BIM model automatic generation method and system based on point cloud data

Abstract

The invention discloses a BIM model automatic generation method and system based on point cloud data, and belongs to the field of building information modelling, and the transmission method comprises the steps: obtaining original point cloud data, and employing a filtering algorithm based on point cloud density adaptive adjustment to carry out the preprocessing of the point cloud data; constructing a voxel octree structure for the preprocessed point cloud data, and performing semantic classification on the point cloud; geometric modeling is carried out based on the segmented point cloud subsets, and a fitting algorithm is adopted to carry out shape completion on a point cloud area; semantic annotation is carried out on the components subjected to geometric reconstruction, a corresponding relation between component types and spatial attributes is constructed, fusion features based on a point feature histogram and a local curvature are adopted, and classification is carried out; a standard BIM component family is converted, and a three-dimensional BIM model is constructed through the mapping relation. According to the method, the voxel octree data structure and the deep semantic segmentation neural network model are combined, division and semantic recognition are performed on the point cloud data, the intelligent degree of the model is improved, and manual intervention is reduced.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of building information modeling, more specifically, relates to a BIM model automatic generation method and system based on point cloud data. BACKGROUND

[0002] With the wide application of building information modeling technology, three-dimensional modeling plays an increasingly important role in architectural design, construction management and operation stage. As an information carrier of the whole life cycle of building, BIM model not only contains the geometric information of building components, but also integrates the multi-dimensional semantic information of components such as materials, processes and progress. The traditional BIM modeling method mainly relies on design drawings for manual modeling, which is not only low in efficiency and long in cycle, but also prone to information errors and model inconsistencies, making it difficult to meet the demand of rapid modeling and updating in large-scale scenarios. In recent years, with the development of perception technologies such as three-dimensional laser scanning and photogrammetry, point cloud data has become the main means to obtain three-dimensional information of building real scene. Point cloud data has the advantages of high precision, information-intensive and true reflection of building space structure, which provides the possibility for the reverse reconstruction of BIM model. Some existing technologies convert point cloud data into grid models or component sets to assist BIM model construction. However, these methods generally have the following problems: traditional filtering methods cannot adaptively adjust according to the distribution characteristics of point cloud, which may lead to loss of key edge information and affect the modeling accuracy. Existing methods usually use rule-driven or semi-automatic methods to classify and identify point clouds, which is difficult to meet the high-precision identification needs of multiple types of components in complex building scenarios. Due to factors such as point cloud occlusion and scanning angle, there are data missing areas in components, and traditional fitting algorithms cannot effectively complete them, resulting in incomplete geometric structure of BIM model. Some researches can generate three-dimensional models, but they are not deeply integrated with BIM family templates, which cannot realize parameterized control and semantic information injection of components, limiting the engineering application value of the model.

[0003] Therefore, there is an urgent need for an efficient, intelligent and automated point cloud to BIM model generation method that can balance geometric modeling and semantic expression, realize accurate identification, geometric reconstruction, family component transformation and BIM model construction at the component level, improve modeling efficiency and application quality, and meet the actual needs of digital building management. SUMMARY

[0004] This section is intended to summarize some aspects of the embodiments of the present application and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of the specification to avoid obscuring the purpose of this section, abstract and title, and such simplifications or omissions cannot be used to limit the scope of the present application.

[0005] In view of the problems of the above-mentioned or existing methods and systems for automatic generation of BIM models based on point cloud data, this invention is proposed.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides a method for automatically generating BIM models based on point cloud data, including: scanning a target building with a 3D laser scanner to obtain raw point cloud data, and preprocessing the point cloud data using a filtering algorithm based on adaptive adjustment of point cloud density;

[0008] A voxel octree structure is constructed for the preprocessed point cloud data, and a semantic segmentation neural network model is used to perform semantic classification of the point cloud.

[0009] Geometric modeling is performed based on the segmented point cloud subsets, shape feature parameters of components are extracted, and a fitting algorithm combining shape prior library and supervised learning training model is used to complete the shape of some missing or occluded point cloud regions.

[0010] Type identification and semantic annotation are performed on the geometrically reconstructed components to construct the correspondence between component types and spatial attributes. The fusion features based on point feature histograms and local curvature are input into a deep convolutional neural network for classification.

[0011] The identified component information is converted into a standard BIM component family, and a three-dimensional BIM model is constructed through the mapping relationship between geometric features and parameter attributes.

[0012] As a preferred embodiment of the automatic BIM model generation method based on point cloud data described in this invention, the method includes: acquiring raw point cloud data and preprocessing the point cloud data using a filtering algorithm based on adaptive adjustment of point cloud density, including:

[0013] The K-nearest neighbor search algorithm is used to calculate the density of points in the local neighborhood of each point in the point cloud. For each point p i Calculate the average distance d from it to its K neighboring points. i d of all points i Construct a global density distribution map;

[0014] Statistical analysis was performed on the density distribution map to calculate its global mean μ and standard deviation σ. The dynamic threshold interval for noise removal was then set as follows:

[0015] T low =μ-α·σ,T high =μ+β·σ

[0016] Here, α and β are adjustable sensitivity parameters;

[0017] For each point pi If its neighborhood average distance d i <T low If the points in the region are too dense, the center point is retained and clustering is simplified; if d i >T high If d i ∈[T low ,T high If ], then retain that point.

[0018] As a preferred embodiment of the automatic BIM model generation method based on point cloud data described in this invention, the method includes: constructing voxels and octree structures from the preprocessed point cloud data, and performing semantic classification of the point cloud using a semantic segmentation neural network model, including:

[0019] Set the minimum voxel side length l min The entire point cloud space is divided into three dimensions at this scale; starting from the top root node, each octet is automatically subdivided into 8 child nodes based on the number of points it contains and their spatial distribution; if the number of points in a child node exceeds a set threshold T... voxel If the recursive subdivision continues until density balance is satisfied or the minimum resolution scale is reached, each leaf node stores the set of points it contains and the corresponding local geometric features.

[0020] The system constructs a neighborhood point set for each point, performs feature aggregation on each neighborhood, obtains local spatial geometric patterns, extracts multi-scale global context information, captures overall spatial structural features, inputs the extracted fusion features into a fully connected layer, and outputs the semantic category probability distribution corresponding to each point. The classification results are then subjected to spatial smoothing and consistency optimization processing, and the classified semantic point cloud data is output in a structured format.

[0021] As a preferred embodiment of the automatic BIM model generation method based on point cloud data described in this invention, the method includes: performing geometric modeling based on segmented point cloud subsets, extracting shape feature parameters of components, and employing a fitting algorithm combining a shape prior library and a supervised learning training model to complete the shape of partially missing or occluded point cloud regions, including:

[0022] For regions with missing points in the local point cloud, a deep neural network model is constructed, taking feature vectors as input and outputting the completed point cloud shape:

[0023]

[0024] Where F(P) represents the local context features extracted by DGCNN, f θ The parameter θ is optimized through supervised training.

[0025] The output point set is merged with the original point set using a weighted fusion strategy:

[0026]

[0027] Where λ is the adjustable fusion factor;

[0028] The trained point cloud completion network is used to further complete the missing regions. The input is the feature vector of the missing component of the point cloud and the matching coarse shape. The output is the completed point cloud that is seamlessly integrated with the original point cloud. The completed point cloud is then fused with the original point cloud.

[0029] As a preferred embodiment of the automatic BIM model generation method based on point cloud data described in this invention, the method includes: performing type identification and semantic annotation on the geometrically reconstructed components to construct a correspondence between component types and spatial attributes, including:

[0030] A component classification model based on supervised learning or graph neural networks is adopted. The input is a feature vector and the output is a component type label. For each component, its spatial attribute information in the building model is extracted. The component's geometric features, identification type and spatial attributes are fused to form a structured semantic annotation, generate semantic description triples, and further construct a mapping database between component type and spatial attributes.

[0031] As a preferred embodiment of the automatic BIM model generation method based on point cloud data described in this invention, the method employs a fusion feature based on point feature histograms and local curvature, inputting it into a deep convolutional neural network for classification, including:

[0032] For each point p i Calculate the covariance matrix in its neighborhood point set:

[0033]

[0034] in, It is the neighborhood centroid;

[0035] The PFH features are fused with local curvature features to construct the final input feature vector for each point. The fused features are then input into a deep neural network suitable for point cloud processing. The feature input layer of the PointCNN-based convolutional neural network receives the F... i X-Conv convolution operations are used to extract local contextual structure from point clouds, and multiple layers are stacked to obtain high-order features. The classification header outputs semantic category labels T. i .

[0036] As a preferred embodiment of the automatic BIM model generation method based on point cloud data described in this invention, the method includes: converting the identified component information into a standard BIM component family, and constructing a three-dimensional BIM model through the mapping relationship between geometric features and parameter attributes, including:

[0037] For each component's point cloud geometric model, fitting and feature extraction are performed to obtain a parameter set θ that can be mapped to a family template. The extracted component parameters are then mapped to the parameter input interface of the family template to generate standardized BIM component family instances. If the parameters in the family template are {P1, P2, ..., P...} n}, then the assignment is performed through the mapping function M;

[0038] Using the spatial location parameters or geometric center information retained during the identification phase, the spatial coordinates and orientation of the components are calculated and three-dimensional positioning is performed. After the component family is instantiated and placed on site, the topological structure diagram between the components is reconstructed based on the spatial relationship between the identified components. The intersection points or coplanar areas are calculated based on the relative spatial relationship between the point clouds to achieve automatic connection. The final output is a complete three-dimensional BIM model data structure.

[0039] An automatic BIM model generation system based on point cloud data includes: a preprocessing module, used to scan the target building with a 3D laser scanner to obtain raw point cloud data, and to preprocess the point cloud data using a filtering algorithm based on adaptive adjustment of point cloud density;

[0040] The component feature segmentation module is used to construct a voxel octree structure from the preprocessed point cloud data and perform semantic classification of the point cloud in conjunction with a semantic segmentation neural network model.

[0041] The component geometry reconstruction module is used to perform geometric modeling based on the segmented point cloud subset, extract the shape feature parameters of the component, and use a fitting algorithm that combines a shape prior library and a supervised learning training model to complete the shape of some missing or occluded point cloud regions.

[0042] The component semantic recognition module is used to identify the type and semantically label the components after geometric reconstruction, construct the correspondence between component type and spatial attributes, and use the fusion features based on point feature histogram and local curvature to input into a deep convolutional neural network for classification.

[0043] The parametric modeling module is used to convert the identified component information into standard BIM component families and construct a 3D BIM model through the mapping relationship between geometric features and parametric attributes.

[0044] A computing device, the computing device comprising:

[0045] At least one processor, memory, and input / output unit;

[0046] The memory is used to store computer programs, and the processor is used to call the computer programs stored in the memory to execute the steps of the automatic generation method of BIM model based on point cloud data.

[0047] A computer-readable storage medium includes instructions that, when executed on a computer, cause the computer to perform steps of a method for automatically generating a BIM model based on point cloud data.

[0048] The beneficial effects of this invention are as follows: This invention employs an adaptive filtering algorithm based on point cloud density distribution, which can effectively remove noise points and outliers while preserving the edge features of key structures, significantly improving the accuracy and robustness of subsequent modeling. By combining a voxel octree data structure with a deep semantic segmentation neural network model, spatial partitioning and semantic recognition of point cloud data are performed, achieving high-precision automatic identification of various types of building components, significantly improving the model's intelligence level and reducing manual intervention. Addressing the occlusion and incompleteness issues in point clouds, a shape prior library is constructed and combined with supervised learning to train the model for shape fitting and completion, reconstructing the geometric contours of missing components and ensuring the integrity and accuracy of the model structure. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0050] Figure 1 A flowchart of a method for automatically generating BIM models based on point cloud data, provided in an embodiment of the present invention.

[0051] Figure 2 This is a schematic diagram of the automatic BIM model generation system based on point cloud data provided in an embodiment of the present invention.

[0052] Figure 3 A schematic diagram of the structure of a medium according to an embodiment of the present invention is shown.

[0053] Figure 4 A schematic diagram of the structure of a computing device according to an embodiment of the present invention is shown.

[0054] In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts. Detailed Implementation

[0055] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0056] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0057] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0058] Example

[0059] The following is for reference. Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the BIM model automatic generation method based on point cloud data provided by the present invention. It should be noted that the embodiments of the present invention can be applied to any applicable scenario.

[0060] Figure 1 The flowchart of the automatic BIM model generation method based on point cloud data provided in an embodiment of the present invention includes:

[0061] S1: The target building is scanned using a 3D laser scanner to obtain raw point cloud data, and the point cloud data is preprocessed using a filtering algorithm based on adaptive adjustment of point cloud density.

[0062] Preferably, the K-nearest neighbor search algorithm is used to calculate the point density in the local neighborhood of each point in the point cloud, and for each point p... i Calculate the average distance d from it to its K neighboring points. i d of all points i Construct a global density distribution map;

[0063] Statistical analysis was performed on the density distribution map to calculate its global mean μ and standard deviation σ. The dynamic threshold interval for noise removal was then set as follows:

[0064] T low =μ-α·σ,T high =μ+β·σ

[0065] Here, α and β are adjustable sensitivity parameters;

[0066] For each point p i If its neighborhood average distance d i <T lowIf the points in the region are too dense, the center point is retained and clustering is simplified; if d i >T high If d i ∈[T low ,T high If ], then retain that point.

[0067] Furthermore, taking the raw point cloud data obtained from scanning a 5-story office building as an example, the total number of point clouds is approximately 6,000,000 points. PCL is used for data import and unified processing. The raw point cloud P, including multiple dense regions, is loaded, and the neighborhood size parameter K = 20 is set. For each point p in the point cloud... i Perform a K-nearest neighbor search to obtain the average Euclidean distance d within the neighborhood. i Construct a complete density list D:

[0068]

[0069] Calculate the global density mean μ and standard deviation σ:

[0070] import numpy as np

[0071] mu=np.mean(d_list)#Example: mu≈0.045m

[0072] sigma=np.std(d_list)#Example: sigma≈0.023m

[0073] Setting parameters α = 1.0 and β = 1.5, we get:

[0074] T low =μ-α·σ=0.045-1.0×0.023=0.022m

[0075] T high =μ+β·σ=0.045+1.5×0.023=0.0795m

[0076] Determine d for each point in turn i Within the density range and execute:

[0077] If d i Regions <0.022m are considered high-density areas; center points are retained, and Voxel Grid is used for cluster simplification. If d i Points with a density >0.0795m are considered noise and should be removed immediately; points with normal density in the middle range should be retained.

[0078] After processing, the point cloud P' has approximately 2,300,000 points remaining.

[0079] S2: Construct a voxel octree structure for the preprocessed point cloud data, and perform semantic classification of the point cloud by combining it with a semantic segmentation neural network model.

[0080] Preferably, the minimum voxel side length is set to l. min The entire point cloud space is divided into three dimensions at this scale; starting from the top root node, each octet is automatically subdivided into 8 child nodes based on the number of points it contains and their spatial distribution; if the number of points in a child node exceeds a set threshold T... voxel If the recursive subdivision continues until density balance is satisfied or the minimum resolution scale is reached, each leaf node stores the set of points it contains and the corresponding local geometric features.

[0081] The system constructs a neighborhood point set for each point, performs feature aggregation on each neighborhood, obtains local spatial geometric patterns, extracts multi-scale global context information, captures overall spatial structural features, inputs the extracted fusion features into a fully connected layer, and outputs the semantic category probability distribution corresponding to each point. The classification results are then subjected to spatial smoothing and consistency optimization processing, and the classified semantic point cloud data is output in a structured format.

[0082] Furthermore, the minimum voxel side length l is set. min =0.1m, using this as the minimum resolution scale, the overall point cloud data space is divided into three dimensions. The initial octree root node covers the entire point cloud bounding box B(x min ,x max ,y min ,y max ,z min ,z max ).

[0083] Set the voxel subdivision threshold T voxel =200. Recursively subdivide from the root node: if the number of points in the current voxel node > T voxel Then it is divided into 8 sub-voxels; each sub-node inherits its spatial boundary, and the point set P in each leaf node is recorded during the partitioning process. i For each point p i Construct its K-neighborhood point set N(p) i K is set to 24; the following geometric features are calculated from its neighborhood for feature encoding, and multi-scale extraction using spherical neighborhood and voxel neighborhood is adopted, with radii r = {0.1m, 0.2m, 0.4m} respectively, to construct multi-scale context descriptors.

[0084] The improved PointNet++ or RandLA-Net model is adopted. The network structure includes: 128-dimensional fused features input at each point of the input layer; hierarchical sampling module: downsampling spatial features; feature propagation module: multi-scale global context backpropagation; and fully connected layer output semantic probability distribution P(Ci|pi)∈RK, where K is the number of categories.

[0085] The initial semantic prediction results are post-processed, and spatial consistency optimization is performed using a CRF model and adjacency relationships between octree nodes to suppress local misclassification and improve semantic boundary consistency. A final semantic label is assigned to each point, and the structured output is as follows:

[0086]

[0087] Alternatively, save it as a structured point cloud format (such as .ply, .pcd) with semantic tag channels.

[0088] S3: Based on the segmented point cloud subset, geometric modeling is performed, the shape feature parameters of the components are extracted, and a fitting algorithm combining the shape prior library and the supervised learning training model is used to complete the shape of some missing or occluded point cloud regions.

[0089] Preferably, for regions with missing point clouds, a deep neural network model is constructed, taking feature vectors as input and outputting the completed shape point cloud:

[0090]

[0091] Where F(P) represents the local context features extracted by DGCNN, f θ The parameter θ is optimized through supervised training.

[0092] The output point set is merged with the original point set using a weighted fusion strategy:

[0093]

[0094] Where λ is the adjustable fusion factor;

[0095] The trained point cloud completion network is used to further complete the missing regions. The input is the feature vector of the missing component of the point cloud and the matching coarse shape. The output is the completed point cloud that is seamlessly integrated with the original point cloud. The completed point cloud is then fused with the original point cloud.

[0096] Furthermore, firstly, the entire point cloud data is spatially segmented to identify component regions with sparse or missing points. A voxel octree structure is used to divide the point cloud space, and the density of points within each voxel node is calculated. Regions below a set threshold are marked as potentially missing regions.

[0097] Further extraction of a subset of the point cloud in this region constitutes the point cloud data to be completed. The points contained therein are usually incompletely distributed or have irregular edges.

[0098] A graph convolution-based feature extraction neural network is used to process the point cloud data of the missing region: a local K-nearest neighbor graph structure (e.g., K=20) is constructed for each point to extract the relative spatial relationship between points; the three-dimensional coordinate information of the points is input into the graph convolution network to extract local context features layer by layer; after multiple convolution and max pooling operations, a global shape encoding vector F(P)∈Rd is obtained, which represents the overall features of the current missing region.

[0099] A shape prior library is constructed from pre-collected standard BIM component samples, and the geometric feature vector of each sample is extracted in the same way. The shape sample that is closest to the current missing region in the prior library is found by Euclidean distance or cosine similarity to provide prior constraints for the coarse shape.

[0100] The feature vector of the current missing region and the matching shape code are used as joint inputs to a pre-trained supervised learning network model for inference. The network outputs a complete point set Pcomp∈RN×3P, where the points are located in the spatial position of the missing region and are geometrically continuous with the original point cloud.

[0101] To improve the naturalness and stability of the completion, the original point set and the completed point set are merged. Finally, the merged point set is remapped to the global coordinate system, aligned with the original point cloud data, and missing areas are replaced to form complete component point cloud data. For multiple component completion tasks, the above operations can be performed in batches to gradually improve the entire point cloud model, providing a complete and continuous geometric data foundation for the subsequent automatic generation of BIM component models.

[0102] S4: Perform type identification and semantic annotation on the geometrically reconstructed components, construct the correspondence between component types and spatial attributes, and use the fusion features based on point feature histograms and local curvature to input into a deep convolutional neural network for classification.

[0103] Preferably, a component classification model based on supervised learning or graph neural networks is adopted. The input is a feature vector, and the output is a component type label. For each component, its spatial attribute information in the building model is extracted. The component's geometric features, identification type, and spatial attributes are fused to form a structured semantic annotation, generate a semantic description triple, and further construct a mapping database between component type and spatial attributes.

[0104] Preferably, for each point p i Calculate the covariance matrix in its neighborhood point set:

[0105]

[0106] in, It is the neighborhood centroid;

[0107] The PFH features are fused with local curvature features to construct the final input feature vector for each point. The fused features are then input into a deep neural network suitable for point cloud processing. The feature input layer of the PointCNN-based convolutional neural network receives the F... i X-Conv convolution operations are used to extract local contextual structure from point clouds, and multiple layers are stacked to obtain high-order features. The classification header outputs semantic category labels T. i .

[0108] Furthermore, PointCNN, a deep convolutional network suitable for point cloud processing, is used for training and inference: the input layer receives the fused feature vector of each point; the X-Conv convolutional module rearranges the order of neighboring points through a learned transformation matrix, realizing learnable local spatial convolution operations; stacked multi-layer X-Conv layers are used to extract contextual structures from low to high order; fully connected layers are used for classification; the output layer outputs the semantic category label of each point; model training is supervised using an existing labeled point cloud dataset, and the objective function is optimized as the cross-entropy loss function. Based on the predicted label of each point, clustering and merging are performed to generate complete component regions. For each identified component, a triplet representation is generated according to the semantic structure, and all component triplets are imported into a structured database system to form a mapping database of component types and spatial attributes, supporting subsequent BIM model generation, visualization retrieval, and semantic querying.

[0109] S5: Convert the identified component information into a standard BIM component family, and construct a 3D BIM model through the mapping relationship between geometric features and parameter attributes.

[0110] Preferably, for the point cloud geometric model of each component, fitting and feature extraction are performed to obtain a parameter set θ that can be mapped to the family template. The extracted component parameters are then mapped to the parameter input interface of the family template to generate a standardized BIM component family instance. If the parameters in the family template are {P1, P2, ..., P...} n}, then the assignment is performed through the mapping function M;

[0111] Using the spatial location parameters or geometric center information retained during the identification phase, the spatial coordinates and orientation of the components are calculated and three-dimensional positioning is performed. After the component family is instantiated and placed on site, the topological structure diagram between the components is reconstructed based on the spatial relationship between the identified components. The intersection points or coplanar areas are calculated based on the relative spatial relationship between the point clouds to achieve automatic connection. The final output is a complete three-dimensional BIM model data structure.

[0112] Furthermore, for each component, its geometric shape is extracted from the point cloud data. Taking a column as an example, RANSAC or least squares fitting algorithms are used to fit a cylindrical shape to its point cloud data. The fitting result is a parametric geometric shape, such as the radius, axial position, and height of the cylinder. For irregular shapes, polynomial surface fitting, B-spline surface fitting, and other methods are used to extract the surface features of the component.

[0113] In BIM software, a series of standardized family templates are predefined, including wall families, column families, door and window families, etc. Each family template contains a certain number of parameters, and each template has a corresponding input interface. A mapping function M is designed to map the geometric feature parameters θ of the components extracted from the point cloud to the parameter interface of the family template. This mapping function can be optimized through a simple linear mapping or by using a regression model based on machine learning. Based on the mapped parameters P1, P2, ..., P... n Standardized BIM component family instances are generated through the API interface of BIM software. Simultaneously, the spatial coordinates and orientation of each component are calculated using the positioning information from point cloud data.

[0114] Based on the spatial location parameters or geometric center information of the components, the spatial coordinates and orientation of the components are calculated. For example, the centroid position in the point cloud data is used as the reference point for the component to calculate its orientation. The topological relationships between components are reconstructed using the point cloud data and the identified spatial positions of the components. For example, if a door is installed on a wall, their connection is confirmed by calculating their spatial relationship. Geometric algorithms, such as intersection calculation or coplanarity detection, are used to determine whether components intersect or are coplanar. If there is a gap between a wall and a window, the intersection points or intersecting areas are calculated and automatically connected. Based on the spatial relationships between components, a topological structure diagram is constructed to represent the connections between them. For example, the connection between a column and a floor slab, or the connection between a wall and doors and windows. The topological diagram generates connection information that can be used for construction or later optimization.

[0115] Place the instantiated BIM components at designated locations on the building model to complete the model construction in 3D space. Combine the details of the point cloud data to further optimize the shape and connections of the components, ensuring seamless integration. After instantiating and placing all components, output a complete 3D BIM model. This model includes information such as the geometry, spatial location, topological relationships, and attributes of all components, conforming to the BIM standard data structure.

[0116] After introducing the method of exemplary embodiments of the present invention, the following references are made. Figure 2 An exemplary embodiment of the present invention provides a BIM model automatic generation system based on point cloud data, the system comprising:

[0117] The preprocessing module is used to scan the target building with a 3D laser scanner to obtain raw point cloud data, and to preprocess the point cloud data using a filtering algorithm based on adaptive adjustment of point cloud density.

[0118] The component feature segmentation module is used to construct a voxel octree structure from the preprocessed point cloud data and perform semantic classification of the point cloud in conjunction with a semantic segmentation neural network model.

[0119] The component geometry reconstruction module is used to perform geometric modeling based on the segmented point cloud subset, extract the shape feature parameters of the component, and use a fitting algorithm that combines a shape prior library and a supervised learning training model to complete the shape of some missing or occluded point cloud regions.

[0120] The component semantic recognition module is used to identify the type and semantically label the components after geometric reconstruction, construct the correspondence between component type and spatial attributes, and use the fusion features based on point feature histogram and local curvature to input into a deep convolutional neural network for classification.

[0121] The parametric modeling module is used to convert the identified component information into standard BIM component families and construct a 3D BIM model through the mapping relationship between geometric features and parametric attributes.

[0122] After introducing the methods and systems of exemplary embodiments of the present invention, the following references are made. Figure 3 A computer-readable storage medium according to exemplary embodiments of the present invention will be described, please refer to... Figure 3 The computer-readable storage medium shown is an optical disc 30, on which a computer program (i.e., a program product) is stored. When the computer program is run by a processor, it implements the steps described in the above-described method implementation. For example, it scans the target building with a 3D laser scanner to obtain raw point cloud data, preprocesses the point cloud data using a filtering algorithm based on adaptive adjustment of point cloud density, constructs a voxel octree structure for the preprocessed point cloud data, and performs semantic classification of the point cloud using a semantic segmentation neural network model, performs geometric modeling based on the segmented point cloud subset, extracts the shape feature parameters of the components, and uses a fitting algorithm combining a shape prior library and a supervised learning training model to complete the shape of partially missing or occluded point cloud regions, performs type identification and semantic annotation on the geometrically reconstructed components, constructs the correspondence between component types and spatial attributes, and uses a fusion feature based on point feature histograms and local curvature to input into a deep convolutional neural network for classification, converts the identified component information into a standard BIM component family, and constructs a 3D BIM model through the mapping relationship between geometric features and parameter attributes. The specific implementation methods of each step will not be repeated here.

[0123] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.

[0124] After introducing the methods, apparatus, and media of exemplary embodiments of the present invention, the following references are made. Figure 4 A computing device for automatically generating BIM models based on point cloud data according to an exemplary embodiment of the present invention.

[0125] Figure 4 A block diagram is shown of an exemplary computing device 40 suitable for implementing embodiments of the present invention. The computing device 40 may be a computer system or a server. Figure 4 The computing device 40 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0126] like Figure 4 As shown, the components of computing device 40 may include, but are not limited to: one or more processors or processing units 401, system memory 402, and bus 403 connecting different system components (including system memory 402 and processing unit 401).

[0127] The computing device 40 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computing device 40, including volatile and non-volatile media, and removable and non-removable media.

[0128] System memory 402 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 4021 and / or cache memory 4022. Computing device 40 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, ROM 4023 may be used to read and write non-removable, non-volatile magnetic media (…). Figure 4 Not shown in the image (usually referred to as a "hard drive"). Although not shown in Figure 4The diagram illustrates that disk drives for reading and writing to removable non-volatile disks (e.g., "floppy disks") and optical disc drives for reading and writing to removable non-volatile optical discs (e.g., CD-ROMs, DVD-ROMs, or other optical media) can be provided. In these cases, each drive can be connected to bus 403 via one or more data media interfaces. System memory 402 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0129] A program / utility 4025 having a set (at least one) of program modules 4024 may be stored, for example, in system memory 402, and such program modules 4024 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment. Program modules 4024 typically perform the functions and / or methods described in the embodiments of the present invention.

[0130] The computing device 40 can also communicate with one or more external devices 404 (such as a keyboard, pointing device, display, etc.). This communication can be performed via the input / output (I / O) interface 405. Furthermore, the computing device 40 can also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 406. Figure 4 As shown, network adapter 406 communicates with other modules of computing device 40 (such as processing unit 401) via bus 403. It should be understood that, although... ​ As not shown, it can be used in conjunction with computing device 40 with other hardware and / or software modules.

[0131] The processing unit 401 executes various functional applications and data processing by running programs stored in the system memory 402. For example, it scans the target building with a 3D laser scanner to obtain raw point cloud data, preprocesses the point cloud data using a filtering algorithm based on adaptive adjustment of point cloud density, constructs a voxel octree structure for the preprocessed point cloud data, and performs semantic classification of the point cloud using a semantic segmentation neural network model, performs geometric modeling based on the segmented point cloud subset, extracts the shape feature parameters of the components, and uses a fitting algorithm combining a shape prior library and a supervised learning training model to complete the shape of partially missing or occluded point cloud regions, performs type identification and semantic annotation on the geometrically reconstructed components, constructs the correspondence between component types and spatial attributes, and uses a fusion feature based on point feature histograms and local curvature to input into a deep convolutional neural network for classification, and converts the identified component information into a standard BIM component family, constructing a 3D BIM model through the mapping relationship between geometric features and parameter attributes.

[0132] The specific implementation methods for each step will not be repeated here. It should be noted that although several units / modules or sub-units / sub-modules of the BIM model automatic generation system based on point cloud data have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0133] In the description of this invention, it should be noted that the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0134] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0135] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0136] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0137] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0138] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0139] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0140] Furthermore, although the operations of the method of the present invention are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

Claims

1. A method for automatically generating BIM models based on point cloud data, characterized in that, include: The target building is scanned by a 3D laser scanner to obtain raw point cloud data, and the point cloud data is preprocessed by a filtering algorithm based on adaptive adjustment of point cloud density. A voxel octree structure is constructed for the preprocessed point cloud data, and a semantic segmentation neural network model is used to perform semantic classification of the point cloud. Geometric modeling is performed based on the segmented point cloud subsets, shape feature parameters of components are extracted, and a fitting algorithm combining shape prior library and supervised learning training model is used to complete the shape of some missing or occluded point cloud regions. Type identification and semantic annotation are performed on the geometrically reconstructed components to construct the correspondence between component types and spatial attributes. The fusion features based on point feature histograms and local curvature are input into a deep convolutional neural network for classification. The identified component information is converted into a standard BIM component family, and a three-dimensional BIM model is constructed through the mapping relationship between geometric features and parameter attributes.

2. The method for automatically generating BIM models based on point cloud data as described in claim 1, characterized in that, The process of acquiring raw point cloud data involves preprocessing the point cloud data using a filtering algorithm based on adaptive adjustment of point cloud density, including: The K-nearest neighbor search algorithm is used to calculate the density of points in the local neighborhood of each point in the point cloud. For each point p i Calculate the average distance d from it to its K neighboring points. i d of all points i Construct a global density distribution map; Statistical analysis was performed on the density distribution map to calculate its global mean μ and standard deviation σ. The dynamic threshold interval for noise removal was then set as follows: T low =μ-a·σ,T high =μ+β·σ Here, α and β are adjustable sensitivity parameters; For each point p i If its neighborhood average distance d i <T low If the points in the region are too dense, the center point is retained and clustering is simplified; if d i >T high If d i ∈[T low ,T high If ], then retain that point.

3. The method for automatically generating BIM models based on point cloud data as described in claim 1, characterized in that, The process of constructing voxels and octree structures from preprocessed point cloud data and performing semantic classification of the point clouds using a semantic segmentation neural network model includes: Set the minimum voxel side length l min The entire point cloud space is divided into three dimensions at this scale; starting from the top root node, each octet is automatically subdivided into 8 child nodes based on the number of points it contains and their spatial distribution; if the number of points in a child node exceeds a set threshold T... voxel If the recursive subdivision continues until density balance is satisfied or the minimum resolution scale is reached, each leaf node stores the set of points it contains and the corresponding local geometric features. The system constructs a neighborhood point set for each point, performs feature aggregation on each neighborhood, obtains local spatial geometric patterns, extracts multi-scale global context information, captures overall spatial structural features, inputs the extracted fusion features into a fully connected layer, and outputs the semantic category probability distribution corresponding to each point. The classification results are then subjected to spatial smoothing and consistency optimization processing, and the classified semantic point cloud data is output in a structured format.

4. The method for automatically generating BIM models based on point cloud data as described in claim 1, characterized in that, The geometric modeling based on the segmented point cloud subsets extracts the shape feature parameters of the components. A fitting algorithm combining a shape prior library and a supervised learning training model is used to complete the shape of partially missing or occluded point cloud regions, including: For regions with missing points in the local point cloud, a deep neural network model is constructed, taking feature vectors as input and outputting the completed point cloud shape: Where F(P) represents the local context features extracted by DGCNN, f θ The parameter θ is optimized through supervised training. The output point set is merged with the original point set using a weighted fusion strategy: Where λ is the adjustable fusion factor; The trained point cloud completion network is used to further complete the missing regions. The input is the feature vector of the missing component of the point cloud and the matching coarse shape. The output is the completed point cloud that is seamlessly integrated with the original point cloud. The completed point cloud is then fused with the original point cloud.

5. The method for automatically generating BIM models based on point cloud data as described in claim 1, characterized in that, The process of identifying and semantically labeling the geometrically reconstructed components, and constructing the correspondence between component types and spatial attributes, includes: A component classification model based on supervised learning or graph neural networks is adopted. The input is a feature vector and the output is a component type label. For each component, its spatial attribute information in the building model is extracted. The component's geometric features, identification type and spatial attributes are fused to form a structured semantic annotation, generate semantic description triples, and further construct a mapping database between component type and spatial attributes.

6. The method for automatically generating BIM models based on point cloud data as described in claim 1, characterized in that, The method employs a fusion feature based on point feature histograms and local curvature, inputting it into a deep convolutional neural network for classification, including: For each point p i Calculate the covariance matrix in its neighborhood point set: in, It is the neighborhood centroid; The PFH features are fused with local curvature features to construct the final input feature vector for each point. The fused features are then input into a deep neural network suitable for point cloud processing. The feature input layer of the PointCNN-based convolutional neural network receives the F... i X-Conv convolution operations are used to extract local contextual structure from point clouds, and multiple layers are stacked to obtain high-order features. The classification header outputs semantic category labels T. i .

7. The method for automatically generating BIM models based on point cloud data as described in claim 1, characterized in that, The process of converting the identified component information into a standard BIM component family and constructing a 3D BIM model through the mapping relationship between geometric features and parameter attributes includes: For each component's point cloud geometric model, fitting and feature extraction are performed to obtain a parameter set θ that can be mapped to a family template. The extracted component parameters are then mapped to the parameter input interface of the family template to generate standardized BIM component family instances. If the parameters in the family template are {P1, P2, ..., P...} n }, then the assignment is performed through the mapping function M; Using the spatial location parameters or geometric center information retained during the identification phase, the spatial coordinates and orientation of the components are calculated and three-dimensional positioning is performed. After the component family is instantiated and placed on site, the topological structure diagram between the components is reconstructed based on the spatial relationship between the identified components. The intersection points or coplanar areas are calculated based on the relative spatial relationship between the point clouds to achieve automatic connection. The final output is a complete three-dimensional BIM model data structure.

8. A BIM model automatic generation system based on point cloud data, characterized in that, include: The preprocessing module is used to scan the target building with a 3D laser scanner to obtain raw point cloud data, and to preprocess the point cloud data using a filtering algorithm based on adaptive adjustment of point cloud density. The component feature segmentation module is used to construct a voxel octree structure from the preprocessed point cloud data and perform semantic classification of the point cloud in conjunction with a semantic segmentation neural network model. The component geometry reconstruction module is used to perform geometric modeling based on the segmented point cloud subset, extract the shape feature parameters of the component, and use a fitting algorithm that combines the shape prior library and the supervised learning training model to complete the shape of some missing or occluded point cloud regions. The component semantic recognition module is used to identify the type and semantically label the components after geometric reconstruction, construct the correspondence between component type and spatial attributes, and use the fusion features based on point feature histogram and local curvature to input into a deep convolutional neural network for classification. The parametric modeling module is used to convert the identified component information into standard BIM component families and construct a 3D BIM model through the mapping relationship between geometric features and parametric attributes.

9. A computing device, the computing device comprising: At least one processor, memory, and input / output unit; The memory is used to store computer programs, and the processor is used to call the computer programs stored in the memory to execute the steps of the automatic generation method of BIM model based on point cloud data as described in any one of claims 1 to 7.

10. A computer-readable storage medium comprising instructions that, when executed on a computer, cause the computer to perform the steps of the method for automatically generating a BIM model based on point cloud data as described in any one of claims 1 to 7.

Images