Data processing method for 3D GS and BIM model space matching

Abstract

In order to improve the accuracy and fineness of three-dimensional scene interactive display in a three-dimensional reconstruction model, the invention provides a data processing method for 3DGS and BIM model space matching, the method comprises three-dimensional reconstruction based on a 3DGS technology, the process comprises data acquisition, data generation and data splicing, then BIM model generation is carried out, and the BIM model space matching is carried out. And then, the obtained 3DGS data and the BIM data are imported into a Unity 3D engine, and 3DGS model space matching and BIM model space matching are carried out in the Unity 3D engine. According to the method, the 3D GS technology and the BIM model are dynamically fused, multi-source data can be subjected to superposition processing, a physical space is accurately restored, and fine three-dimensional scene interactive display is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a data processing method for spatial matching of 3DGS and BIM models. Background Art

[0002] 3D Gaussian Splatting (3DGS) is a rasterization technology used to represent 3D scenes and render new perspectives. This technology not only reduces training time, but also generates more efficient 3D representations and has excellent perspective synthesis capabilities. Gaussian splatting is essentially a rendering method that adds additional parameters to each point in the point cloud and renders a point in a rasterized manner during rendering to achieve sufficient effects.

[0003] Building Information Modeling (BIM) is an engineering data model based on three-dimensional digital technology that integrates relevant information at all stages of a construction project and can provide all the basic data information of a building throughout its entire life cycle.

[0004] Currently, there is no technical solution to integrate 3DGS and BIM models, and it is impossible to realize the superimposed processing of 3DGS and BIM data, which may result in the interactive display of three-dimensional scenes not being detailed enough. Summary of the invention

[0005] In order to overcome the technical problem that the prior art fails to integrate 3DGS and BIM models, the purpose of the present invention is to provide a data processing method for spatial matching of 3DGS and BIM models, thereby improving the accuracy and refinement of the existing three-dimensional scene interactive display.

[0006] A data processing method for spatial matching of 3DGS and BIM models, comprising the following steps:

[0007] Step 1: Perform three-dimensional reconstruction based on 3DGS technology, including: 1) Data acquisition: Acquire 3D point cloud data or multi-view image data for three-dimensional reconstruction; 2) Data generation: Perform rasterization, Gaussian weighting and color mixing processing on the acquired data, and generate a single-segment Gaussian splash after the processing is completed; 3) Data splicing: Perform data splicing on the generated single-segment Gaussian splash to form multiple-segment Gaussian splash, and export the spliced ​​Gaussian splash scene as a .ply format file;

[0008] Step 2: BIM model generation, including: creating a BIM model using BIM modeling software, wherein the BIM model includes geometric elements and non-geometric attribute data, and processing the geometric elements and non-geometric attribute data into SQLite database data;

[0009] Step 3: Import 3DGS data and BIM data into Unity3D engine;

[0010] Step 4: Spatial matching of 3DGS and BIM models, including: 1) converting the coordinate systems of the two models into the same coordinate system; 2) aligning the two models; 3) adjusting the lighting and shadows of the Gaussian splash effect to ensure that the effect is natural and coordinated with the scene; 4) displaying and hiding the layers according to the professional classification of the BIM model; 5) confirming that the display and interaction effects of the model are correct, and exporting the executable file.

[0011] Furthermore, the data collection in step 1) includes: acquiring the data through a depth sensor, a lidar, a fisheye camera or multi-angle shooting; and representing each 3D space-aligned pixel point with a Gaussian distribution, defining its position (x, y, z) and covariance matrix.

[0012] Furthermore, the data generation in step 1) 2) includes rasterization: rendering each pixel point as a disk or a sphere, and the radius of the disk or the sphere is determined by the radius parameter of the point; Gaussian weighting: using a Gaussian function to weight the rendered disk or sphere so that its edge presents a smooth transition effect; color mixing: mixing the rendering result with the background image or color to generate the final image.

[0013] Furthermore, the data generation in step 1) includes data splicing, and the steps of data splicing are as follows:

[0014] a. Determine the Gaussian splash model of the input scene. Multiple Gaussian splash scenes will be entered into the editor in .ply format;

[0015] b. Spatial mapping of each scene: Translation: Assuming that each scene has a center, reposition each scene by moving the center point in the coordinate system; Rotation: Rotate the scene through the rotation matrix;

[0016] Translation transformation: x'=x+△x, y'=y+△y

[0017] Rotation transformation:

[0018] c. Processing gaps and overlapping areas: When stitching, the boundaries of the scenes may overlap or have gaps. If there are gaps, stitch them directly and stitch the scenes adjacent to each other. The boundary area will naturally form a transition. If there are overlaps, merge them and perform Gaussian weighted averaging in the overlapping area to make the transition smoother. The Gaussian weighted formula is as follows: G blend (x,y)=w1*G1(x,y)+w2*G2(x,y)

[0019] d. Adjust the global scope: Boundary expansion: In order to accommodate the stitching of multiple scenes, the boundary range of the entire large scene needs to be adjusted to ensure that all Gaussian scenes can be covered; Interpolation and sampling: If the Gaussian splash scene is a discrete data point, use the interpolation method to expand this data to a larger spatial range;

[0020] e. Generate the global Gaussian scene after stitching: Calculate the pixel value of the new scene: For each pixel point, calculate the value of all superimposed scenes. If there are multiple overlapping Gaussian splash points at a certain position, the final value is obtained by summing or weighted averaging;

[0021] f. Post-processing and optimization: Smoothing: Use filters to smooth the stitched scene to eliminate sharp changes in transition areas; Brightness and contrast adjustment: Adjust the brightness and contrast of the entire scene to ensure consistent visual effects;

[0022] g. Visualization and output: Output the stitched large scene: Save the stitched Gaussian splash scene as an image or data file, or use it for subsequent processing and analysis, and export it in .ply format.

[0023] Furthermore, the BIM model in step 2 includes geometric elements and non-geometric attribute data, wherein the geometric elements include: walls, floors and roofs, door and window components, beams, columns and structural frames, stairs, handrails and rods, furniture, equipment and other components; the non-geometric attributes include: material information, construction stage attributes, maintenance information, cost attributes, associated documents and file management.

[0024] Furthermore, the step 2 performs feature-preserving processing on the geometric elements of the BIM model, and converts the non-geometric attributes into structured Json data.

[0025] Furthermore, the geometric element processing includes: detail level LOD adjustment: adjusting the detail level of the model to reduce the complexity of the model; merging elements: merging multiple small geometric elements into a large element to reduce the number of elements in the model; deleting unnecessary elements: removing invisible or unimportant elements in the model; using proxy models: using proxy models to replace complex elements;

[0026] The non-geometric attribute processing includes: attribute data Jsonization: ensuring that all non-geometric attribute data are retained and translated into Json format; data mapping: establishing a mapping relationship between geometric elements and original non-geometric attribute data through UDID to restore complete information.

[0027] Furthermore, the step 3 of importing 3DGS data includes: importing the converted data file into the Unity3D project, and in the Unity3D editor, adding the .ply format file exported by 3DGS to the Asset folder of the project; using a point cloud renderer to render three-dimensional Gaussian splash; then configuring and adjusting parameters of the point cloud renderer according to the imported point cloud data file; and finally performing performance optimization, using LOD technology or block rendering to optimize the large number of points contained in the point cloud data.

[0028] Furthermore, the step 3 of importing BIM data includes: importing the converted BIM model file SQLite database into the Unity3D project, and in the Unity3D editor, adding the SQLite database to the Asset folder of the project; using the Material Editor and Shader Graph to configure materials for the model, and applying textures for visual enhancement; reading and displaying parameters and properties in the BIM model, writing scripts to parse the attribute data of the BIM model, and displaying it in Unity3D; using LOD technology, block loading, and light mapping to improve rendering performance and frame rate.

[0029] Furthermore, the step 4 (2) of aligning the two models includes: manually or automatically aligning the two models using an alignment tool, and aligning the two models by adjusting the position, rotation and scaling of the models. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 The multi-viewing angle pictures collected by the panoramic camera of the present invention;

[0031] Figure 2 It is a sparse point cloud image generated by processing the multi-view image of the present invention through COLMAP;

[0032] Figure 3 A single-segment rendering model diagram generated by the present invention;

[0033] Figure 4 A multi-segment rendering model diagram generated by splicing in the same coordinate system of the present invention;

[0034] Figure 5 Importing the merged Gaussian splash rendering model graph into the Unity3D compiler of the present invention;

[0035] Figure 6 Importing a BIM model diagram into the Unity3D compiler of the present invention;

[0036] Figure 7 This is an overlay image of the Gaussian splash rendering model of the present invention and the BIM model. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0038] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0039] The three-dimensional reconstruction process based on 3DGS technology has the following steps:

[0040] Step 1: Data Collection

[0041] First, you need to obtain 3D point cloud data or multi-view image data for 3D reconstruction. These data can be obtained through depth sensors, lidar, fisheye cameras, or multi-angle shooting.

[0042] Exemplarily, this embodiment uses a panoramic camera (composed of two fisheye cameras) containing a laser radar device. The laser radar device and its attached panoramic camera can be handheld for shooting. The laser radar module of the device will be used for SLAM positioning of indoor and outdoor laser radars in buildings (SLAM is the abbreviation of Simultaneous localization and mapping, meaning "simultaneous positioning and mapping"), giving accurate positioning data to the panoramic photos taken by each panoramic camera. The two fisheye cameras distributed on both sides of the device will take panoramic pictures. The captured panoramic pictures contain a field of view of 180° FOV and can be aligned with the positioning posture based on the laser radar in terms of timestamp.

[0043] For example, in addition to data collection by specific LiDAR + dual fisheye devices, data collection by Gaussian splash can also rely on panoramic cameras for data collection. Taking the insta360 panoramic camera as an example, it can shoot dual fisheye videos. Through the SDK it provides, firstly, the dual fisheye videos are synthesized into panoramic videos, and then frames are taken from the panoramic videos, and the panoramic photos in these selected frames are split into photos of different perspectives, which is convenient for subsequent calculations, such as Figure 1 shown.

[0044] The insta360 panoramic camera cannot provide positioning information like the above-mentioned lidar devices, so the positioning information of each photo will be calculated and extracted from the perspective photos split from the panoramic photo. This uses the structure from motion (SFM) method, which mainly includes feature point extraction, feature point matching, trajectory calculation and storage, and trajectory conversion into world coordinates.

[0045] After the shooting is completed, the photo data about the 3D space and the spatial position of each photo are obtained.

[0046] Then, each 3D space-aligned pixel point is represented by a Gaussian distribution, defining its position (x, y, z) and covariance matrix to describe its shape and size. The parameters of the Gaussian distribution can be initialized based on the spatial position and local density of the point cloud. Its position (x, y, z) is calculated and generated by LiDAR or panoramic video frame COLMAP (a SFM method), where COLMAP is the abbreviation of "COLLISION-MAPping" and is an open source computer vision software mainly used for 3D reconstruction from a series of 2D images.

[0047] Among them, Figure 2 As shown in the figure, when applying the COLMAP method, the sparse point cloud is generated by the SFM method. Specifically, the scene is first initially reconstructed using SFM to generate a sparse point cloud. The sparse point cloud is obtained by image registration and feature matching from different perspectives, representing the key points in the scene and their three-dimensional coordinates. Then, a Gaussian distribution is further established on these key points to achieve a three-dimensional representation of the scene.

[0048] Step 2: Data Generation

[0049] Rasterization: Renders each point as a disk or sphere whose radius is determined by the radius parameter of the point.

[0050] Gaussian weighting: Use a Gaussian function to weight the rendered disk or sphere so that its edges present a smooth transition effect. The weight of the Gaussian function is determined by the distance between the point and the pixel center and the radius parameter of the point.

[0051] Color blending: Blend the rendering result with the background image or color to generate the final image.

[0052] For example, after the rendering is completed, some post-processing steps may be performed, such as adding blur effects, adjusting colors and brightness, etc., to enhance the visual effect. The above steps may generate a single-segment Gaussian splash, wherein the single-segment Gaussian splash model is shown in FIG. Figure 3 shown.

[0053] Step 3: Data splicing

[0054] The above steps can generate a single-segment Gaussian splash. When forming multiple-segment Gaussian splashes, data splicing is required. Data splicing aligns and merges the processed data in spatial and temporal dimensions to ensure that the spliced ​​data is continuous and complete.

[0055] The data splicing steps are as follows:

[0056] 1. Determine the Gaussian splash model of the input scene. Multiple Gaussian splash scenes will be entered into the editor in .ply format.

[0057] 2. Spatial mapping of each scene, translation: Assuming that each scene has a center, each scene can be repositioned by moving the center point in the coordinate system. For example, (x_0, y_0) is the center point of the scene, and it is transformed to the target position (x{\prime}, y{\prime}) through translation transformation; rotation: If the scene needs to be rotated, it can be achieved through the rotation matrix:

[0058] Translation transformation: x'=x+△x, y'=y+△y

[0059] Rotation transformation:

[0060] 3. Dealing with boundaries and overlapping areas

[0061] When stitching, the boundaries of the scenes may overlap or have gaps. Different strategies can be adopted:

[0062] Direct stitching: Simply stitch the scenes together, and the boundary area will form a natural transition.

[0063] Blending: Perform weighted averaging in overlapping areas to make the transition smoother. Common strategies are to use linear interpolation or Gaussian weighting:

[0064] Gaussian weighted formula: G blend (x,y)=w1*G1(x,y)+w2*G2(x,y)

[0065] 4. Adjust the global scope

[0066] Boundary expansion: In order to accommodate the stitching of multiple scenes, it may be necessary to adjust the boundary range of the entire large scene to ensure that all Gaussian scenes are covered.

[0067] Interpolation and sampling: If the Gaussian splat scene is a discrete data point, you can use interpolation methods to spread this data to a larger spatial range. For example, use bilinear interpolation or bicubic interpolation.

[0068] 5. Global Gaussian scene generation after stitching

[0069] Calculate the pixel value of the new scene: For each pixel, calculate the value of all superimposed scenes. If there are multiple overlapping Gaussian splash points at a certain location, the final value is obtained by summing or weighted averaging.

[0070] 6. Post-processing and optimization

[0071] Smoothing: Use a filter (such as Gaussian filtering) to smooth the stitched scene to eliminate sharp changes in transition areas.

[0072] Brightness and contrast adjustment: Adjust the brightness and contrast of the entire scene as needed to ensure consistent visual effects.

[0073] 7. Visualization and Output

[0074] Output the stitched large scene: Save the stitched Gaussian splash scene as an image or data file, or use it for subsequent processing and analysis. Export it as a .ply format file to facilitate subsequent superposition with the BIM model. The finished stitched rendering model is as follows: Figure 4 shown.

[0075] The generation of Building Information Model (BIM) includes the following steps:

[0076] Step 1: BIM Modeling

[0077] Users create BIM models through various modeling software, such as Autodesk Revit, Autodesk Naviswork, IFC, AvevaPDMS and other BIM modeling software. The geometric elements and non-geometric attributes in the BIM model are fully created. Typical geometric elements include: walls, floors and roofs, door and window components, beams, columns and structural frames, stairs, handrails and rods, furniture, equipment and other components. Typical non-geometric attributes include: material information, construction stage attributes, maintenance information, cost attributes, associated documents and file management, etc.

[0078] In BIM modeling software, you can switch between different view modes, such as 3D view, plan view, elevation view, etc., to verify the correctness of geometry and attribute information. After modeling, export the BIM model to common file formats, such as IFC, RVT, DWG, etc. Through the above steps, the creation of geometric elements and the assignment of non-geometric attributes constitute a complete BIM modeling process, which can provide comprehensive data information support for various stages such as building design, construction, and maintenance.

[0079] Step 2: Data processing

[0080] Model processing is performed using a cloud server or a computer room workstation. BIM data is processed into SQLite database data through a specific algorithm, in which the geometric elements of the BIM model are processed to preserve their features, and non-geometric attributes are converted into structured Json data. Feature-preserving processing means that for a Mesh model composed of triangular or polygonal faces, its geometric features are retained while the number of triangular or polygonal faces is reduced, so that it will not be broken or have lost or misaligned geometric elements.

[0081] Specifically, the geometric elements are processed as follows:

[0082] Level of Detail (LOD) Adjustment: Adjust the level of detail of the model to reduce the complexity of the model. LOD (Level of Detail) is a standard used in BIM models to describe the level of detail of the model.

[0083] Merge Elements: Merge multiple small geometric elements into one large element to reduce the number of elements in the model.

[0084] Delete unnecessary elements: Remove invisible or unimportant elements in the model, such as internal structures, temporary supports, etc.

[0085] Use proxy models: For complex elements, simple proxy models can be used instead to reduce the complexity of the model.

[0086] Non-geometric attributes are handled as:

[0087] Attribute data JSON conversion: Ensure that all non-geometric information, such as material, cost, schedule and other attribute data, is retained and translated into JSON format. JSON (JavaScript Object Notation) is a lightweight data exchange format.

[0088] Data mapping: A mapping relationship between geometric elements and original non-geometric attribute data is established through UDID so that complete information can be restored when needed. UDID (Unique Device Identifier) ​​refers to a unique device identifier.

[0089] After the above processing, we store both geometric and non-geometric information in a SQLite database and link the files through specific indexes. In order to maintain the best performance, we will also perform LZMA file format compression.

[0090] The process of importing two types of model data into the Unity3D engine is as follows:

[0091] Step 1: Import 3DGS data

[0092] Import the converted data file into the Unity3D project. In the Unity3D editor, you can use the file import function to add the 3DGS (.ply format) data file into the asset folder of the project.

[0093] Create a point cloud renderer: In Unity3D, you need to write or use an existing point cloud renderer to render 3D Gaussian splash. The point cloud renderer can be implemented using Unity's Mesh, Particle System, or Shader Graph.

[0094] Parameter configuration: According to the imported point cloud data file, configure and adjust the parameters of the point cloud renderer to ensure that the color, size, opacity and other characteristics of the points can be displayed and rendered correctly.

[0095] Performance optimization: Since point cloud data may contain a large number of points, rendering may need to be optimized, such as using LOD technology or block rendering to improve rendering performance and frame rate.

[0096] After the above steps, import 3DGS into Unity3D engine, such as Figure 5 shown.

[0097] Step 2: Import BIM data

[0098] Import the converted BIM model file (SQLite database) into the Unity3D project. In the Unity3D editor, use the file import function to add the BIM model file to the project's Asset folder.

[0099] Material and texture configuration: Add or adjust materials and textures for BIM models. In Unity3D, you can use the Material Editor and Shader Graph to configure appropriate materials for the model and apply textures to enhance the visual effect.

[0100] Model parsing: It is necessary to read and display the parameters and properties in the BIM model, write scripts to parse the attribute data of the BIM model, and display it in Unity3D.

[0101] Performance optimization: Optimize performance for large-scale BIM model projects. For example, using LOD technology, block loading, light mapping and other methods can improve rendering performance and frame rate.

[0102] After the above steps, import the BIM model into the Unity3D engine, such as Figure 6 shown.

[0103] The steps for matching 3DGS and BIM model space in Unity3D engine are as follows:

[0104] Step 1: Coordinate System Calibration

[0105] Check whether the coordinate systems of the two models are consistent. If not, they need to be converted to the same coordinate system. In the previous step, they have been imported into the Unity3D engine, so the coordinate systems are unified.

[0106] Step 2: Positioning and Alignment

[0107] In the software, use the alignment tools to manually or automatically align two models. This involves adjusting the position, rotation, and scale of the models.

[0108] Currently, alignment can be done visually using specific points or features in the model as reference points; it is also possible to use a reference model (such as a CAD drawing or field scan data) that can be imported as a reference to assist in alignment. Through scripting, it is also possible to create some automated alignment tools: these tools are usually based on feature matching or geometry matching algorithms.

[0109] Step 3: Adjust lighting and environment

[0110] Depending on the lighting and environment settings of your scene, adjust the lighting and shadows of the Gaussian Splatter effect to ensure that the effect is natural and blends in with the scene.

[0111] Step 4: Professional classification

[0112] According to the professional classification of BIM models, different layers are displayed and hidden, such as electromechanical models, architectural models, structural models, etc.

[0113] Step 5: Export the desktop application

[0114] Confirm that the display and interaction effects of the model are correct, and adjust and optimize as needed. Then use the application publishing function of Unity3D to export an executable file that can be executed by Windows PC, and superimpose the 3DGS and BIM models. Figure 7 shown.

[0115] By dynamically integrating 3DGS technology with BIM models, the present invention can superimpose and process multi-source data, accurately restore physical space, and achieve detailed interactive display of three-dimensional scenes. It can also assist in judging the feasibility of solutions, assist in engineering project management and decision-making, and ensure that project results meet expectations.

[0116] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A data processing method for spatial matching of 3DGS and BIM models, characterized in that: The steps include: Step 1: Perform three-dimensional reconstruction based on 3DGS technology, including: 1) Data acquisition: Acquire 3D point cloud data or multi-view image data for three-dimensional reconstruction; 2) Data generation: Perform rasterization, Gaussian weighting and color mixing processing on the acquired data, and generate a single-segment Gaussian splash after the processing is completed; 3) Data splicing: Perform data splicing on the generated single-segment Gaussian splash to form multiple-segment Gaussian splash, and export the spliced ​​Gaussian splash scene as a .ply format file; Step 2: BIM model generation, including: creating a BIM model using BIM modeling software, wherein the BIM model includes geometric elements and non-geometric attribute data, and processing the geometric elements and non-geometric attribute data into SQLite database data; Step 3: Import 3DGS data and BIM data into Unity3D engine; Step 4: Spatial matching of 3DGS and BIM models, including: 1) converting the coordinate systems of the two models into the same coordinate system; 2) aligning the two models; 3) adjusting the lighting and shadows of the Gaussian splash effect to ensure that the effect is natural and coordinated with the scene; 4) displaying and hiding the layers according to the professional classification of the BIM model; 5) confirming that the display and interaction effects of the model are correct, and exporting the executable file.

2. The method according to claim 1, characterized in that 1) Data acquisition in step 1 includes: acquiring the data through a depth sensor, a laser radar, a fisheye camera or multi-angle shooting; and representing each 3D space-aligned pixel point with a Gaussian distribution, defining its position (x, y, z) and covariance matrix.

3. The method according to claim 1, characterized in that The data generation in step 2) of step 1 includes rasterization: rendering each pixel point as a disk or a sphere, and the radius of the disk or the sphere is determined by the radius parameter of the point; Gaussian weighting: using a Gaussian function to weight the rendered disk or sphere so that its edge presents a smooth transition effect; color mixing: mixing the rendering result with the background image or color to generate the final image.

4. The method according to claim 1, characterized in that: 3) Data generation in step 1 includes data splicing, and the steps of data splicing are as follows: a. Determine the Gaussian splash model of the input scene. Multiple Gaussian splash scenes will be entered into the editor in .ply format; b. Spatial mapping of each scene: Translation: Assuming each scene has a center, reposition each scene by moving the center point in the coordinate system; Rotation: Rotate the scene through the rotation matrix; Translation transformation formula: x'=x+△x, y'=y+△y Rotation transformation formula: c. Processing gaps and overlapping areas: When stitching, the boundaries of the scenes will overlap or have gaps. If there are gaps, stitch them directly and stitch the scenes adjacent to each other. The boundary area will naturally form a transition. If there are overlaps, merge them and perform Gaussian weighted averaging in the overlapping area to make the transition smoother. The Gaussian weighted formula is as follows: G blend (x,y)=w1*G1(x,y)+w2*G2(x,y) d. Adjust the global scope: Boundary expansion: In order to accommodate the stitching of multiple scenes, the boundary range of the entire large scene needs to be adjusted to ensure that all Gaussian scenes can be covered; Interpolation and sampling: If the Gaussian splash scene is a discrete data point, use the interpolation method to expand this data to a larger spatial range; e. Generate the global Gaussian scene after stitching: Calculate the pixel value of the new scene: For each pixel point, calculate the value of all superimposed scenes. If there are multiple overlapping Gaussian splash points at a certain position, the final value is obtained by summing or weighted averaging; f. Post-processing and optimization: Smoothing: Use filters to smooth the stitched scene to eliminate sharp changes in transition areas; Brightness and contrast adjustment: Adjust the brightness and contrast of the entire scene to ensure consistent visual effects; g. Visualization and output: Output the stitched large scene: Save the stitched Gaussian splash scene as an image or data file, or use it for subsequent processing and analysis, and export it in .ply format.

5. The method according to claim 1, characterized in that The BIM model in step 2 includes geometric elements and non-geometric attribute data, wherein the geometric elements include: walls, floors and roofs, door and window components, beams, columns and structural frames, stairs, handrails and rods, furniture, equipment and other components; the non-geometric attributes include: material information, construction stage attributes, maintenance information, cost attributes, associated documents and file management.

6. The method according to claim 1, characterized in that The step 2 performs feature-preserving processing on the geometric elements of the BIM model, and converts the non-geometric attributes into structured Json data.

7. The method according to claim 6, characterized in that The geometric element processing includes: Detail level LOD adjustment: adjust the detail level of the model to reduce the complexity of the model; merge elements: merge multiple small geometric elements into a large element to reduce the number of elements in the model; delete unnecessary elements: remove invisible or unimportant elements in the model; use proxy models: use proxy models to replace complex elements; The non-geometric attribute processing includes: Attribute data Jsonization: ensure that all non-geometric attribute data are retained and translated into Json format; data mapping: establish a mapping relationship between geometric elements and original non-geometric attribute data through UDID to restore complete information.

8. The method according to claim 1, characterized in that The step 3 of importing 3DGS data includes: importing the converted data file into the Unity3D project, and in the Unity3D editor, adding the .ply format file exported by 3DGS to the Asset folder of the project; using a point cloud renderer to render three-dimensional Gaussian splash; then configuring and adjusting the parameters of the point cloud renderer according to the imported point cloud data file; and finally performing performance optimization, using LOD technology or block rendering to optimize the large number of points contained in the point cloud data.

9. The method according to claim 1, characterized in that: The step 3 of importing BIM data includes: importing the converted BIM model file SQLite database into the Unity3D project, and in the Unity3D editor, adding the SQLite database to the Asset folder of the project; using the material editor and Shader Graph to configure materials for the model, and applying textures for visual enhancement; reading and displaying parameters and properties in the BIM model, writing scripts to parse the attribute data of the BIM model, and displaying it in Unity3D; using LOD technology, block loading, and light mapping to improve rendering performance and frame rate.

10. The method according to claim 1, characterized in that The step 4 (2) aligning the two models includes: manually or automatically aligning the two models using an alignment tool, and aligning the two models by adjusting the position, rotation and scaling of the models.

Images