling software was used to reconstruct the hierarchical relationship of tree branches and canopy structure based on real-world images and morphological characteristics, and physical wind field parameters were configured to achieve high-fidelity dynamic simulation. To achieve high-fidelity visualization applications and multi-dimensional interactive integration, the building entity model and vegetation model are imported into the Unreal Engine environment. Physically realistic rendering technology is applied for real-time lighting processing and light and shadow baking. A dynamic weather system, day and night cycle switching logic, and a high-quality dynamic character model library are integrated to construct an immersive digital humanistic interactive scene of historical buildings. The construction of a unified space-time reference frame and attribute association logic includes: Semantically link the plane coordinate system, elevation datum, and spherical coordinate system to form a unified spatial datum; A time reference system is constructed using the Gregorian calendar and standard time zones; The point cloud data collected in batches is registered by using an iterative nearest point algorithm to ensure that the stitching error is controlled within a preset threshold. Semantic relationships between entities are constructed based on graph neural networks, and each data point is spatially encoded and matched with entity registration information. Performing parametric fine modeling of the historical building ontology includes: The model is divided into levels according to the level of detail. Level 1 shows the overall outline of the building, Level 2 shows the main structure and ignores non-structural auxiliary components, Level 3 reflects the exterior facade construction and material distinction, and Level 4 realistically reproduces the tile arrangement, complex decoration and damaged details. For components made using traditional techniques, the process is recorded in the model, and structural and decorative layers are modeled separately. Using texture mapping to replace complex geometric representations to optimize model performance; Performing parametric fine modeling of historical building entities also includes: When there is a conflict between the survey information and the on-site images, the internal structure is modeled based on the survey information first, and the exterior is modeled based on the actual on-site conditions. If the on-site reconstruction damages the original appearance, the original historical appearance is restored based on the survey information. Attached components that are not reflected in the survey information and are not part of the building structure shall not be modeled and represented. Achieving high-fidelity visualization applications and multi-dimensional interactive integration includes: The application of physically based rendering techniques enables dynamic shadows, screen-space reflections, and lighting channel processing. Use sequence editor tools to create non-linear real-time animation content; AI-driven interactive responses are achieved through Unreal Engine's built-in scene query system and behavior tree logic. Vegetation modeling employs parametric growth logic, determining the growth trajectory of the main trunk in virtual space through hand-drawing and adjusting the trunk's diameter and curvature using control points. A branching system is generated by adding multi-level geometric components, comprising at least three levels: main branches, secondary branches, and twigs. The leaf system is generated using instantiation rendering technology, with each leaf containing independent deflection angle parameters. To achieve dynamic realism of the vegetation in the digital twin environment, a physical wind field model is introduced to perform stress analysis; the wind load on the vegetation is calculated using the following simplified formula: in, For wind load, air density, For real-time wind speed, The drag coefficient of the vegetation canopy. The windward area of ​​the blade system is defined; the wind force field intensity and gust frequency are controlled in real time by adjusting the slider in the software; in terms of material representation, a multi-layer material including diffuse map, normal map and alpha mask map is imported, and the light transmission characteristics of the blade are restored by subsurface scattering shader; reinforcement learning agent is introduced to optimize vegetation distribution. The agent adopts a deep deterministic policy gradient algorithm. The state space includes terrain slope, light intensity and historical vegetation distribution records. The action space is the coordinate vector of vegetation placement. The reward function is set as a weighted product of the rationality of vegetation distribution and landscape harmony.

2. The method for refined BIM modeling of historical buildings based on multi-source data fusion according to claim 1, characterized in that, The process of collecting multi-source heterogeneous basic data of historical buildings includes: Obtain surrounding environmental information covering the main entrance of the building, surrounding structures, roads, squares, waterways, mountains, and green spaces; Obtain roof information including roof type, roof structure, and additions / alterations; Facade information covering all visible facades and collecting information on materials, decorations, structural details, doors and windows, and damage and deformation; Collect information on the structural layout, interior arrangement, and structural details of accessible indoor spaces.

3. The method for refined BIM modeling of historical buildings based on multi-source data fusion according to claim 1, characterized in that, Parametric reconstruction and environmental integration of vegetation systems include: The growth trajectory of the main stem is defined by hand and its shape is adjusted using control points. Add at least two different levels of geometric components to construct a layered structure of branches and leaves; Import color maps, normal maps, and transparency maps for material filling, and configure wind field parameters to achieve a dynamic swaying effect.

4. The method for refined BIM modeling of historical buildings based on multi-source data fusion according to claim 1, characterized in that, Achieving high-fidelity visualization applications and multi-dimensional interactive integration also includes: Global illumination technology is used to simulate the multiple reflection paths of light between complex components; Light intensity and color performance can be controlled by curves and gradient parameters; Integrate a dynamic weather system to support seasonal changes and geographic coordinate-based day-night cycle switching.

5. The method for refined BIM modeling of historical buildings based on multi-source data fusion according to claim 1, characterized in that, It also includes the logic for quality control of results, specifically including: Detailed modeling of the protected areas was performed, and all areas were randomly sampled for inspection. Perform detailed modeling of the main structure and conduct random inspections at a rate no less than the preset proportion; Ensure that the deviation between the model's planar positioning and the survey drawing, as well as the error between the component dimensions and the actual dimensions, are all controlled within the allowable range.

6. The method for refined BIM modeling of historical buildings based on multi-source data fusion according to claim 1, characterized in that, The method also includes model naming conventions, specifically including: Use a combination of Latin characters, numbers, and underscores in the naming convention, without spaces between characters; Basement floors are represented by the character B, and above-ground floors are represented by the character F; The file name consists of the building name, project code, zoning system, professional code, type, elevation, and description information in that order, separated by hyphens.

Images