Point cloud building component modeling method and system based on feature extraction

Abstract

The invention relates to the technical field of three-dimensional modeling in engineering surveying, in particular to a point cloud building component modeling method and system based on feature extraction. Three-dimensional point cloud data of building components are obtained, a multi-scale local neighborhood calculation covariance matrix is constructed with each point as the center, eigenvalues are decomposed, an edge probability graph is generated, corresponding component categories are semantically output, and corresponding key feature point sets are screened; taking the feature point cloud as a control point, constructing a three-level B-spline surface model, and adjusting the corresponding spatial distribution density; a weighted complete graph is formed through a key feature point set, a non-planar area is identified and processed through a Kurtowski theorem, a QEM algorithm is applied to carry out lightweight processing on a three-dimensional model, screening and optimization are carried out through automatically extracting key feature points of building components, and a concise and accurate lightweight model is constructed based on feature point topological optimization and a graph theory algorithm. The automation degree and efficiency of modeling are improved, and the contradiction between model lightweight and precision is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional modeling in engineering surveying, and particularly relates to a method and system for point cloud building component modeling based on feature extraction. Background Art

[0002] Currently, laser point cloud technology has been widely applied in fields such as building component modeling and three-dimensional reconstruction, and can quickly obtain the spatial geometric information of building entities, becoming the core data source for promoting engineering applications such as building information modeling (BIM), digital twin, and urban modeling. The high-density three-dimensional point cloud obtained through laser scanning provides rich data support for the digital expression of building components. However, in the actual modeling process, most existing methods rely on manual modeling or semi-automatic modeling methods, and there are still many problems. For example, the modeling process highly depends on manual operations, the operation process is cumbersome, the overall efficiency is low, and it is difficult to guarantee the accuracy and consistency of component results. At the same time, the lack of a standardized feature point extraction and recognition mechanism leads to problems such as structural misjudgment and blurred boundaries in the modeling results.

[0003] In recent years, to improve the level of modeling automation, the academic and industrial communities have gradually introduced point cloud recognition technologies based on deep learning, as well as methods of multi-source data fusion such as RGB-D, normal vectors, semantic labels, etc., for assisting in the classification and recognition of building components. Although the above methods have made certain progress in component segmentation and semantic annotation, there are still key bottlenecks in the structural modeling of feature points: ① The lack of systematic analysis and optimization of the topological structure between feature points leads to incomplete modeling structures and inaccurate geometric expressions; ② There are a large amount of redundant data or invalid point connections in the model, affecting the lightweight expression of the overall model; it is impossible to achieve differential modeling control according to component complexity and semantic importance, which not only affects the reconstruction efficiency but also increases the subsequent processing cost; ③ The lack of a general lightweight modeling mechanism makes it difficult for the model to be directly applied to actual scenarios such as BIM systems, GIS platforms, three-dimensional visualization, and operation and maintenance analysis. Summary of the Invention

[0004] Aiming at the problems of incomplete topological structure, serious model redundancy, and lack of lightweight control in the prior art, the present invention proposes a method and system for point cloud building component modeling based on feature extraction, which integrates deep learning and geometric analysis to automatically extract key feature points of building components, combines adaptive reconstruction and multi-scale detail enhancement technologies to adjust the model granularity and structural accuracy as needed; further constructs spatial connection relationships through feature point topology optimization and graph theory algorithms to generate a lightweight model with a simple structure and accurate expression, effectively improving the automation degree and application efficiency of building component modeling, and being applicable to engineering scenarios such as BIM, GIS, and three-dimensional visualization.

[0005] The present invention provides a method for modeling building components based on feature extraction, and the steps include: Perform laser scanning on the target building component area to obtain the three-dimensional point cloud data of the building component and perform data preprocessing; Based on the preprocessed three-dimensional point cloud data, construct multi-scale local neighborhoods centered on each point, calculate the covariance matrix for the point sets within each scale local neighborhood and decompose the eigenvalues, extract the normalized curvature index, perform normalized mapping through the Sigmoid function to generate an edge probability map for screening candidate edge feature points, perform semantic segmentation on the candidate edge feature points through an improved PointNet++ network to output the corresponding component categories, and screen the corresponding key feature point sets based on the component categories; Using the feature point cloud in the key feature point set as control points, construct a three-level B-spline surface model to fit the geometric shape of the target component, and set a density adjustment function based on the importance of each feature point cloud within a preset local neighborhood to dynamically adjust the corresponding spatial distribution density, for adaptively reconstructing the three-dimensional model of the target component and introducing the Laplace deformation algorithm for detail enhancement; Based on the adaptively reconstructed three-dimensional model, form a weighted complete graph through the key feature point set, use the improved Kruskal algorithm to generate a minimum spanning tree to determine the preliminary connection relationship, introduce the simulated annealing algorithm to iteratively optimize the connection relationship through the energy function, perform developability verification on the connection result through Kuratowski's theorem, identify and process non-planar regions, and apply the QEM algorithm to lightweight the three-dimensional model to complete the three-dimensional modeling of the target component.

[0006] Further, based on the preprocessed three-dimensional point cloud data, construct multi-scale local neighborhoods centered on each point, calculate the covariance matrix for the point sets within each scale local neighborhood and decompose the eigenvalues, extract the normalized curvature index, perform normalized mapping through the Sigmoid function to generate an edge probability map for screening candidate edge feature points, including: Calculate the covariance matrix of each point in the multi-scale local neighborhood and decompose it to obtain the eigenvalues in the three-axis directions; Calculate the normalized curvature index based on the minimum eigenvalue in the three-axis directions; Based on the normalized curvature index of each point in the multi-scale local neighborhood, construct a multi-scale edge response function through weighted fusion, perform normalized mapping through the Sigmoid function, and generate an edge probability map; Screen the points in the edge probability map whose edge response values are greater than the preset threshold to obtain candidate edge feature points.

[0007] Further, perform semantic segmentation on the candidate edge feature points through an improved PointNet++ network to output the corresponding component categories, and screen the corresponding key feature point sets based on the component categories, including, Taking the 9D feature vector composed of the spatial coordinates, normal vectors, and RGB color information of the candidate edge feature points as the input, an improved PointNet++ network is used for segmentation. The candidate edge feature points with semantic confidence higher than the set confidence threshold are screened, and the corresponding component categories are labeled and then output. According to the output component categories, corresponding feature enhancement is performed, and the corresponding key feature point sets are extracted.

[0008] Furthermore, according to the output component categories, corresponding feature enhancement is performed, and the corresponding key feature point sets are extracted. When the component category is a planar component, a plane fitting residual function is constructed based on the extracted candidate edge feature points. The optimal plane parameters are iteratively solved through the L-M algorithm, and corner detection is performed by combining the convex hull vertex spacing of the plane projection point set. The candidate edge feature points at the structural turning positions are screened to form the key feature point set. When the component category is a cylindrical component, a cylindrical axis energy function is constructed, the endpoint positions are optimized by the conjugate gradient method, a multi-constraint scoring function is constructed by introducing geometric deviation, semantic confidence, and point spacing, and the corresponding candidate edge feature points are screened through the branch and bound algorithm to form the key feature point set.

[0009] Furthermore, taking the feature point cloud in the key feature point set as the control points, a three-level B-spline surface model is constructed to fit the geometric shape of the target component, including: An initial control grid is constructed based on the key feature point set, and a reconstruction is performed through the B-spline surface model using a layer-by-layer refinement mechanism. The levels include a coarse-grained layer, a medium-grained layer, and a fine-grained layer. The B-spline surface model is ; where is the model level, is the layer, and the B-spline surface generated through the parameter coordinates is , are the B-spline basis functions along the u and v directions, used to construct the shape control function of the surface, is the control point coordinate corresponding to the i,j control grid node in the l-th layer, , are the number of control points along the u and v directions in the l-th level.

[0010] Furthermore, the density adjustment function is ; where is the number of points contained in the spherical neighborhood with a preset radius centered at (x,y,z); is the volume of the spherical neighborhood; is the curvature weight; is the semantic weight corresponding to the component category.

[0011] Furthermore, based on the three-dimensional model of adaptive reconstruction, a weighted complete graph is formed through the key feature point set, and the improved Kruskal algorithm is used to generate a minimum spanning tree to determine the preliminary connection relationship. The simulated annealing algorithm is introduced to iteratively optimize the connection relationship through the energy function, including mapping the key feature point set to graph nodes, constructing weighted connection edges between any two points to generate a weighted complete graph, and the weight of the connection edge is calculated as follows:

[0012] where and are the three-dimensional coordinates of the key feature points; global average point ; is the included angle between the normal vectors of adjacent patches; is the semantic difference factor , is the component confidence; and are the weight coefficients; Using the improved Kruskal algorithm, the edge set is sorted by ascending order of the weighted complete graph, and the union-find data structure is used to monitor the connected components to generate a minimum spanning tree and retain the redundant edge set; Introduce the simulated annealing algorithm for iterative optimization of the connection path for the minimum spanning tree structure, and output the corresponding topological connection structure.

[0013] Furthermore, the developability of the connection result is verified through Kuratowski's theorem, and the non-planar regions are identified and processed, including Based on the optimized topological connection structure, the developability is verified through Kuratowski's theorem to identify the non-planar regions that are not developable in the topological connection structure; Topologically divide the connection structure of the non-planar region and reconstruct it into a corresponding developable surface subgraph using NURBS surfaces.

[0014] Furthermore, apply the QEM algorithm to lightweight the three-dimensional model and complete the three-dimensional modeling of the target component, including Based on the edge collapse cost function compress the non-critical regions and critical regions correspondingly, and optimize the multi-level texture storage structure using BC7 texture compression and JPEG-XL encoding; where v is the position of the vertex after edge collapse, and are the quadratic error matrices of adjacent triangular patches; Perform double verification on the reconstruction accuracy and geometric integrity of the lightweight 3D model.

[0015] Based on the same inventive concept, the present invention also provides a point cloud building component modeling system based on feature extraction, which adopts the point cloud building component modeling method as described above, including A data acquisition module, configured to perform laser scanning on the target building component area, obtain the 3D point cloud data of the building component, and perform data preprocessing; A feature extraction module, configured to construct a multi-scale local neighborhood centered on each point based on the preprocessed 3D point cloud data, calculate the covariance matrix of the point set within each scale local neighborhood and decompose the eigenvalues, extract the normalized curvature index, perform normalized mapping through the Sigmoid function to generate an edge probability map for screening candidate edge feature points, perform semantic segmentation on the candidate edge feature points through an improved PointNet++ network to output the corresponding component category, and screen the corresponding key feature point set based on the component category; An adaptive reconstruction module, configured to use the feature point cloud in the key feature point set as control points, construct a three-level B-spline surface model to fit the geometric shape of the target component, and dynamically adjust the corresponding spatial distribution density by setting a density adjustment function according to the importance of each feature point cloud within a preset local neighborhood, for adaptively reconstructing the 3D model of the target component and introducing the Laplacian deformation algorithm for detail enhancement; A lightweight processing module, configured to form a weighted complete graph based on the adaptively reconstructed 3D model through the key feature point set, use the improved Kruskal algorithm to generate a minimum spanning tree to determine the preliminary connection relationship, introduce the simulated annealing algorithm to iteratively optimize the connection relationship through an energy function, perform developability verification on the connection result through Kuratowski's theorem, identify and process non-planar regions, and apply the QEM algorithm to perform lightweight processing on the 3D model to complete the 3D modeling of the target component.

[0016] Compared with the prior art, the present invention has at least one of the following technical effects: Through the building component recognition technology that combines deep learning and geometric features, the present invention can automatically extract the key feature points of building components, use the adaptive reconstruction and multi-scale detail enhancement technologies to screen and optimize the feature points, and dynamically adjust the detail level of the model according to the actual engineering needs. Finally, based on the topological optimization of feature points and graph theory algorithms, a simple and accurate lightweight model is constructed. The present invention improves the automation degree and efficiency of building component modeling, effectively solves the contradiction between model lightweight and accuracy, and provides strong technical support for the digital and intelligent transformation of the construction industry. Description of the Drawings

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments: Figure 1 It is a step flowchart of the point cloud building component modeling method of the present invention. Specific embodiments

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present application. The following will illustrate the specific embodiments of the present invention in conjunction with the drawings and embodiments.

[0019] The first embodiment Currently, in the process of three-dimensional modeling of building components, although laser point cloud technology has been widely adopted, there are still significant deficiencies in the identification and efficient modeling of complex components. Most existing modeling methods rely on manual modeling or semi-automated operations, which not only involve a large amount of work and long time consumption, but also lack effective means in terms of detail expression, geometric accuracy, and model redundancy control. For example, when scanning and modeling the exterior facade of a large building, the point cloud data often contains a large number of invalid redundant points, and feature extraction relies on manual screening, which severely restricts the modeling efficiency and the lightweight effect of the model.

[0020] Furthermore, although existing deep learning-based point cloud recognition methods can achieve preliminary semantic segmentation and classification, their ability to extract key feature points inside components is limited. In particular, they lack the modeling and analysis of the geometric topological relationships between feature points, resulting in discontinuous modeling structures or missing connection relationships. At the same time, most existing solutions are single-scale processing and fail to dynamically adjust the modeling details according to the complexity of the components and semantic information, making it difficult to balance model accuracy and computational performance and unable to meet the requirements of scenarios such as BIM system integration, operation and maintenance analysis, and even deployable structure recognition in actual engineering. For example, In a large commercial complex construction project in a certain city, the project team needs to digitally model the main building structure for subsequent operation and maintenance management and BIM system integration. However, after the laser scanning equipment on-site obtains the point cloud data, it still mainly relies on manual marking and modeling of components such as walls, beams, and columns. This process is not only inefficient and time-consuming, but also often has problems such as unstable modeling accuracy, missing details, and duplicate modeling due to the lack of a unified component feature extraction mechanism.

[0021] Furthermore, when faced with multi-layered and interleaved structural components (such as arch beams and diagonal walls), traditional methods cannot automatically identify their geometric categories and topological relationships, resulting in a large amount of redundant data, connection errors, or missing key structures in the model, seriously affecting the integrity of the 3D model and the subsequent application effects. In addition, most current point cloud modeling tools lack a balancing control mechanism between model complexity and semantic importance, and cannot automatically adjust the modeling accuracy and point cloud density for different components, causing problems of increased storage and rendering burdens.

[0022] Therefore, this embodiment proposes a point cloud building component modeling method based on feature extraction, mainly a lightweight modeling method for laser point cloud building components based on feature extraction and geometric topology analysis. This method integrates an improved PointNet++ semantic segmentation network and a multi-scale curvature feature extraction mechanism, and can automatically identify edge feature points and component categories; by constructing a topological connection graph and introducing Kruskal's minimum spanning tree and simulated annealing optimization, the automatic simplification of the structural relationships of feature points is realized; combined with developability verification and Laplacian deformation methods, the rationality of the model structure and the geometric expression accuracy are improved; finally, the QEM algorithm is used to achieve lightweight compression of the 3D model, while ensuring the integrity of component semantics and geometric features, greatly improving the modeling efficiency and model applicability, and adapting to the integration requirements of various engineering platforms. The specific implementation method is as follows: The present invention provides a point cloud building component modeling method based on feature extraction, and the steps include: Using a 3D laser scanner to perform laser scanning on the target building component area, obtaining 3D point cloud data of the building components and performing data preprocessing; Specifically, the laser scanner can perform high-density data collection at the construction site to obtain 3D point cloud data, and perform preprocessing operations such as noise removal, point cloud registration, and missing data filling on the original point cloud data to obtain the point cloud set to be processed ; Based on the preprocessed 3D point cloud data, a multi-scale local neighborhood is constructed with each point as the center, the covariance matrix of the point set within each scale local neighborhood is calculated and the eigenvalues are decomposed, the normalized curvature index is extracted, and normalized mapping is performed through the Sigmoid function to generate an edge probability map for screening candidate edge feature points. The candidate edge feature points are output with the corresponding component categories through the improved PointNet++ network, and the corresponding key feature point set is screened based on the component categories; Specifically, an adaptive feature point extraction method based on geometric shape and semantic information is used to automatically extract the key feature points of building components in the laser point cloud. According to the geometric shape characteristics of the components, by calculating the eigenvalue such as the curvature of each point in the point cloud, multiple candidate edge feature points are identified. Combining the semantic information of the components and the geometric features of the components, through screening and optimizing the candidate edge feature points, a key feature point set that has a decisive impact on model construction is extracted.

[0023] Taking the feature point cloud in the key feature point set as control points, a three-level B-spline surface model is constructed to fit the geometric shape of the target component, and a density adjustment function is set through the importance of each feature point cloud in the preset local neighborhood to dynamically adjust the corresponding spatial distribution density, which is used to adaptively reconstruct the three-dimensional model of the target component and introduce the Laplace deformation algorithm for detail enhancement. Specifically, this feature point cloud-driven adaptive reconstruction technology mainly adjusts the number and positional relationship of feature points dynamically through multi-level and multi-scale refinement processing means such as interpolation, smoothing, and texture mapping according to the local geometric features and global positional relationships of the feature point cloud, ensuring that the model accurately restores the detailed information of the building component while maintaining the lightweight characteristics and enhancing the realism of the model.

[0024] Based on the three-dimensional model of adaptive reconstruction, a weighted complete graph is formed through the key feature point set, and the improved Kruskal algorithm is used to generate a minimum spanning tree to determine the preliminary connection relationship. The simulated annealing algorithm is introduced to iteratively optimize the connection relationship through the energy function. The connection result is verified for developability through Kuratowski's theorem, non-planar regions are identified and processed, and the QEM algorithm is applied to lightweight the three-dimensional model to complete the three-dimensional modeling of the target component.

[0025] Specifically, during the model reconstruction process, the graph theory algorithm is used to perform topological optimization on the feature points. The feature points are regarded as nodes in the graph, and by calculating the spatial positional relationships such as the distance and angle between the nodes, a connection relationship graph between the feature points is constructed. The simulated annealing algorithm is used to optimize the algorithm, adjust the connection relationship between the feature points, simplify the model structure, optimize the connection path between the feature points, find the optimal or approximate optimal solution of the connection path between the feature points, and use planar graph determination algorithms such as Kuratowski's theorem in graph theory to ensure that the constructed model is composed of non-crossing planes or surfaces, thus avoiding incorrect model structures.

[0026] Furthermore, based on the preprocessed three-dimensional point cloud data, a multi-scale local neighborhood is constructed with each point as the center. The covariance matrix of the point set in each scale local neighborhood is calculated and the eigenvalues are decomposed. The normalized curvature index is extracted and normalized mapping is generated through the Sigmoid function to generate an edge probability map for screening candidate edge feature points, including: Calculate the covariance matrix of each point in the multi-scale local neighborhood, and decompose it to obtain the eigenvalues in the three-axis directions; Calculate the normalized curvature index based on the minimum eigenvalue in the three-axis directions; Based on the normalized curvature index of each point in the multi-scale local neighborhood, construct a multi-scale edge response function by weighted fusion, and perform normalized mapping through the Sigmoid function to generate an edge probability map; Screen the points in the edge probability map whose edge response values are greater than the preset threshold to obtain candidate edge feature points.

[0027] Specifically, calculating the covariance matrix of each point in the multi-scale local neighborhood and decomposing it to obtain the eigenvalues in the three-axis directions includes: (1) Taking each point as the center, construct local neighborhoods of three scales respectively, set the neighborhood radii to 5mm, 15mm, and 30mm respectively. At each scale, calculate the covariance matrix for the neighborhood point set, perform eigenvalue decomposition, and obtain the eigenvalues on the three main axis directions , which are used to measure the local curvature characteristics of this point. The calculation formula for the local neighborhood covariance matrix C is: ; where k is the number of neighborhood points; is the neighborhood point coordinate is the neighborhood centroid coordinate ; Extract the minimum eigenvalue Calculate the normalized curvature index: ; are the eigenvalues of the covariance matrix C; m is the neighborhood radius, in millimeters.

[0028] (2) Construct an edge probability map through three-scale feature fusion, where the edge response function is:

[0029] where is the dynamic threshold, is the global average curvature, is the variance of the curvature at this size; is the probability function for probability normalization. Therefore, is an adaptive edge judgment baseline; When the preset threshold is 0.7, > 0.7 is judged as an edge point, and the edge points that meet the requirements are extracted as candidate edge feature points.

[0030] Further, the improved PointNet++ network is used to output the corresponding component categories for the candidate edge feature points of semantic segmentation, and the corresponding key feature point sets are filtered based on the component categories, including: Taking the 9D feature vector composed of the spatial coordinates, normal vectors, and RGB color information of the candidate edge feature points as the input, using the improved PointNet++ network for segmentation, filtering the candidate edge feature points with semantic confidence higher than the set confidence threshold, labeling the corresponding component categories, and then outputting. Perform corresponding feature enhancement according to the output component categories, and extract the corresponding key feature point sets.

[0031] Further, perform corresponding feature enhancement according to the output component categories, and extract the corresponding key feature point sets; When the component category is a planar component, a plane fitting residual function is constructed based on the extracted candidate edge feature points. The optimal plane parameters are iteratively solved by the L-M algorithm, and corner detection is performed by combining the vertex spacing of the convex hull of the plane projection point set. The candidate edge feature points at the structural turning positions are filtered to form the key feature point set; When the component category is a cylindrical component, a column axis energy function is constructed, the endpoint positions are optimized by the conjugate gradient method, a multi-constraint scoring function is constructed by introducing geometric deviation, semantic confidence, and point spacing, and the corresponding candidate edge feature points are filtered by the branch and bound algorithm to form the key feature point set.

[0032] Specifically, based on the extracted candidate edge feature points, the improved PointNet++ network is used for semantic segmentation of the point cloud data; the network input includes the 9D feature vector composed of the spatial coordinates, normal vectors, and RGB color information of each point, and the output includes the component category label and semantic confidence value of each point.

[0033] Based on the component categories, perform differential geometric analysis optimization to extract the key feature point sets, (1) When the component category is a planar component, such as building components like walls and floors: Construct a plane fitting residual function:

[0034] Where, , , is the three-dimensional coordinate of the i-th point, a, b, c, d are the number of plane equations, is the absolute value of the distance from the i-th point to the plane, that is, the perpendicular distance, is the modulus of the plane normal vector, used to normalize the distance, represents the sum of the distances from all points to the fitted plane, that is, the "fitting residual", and the smaller the value, the better the fitting.

[0035] The optimal plane equation is iteratively solved using the L-M (Levenberg-Marquardt) algorithm: This iterative solution of the optimal plane equation is the fitting solution target.

[0036] Calculate the distance between the vertices of the convex hull of the plane projection point set for corner detection, and screen the points that meet the following conditions: ; , are two adjacent convex hull corner points, is the distance between points, is the diagonal length of the model bounding box. If the distance between adjacent corner points is greater than 15% of the overall component length, it is recognized as an effective corner point at the structural corner and retained as a feature point.

[0037] (2) When the component category is a cylindrical surface component, such as building components like beams and columns: Construct the geometric expression for cylindrical surface fitting, , where, , is the center coordinate of the cross-section of the cylinder, r is the radius of the cylinder, is used to define the height interval of the cylinder in the vertical direction.

[0038] Construct the axis energy function of the column to optimize the endpoints:

[0039] where, α and β are the orthogonality and continuity weight coefficients, α is taken as 0.7, β is taken as 0.3, orthogonality means that the column axis should be orthogonal to the cross-section, and continuity means that the column axis should be smooth and continuous in the vertical direction, which is used to optimize the column centerline trend, that is, whether the column centerline is straight and symmetric.

[0040] Solve the axis parameters by the conjugate gradient method

[0041] Construct a multi-constraint optimization function for feature point optimization and screening:

[0042] where, N is the number of candidate feature points, is the i-th candidate feature point, is the geometric feature deviation, which measures whether the local part is a turning part through the curvature difference; is the semantic confidence; is the point spacing constraint term; the weight coefficients are set as: ω1 = 0.6, ω2 = 0.3, ω3 = 0.1 The branch and bound algorithm is used for global optimization, and the termination conditions are set as: ; is the objective function of the k-th iteration, Furthermore, taking the feature point cloud in the key feature point set as the control points, a three-level B-spline surface model is constructed to fit the geometric shape of the target component, including: Based on the key feature point set, an initial control grid is constructed. The layer-by-layer refinement mechanism is adopted for reconstruction through the B-spline surface model. The levels include the coarse-grained layer, the medium-grained layer, and the fine-grained layer. The main method is to adopt a three-level progressive B-spline surface reconstruction method to construct a multi-resolution model from coarse to fine, dynamically balancing the model accuracy and lightweight requirements. The B-spline surface model is ; Among them, is the model level, is the layer. On this layer, the B-spline surface generated through the parametric coordinates is , are the B-spline basis functions along the u and v directions, which are used to construct the shape control function of the surface, are the control point coordinates corresponding to the i,j-th control grid node in the l-th layer, 、 are the number of control points along the u and v directions in the l-th level.

[0043] Node vector configuration: The first layer 50mm, suitable for overall contour construction; The second layer 20mm, supplementing the main structural details; The third layer 5mm, used to depict the high-precision feature edges.

[0044] Furthermore, based on the local geometric complexity and semantic importance, the feature point density is dynamically adjusted to achieve multi-scale modeling with high fidelity in important regions and lightweight in secondary regions. A density adjustment function is established, and the density adjustment function is ; Among them, is the number of points contained in the spherical neighborhood with a preset radius centered at (x,y,z); is the volume of the spherical neighborhood; is the curvature weight; is the semantic weight corresponding to the component category. Among them, for the cylindrical component = 1.5, planar component = 0.7. For example, columnar components such as columns and beams should be built more clearly, while planar components such as wall decorations and panel patterns can be built coarser to save data volume.

[0045] Use the Laplace deformation algorithm to enhance the sharpness of the feature edges while maintaining the smooth continuity of the surface, and construct the deformation equation:

[0046] Among them, is the Laplace matrix, and the element ; is the hard constraint weight to ensure that the displacement deviation of the feature points ; is a diagonal matrix. When is a feature point , otherwise is the deformation constraint vector, which is calculated from the difference between the initial surface and the target position.

[0047] Furthermore, based on the three-dimensional model of adaptive reconstruction, form a weighted complete graph through the key feature point set, use the improved Kruskal algorithm to generate a minimum spanning tree to determine the preliminary connection relationship, and introduce the simulated annealing algorithm to iteratively optimize the connection relationship through the energy function, including, Map the key feature point set to graph nodes, construct weighted connection edges between any two points to generate a weighted complete graph, and the weight of the connection edge is calculated as follows:

[0048] Among them, , are the three-dimensional coordinates of the key feature points; Global average point ; is the included angle between the normal vectors of adjacent patches; is the semantic difference factor , is the component confidence; , are the weight coefficients; Use the improved Kruskal algorithm to sort the edge set in ascending order for the weighted complete graph, use the union-find data structure to monitor the connected components, generate a minimum spanning tree and retain the redundant edge set; Introduce the simulated annealing algorithm for the minimum spanning tree structure to iteratively optimize the connection path and output the corresponding topological connection structure.

[0049] Furthermore, verify the developability of the connection result through Kuratowski's theorem, identify and process non-planar regions, including, Based on the optimized topological connection structure, the developability is verified through Kuratowski's theorem to identify the non-developable non-planar regions in the topological connection structure; Topologically divide the connection structure of the non-planar region and reconstruct it into a corresponding developable surface sub-graph using NURBS surfaces.

[0050] Furthermore, apply the QEM algorithm to lightweight the 3D model and complete the 3D modeling of the target component, including, Based on the edge collapse cost function Correspondingly compress the non-critical and critical regions, and optimize the multi-level texture storage structure using BC7 texture compression and JPEG-XL encoding; where v is the position of the vertex after edge collapse, 、 is the quadratic error matrix of adjacent triangular patches; Double verify the reconstruction accuracy and geometric integrity of the lightweight 3D model.

[0051] Specifically, during the model reconstruction process, use graph theory algorithms to topologically optimize the feature points. Treat the feature points as nodes in the graph, and construct a connection relationship graph between the feature points by calculating the spatial position relationships such as the distance and angle between the nodes. Optimize the algorithm through the simulated annealing algorithm, adjust the connection relationships between the feature points, simplify the model structure, optimize the connection paths between the feature points, and find the optimal or approximate optimal solution for the connection paths between the feature points. Ensure that the constructed model is composed of non-crossing planes or surfaces, thus avoiding incorrect model structures.

[0052] Map the feature point set to graph nodes, construct a weighted complete graph through geometric relationships and semantic constraints, and use improved graph theory algorithms to optimize the connection relationships to generate a lightweight topological structure.

[0053]

[0054] Among them, 、 are the three-dimensional coordinates of the feature points; Global average point ; is the included angle between the normal vectors of adjacent patches; is the semantic difference factor , is the component confidence; 、 are the weight coefficients, = 0.7.

[0055] The improved Kruskal algorithm is used to generate the minimum spanning tree, and the redundant edge set is retained. First, the edge set is sorted in ascending order. Secondly, the Union-Find data structure is used to monitor the connected components to generate the minimum spanning tree (MST), and the redundant edge set is retained.

[0056] Simulated annealing optimization is performed to construct an energy function to optimize the connection relationship:

[0057] In the formula, is the node degree; is the number of crossing edges in the projection plane.

[0058] Set the initial temperature T0 = 1000, the annealing coefficient α = 0.95, and the Markov chain length L = 1000. Perform state transition operations: edge exchange (randomly replace two edges with non-crossing new edges), node perturbation (apply Gaussian noise displacement), and topological mutation (reverse the connection of the local subgraph with a 10% probability) Apply Kuratowski's theorem to detect non-planar structures: construct an adjacency matrix , if there exists or subgraph, it is determined to be non-planar; the non-planar region is segmented to generate surface patches that satisfy the Euler characteristic ; use NURBS surface to reconstruct the non-developable region:

[0059] Among them, is the weight of the control point with inverse distance weighting, is the B-spline basis function.

[0060] Geometric simplification is performed using the QEM algorithm for lightweight processing, and the edge collapse cost is defined:

[0061] , is the quadratic error matrix of adjacent triangular patches. Iteratively collapse the low-cost edges, with a simplification rate of 70% for the secondary region and 30% for the important region. Texture optimization constructs anisotropic Mipmap, compresses it using the BC7 format, and jointly encodes it with JPEG-XL.

[0062] Verify the reconstruction quality, and perform double verification of the reconstruction accuracy and geometric integrity on the lightweight model, where the geometric error satisfies , and the topological integrity evaluation index: .

[0063] Second Embodiment Based on the same inventive concept, the present invention also provides a point cloud building component modeling system based on feature extraction, which adopts the point cloud building component modeling method as described above, including A data acquisition module, which is used to perform laser scanning on the target building component area, obtain the three-dimensional point cloud data of the building component and perform data preprocessing; A feature extraction module, which is used to construct multi-scale local neighborhoods centered on each point based on the preprocessed three-dimensional point cloud data, calculate the covariance matrix of the point sets in each scale local neighborhood and decompose the eigenvalues, extract the normalized curvature index, perform normalized mapping through the Sigmoid function to generate an edge probability map for screening candidate edge feature points, perform semantic segmentation on the candidate edge feature points through an improved PointNet++ network to output the corresponding component categories, and screen the corresponding key feature point sets based on the component categories; An adaptive reconstruction module, which is used to use the feature point cloud in the key feature point set as control points to construct a three-level B-spline surface model to fit the geometric shape of the target component, and dynamically adjust the corresponding spatial distribution density by setting a density adjustment function according to the importance of each feature point cloud in the preset local neighborhood for adaptively reconstructing the three-dimensional model of the target component and introducing the Laplace deformation algorithm for detail enhancement; A lightweight processing module, which is used to form a weighted complete graph based on the adaptively reconstructed three-dimensional model through the key feature point set, adopt an improved Kruskal algorithm to generate a minimum spanning tree to determine the preliminary connection relationship, introduce a simulated annealing algorithm to iteratively optimize the connection relationship through an energy function, verify the developability of the connection result through Kuratowski's theorem, identify and process non-planar regions, and apply the QEM algorithm to perform lightweight processing on the three-dimensional model to complete the three-dimensional modeling of the target component.

[0064] Although the present invention has been disclosed above with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solution of the present invention by using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the technical solution of the present invention shall fall within the protection scope of the technical solution of the present invention.

Claims

1. A point cloud building component modeling method based on feature extraction, characterized in that the steps Including: Performing laser scanning on the target building component area to obtain three-dimensional point cloud data of the building components and performing data preprocessing; Based on the preprocessed three-dimensional point cloud data, constructing multi-scale local neighborhoods centered on each point, calculating the covariance matrix for the point sets within each scale local neighborhood and decomposing the eigenvalues, extracting the normalized curvature index and performing normalized mapping through the Sigmoid function to generate an edge probability map for screening candidate edge feature points, performing semantic segmentation on the candidate edge feature points through an improved PointNet++ network to output the corresponding component categories, and screening the corresponding key feature point sets based on the component categories; Using the feature point cloud in the key feature point set as control points, constructing a three-level B-spline surface model to fit the geometric shape of the target component, and dynamically adjusting the corresponding spatial distribution density by setting a density adjustment function according to the importance of each feature point cloud within a preset local neighborhood for adaptively reconstructing the three-dimensional model of the target component and introducing the Laplace deformation algorithm for detail enhancement; Based on the adaptively reconstructed three-dimensional model, forming a weighted complete graph through the key feature point set, using the improved Kruskal algorithm to generate a minimum spanning tree for determining the preliminary connection relationship, introducing the simulated annealing algorithm to iteratively optimize the connection relationship through an energy function, performing developability verification on the connection result through Kuratowski's theorem, identifying and processing non-planar regions, and applying the QEM algorithm to perform lightweight processing on the three-dimensional model to complete the three-dimensional modeling of the target component.

2. The point cloud building component modeling method according to claim 1, wherein Based on the preprocessed three-dimensional point cloud data, constructing multi-scale local neighborhoods centered on each point, calculating the covariance matrix for the point sets within each scale local neighborhood and decomposing the eigenvalues, extracting the normalized curvature index and performing normalized mapping through the Sigmoid function to generate an edge probability map for screening candidate edge feature points, including: Calculating the covariance matrix of each point in the multi-scale local neighborhoods and decomposing to obtain the eigenvalues in the three-axis directions; Calculating the normalized curvature index based on the minimum eigenvalue in the three-axis directions; Based on the normalized curvature index of each point in the multi-scale local neighborhoods, constructing a multi-scale edge response function through weighted fusion, and performing normalized mapping through the Sigmoid function to generate the edge probability map; Screening the points in the edge probability map with edge response values greater than a preset threshold to obtain the candidate edge feature points.

3. The point cloud building component modeling method according to claim 2, characterized in that, Performing semantic segmentation on the candidate edge feature points through an improved PointNet++ network to output the corresponding component categories, and screening the corresponding key feature point sets based on the component categories, Including, Using a 9D feature vector composed of the spatial coordinates, normal vectors, and RGB color information of the candidate edge feature points as input, performing segmentation using the improved PointNet++ network, screening the candidate edge feature points with semantic confidence higher than a set confidence threshold, and outputting after labeling the corresponding component categories; Perform corresponding feature enhancement according to the output component category, and extract the corresponding key feature point set.

4. The point cloud building component modeling method according to claim 3, characterized in that, Perform corresponding feature enhancement according to the output component category, and extract the corresponding key feature point set; When the component category is a planar component, construct a plane fitting residual function based on the extracted candidate edge feature points, iteratively solve the optimal plane parameters through the L-M algorithm, and perform corner detection by combining the vertex spacing of the convex hull of the plane projection point set, and screen out the candidate edge feature points at the structural turning positions to form the key feature point set; When the component category is a cylindrical component, construct a cylindrical axis energy function, optimize the endpoint positions by the conjugate gradient method, introduce a geometric deviation degree, the semantic confidence degree and the point spacing to construct a multi-constraint scoring function, and screen out the corresponding candidate edge feature points through the branch and bound algorithm to form the key feature point set.

5. The point cloud building component modeling method according to claim 4, characterized in that Use the feature point cloud in the key feature point set as control points to construct a three-level B-spline surface model to fit the geometric shape of the target component, including: Construct an initial control grid based on the key feature point set, and use a layer-by-layer refinement mechanism to reconstruct through the B-spline surface model. The layers include a coarse-grained layer, a medium-grained layer and a fine-grained layer. The B-spline surface model is ； Among them, is the model level, is on the layer, and is a B-spline surface generated by parameter coordinates . , are B-spline basis functions in the u and v directions, and are used to construct the shape control function of the surface, is the control point coordinate corresponding to the i,j control grid node in the layer, , are the numbers of control points in the u and v directions in the l-th level.

6. The point cloud building component modeling method according to claim 5, characterized in that The density adjustment function is ; where is the number of points contained in a spherical neighborhood with a preset radius centered at (x, y, z); is the volume of the spherical neighborhood; is the curvature weight; is the semantic weight corresponding to the component category.

7. The point cloud building component modeling method according to claim 6, characterized in that, Based on the adaptively reconstructed 3D model, form a weighted complete graph through the key feature point set, use the improved Kruskal algorithm to generate a minimum spanning tree to determine the preliminary connection relationship, and introduce the simulated annealing algorithm to iteratively optimize the connection relationship through the energy function, including Map the set of key feature points to graph nodes, construct weighted connection edges between any two points to generate the weighted complete graph, and the weight of the connection edge is calculated as follows: Among them, and are the three-dimensional coordinates of the key feature points; Global average point ; is the included angle between the normal vectors of adjacent patches; is the semantic difference factor , is the component confidence; and are the weight coefficients; Use the improved Kruskal algorithm to sort the edge set of the weighted complete graph in ascending order, use the union-find data structure to monitor the connected components, generate the minimum spanning tree and retain the redundant edge set; Introduce the simulated annealing algorithm to iteratively optimize the connection path for the minimum spanning tree structure, and output the corresponding topological connection structure.

8. The point cloud building component modeling method according to claim 7, characterized in that Verify the developability of the connection result through Kuratowski's theorem, and identify and process non-planar regions, including Based on the optimized topological connection structure, perform the developability verification through Kuratowski's theorem, and identify the non-planar regions that are not developable in the topological connection structure; Topologically divide the connection structure of the non-planar region, and reconstruct it into a corresponding developable surface subgraph using NURBS surfaces.

9. The point cloud building component modeling method according to claim 8, characterized in that, Apply the QEM algorithm to lightweight the 3D model, and complete the 3D modeling of the target component, including Based on the edge collapse cost function Correspondingly compress the non-critical area and the critical area, and adopt BC7 texture compression and JPEG-XL encoding to optimize the multi-level texture storage structure; where v is the position of the vertex after edge collapse, , is the quadratic error matrix of adjacent triangular patches; Double-check the reconstruction accuracy and geometric integrity of the lightweighted 3D model.

10. A point cloud building component modeling system based on feature extraction, which adopts the point cloud building component modeling method described in any one of claims 1 to 9, and is characterized in that, including A data acquisition module for laser scanning the target building component area, obtaining the 3D point cloud data of the building component and performing data preprocessing; A feature extraction module, which is used to construct multi-scale local neighborhoods centered on each point based on the preprocessed three-dimensional point cloud data, calculate the covariance matrix and decompose the eigenvalues for the point sets within each scale local neighborhood, extract the normalized curvature index, perform normalized mapping through the Sigmoid function to generate an edge probability map for screening candidate edge feature points, perform semantic segmentation on the candidate edge feature points through an improved PointNet++ network to output the corresponding component categories, and screen the corresponding key feature point sets based on the component categories; An adaptive reconstruction module, which is used to use the feature point cloud in the key feature point set as control points to construct a three-level B-spline surface model to fit the geometric shape of the target component, and dynamically adjust the corresponding spatial distribution density by setting a density adjustment function based on the importance of each feature point cloud within a preset local neighborhood, for adaptively reconstructing the three-dimensional model of the target component and introducing the Laplace deformation algorithm for detail enhancement; A lightweight processing module, which is used to form a weighted complete graph based on the three-dimensional model of the adaptive reconstruction through the key feature point set, use the improved Kruskal algorithm to generate a minimum spanning tree to determine the preliminary connection relationship, introduce the simulated annealing algorithm to iteratively optimize the connection relationship through an energy function, verify the developability of the connection result through Kuratowski's theorem, identify and process non-planar regions, and apply the QEM algorithm to perform lightweight processing on the three-dimensional model to complete the three-dimensional modeling of the target component.

Images