A virtual assembly method of fabricated steel structure based on BIM platform

Abstract

The application discloses a kind of based on BIM platform's fabricated steel structure virtual assembly method, it is related to virtual assembly technical field, can solve the technical problem of model boundary distortion, insufficient precision in three-dimensional model reconstruction and virtual assembly process, including: obtaining the original point cloud data set of steel structure;According to the spatial distribution characteristics of adjacent point cloud, the structure distribution synchronism of each point cloud data is determined;According to the distance distribution characteristics of adjacent point cloud to fitting plane and structure distribution synchronism, the spatial plane synchronism of each point cloud data is determined;According to the direction correlation between normal vector and the unit vector of spatial vertical direction and spatial plane synchronism, the platform structure of each point cloud data is determined Understand degree, and accordingly determine the pure point cloud data set of steel structure from original point cloud data set;According to pure point cloud data set, steel structure three-dimensional model is obtained by reconstruction, and steel structure three-dimensional model is assembled with BIM building model in virtual environment.

Description

Technical Field

[0001] This invention relates to the field of virtual assembly technology, and more specifically to a virtual assembly method for prefabricated steel structures based on a BIM platform. Background Technology

[0002] Building Information Modeling (BIM) technology has significantly improved the efficiency and precision of collaborative design, construction, and management of building projects by creating and managing digital information models. Among these technologies, virtual assembly technology can simulate the assembly process of steel structures in digital space before actual construction. This allows for interference detection, verification of design rationality, and optimization of construction plans, which is crucial for ensuring the quality, controlling costs, and managing schedules of large and complex steel structure projects.

[0003] In the virtual assembly process, one of the core steps is to acquire and process the actual 3D point cloud data of the steel structure to reconstruct an accurate digital model. Existing technologies typically use 3D laser scanning to acquire the original point cloud containing the steel structure and its supporting platform (such as an inspection platform), and attempt to separate the background using methods such as pass-through filtering to extract the pure steel structure point cloud. However, steel structure components in engineering projects have diverse shapes and complex surfaces, while the supporting platform is often a planar structure. Traditional background separation methods based on fixed thresholds are difficult to dynamically adapt to such complex scenarios. Summary of the Invention

[0004] To address the technical problem of model boundary distortion and insufficient accuracy in subsequent 3D model reconstruction and virtual assembly caused by using a fixed height threshold for point cloud background separation of complex steel structures and their supporting platforms, this invention aims to provide a virtual assembly method for prefabricated steel structures based on a BIM platform. The specific technical solution adopted is as follows: In a first aspect, the present invention provides a virtual assembly method for prefabricated steel structures based on a BIM platform, comprising: acquiring an original point cloud dataset of the steel structure; determining the structural distribution synchronicity of each point cloud data based on the spatial distribution characteristics of neighboring point clouds in the original point cloud dataset; wherein, structural distribution synchronicity is used to characterize the consistency of the normal vector direction within the local neighborhood of the point cloud data and the orthogonality of the connection relationship between point clouds; determining the spatial plane synchronicity of each point cloud data based on the distance distribution characteristics of neighboring point clouds to the fitting plane and the structural distribution synchronicity; wherein, spatial plane synchronicity is used to characterize the degree to which the local neighborhood of the point cloud data approaches the same plane and the consistency of its distribution direction; determining the platform structural commonality of each point cloud data based on the directional correlation between the normal vector and the unit vector in the vertical direction of space and the spatial plane synchronicity, and determining the pure point cloud dataset of the steel structure from the original point cloud dataset based on the platform structural commonality; reconstructing a three-dimensional model of the steel structure based on the pure point cloud dataset of the steel structure, and assembling the three-dimensional model of the steel structure with the BIM building model in a virtual environment.

[0005] Secondly, the present invention provides a virtual assembly system for prefabricated steel structures based on a BIM platform, comprising: a data acquisition module, a structural distribution identification module, a spatial plane identification module, a platform structure identification module, and a model reconstruction assembly module. The system comprises the following modules: a data acquisition module for acquiring the original point cloud dataset of the steel structure; a structural distribution identification module for determining the structural distribution synchronization of each point cloud dataset based on the spatial distribution characteristics of its neighboring point clouds; where structural distribution synchronization characterizes the consistency of the normal vector direction within the local neighborhood of the point cloud dataset and the orthogonality of the connection relationships between point clouds; a spatial plane identification module for determining the spatial plane synchronization degree of each point cloud dataset based on the distance distribution characteristics of its neighboring point clouds to the fitting plane and the structural distribution synchronization; where spatial plane synchronization characterizes the degree to which the local neighborhood of the point cloud dataset approaches the same plane and the consistency of its distribution direction; a platform structure identification module for determining the platform structure commonality of each point cloud dataset based on the directional correlation between the normal vector and the unit vector in the vertical direction of space and the spatial plane synchronization degree, and determining the pure point cloud dataset of the steel structure from the original point cloud dataset based on the platform structure commonality; and a model reconstruction and assembly module for reconstructing a 3D model of the steel structure based on the pure point cloud dataset of the steel structure and assembling the 3D model of the steel structure with the BIM building model in a virtual environment.

[0006] Thirdly, the present invention provides an electronic device, comprising: a processor and a memory; wherein the memory is used to store one or more programs, the one or more programs including computer-executable instructions, and when the electronic device is running, the processor executes the computer-executable instructions stored in the memory to cause the electronic device to perform the virtual assembly method for prefabricated steel structures based on a BIM platform as described in the first aspect and any possible implementation thereof.

[0007] This invention offers the following advantages: By calculating and fusing the structural distribution synchronization and spatial planar synchronization of local geometric distribution and directional features, and further combining vertical constraints to construct platform structural commonality, high-precision and adaptive separation of point clouds of complex steel structures and simple horizontal background platforms is achieved. This method overcomes the limitations of traditional fixed threshold filtering in the face of complex shapes, effectively eliminates background interference, and ensures the boundary integrity and accuracy of subsequent 3D reconstruction models, thereby improving the reliability and practicality of BIM-based virtual assembly results. Attached Figure Description

[0008] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic diagram of the architecture of a virtual assembly system for prefabricated steel structures based on a BIM platform, provided in one embodiment of the present invention. Figure 2 This is a flowchart illustrating a virtual assembly method for prefabricated steel structures based on a BIM platform, as provided in one embodiment of the present invention. Detailed Implementation

[0010] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0011] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0012] The following description, in conjunction with the accompanying drawings, details a specific scheme for a virtual assembly method for prefabricated steel structures based on a BIM platform provided by the present invention.

[0013] For example, such as Figure 1 The diagram shown is an architectural schematic of a virtual assembly system for prefabricated steel structures based on a BIM platform (hereinafter referred to as virtual assembly system 10) according to an embodiment of the present invention. The virtual assembly system 10 includes: a data acquisition module 11, a structural distribution identification module 12, a spatial plane identification module 13, a platform structure identification module 14, and a model reconstruction and assembly module 15. The modules are described in detail below: (1) Data acquisition module 11.

[0014] The data acquisition module 11 is responsible for collecting the three-dimensional geometric information of the steel structure in the real physical space, providing the most basic raw data for the entire virtual assembly process. This data acquisition module 11 interacts with the hardware scanning device to complete the data collection and initial aggregation.

[0015] Optionally, the data acquisition module 11 is used to acquire the original point cloud dataset of the steel structure.

[0016] Specifically, the data acquisition module 11 controls a 3D laser scanning device to perform a surround scan of the steel structure to be assembled. During the scan, the steel structure is placed on a stable support platform. The point cloud data collected by the device includes points on the surface of the steel structure and background points on the support platform below, which together form an initial point cloud set, i.e., the original point cloud dataset. The data acquisition module 11 outputs this dataset to the structure distribution recognition module 12 to initiate subsequent processing.

[0017] (2) Structure distribution identification module 12.

[0018] The structure distribution recognition module 12 is responsible for receiving the original point cloud dataset and analyzing the geometric distribution pattern of each point cloud data in its local neighborhood, quantifying the consistency characteristics of its structure, and providing the first dimension of discrimination criteria for distinguishing between the complex steel structure and the simple background platform.

[0019] Optionally, the structure distribution recognition module 12 is used to determine the structural distribution synchronization of each point cloud data based on the spatial distribution characteristics of the neighboring point clouds of each point cloud data in the original point cloud dataset. The structural distribution synchronization characterizes the consistency of the normal vector directions within the local neighborhood of the point cloud data and the orthogonality of the connectivity relationships between point clouds.

[0020] Specifically, the structure distribution identification module 12 first finds the set of neighboring points for each point cloud data. Next, it estimates the normal vectors of the point cloud data and its neighboring points. Then, based on the angle between the normal vector of the point cloud data and the direction of the lines connecting it to its neighboring points, the module calculates a first feature parameter reflecting orthogonality. Simultaneously, based on the directional similarity between the normal vector of the point cloud data and the normal vectors of its neighboring points, it calculates a second feature parameter reflecting parallelism. Finally, the module combines the first and second feature parameters to generate a structural distribution synchronization value for the point cloud data. A higher value indicates that the local region containing the point is closer to a simple, uniformly oriented plane.

[0021] The structure distribution recognition module 12 outputs the calculated structure distribution of each point cloud data to the spatial plane recognition module 13 in sync.

[0022] (3) Spatial plane recognition module 13.

[0023] The spatial plane recognition module 13 is responsible for combining the local plane fitting degree of the point cloud with the structural distribution synchronization to further evaluate the plane consistency strength of the local region to which each point cloud data belongs in space.

[0024] Optionally, the spatial plane recognition module 13 is used to determine the spatial plane synchronization degree of each point cloud data based on the distance distribution characteristics of neighboring point clouds to the fitting plane and the structural distribution synchronization of each point cloud data. The spatial plane synchronization degree characterizes the degree to which the local neighborhoods of the point cloud data converge to the same plane and the consistency of their distribution directions.

[0025] Specifically, the spatial plane recognition module 13 performs plane fitting on the neighboring point cloud set for each point cloud data to obtain a fitting plane representing the best fit for that local region. Then, the spatial plane recognition module 13 calculates the distances from all points in the neighboring point cloud set to the fitting plane and calculates a plane convergence parameter based on the statistical characteristics of these distances (such as the ratio of the mean to the global maximum). Finally, the spatial plane recognition module 13 multiplies the plane convergence parameter with the structural distribution synchronization from the structural distribution recognition module 12 to obtain the spatial plane synchronization degree of the point cloud data. The spatial plane synchronization degree, as a fusion feature, combines local geometry with directional distribution consistency; a high value strongly suggests that the point may be located on a large plane with consistent orientation.

[0026] The spatial plane recognition module 13 outputs the spatial plane synchronization of all point cloud data to the platform structure recognition module 14.

[0027] (4) Platform structure identification module 14.

[0028] The platform structure recognition module 14 is responsible for comprehensively utilizing spatial planar synchronization and vertical constraints to accurately identify the background point cloud belonging to the horizontal bearing platform, and performing filtering operations to finally output pure point cloud data containing only the steel structure.

[0029] Optionally, the platform structure identification module 14 is used to determine the platform structure generality of each point cloud data based on the directional correlation between the normal vector of each point cloud data and the unit vector in the vertical direction of space, as well as the spatial plane synchronization degree, and to determine the pure point cloud dataset of the steel structure from the original point cloud dataset based on the platform structure generality.

[0030] For example, the platform structure recognition module 14, based on its internal workflow, can be divided into three collaborative sub-modules: a general knowledge calculation sub-module 141, a background segmentation sub-module 142, and an adaptive filtering sub-module 143. The three sub-modules are described below: (4.1) General knowledge calculation submodule 141.

[0031] Optionally, the general knowledge calculation submodule 141 is used to calculate the platform structure general knowledge of each point cloud data.

[0032] Specifically, the general knowledge calculation submodule 141 first calculates the cosine similarity between the normal vector of each point cloud data and the unit vector in the vertical direction of space, obtaining the directional correlation parameter. The higher the directional correlation parameter, the closer the local plane where the point is located is to being horizontal. Then, the general knowledge calculation submodule 141 multiplies the directional correlation parameter with the spatial plane synchronization degree from the spatial plane recognition module 13 to obtain the final platform structure general knowledge. The platform structure general knowledge integrates three key criteria: "whether it is on a plane", "whether the planes are oriented in the same direction", and "whether the planes are horizontal", and is a powerful indicator for identifying the background platform.

[0033] (4.2) Background segmentation submodule 142.

[0034] Optionally, the background segmentation submodule 142 is used to determine the background point cloud dataset based on platform architecture knowledge.

[0035] Specifically, the background segmentation submodule 142 first performs statistical analysis on the platform structure commonality of all point cloud data, automatically calculating a segmentation threshold using methods such as Otsu's inter-class variance. Then, it filters point clouds with commonality higher than this threshold, forming a first candidate point cloud set, which contains all points suspected to be on the horizontal plane. Finally, the background segmentation submodule 142 performs density-based spatial clustering on the first candidate point cloud set. Before clustering, it adaptively determines the neighborhood radius of the clustering algorithm based on the average Euclidean distance between point clouds. After clustering, multiple point cloud clusters are obtained. The background segmentation submodule 142 calculates the average height of all points in each cluster in the vertical direction and determines the cluster with the smallest average height as the surface of the platform, i.e., the background point cloud dataset.

[0036] (4.3) Adaptive filtering submodule 143.

[0037] Optionally, the adaptive filtering submodule 143 is used to filter out the background from the background point cloud dataset to obtain a clean point cloud dataset.

[0038] Specifically, the adaptive filtering submodule 143 receives the background point cloud dataset determined by the background segmentation submodule 142 and performs plane fitting on it to obtain a reference plane that accurately describes the spatial position of the platform. Since the platform may not be perfectly level, this reference plane is tilted in space. Based on this reference plane, the adaptive filtering submodule 143 derives a dynamic height threshold that varies with spatial position, and performs pass-through filtering on the original point cloud dataset using this threshold, removing all point clouds located near the reference plane (i.e., the background), and finally outputs a clean point cloud dataset of the steel structure.

[0039] The platform structure recognition module 14 outputs the pure point cloud dataset to the model reconstruction and assembly module 15.

[0040] (5) Model reconstruction assembly module 15.

[0041] The model reconstruction and assembly module 15 is responsible for post-processing the clean steel structure point cloud, reconstructing the 3D model, and finally completing the precise assembly in the BIM virtual environment.

[0042] Optionally, the model reconstruction and assembly module 15 is used to reconstruct a three-dimensional model of the steel structure based on the pure point cloud dataset of the steel structure, and assemble the three-dimensional model of the steel structure with the BIM building model in a virtual environment.

[0043] For example, the model reconstruction and assembly module 15 can be divided into two sub-modules: the point cloud optimization sub-module 151 and the assembly execution sub-module 152. These are described below: (5.1) Point cloud optimization submodule 151.

[0044] Optionally, the point cloud optimization submodule 151 is used to process the clean point cloud dataset and reconstruct the 3D model.

[0045] Specifically, the point cloud optimization submodule 151 first performs statistical filtering on the clean point cloud dataset to remove possible outlier noise points. Then, it downsamples the denoised point cloud to reduce the data volume and improve processing efficiency while preserving model features. Finally, it uses algorithms such as Poisson surface reconstruction to reconstruct a continuous, watertight 3D surface model of the steel structure from the processed point cloud.

[0046] (5.2) Assembly execution submodule 152.

[0047] Optionally, the assembly execution submodule 152 is used to virtually assemble the reconstructed model with the BIM model.

[0048] Specifically, the assembly execution submodule 152 first performs iterative nearest-point registration and comparison between the reconstructed 3D steel structure model and a standard-sized 3D steel structure model to verify whether its manufacturing dimensions meet the accuracy requirements. After successful verification, the assembly execution submodule 152 uses the predetermined target assembly position in the BIM building model as a reference and again utilizes the registration algorithm to calculate the precise position and orientation of the reconstructed 3D steel structure model in the virtual scene, i.e., the assembly pose. Finally, based on this pose, the assembly execution submodule 152 assembles the 3D steel structure model with the overall building information model in the BIM platform, completing the entire virtual assembly process, thereby verifying the feasibility of the design and the accuracy of the assembly before actual construction.

[0049] The above provides an introduction to the BIM-based virtual assembly system for prefabricated steel structures 10 and its included modules.

[0050] For example, such as Figure 2 The diagram shown is a flowchart illustrating a virtual assembly method for prefabricated steel structures based on a BIM platform, according to an embodiment of the present invention. The method includes the following steps: S201. Obtain the original point cloud dataset of the steel structure.

[0051] For example, this step can be performed by the data acquisition module 11 in the virtual assembly system 10 described above.

[0052] Specifically, the data acquisition module 11 controls a 3D laser scanning device to scan the steel structure to be assembled, which is placed on the support platform (i.e., the inspection platform). During the scanning process, the device collects the 3D spatial coordinates of the steel structure surface and the surface of the inspection platform below it from multiple angles. These coordinate points combine information about the target object and the background environment. The data acquisition module 11 integrates all the collected point cloud data to form an initial, unprocessed raw point cloud dataset, which is then output to subsequent modules as the starting point for processing.

[0053] S202. Based on the spatial distribution characteristics of the neighboring point clouds of each point cloud data in the original point cloud dataset, determine the structural distribution synchronization of each point cloud data. Structural distribution synchronization characterizes the consistency of the normal vector directions within the local neighborhood of the point cloud data and the orthogonality of the connectivity relationships between point clouds.

[0054] For example, this step can be performed by the structure distribution recognition module 12 in the virtual assembly system 10 described above. Specifically, the structure distribution recognition module 12 receives the original point cloud dataset. For each point cloud data, it first determines its neighboring point cloud set in space. Next, it estimates the normal vectors of the point cloud data and each of its neighboring point cloud data. Then, the structure distribution recognition module 12 analyzes the angular relationship between the normal vector of the point cloud data and the direction of the line connecting it to each neighboring point cloud data, calculating a first feature parameter reflecting local orthogonality. Simultaneously, it analyzes the directional similarity between the normal vector of the point cloud data and the normal vectors of each neighboring point cloud data, calculating a second feature parameter reflecting directional consistency. Finally, the structure distribution recognition module 12 combines the first feature parameter and the second feature parameter to determine the structural distribution synchronization of the point cloud data. It should be noted that the specific process for determining the structural distribution synchronization of each point cloud data is described in S301-S304 below, and will not be repeated here.

[0055] In another possible implementation, when determining the structural distribution synchronization of each point cloud data, the structural distribution identification module 12 can also directly perform principal component analysis on the neighboring point cloud set of each point cloud data, take the eigenvector corresponding to the minimum eigenvalue as the approximation of the normal vector, and characterize the directional consistency by calculating the variance between the normal vectors of all neighboring points. At the same time, the average distance from the point cloud to the local least squares fitting plane is used to indirectly reflect the orthogonality. Finally, the two indices are normalized and multiplied to obtain the structural distribution synchronization.

[0056] Alternatively, in another possible implementation, the structure distribution identification module 12 can also calculate the Gaussian curvature and average curvature of the local neighborhood of each point cloud data, and use the statistical distribution of curvature (such as variance) as a measure of structural complexity. The simpler the structure and the closer it is to a plane, the smaller the curvature variance, and the higher the synchronicity of the mapped structural distribution.

[0057] Therefore, the structure distribution identification module 12 assigns a feature index to each point cloud data by quantifying the consistency of the normal vector direction in the local neighborhood of each point cloud and the orthogonality of its connection with neighboring points, which can effectively distinguish between the complex steel structure surface and the simple background platform surface.

[0058] S203. Based on the distance distribution characteristics of neighboring point clouds to the fitting plane and the structural distribution synchronicity of each point cloud data, determine the spatial plane synchronization degree of each point cloud data. The spatial plane synchronization degree characterizes the degree to which the local neighborhoods of the point cloud data converge to the same plane and the consistency of their distribution directions.

[0059] For example, this step can be performed by the spatial plane recognition module 13 in the virtual assembly system 10 described above. Specifically, the spatial plane recognition module 13 performs plane fitting on each point cloud data set and its neighboring point cloud sets to obtain the best-fitting plane characterizing the local region. Then, the spatial plane recognition module 13 calculates the distances from all point cloud data sets in the neighboring point cloud sets to the fitted plane, and calculates a plane convergence parameter based on the statistical characteristics of these distances (such as the relationship between the mean and the global maximum). Finally, the spatial plane recognition module 13 combines the calculated plane convergence parameter with the point structure distribution synchronization received from the structure distribution recognition module 12 to determine the spatial plane synchronization degree of the point cloud data. It should be noted that the specific process for determining the spatial plane synchronization degree of each point cloud data set is described in S401-S403 below, and will not be repeated here.

[0060] In another possible implementation, when determining the spatial plane synchronization degree of each point cloud data, the spatial plane identification module 13 can also iteratively sample and fit planes from the neighboring point cloud set, and finally select the plane with the most interior points and the smallest average distance from the interior points to the plane as the fitting result, and then combine the structural distribution synchronization to calculate the spatial plane synchronization degree. This method has better robustness to outliers in the neighboring point cloud set.

[0061] Alternatively, in another possible implementation, when calculating the plane convergence, the spatial plane identification module 13 not only considers the average distance, but also introduces the sum of squared residuals or the coefficient of determination (R²) of the fitted plane as a measure of goodness of fit. The goodness of fit is weighted and combined with the average distance, and then combined with the synchronicity of the structural distribution to form the spatial plane synchronization degree.

[0062] Therefore, the spatial plane recognition module 13 obtains a more powerful comprehensive feature—spatial plane synchronization—by fusing the distance distribution characteristics of neighboring point clouds to the fitted plane and the structural distribution synchronization reflecting the consistency of local orientation of each point cloud data. The higher this feature value, the more likely the point is to be located on a plane with consistent orientation and flatness, which provides a key basis for the subsequent accurate identification of the horizontal detection platform.

[0063] S204. Based on the directional correlation between the normal vector of each point cloud data and the unit vector in the vertical direction of space, as well as the spatial plane synchronization degree, determine the platform structure commonality of each point cloud data, and determine the pure point cloud dataset of the steel structure from the original point cloud dataset based on the platform structure commonality.

[0064] For example, this step can be performed by the platform structure recognition module 14 in the virtual assembly system 10 described above. Specifically, the platform structure recognition module 14 first calculates the direction cosine similarity between the normal vector of each point cloud data and the unit vector in the vertical direction (Z-axis) of space to obtain the direction correlation parameter. Secondly, it multiplies this direction correlation parameter by the spatial plane synchronization degree of the point received from the spatial plane recognition module 13 to obtain the platform structure recognition degree of the point cloud data. Thirdly, based on the platform structure recognition degree of all point cloud data, the platform structure recognition module 14 determines the segmentation threshold using a preset threshold segmentation algorithm, and accordingly initially divides the point cloud into candidate sets. Then, it performs density-based spatial clustering on the candidate sets and filters out the background point cloud dataset representing the detection platform based on the spatial height characteristics of each cluster. Finally, it fits a reference plane based on the background point cloud dataset and uses the dynamic height threshold defined by this plane to perform pass-through filtering on the original point cloud dataset, filtering out the background point cloud and outputting a clean point cloud dataset of the steel structure. It should be noted that the specific process for determining the pure point cloud dataset of the steel structure is described in S401-S405 below, and will not be repeated here.

[0065] In another possible implementation, after determining the clean point cloud dataset of the steel structure, the platform structure recognition module 14 can further perform morphological closing operations on the background point cloud to fill small holes that may be caused by noise, after initially segmenting the background platform point cloud through thresholding and clustering. Then, it extracts the largest connected component as the final background region, and performs plane fitting and filtering. This method can further improve the completeness and robustness of background region extraction.

[0066] Alternatively, in another possible implementation, the platform structure recognition module 14 not only fits the plane of the entire background platform on a large scale, but also divides the point cloud into blocks and fits local planes on multiple smaller scales. Then, it comprehensively evaluates the consistency of the normal vectors and distances between each local plane and the global large plane, as well as the spatial plane synchronization of the point cloud in that local area. It uses voting or weighting to more finely determine whether each point belongs to the background, which is especially suitable for detecting scenarios where there are small undulations or stains on the platform surface.

[0067] Therefore, the platform structure recognition module 14, by introducing vertical constraints and combining them with spatial plane synchronization, constructs a platform structure general recognition index, which is specifically used to identify horizontally bearing platforms, providing a basis for subsequent high-precision 3D reconstruction.

[0068] S205. Based on the pure point cloud dataset of the steel structure, reconstruct the three-dimensional model of the steel structure, and assemble the three-dimensional model of the steel structure with the BIM building model in a virtual environment.

[0069] For example, this step can be performed by the model reconstruction assembly module 15 in the virtual assembly system 10 described above, and specifically includes the following steps: (1) Denoising and filtering are performed on the clean point cloud dataset, and downsampling is performed on the denoised and filtered clean point cloud dataset.

[0070] Optionally, this sub-step can be performed by the point cloud optimization sub-module 151 in the model reconstruction assembly module 15.

[0071] Specifically, the point cloud optimization submodule 151 performs statistical filtering on the received clean point cloud dataset of the steel structure. By analyzing the distance distribution between each point and its neighbors, outliers with excessively large distance standard deviations are removed to eliminate noise data introduced during the scanning process. Then, the point cloud optimization submodule 151 performs voxel mesh downsampling on the denoised point cloud data. This involves creating a uniform voxel mesh in three-dimensional space and using the centroid (or the first point) of all points within each voxel to represent that voxel. This significantly reduces the total amount of point cloud data while effectively preserving the macroscopic geometric features of the steel structure, thereby improving the efficiency of subsequent processing.

[0072] (2) Calculate the downsampled pure point cloud dataset according to the preset surface reconstruction algorithm to generate a continuous three-dimensional surface model of the steel structure.

[0073] For example, the point cloud optimization submodule 151 inputs the downsampled point cloud data into the Poisson surface reconstruction algorithm. This algorithm reconstructs the discrete point cloud data into a continuous, closed triangular mesh surface model by constructing a spatial octree structure of the point cloud, calculating the indicator function, and extracting its isosurface, thereby obtaining a high-fidelity three-dimensional surface model of the steel structure.

[0074] (3) Register and compare the three-dimensional model of the steel structure with the standard three-dimensional model of the steel structure to determine whether the dimensions of the three-dimensional model of the steel structure conform to the verification results.

[0075] Optionally, this sub-step can be performed by the assembly execution sub-module 152 in the model reconstruction assembly module 15. Specifically, the assembly execution sub-module 152 registers the reconstructed three-dimensional steel structure model with a preset three-dimensional steel structure model having standard design dimensions.

[0076] For example, the assembly execution submodule 152 employs an iterative nearest-point algorithm to minimize the average distance between corresponding points of two models through iterative calculation, thereby achieving precise positioning. By calculating the residual distance between corresponding points of the two models after registration and comparing it with a preset tolerance threshold, it is determined whether the manufacturing dimensions of the reconstructed model meet the accuracy requirements, and the verification result is output.

[0077] (4) If the verification result is passed, the assembly pose of the steel structure three-dimensional model in the BIM building model is calculated by using a preset registration algorithm based on the target assembly position information in the BIM building model.

[0078] In this step, after receiving the instruction that the dimensions have been verified, the assembly execution submodule 152 reads the predefined target installation position and theoretical orientation (usually in the form of coordinates and rotation matrices) of the steel structure component from the BIM building information model. Following this, the assembly execution submodule 152 again calls the iterative nearest-point algorithm, using the local structure of the BIM model at the target installation position as a reference, to calculate the translation and rotation parameters required for the reconstructed 3D steel structure model to accurately align with it, i.e., the final assembly orientation of the component in the overall virtual building environment.

[0079] (5) Based on the assembly posture, complete the assembly of the three-dimensional steel structure model and the BIM building model in the virtual environment.

[0080] Furthermore, the assembly execution submodule 152 applies the calculated assembly pose (translation vector and rotation matrix) to the reconstructed 3D steel structure model, driving it to move and rotate to a specified position within the BIM building model. Finally, in the virtual 3D scene, the system visually integrates and renders the positioned steel structure model with surrounding BIM model components such as beams, columns, and slabs, thus completing the final presentation of the entire virtual assembly process for interference checks, process simulation, and scheme verification.

[0081] Thus, the model reconstruction and assembly module 15, through a series of operations such as point cloud post-processing, surface reconstruction, dimension verification and high-precision pose registration, realizes the transformation from discrete point cloud to solid 3D model that can be used for engineering analysis, and finally accurately integrates it into the overall BIM digital environment, completing a complete and reliable digital workflow from physical entity scanning to virtual space assembly.

[0082] Based on the above technical solution, this invention achieves high-precision, adaptive separation of point clouds between complex steel structures and simple horizontal background platforms by calculating and fusing the structural distribution synchronization and spatial planar synchronization of local geometric distribution and directional features, and further combining vertical constraints to construct platform structural recognizability. This method overcomes the limitations of traditional fixed threshold filtering in the face of complex shapes, effectively eliminates background interference, ensures the boundary integrity and accuracy of the subsequent 3D reconstruction model, and thus improves the reliability and practicality of BIM-based virtual assembly results.

[0083] For example, in another virtual assembly method for prefabricated steel structures based on a BIM platform provided in one embodiment of the present invention, the structural distribution synchronization of each point cloud data in the original point cloud dataset is determined according to the spatial distribution characteristics of neighboring point clouds. This specifically includes the following steps: S301. Determine the nearest point cloud set for each point cloud data in the original point cloud dataset.

[0084] Specifically, the structure distribution recognition module 12 targets each point cloud data point in the original point cloud dataset. A fast nearest neighbor search algorithm based on spatial partitioning (such as the KD-Tree algorithm) is used to search for the nearest neighbor in the entire dataset. The point cloud data consists of the K nearest points in Euclidean space. Neighborhood cluster For example, the number of neighboring points K can be preset to 30. This value is based on experience and aims to balance the accuracy of describing local geometric features with computational efficiency: too few points may not stably represent the local structure, while too many points will introduce more interference from irrelevant regions and increase the computational burden.

[0085] S302. Determine the first feature parameter based on the normal vector of each point cloud data and the angle relationship between the direction of the line connecting each point cloud data to each point cloud data in the neighboring point cloud set.

[0086] In this step, the structure distribution recognition module 12 first calculates the point cloud data. Gathering at its neighboring points Each point in The direction vector of the line connecting them.

[0087] In addition, the structure distribution recognition module 12 calculates point cloud data. normal vector For example, regarding point cloud data and its neighboring points cloud cluster Principal component analysis was used to calculate the covariance matrix of the point set, and eigenvalue decomposition was performed on the covariance matrix. The eigenvector corresponding to the smallest eigenvalue was taken as the point cloud data. normal vector Approximate to .

[0088] After that, calculate the point cloud data. normal vector direction vector of each line (from point to The angle between ) Based on the aforementioned included angle, the structure distribution recognition module 12 calculates the first feature parameter using the following formula: In the above formula, Representing point cloud data The first feature parameter, K, represents the number of elements in the neighboring point cloud set (i.e., the preset value in S301 mentioned above, such as 30).

[0089] It should be noted that the above formula quantifies the orthogonality of local geometric relationships. Ideally, if Located on a smooth surface, its normal vector should be perpendicular to the direction of the line connecting many neighboring points (i.e., the included angle is θ). The formula calculates each included angle. and The absolute deviation, and used Subtract this deviation, and then perform normalization and summation. When most of the included angles are close to 90 degrees, the absolute value of the deviation is small, and the numerator term... The value will be larger after summing. A larger value indicates that the normal vector in the local neighborhood of that point tends to be orthogonal to the line connecting the points, and the structure distribution has a high degree of flatness. Conversely, in regions with complex structures and high curvature, The value is relatively small.

[0090] S303. Determine the second feature parameter based on the directional similarity between the normal vector of the point cloud data and the normal vectors of each point cloud data in the neighboring point cloud sets.

[0091] Furthermore, the structure distribution recognition module 12 calculates the point cloud data. normal vector It is clustered with neighboring points Each point in normal vector The degree of directional similarity between them. Here, cosine similarity is used as the metric, and their sum is calculated as a fundamental component of the second feature parameter.

[0092] Specifically, the sum of the above That is, the unnormalized form of the second characteristic parameter, where This represents the cosine similarity between the two vectors within the parentheses.

[0093] S304. Determine the structural distribution synchronization of the point cloud data based on the first feature parameter and the second feature parameter.

[0094] For example, the structure distribution recognition module 12, combining the first and second feature parameters obtained from the aforementioned calculations, calculates the structural distribution synchronization of the point cloud data using the following formula: In the above formula, Representing point cloud data Synchronicity of structural distribution Representing point cloud data The first characteristic parameter, Representing point cloud data The second characteristic parameter (unnormalized) is K, which represents the number of elements in the nearest point cloud set.

[0095] It should be noted that the structural distribution synchronization in the formula It is a characteristic of orthogonality of direction ( The formula combines and averages the features of consistency with the normal vector direction (cosine similarity sum). It fuses the two features through multiplication: only when... The local neighborhood also exhibits high orthogonality. When the direction of the normal vector is consistent (large) and has a high similarity, The value of is high only then. This precisely characterizes the definition of "structural distribution synchronicity"—the normal vectors in the local neighborhood are aligned and the connections between point clouds tend to be orthogonal. Such regions are very likely to correspond to a simple, near-planar background platform surface. Conversely, on complex steel structure surfaces, both eigenvalues ​​are usually low, or one of them is low, leading to The value is relatively small. Dividing by K is to make the result independent of the number of neighboring points, forming a standardized metric.

[0096] Based on the above technical solution, this embodiment of the invention quantitatively calculates a local geometric feature descriptor called "structural distribution synchronization" for each point cloud data. This descriptor innovatively integrates information from two dimensions: the consistency of normal vector direction and the orthogonality of inter-point connections. It can accurately distinguish between the complex and varied surfaces of steel structures and the simple, flat surfaces of background platforms. This provides a crucial and reliable basis for accurately identifying and filtering background point clouds in subsequent processes, and is an important foundation for the entire method to achieve high-precision virtual assembly.

[0097] For example, in another virtual assembly method for prefabricated steel structures based on a BIM platform provided in one embodiment of the present invention, for each point cloud data, the spatial plane synchronization degree of each point cloud data is determined according to the distance distribution characteristics of neighboring point clouds to the fitting plane and the structural distribution synchronization, specifically including the following steps: S401. Fit the neighboring point cloud set of each point cloud data to obtain the fitting plane.

[0098] Specifically, the spatial plane recognition module 13 performs each point cloud data... Neighborhood cluster (Determined by the aforementioned step S301) Perform plane fitting to find a spatial plane that best represents the distribution trend of the local point set.

[0099] For example, this embodiment uses the least squares method for fitting. This algorithm solves for the optimal plane equation parameters (typically expressed as...) by minimizing the sum of the squares of the perpendicular distances from all points in the neighboring point cloud to the plane to be determined. Ax + By + Cz + D =0 (normal vector and intercept). Finally, the spatial plane recognition module 13 performs a process for each point cloud data. The calculation yields a unique plane that represents the best fit in its local neighborhood, known as the fitting plane.

[0100] S402. Calculate the distance from all point cloud data in the neighboring point cloud set to the fitting plane, and determine the plane proximity parameter for each point cloud data.

[0101] Furthermore, the spatial plane recognition module 13 performs a certain function for each point cloud data. Calculate its nearest neighbor cloud. Each point in vertical distance to the fitted plane Then, based on this set of distances { | j =1,2,…, K}, calculate its mean Next, the spatial plane recognition module 13 combines a global reference value—the maximum value among the plane distances of neighboring point cloud sets corresponding to all point cloud data (traversing the entire original point cloud dataset), denoted as Ultimately, point cloud data Plane approach parameter = .

[0102] S403. Determine the spatial plane synchronization degree of each point cloud data based on the plane proximity parameter and the synchronization of the structure distribution.

[0103] For example, the spatial plane recognition module 13 calculates point cloud data using the following formula. Spatial planar synchronization: In the above formula, Representing point cloud data Spatial planar synchronization This indicates the point cloud data determined by S402. The plane proximity parameter, This indicates the point cloud data determined by S304. The structural distribution synchronization.

[0104] It should be noted that the spatial plane synchronization degree in the formula It is the product of planar convergence and structural distribution synchronicity. This computational design integrates two key types of information: 1. Planar geometric approximation: reflected by the planar convergence parameter, directly measuring whether local point clouds are spatially clustered within a single plane. 2. Local structural orientation consistency: determined by structural distribution synchronicity. This reflects the consistency of local normal vector directions and the orthogonality of geometric connections, and is an important indicator for judging whether a structure is simple and regular. Multiplication means that only when a local region containing point cloud data simultaneously satisfies two conditions—that it is spatially highly similar to a plane (high plane similarity parameter) and that its local structural directions are consistent and regular (high structural distribution synchronization)—is its spatial plane synchronization considered complete. Only then can very high values ​​be achieved. This is a typical characteristic of simple, flat, and uniformly oriented background areas such as detection platforms. However, for complex steel structure surfaces, even if some local areas may approximate a plane (with a not low plane approximation parameter), the synchronicity of their structural distribution is often low (the normal vector direction varies), leading to the final... The value will not be very high, thus effectively distinguishing it from the background area.

[0105] Based on the above technical solution, this embodiment of the invention integrates the planar fitting approximation degree (geometric features) of local point clouds with the directional distribution consistency of local structures (topological / directional features) to calculate a comprehensive index of spatial planar synchronization. This index can more accurately and robustly identify regions in the scene that are not only flat but also have consistent structural regularity (mainly background platforms), providing a stronger and more discriminative basis for distinguishing targets from the background than a single geometric feature.

[0106] For example, in another virtual assembly method for prefabricated steel structures based on a BIM platform provided in one embodiment of the present invention, for each point cloud data, the platform structure commonality of each point cloud data is determined according to the directional correlation between the normal vector and the unit vector in the vertical direction of space and the spatial plane synchronization degree, and the pure point cloud dataset of the steel structure is determined from the original point cloud dataset according to the platform structure commonality. The specific steps include: S401. Calculate the cosine similarity between the normal vector of each point cloud data and the unit vector in the vertical direction of space, and use it as the directional correlation parameter.

[0107] Optionally, this step can be performed by the general knowledge calculation submodule 141 in the platform structure recognition module 14.

[0108] Specifically, the general knowledge calculation submodule 141 acquires each point cloud data. normal vector (Calculated and stored in S302). Simultaneously, define a unit vector in the vertical direction of space (usually aligned with the direction of gravity or the Z-axis of the world coordinate system). The general knowledge calculation submodule 141 calculates the cosine similarity between the two vectors to obtain the directional relevance parameter. ,in This represents the cosine similarity between the two vectors within the parentheses.

[0109] S402. Determine the platform structure commonality of point cloud data based on the directional correlation parameter and spatial plane synchronization degree.

[0110] Furthermore, the general knowledge calculation submodule 141 calculates the point cloud data using the following formula. Platform architecture general knowledge: In the above formula, Representing point cloud data Platform structure general knowledge This represents the directional correlation parameter determined by S401. Representing point cloud data Spatial plane synchronization.

[0111] It should be noted that the platform structure has a high degree of general understanding. It is the product of the directional correlation parameter (reflecting the levelness of the local surface) and the spatial planar synchronization degree (reflecting the flatness and regularity of the local surface). This calculation mechanism requires that a point cloud data point must simultaneously meet two conditions to obtain high platform structure recognition: 1. The local surface it occupies is nearly horizontal (high directional correlation parameter). 2. The local region it occupies is flat and structurally regular (high spatial planar synchronization degree). This is precisely the typical characteristic of point clouds for detecting platform surfaces. However, for points on steel components, even if some areas may be relatively flat (not low spatial planar synchronization degree), because their normal vectors are often not parallel to the vertical direction (low directional correlation parameter), the final... The value will be lower. Therefore, this indicator can accurately distinguish between horizontally placed, simple-structured testing platforms and complex-shaped steel structures.

[0112] S403. Based on the platform structure commonality of all point cloud data, determine the segmentation threshold, and classify the point cloud data with platform structure commonality greater than the segmentation threshold as the first candidate point cloud set.

[0113] Optionally, this step can be performed by the background segmentation submodule 142 in the platform structure recognition module 14.

[0114] Specifically, the background segmentation submodule 142 collects platform architecture knowledge of all point cloud data in the original point cloud dataset. | i =1, 2, ..., N}, where N is the total number of point clouds.

[0115] For example, this embodiment uses the maximum inter-class variance method to traverse all possible thresholds, calculate the inter-class variance of the foreground (candidate platform points) and background (other points) data, and select the threshold that maximizes the inter-class variance as the optimal segmentation threshold, which can maximize the separation of the two types of data.

[0116] Understandably, the Otsu's method is an adaptive threshold selection method suitable for image (or data) segmentation using bimodal histograms. Its advantage lies in the fact that it does not require pre-setting a threshold but automatically determines it based on the data's inherent distribution characteristics. When processing numerical distributions such as the commonality of point cloud platform structures, if there is a clear distinction between platform points and structure points, this method can find an effective threshold to separate them. If the data distribution does not exhibit a clear bimodality, this method can also find a relatively reasonable balance point. This method is a well-known algorithm in the field.

[0117] After this, the background segmentation submodule 142 will... Point cloud data exceeding the segmentation threshold are filtered out to form the first candidate point cloud set, which contains all candidate background points with high-level, flat features.

[0118] S404. Perform spatial clustering on the first candidate point cloud set according to the preset spatial clustering algorithm to obtain multiple point cloud clusters, and determine the background point cloud dataset according to the spatial height features of each of the multiple point cloud clusters.

[0119] For example, the background segmentation submodule 142 performs spatial clustering to obtain multiple point cloud clusters, specifically including the following sub-steps: (1) Calculate the Euclidean distance between any two point cloud data in the first candidate point cloud set, and determine the neighborhood radius of the spatial clustering algorithm based on the average value of all Euclidean distances.

[0120] Specifically, the background segmentation submodule 142 first calculates the 3D Euclidean distance between all pairs of points in the first candidate point cloud. Assuming there are M points in the point cloud, then there are a total of... Each distance value is calculated. The background segmentation submodule 142 calculates the arithmetic mean of these distances. Then, this average value... It is directly used as the neighborhood radius parameter of the preset spatial clustering algorithm.

[0121] For example, the aforementioned preset spatial clustering algorithm can specifically be a density-based spatial clustering algorithm (Density-Based Spatial Clustering of Applications with Noise, DBSCAN).

[0122] (2) Using the neighborhood radius and the preset minimum number of points as input parameters for the spatial clustering algorithm, perform clustering analysis on the first candidate point cloud and output multiple point cloud clusters.

[0123] Furthermore, the background segmentation submodule 142 sets another key parameter for the DBSCAN algorithm—the minimum number of points required within the neighborhood of the core point. For example, the minimum number of points can be preset to 10. This value is based on experience: too small a value leads to sensitivity to noise and the generation of too many small clusters; too large a value may merge clusters that should be separate, or cause boundary points to be misclassified as noise. Typically, the minimum number of points is related to the data dimension. For 3D point clouds, a rule of thumb is to use more than twice the data dimension (i.e., ≥6). In this embodiment, the minimum number of points is set to 10, which is a compromise.

[0124] Therefore, using the neighborhood radius (i.e.) The background segmentation submodule 142 uses the minimum number of points (10 in the previous example) as input parameters for the spatial clustering algorithm, and performs DBSCAN clustering on the first candidate point cloud. The algorithm groups points that are sufficiently dense in spatial location (the number of points in the neighborhood is greater than the minimum number of points) into the same cluster, and marks sparse points as noise (in this embodiment, noise points are not considered as valid clusters). Finally, the algorithm outputs several point cloud clusters. .

[0125] After obtaining multiple point cloud clusters, the background segmentation submodule 142 calculates the average coordinates of all point cloud data in each cluster in the vertical direction (Z-axis). Since the detection platform is the base supporting the steel structure, it is typically the lowest in terms of spatial height. Therefore, the background segmentation submodule 142 will have the minimum mean Z-axis coordinate. The point cloud cluster was identified as the background point cloud dataset representing the surface of the detection platform.

[0126] S405. Based on the background point cloud dataset, filter out the background point cloud data from the original point cloud dataset to obtain the clean point cloud dataset of the steel structure.

[0127] Optionally, this step can be performed by the adaptive filtering submodule 143 in the platform structure recognition module 14.

[0128] For example, the adaptive filtering submodule 143 filters out background point cloud data from the original point cloud dataset based on the background point cloud dataset, specifically including the following sub-steps: (1) Perform plane fitting on the background point cloud dataset to obtain the reference plane.

[0129] Specifically, the adaptive filtering submodule 143 receives the background point cloud dataset determined by the background segmentation submodule 142. Since the detection platform surface may not be perfectly level, or there may be a deviation between the scan coordinate system and the world coordinate system, the adaptive filtering submodule 143 uses the RANSAC algorithm to perform robust plane fitting on the background point cloud dataset. The RANSAC algorithm iteratively fits a plane to a randomly sampled point set and counts the number of interior points (points that conform to the plane). Finally, it selects the plane model with the most interior points as the optimal fitting result, thus effectively resisting noise and outliers that may exist in the data. The fitted plane equation is... This is the reference plane.

[0130] (2) Perform pass-through filtering on the original point cloud dataset based on the dynamic height threshold range of the reference plane in three-dimensional space.

[0131] In this sub-step, the adaptive filtering submodule 143 filters the original point cloud dataset based on the fitted reference plane. Its core idea is to filter out points located near the reference plane (belonging to the background) and retain points far from the plane (belonging to the steel structure). In specific implementation, for each point in the original point cloud dataset... Calculate its perpendicular distance to the reference plane. .distance The sign of the point indicates which side of the reference plane the point is on (for example, the side the normal vector points to is considered positive).

[0132] Subsequently, the adaptive filtering submodule 143 sets a dynamic height threshold range. For example, it can be set =0.001 meters, =0.05 meters. Threshold value selection rules: It is a very small positive value (e.g., 1 mm) used to filter out points that are close to the platform surface and may be caused by contact or noise. The settings should take into account the actual gap between the steel structure components and the platform, the point cloud noise level, and the possible thickness of the platform, and are typically set to the order of a few centimeters (e.g., 5 centimeters). Points outside this range are considered as steel structure point clouds.

[0133] Finally, the adaptive filtering submodule 143 performs pass-through filtering: retaining all that meet the requirements. < or > The point cloud data refers to points located sufficiently far above and below the reference plane; filtering out points that meet the following conditions... ≤ ≤≤ The point (i.e., the point located in a thin layer near the reference plane, i.e., the background point).

[0134] (3) The filtered point cloud data is used as the pure point cloud dataset of the steel structure.

[0135] Furthermore, the adaptive filtering submodule 143 outputs all the point cloud data retained after the pass-through filtering operation, which is the clean point cloud dataset of the steel structure. This dataset has largely removed the interference from the background of the detection platform and mainly contains the surface point cloud of the steel structure itself, providing clean and high-quality input for subsequent 3D reconstruction.

[0136] Based on the above technical solution, this embodiment of the invention achieves accurate initial screening of background platform points by constructing a platform structure recognition index that integrates levelness and flatness regularity. Subsequently, adaptive threshold segmentation and density-based spatial clustering are used to robustly extract the complete point cloud of the background platform surface. Finally, dynamic threshold filtering based on robust plane fitting completely separates the background from the target, resulting in a high-quality, clean point cloud of the steel structure. This series of interconnected processes effectively solves the problem that traditional fixed threshold methods cannot adapt to complex shapes and tilted platform scenarios, significantly improving the accuracy, completeness, and automation of background filtering, laying a solid data foundation for high-precision virtual assembly.

[0137] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0138] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A virtual assembly method for prefabricated steel structures based on a BIM platform, characterized in that, The method includes: Obtain the original point cloud dataset of the steel structure; A set of neighboring point clouds is determined for each point cloud data in the original point cloud dataset; a first feature parameter is determined for each point cloud data based on the relationship between the normal vector of each point cloud data and the angular relationship between the line directions connecting each point cloud data to each point cloud data in the neighboring point cloud dataset; a second feature parameter is determined for each point cloud data based on the directional similarity between the normal vector of each point cloud data and the normal vectors of each point cloud data in the neighboring point cloud dataset; and the structural distribution synchronization of each point cloud data is determined based on the first feature parameter and the second feature parameter; wherein, the structural distribution synchronization is used to characterize the consistency of the normal vector directions within the local neighborhood of the point cloud data and the orthogonality of the connection relationships between point clouds; Fit a neighboring point cloud set for each point cloud data to obtain a fitting plane; calculate the distance from all point cloud data in the neighboring point cloud set to the fitting plane, and determine the plane proximity parameter for each point cloud data, including: for each point cloud data Calculate its nearest neighbor cloud. Each point in vertical distance to the fitted plane The mean value is calculated based on this set of vertical distances. Iterate through all the neighboring point cloud sets in the original point cloud dataset to find the maximum value among them, and denote it as... Point cloud data Plane approach parameter = Based on the planar proximity parameter and the structural distribution synchronization, the spatial planar synchronization degree of each point cloud data is determined; wherein, the spatial planar synchronization degree is used to characterize the degree to which the local neighborhood of the point cloud data approaches the same plane and the consistency of its distribution direction; Calculate the cosine similarity between the normal vector of each point cloud data and the unit vector in the vertical direction of space, as a directional correlation parameter; determine the platform structure commonality of the point cloud data based on the directional correlation parameter and the spatial plane synchronization degree, and determine the pure point cloud dataset of the steel structure from the original point cloud dataset based on the platform structure commonality. Based on the pure point cloud dataset of the steel structure, a three-dimensional model of the steel structure is reconstructed, and the three-dimensional model of the steel structure is assembled with the BIM building model in a virtual environment.

2. The virtual assembly method for prefabricated steel structures based on a BIM platform according to claim 1, characterized in that, Based on the platform structure generality, a clean point cloud dataset of the steel structure is determined from the original point cloud dataset, specifically including: Based on the platform structure commonality of all point cloud data, a segmentation threshold is determined, and point cloud data with a platform structure commonality greater than the segmentation threshold are classified into the first candidate point cloud set. The first candidate point cloud set is spatially clustered according to a preset spatial clustering algorithm to obtain multiple point cloud clusters, and the background point cloud dataset is determined according to the spatial height features of each of the multiple point cloud clusters. Based on the background point cloud dataset, background point cloud data is filtered out from the original point cloud dataset to obtain the clean point cloud dataset of the steel structure.

3. The virtual assembly method for prefabricated steel structures based on a BIM platform according to claim 2, characterized in that, The first candidate point cloud set is spatially clustered according to a preset spatial clustering algorithm to obtain multiple point cloud clusters, specifically including: Calculate the Euclidean distance between any two point cloud data in the first candidate point cloud set, and determine the neighborhood radius of the spatial clustering algorithm based on the average of all Euclidean distances; Using the neighborhood radius and the preset minimum number of points as input parameters for the spatial clustering algorithm, cluster analysis is performed on the first candidate point cloud to output the multiple point cloud clusters.

4. The virtual assembly method for prefabricated steel structures based on a BIM platform according to claim 2, characterized in that, Based on the background point cloud dataset, background point cloud data is filtered out from the original point cloud dataset, specifically including: A reference plane is obtained by performing plane fitting on the background point cloud dataset; Based on the dynamic height threshold range of the reference plane in three-dimensional space, a pass-through filtering operation is performed on the original point cloud dataset; The filtered and retained point cloud data is used as the pure point cloud dataset of the steel structure.

5. The virtual assembly method for prefabricated steel structures based on a BIM platform according to claim 1, characterized in that, Based on the pure point cloud dataset of the steel structure, a 3D model of the steel structure is reconstructed, specifically including: The clean point cloud dataset is subjected to denoising filtering, and the denoised and filtered clean point cloud dataset is then downsampled. The pre-set surface reconstruction algorithm is used to calculate the downsampled clean point cloud dataset to generate a continuous three-dimensional surface model of the steel structure.

6. The virtual assembly method for prefabricated steel structures based on a BIM platform according to claim 1, characterized in that, Assemble the steel structure 3D model with the BIM building model in a virtual environment, specifically including: The three-dimensional model of the steel structure is registered and compared with the standard three-dimensional model of the steel structure to determine whether the dimensions of the three-dimensional model of the steel structure conform to the verification results. If the verification result is passed, the assembly pose of the steel structure 3D model in the BIM building model is calculated by a preset registration algorithm based on the target assembly position information in the BIM building model. Based on the assembly posture, the three-dimensional model of the steel structure and the BIM building model are assembled in the virtual environment.

7. The virtual assembly method for prefabricated steel structures based on a BIM platform according to claim 1, characterized in that, Obtain the original point cloud dataset of the steel structure, specifically including: The steel structure is scanned using a 3D laser scanning device to obtain an initial point cloud set containing the steel structure and the supporting platform, which serves as the original point cloud dataset.

Images