A high-precision house model automatic generation method for a construction site

Abstract

This invention relates to the field of Building Information Modeling (BIM) technology, specifically to a method for automatically generating high-precision building models at construction sites, solving the technical problem of low modeling accuracy in existing technologies. The method includes: acquiring registered point cloud data of the construction site; extracting reference surface sample data from the registered point cloud data, and performing feature analysis on the registered point cloud data in conjunction with the reference surface sample data to determine the wall belonging weight of each sampling point in the registered point cloud data; the reference surface sample data is used to characterize the set of sampling points representing the original concrete forming surface; the wall belonging weight is used to characterize the probability that the corresponding sampling point belongs to a wall structure; and weighted parameter fitting is performed on the registered point cloud data based on the wall belonging weight of each sampling point in the registered point cloud data to generate a building information model.

Description

Technical Field

[0001] This invention relates to the field of building information modeling technology, specifically to a method for automatically generating high-precision house models at construction sites. Background Technology

[0002] With the continuous advancement of digital transformation in the construction industry, Building Information Modeling (BIM) technology has been widely applied throughout the entire lifecycle of building engineering, including design, construction, and operation and maintenance. In the main structural construction and digital quality acceptance phases of building engineering, acquiring 3D point cloud data of the construction site using 3D laser scanning technology and generating a BIM model based on this point cloud data is a core technological step in achieving automated assessment of construction quality and improving the digital level of project management. The accuracy and automation level of the generated model directly determine the efficiency and reliability of construction quality acceptance.

[0003] Currently, the mainstream methods for automatically generating building models in the industry are typically based on laser scanning point cloud data collected at the construction site. These data are then used to extract geometric parameters of the walls through planar fitting algorithms to generate the corresponding building information model. However, in actual construction sites of cast-in-place concrete structures, the complex on-site conditions place extremely high demands on the accuracy and robustness of the model generation. Existing technical solutions struggle to simultaneously guarantee the degree of automation and geometric accuracy of model generation under complex conditions, failing to meet the application requirements of high-precision, high-efficiency automated modeling for digital acceptance of construction quality in building engineering. Summary of the Invention

[0004] To address the technical problem of low modeling accuracy in existing technologies, the present invention aims to provide a method for automatically generating high-precision building models at construction sites. The specific technical solution adopted is as follows: This application provides a method for automatically generating high-precision house models at construction sites, including: Obtain registered point cloud data from the construction site; Based on the registered point cloud data, reference surface sample data is extracted, and feature analysis is performed on the registered point cloud data in combination with the reference surface sample data to determine the wall belonging weight of each sampling point in the registered point cloud data; the reference surface sample data is used to characterize the set of sampling points of the original concrete forming surface; the wall belonging weight is used to characterize the probability that the corresponding sampling point belongs to the wall structure. The registered point cloud data is fitted with weighted parameters based on the wall affiliation weights of each sampling point in the registered point cloud data to generate a building information model.

[0005] In one possible implementation, the method includes: Obtain the original point cloud data and design parameters of the construction site. The design parameters include the coordinates of the wall design center point and the design plane normal vector. Based on the design parameters, the original point cloud data is transformed to generate registered point cloud data.

[0006] In one possible implementation, the method includes: Voxel downsampling is performed on the registered point cloud data to generate sparse point cloud data. Statistical analysis is performed on the vertical deviation distance of each sampling point in the sparse point cloud data to extract reference surface sample data from the sparse point cloud data; the vertical deviation distance is used to characterize the signed vertical distance of the sampling point relative to the design reference surface. A surface connectivity graph is constructed based on sparse point cloud data, and spectral decomposition is performed on the surface connectivity graph to generate connectivity feature vectors. The connectivity feature vectors are used to characterize the numerical distribution of each sampling point in the topological feature space of the sparse point cloud data. Supervised probability estimation is performed based on connectivity feature vectors and reference surface sample data to determine the wall affiliation weights of each sampling point in the registered point cloud data.

[0007] In one possible implementation, the method includes: Calculate the vertical deviation distance of each sampling point in the sparse point cloud data relative to the design reference plane; A frequency distribution histogram is constructed based on the vertical deviation distance of each sampling point, and local peak intervals are obtained from the frequency distribution histogram. Reference surface sample data is constructed based on sampling points whose vertical deviation distance is located within the local peak range.

[0008] In one possible implementation, the method includes: Obtain the local neighborhood point set of each sampling point in the sparse point cloud data, and perform local plane fitting based on the local neighborhood point set of each sampling point to obtain the local normal vector of each sampling point; A surface connectivity graph is constructed based on the three-dimensional coordinates, vertical deviation distance, and local normal vectors of each sampling point in the sparse point cloud data. Perform spectral decomposition on the surface connectivity graph to generate connectivity feature vectors.

[0009] In one possible implementation, the method includes: Extract the feature values ​​corresponding to the reference surface sample data from the connectivity feature vector to construct the observation sample set; Kernel density estimation is performed based on the observed sample set to construct a probability density function; the probability density function is used to characterize the probability density of the distribution of the feature values ​​of the sampling points in the feature space; Determine the probability density of wall attribution for each sampling point in the sparse point cloud data based on the probability density function. The wall affiliation weights of each sampling point in the registered point cloud data are determined based on the wall affiliation probability density of each sampling point in the sparse point cloud data.

[0010] In one possible implementation, the method includes: The wall ownership probability density of each sampling point in the sparse point cloud data is normalized to obtain the sparse weight of each sampling point in the sparse point cloud data. Based on the mapping relationship between sparse point cloud data and the sampling points in the registered point cloud data, the sparse weights of each sampling point in the sparse point cloud data are mapped to each sampling point in the registered point cloud data, thus obtaining the wall belonging weights of each sampling point in the registered point cloud data.

[0011] In one possible implementation, the method includes: Based on the wall affiliation weights of each sampling point in the registered point cloud data, a weighted least squares fit is performed on the registered point cloud data to determine the wall position correction parameters; the wall position correction parameters are used to characterize the intercept of the fitted plane in the normal direction of the design plane. The fitting plane equation is determined based on the wall location correction parameters, and a building information model is generated by combining the preset wall thickness parameters.

[0012] In one possible implementation, the method includes: A weighted least squares objective function is constructed to characterize the sum of squared weighted distances from each sampling point in the registered point cloud data to the fitting plane. The normal vector constraint of the fitting plane is the design plane normal vector. The weighted least squares objective function is optimized to determine the wall position correction parameters.

[0013] In one possible implementation, the method further includes: Obtain the positional residual values ​​from each sampling point in the registered point cloud data to the corresponding location in the building information model; Quality defect detection is performed based on the wall attribution weights and location residuals of each sampling point in the registered point cloud data.

[0014] The present invention has the following beneficial effects: Based on the above technical solution, this application acquires registered point cloud data from the construction site to provide basic data under a unified coordinate reference for subsequent model generation. Then, it extracts reference surface sample data from the registered point cloud data and performs feature analysis on the registered point cloud data in conjunction with the reference surface sample data to determine the wall belonging weight of each sampling point in the registered point cloud data. Guided by the reference surface sample data, it effectively distinguishes wall structures from interference areas and quantifies the confidence level that each sampling point belongs to a wall structure. Finally, based on the wall belonging weight of each sampling point in the registered point cloud data, it performs weighted parameter fitting to generate a building information model. The weighted constraint of the wall belonging weight suppresses the adverse effects of interference data on the fitting results, ensuring the geometric accuracy of the generated building information model. Therefore, the technical solution provided by this application can accurately extract the wall structure base surface and generate a high-precision model that meets acceptance specifications under complex construction site conditions, effectively improving the accuracy and automation level of digital diagnosis of construction quality. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly described below.

[0016] Figure 1 This is a flowchart illustrating a method for automatically generating a high-precision house model at a construction site, provided in one embodiment of the present invention. Figure 2 This is a schematic diagram of the hardware structure of a high-precision automatic house model generation device for construction sites, provided in one embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a high-precision automatic generation method for building models at construction sites proposed in accordance with the present invention.

[0018] The following describes in detail, with reference to the accompanying drawings, a specific scheme for an automatic method for generating high-precision house models at construction sites provided by the present invention.

[0019] Please see Figure 1 The diagram illustrates a method flowchart for automatically generating a high-precision house model at a construction site according to an embodiment of the present invention. The method includes the following steps: Step 101: Obtain the registered point cloud data of the construction site.

[0020] Among them, the registered point cloud data refers to the set of three-dimensional laser scan point clouds that have been aligned with the architectural design coordinate system after coordinate transformation. It is the data input basis for subsequent processing and includes multiple sampling points corresponding to the wall structure at the construction site.

[0021] In one possible implementation, this application can obtain the original point cloud data and design parameters of the construction site, perform coordinate transformation on the original point cloud data according to the design parameters, and generate registered point cloud data.

[0022] In this context, design parameters refer to the predefined standard geometric parameters of the wall to be modeled in the architectural design model. These parameters include the coordinates of the wall's design center point and the design plane normal vector. The coordinates of the wall's design center point refer to the three-dimensional coordinates of the geometric center of the wall to be modeled in the design coordinate system, and the design plane normal vector refers to the standard normal vector of the design plane of the wall to be modeled in the design coordinate system.

[0023] For example, the above-mentioned raw point cloud data can be collected at the construction site by a 3D laser scanning device. It is a 3D point cloud collection that has not undergone coordinate transformation processing and includes 3D sampling data of the wall structure and surrounding environment at the construction site.

[0024] In some embodiments, coordinate transformation can be achieved based on rigid body transformation of rotation and translation components. Through rigid body transformation, the original point cloud data can be transformed from the local coordinate system of the scanning device to the architectural design coordinate system, so that the geometric reference of the original point cloud data is consistent with the geometric reference of the design parameters.

[0025] For example, this application can perform conventional denoising preprocessing on the original point cloud data, such as removing non-target noise points like ground points through pass-through filtering. Next, the coordinates of at least three non-collinear control points (which have known coordinates in the design coordinate system) are obtained from the construction site. The 3D coordinates of these control points are then manually or automatically identified in the filtered original point cloud data. Afterward, the rigid body transformation matrix (rotation matrix and translation vector) is solved using the Singular Value Decomposition (SVD) algorithm. The original point cloud data is then transformed to the design coordinate system using the rigid body transformation matrix. The dataset after this transformation is defined as the registered point cloud data, thus achieving alignment between the scanned data and the design coordinate system.

[0026] Step 102: Extract reference surface sample data from the registered point cloud data, and perform feature analysis on the registered point cloud data in combination with the reference surface sample data to determine the wall belonging weight of each sampling point in the registered point cloud data.

[0027] The reference surface sample data is used to characterize the set of sampling points on the original concrete forming surface. This reference surface sample data can serve as high-confidence sample data to guide feature analysis and provide a supervisory signal for subsequent probability calculations. The wall belonging weight is used to characterize the probability that the corresponding sampling point belongs to the wall structure. The larger the wall belonging weight, the higher the confidence that the corresponding sampling point belongs to the wall structure itself; the smaller the wall belonging weight, the higher the confidence that the corresponding sampling point belongs to the interference area.

[0028] The above steps extract reference surface sample data that can represent the real structure of the wall. Using the reference surface sample data as a reference, feature analysis is performed on the registered point cloud data. This can effectively separate the wall structure that is difficult to distinguish in geometric space from the interference area, and quantify the probability that each sampling point belongs to the wall structure, i.e. the wall belonging weight. This provides a reliable weight basis for subsequent model fitting and avoids the influence of interference data on the model fitting process.

[0029] Step 103: Based on the wall ownership weights of each sampling point in the registered point cloud data, perform weighted parameter fitting on the registered point cloud data to generate a building information model.

[0030] It should be noted that this application determines the wall belonging weight by extracting reference surface sample data and combining it with feature analysis, transforming the structural identification result into a continuous probability value, thus avoiding the edge jaggedness problem caused by hard threshold segmentation. In this way, this application can use the wall belonging weight as the weight basis for the fitting process, and use the wall belonging weight to weight and constrain the parameter fitting process, suppressing the influence of sampling points in the interference area on the fitting result, thereby realizing the planar fitting of the registered point cloud data, ensuring that the geometric parameters obtained by fitting can truly reflect the actual position of the wall structure, and thus generating a high-precision building information model.

[0031] Based on the above technical solution, this application acquires registered point cloud data from the construction site to provide basic data under a unified coordinate reference for subsequent model generation. Then, it extracts reference surface sample data from the registered point cloud data and performs feature analysis on the registered point cloud data in conjunction with the reference surface sample data to determine the wall belonging weight of each sampling point in the registered point cloud data. Guided by the reference surface sample data, it effectively distinguishes wall structures from interference areas and quantifies the confidence level that each sampling point belongs to a wall structure. Finally, based on the wall belonging weight of each sampling point in the registered point cloud data, it performs weighted parameter fitting to generate a building information model. The weighted constraint of the wall belonging weight suppresses the adverse effects of interference data on the fitting results, ensuring the geometric accuracy of the generated building information model. Therefore, the technical solution provided by this application can accurately extract the wall structure base surface and generate a high-precision model that meets acceptance specifications under complex construction site conditions, effectively improving the accuracy and automation level of digital diagnosis of construction quality.

[0032] As a possible embodiment of this application, the above-mentioned "extracting reference surface sample data based on the registered point cloud data, and performing feature analysis on the registered point cloud data in combination with the reference surface sample data to determine the wall belonging weight of each sampling point in the registered point cloud data" can be achieved through the following steps: Step 201: Perform voxel downsampling on the registered point cloud data to generate sparse point cloud data.

[0033] Voxel downsampling refers to a process that spatially partitions and downsamples registered point cloud data using a 3D voxel mesh. This significantly reduces the number of sampling points while preserving the macroscopic topological structure and core geometric features of the registered point cloud data, improving the computational efficiency of subsequent algorithms and avoiding the problems of low computational efficiency and excessive memory consumption caused by excessive data volume. Sparse point cloud data refers to a set of point clouds obtained after voxel downsampling, with a significantly reduced number of sampling points.

[0034] For example, this application can pre-set the side length of the voxel grid, dividing the three-dimensional space corresponding to the registered point cloud data into a solid grid with equal side lengths. The centroid of all sampling points within each voxel grid is calculated, and this centroid is used as the sampling point corresponding to that voxel grid in the sparse point cloud data. In this way, the data size can be significantly reduced while preserving the macroscopic topological structure of the data, thus ensuring the computational efficiency of the algorithm.

[0035] Step 202: Perform statistical analysis based on the vertical deviation distance of each sampling point in the sparse point cloud data, and extract the reference surface sample data from the sparse point cloud data.

[0036] The vertical deviation distance is used to characterize the signed vertical distance of a sampling point relative to the design reference plane. This design reference plane is the plane passing through the design center point of the wall and whose normal vector is the normal vector of the design plane. It serves as the zero-deviation reference for calculating the vertical deviation distance. The sign of the vertical deviation distance directly reflects the concavity or convexity of the sampling point relative to the design reference plane; for example, a positive value indicates that the sampling point is located outside the design reference plane, while a negative value indicates that the sampling point is located inside the design reference plane. This step, through statistical analysis of the vertical deviation distance, can filter out the sampling points that best represent the original concrete forming surface from the sparse point cloud data, obtaining high-confidence reference plane sample data.

[0037] In one possible implementation, this application can calculate the vertical deviation distance of each sampling point in the sparse point cloud data relative to the design reference plane, construct a frequency distribution histogram based on the vertical deviation distance of each sampling point, obtain local peak intervals from the frequency distribution histogram, and then construct reference plane sample data based on the sampling points whose vertical deviation distances are located within the local peak intervals.

[0038] For example, the vertical deviation distance satisfies the following formula: in, For the first The vertical deviation distance of each sampling point relative to the design reference plane. For the first The three-dimensional coordinates of each sampling point The coordinates of the center point of the wall design. To design the plane normal vector.

[0039] The dot product operation in the above formula represents a vector projection operation, used to calculate the signed perpendicular distance from the sampling point to the design reference plane, through vector subtraction. The direction vector pointing from the design center point to the sampling point is obtained. This direction vector is then projected onto the direction of the design plane's normal vector using a dot product operation to obtain the vertical distance of the sampling point relative to the design reference plane. A positive result indicates that the sampling point is located outside the design reference plane; a negative result indicates that the sampling point is located inside the design reference plane. The vertical deviation distance quantifies the subtle undulations of the wall surface and is a fundamental parameter for subsequent statistical analysis.

[0040] After obtaining the vertical deviation distances of the aforementioned sampling points, this application can perform statistical analysis on the vertical deviation distances of each sampling point to construct a frequency distribution histogram. To ensure statistical sensitivity, the group interval of the histogram is set to the smaller value between the scanner's measurement accuracy and the allowable deviation of the construction acceptance specifications. For example, when the scanner's distance measurement accuracy is 2mm and the allowable deviation for wall flatness is 8mm, the group interval can be 2mm. Subsequently, the histogram intervals are scanned in ascending order of distance values ​​(i.e., from the inner side of the wall to the outer side) to identify the local peak interval where the proportion of the first sampling point exceeds the total number of points threshold.

[0041] The frequency distribution histogram is a statistical histogram constructed with vertical deviation distance as the x-axis and the frequency of sampling points as the y-axis. It can intuitively reflect the overall distribution pattern of vertical deviation distance. Local peak intervals represent concentrated distribution areas of vertical deviation distance. For example, the preset threshold can be set to 5%, meaning that intervals where the number of sampling points exceeds 5% of the total number of points are identified as local peak intervals.

[0042] It should be noted that in engineering scenarios where bulging and attachment interference mainly manifest as outward convexity (increased distance value), the original base surface is always located at the bottom layer (i.e., the smallest distance value) in spatial superposition. Even if the wall has a large area of ​​bulging, resulting in more points in the bulging area than on the base surface, the original base surface will still appear as the first significant statistical peak on the left side of the histogram. Therefore, this application adopts the method of identifying the local peak interval where the proportion of the first sampling point exceeds the threshold of the total number of points, rather than the maximum peak interval, which can effectively avoid misjudgment caused by excessively large defect areas.

[0043] Finally, this application uses all sampling points whose vertical deviation distance falls within the local peak range as components of the reference surface sample data to obtain a high-confidence sample set that can represent the original concrete forming surface.

[0044] Step 203: Construct a surface connectivity graph based on sparse point cloud data, and perform spectral decomposition on the surface connectivity graph to generate connectivity feature vectors.

[0045] The connectivity feature vector is used to characterize the numerical distribution of each sampling point in the sparse point cloud data in the topological feature space. The surface connectivity graph is used to describe the local connectivity relationships between each sampling point in the sparse point cloud data. Through the surface connectivity graph, the geometric and topological features of the point cloud data can be transformed into numerical features of the graph structure.

[0046] For example, this application can perform eigenvalue decomposition on the Laplacian matrix of a surface connected graph based on spectral graph theory to obtain a connectivity feature vector that reflects the topology of the point cloud. Sampling points belonging to the same connected component will have highly clustered values ​​in the connectivity feature vector. In this way, the nonlinear deformation problem that is difficult to distinguish in geometric space can be transformed into a numerical clustering problem in feature space.

[0047] In one possible implementation, this application can obtain the local neighborhood point set of each sampling point in the sparse point cloud data, and perform local plane fitting based on the local neighborhood point set of each sampling point to obtain the local normal vector of each sampling point.

[0048] For example, this application can build a kd-tree spatial index based on sparse point cloud data. For each sampling point in the sparse point cloud data, search for its nearest Euclidean distance. Each set of neighboring points constitutes a local neighborhood point set. For example... It can be set to 15 to 20, which can provide robust local plane fitting results while ensuring computational efficiency.

[0049] Based on this, principal component analysis is used to perform local plane fitting on the neighboring points within the local neighborhood of the sampling point, and the local normal vector of the sampling point is calculated. This local normal vector reflects the spatial orientation of the small region where the sampling point is located, and is an important basis for subsequent judgment on whether the surface has undergone folding or abrupt changes.

[0050] Subsequently, a surface connectivity graph is constructed based on the three-dimensional coordinates, vertical deviation distance, and local normal vectors of each sampling point in the sparse point cloud data.

[0051] The edge weights of the surface connectivity graph are determined by the three-dimensional coordinates, vertical deviation distance, and local normal vector of each sampling point. This comprehensively considers the spatial proximity, depth continuity, and normal consistency between sampling points, numerically severing the connection between the wall structure and the interference facilities, and accurately distinguishing between continuous wall structures and broken interference areas.

[0052] For example, the edge weights satisfy the following formula: in, For the first The sampling point and the first Edge weights between sampling points For the first The three-dimensional coordinates of each sampling point For the first The three-dimensional coordinates of each sampling point The scale parameter for spatial proximity terms controls the influence of spatial distance on connectivity strength and can be set as the average spacing between sampling points in sparse point cloud data. to times. For the first Vertical deviation distance of each sampling point For the first Vertical deviation distance of each sampling point This is a scale parameter for the depth continuity term, used to control the influence of depth difference on connectivity strength. It can be set to the surface flatness deviation value allowed by the construction and acceptance specifications (e.g., ...). ). For the first Local normal vector of each sampling point For the first Local normal vector of each sampling point This is the scale parameter for the normal consistency term, used to control the degree of influence of normal differences on connectivity strength. It can be set as the angle threshold corresponding to the allowable perpendicularity deviation (e.g., ...). radian).

[0053] In the above formula Used to measure the distance difference between two points in three-dimensional space, ensuring that the local mesh maintains basic connectivity even with uneven data density. This measure assesses the depth difference between two points in a direction perpendicular to the design plane. When the depth difference between the two points significantly exceeds the allowable value specified in the code, this result rapidly approaches zero, numerically severing the connection between the wall and the protruding structure. This formula measures the difference in surface curvature between two points. The value decreases when the surface undergoes sharp angles or abrupt changes, further enhancing the ability to identify areas of geometric abrupt change. By incorporating multidimensional constraints, the formula explicitly severs the connection between the edge of the electromechanical facility and the wall body numerically, quantifying the microscopic fracture relationship through a connection strength matrix.

[0054] It should be noted that for any two sampling points, the edge weight is calculated using the above formula only if the two sampling points are spatially adjacent; otherwise, the edge weight can be directly reset to zero. Whether they are adjacent can be determined by the aforementioned local neighborhood point set.

[0055] Thus, this application can construct a surface connectivity graph using the proximity relationships and edge weights of each sampling point in the aforementioned sparse point cloud data. Furthermore, this application can traverse the edge weights of each pair of points in the surface connectivity graph to generate... A sparse symmetric matrix of dimension is defined as the surface connectivity strength matrix, where This represents the number of sampling points in the sparse point cloud data.

[0056] Finally, spectral decomposition is performed on the surface connectivity graph to generate connectivity feature vectors.

[0057] The core of spectral decomposition is to solve for the eigenvalues ​​of the Laplacian matrix of the surface-connected graph and extract the eigenvectors that reflect the graph's topological structure.

[0058] For example, this application can first calculate the degree matrix based on the surface connectivity strength matrix. The degree matrix is ​​a diagonal matrix, and the elements on its diagonal represent the total connection strength between each sampling point and other sampling points.

[0059] For example, the total connection strength satisfies the following formula: in, The first degree matrix on the diagonal The total connection strength between each sampling point and other sampling points The number of sampling points for sparse point cloud data. For the first The sampling point and the first Edge weights between sampling points.

[0060] It should be noted that the depth difference term in the above edge weight formula... and normal difference term This results in extremely small edge weights (approaching zero but strictly positive) between the interfering area (such as electromechanical facilities) and the wall itself, thus numerically creating a "soft cutoff." At this point, the surface connectivity graph remains a single connected component with only one zero eigenvalue. Therefore, the eigenvalues ​​and eigenvectors of the symmetric normalized Laplacian matrix can be solved using the Lanczos iterative algorithm. The eigenvector corresponding to the second smallest eigenvalue (i.e., the smallest non-zero eigenvalue) is extracted as the connectivity eigenvector (i.e., the Fiedler vector). The numerical distribution of this eigenvector reflects the cut structure of the point cloud at weak connections: the eigenvalues ​​of sampling points belonging to the wall itself are highly clustered, while the eigenvalues ​​of sampling points belonging to the interfering facilities deviate significantly from this clustering center.

[0061] For example, the symmetric normalized Laplace matrix satisfies the following formula: in, It is a symmetric normalized Laplace matrix. The identity matrix has the same dimensions as the surface connectivity strength matrix. The dimensions are consistent. For degree matrix, Let be the surface connectivity strength matrix.

[0062] It should be noted that in actual calculations, if a sampling point is an isolated point (i.e., the edge weights with all its neighbors are zero), then the degree matrix... The corresponding diagonal element ,lead to It cannot be calculated directly. To ensure the stability of the numerical calculation, a minimal perturbation term (e.g., ...) is added to all diagonal elements of the degree matrix. This ensures that the degree matrix is ​​strictly positive definite. This perturbation does not affect the principal structure of subsequent eigenvectors (because isolated points are unconnected in the graph, and their corresponding eigenvalues ​​can be ignored after the perturbation).

[0063] The above formula, through symmetric normalization, can effectively eliminate the influence of uneven point cloud sampling density on the graph structure. After spectral decomposition, feature vectors reflecting the topological connectivity of the point cloud can be extracted, realizing the mapping of topological relationships in geometric space to numerical distribution in feature space, providing core feature basis for subsequent feature clustering and probability estimation.

[0064] Step 204: Perform supervised probability estimation based on connectivity feature vectors and reference surface sample data to determine the wall affiliation weights of each sampling point in the registered point cloud data.

[0065] Supervised probability estimation refers to a method of estimating the probability density of the numerical distribution in the connectivity feature vector using the baseline sample data as the true sample. It can quantitatively calculate the probability that each sampling point belongs to the wall structure, avoid the category uncertainty problem in unsupervised clustering algorithms, and ensure the stability and robustness of weight calculation.

[0066] Based on the above technical solution, this application generates sparse point cloud data by voxel downsampling the registered point cloud data. This improves subsequent computational efficiency while preserving core geometric features. Then, statistical analysis is performed on the vertical deviation distance of each sampling point in the sparse point cloud data to extract high-confidence reference surface sample data, providing reliable ground truth samples for subsequent probability estimation. Next, a surface connectivity graph is constructed based on the sparse point cloud data and spectral decomposition is performed to generate connectivity feature vectors. This transforms the indistinguishable walls and interference areas in geometric space into distinguishable numerical clusters in feature space. Finally, supervised probability estimation is performed based on the connectivity feature vectors and reference surface sample data to determine the wall affiliation weights for each sampling point, effectively separating the wall structure from the interference area and ensuring the accuracy and stability of the weight calculation. This solution can accurately distinguish the wall structure itself from the interference area under complex working conditions, providing a reliable weight basis for subsequent weighted fitting and further improving the accuracy and robustness of the generated building information model.

[0067] As a possible embodiment of this application, the above-mentioned "supervised probability estimation based on connectivity feature vectors and reference surface sample data to determine the wall belonging weight of each sampling point in the registered point cloud data" can be achieved through the following steps: Step 301: Extract the feature values ​​corresponding to the reference surface sample data from the connectivity feature vector to construct the observation sample set.

[0068] For example, this application can obtain the feature values ​​corresponding to all sampling points in the reference surface sample data from the connectivity feature vector, and gather all the extracted feature values ​​to form an observation sample set. This observation sample set represents the numerical distribution of the concrete base surface known to be most certain in the feature space, and is the true value sample for supervised probability estimation.

[0069] Step 302: Perform kernel density estimation based on the observed sample set and construct the probability density function.

[0070] Among them, the Kernel Density Estimation (KDE) algorithm can fit the true data distribution shape by superimposing kernel functions without assuming the shape of the data distribution. The probability density function is used to characterize the probability density of the distribution of the feature values ​​of the sampling points in the feature space. The larger the output value of the probability density function, the higher the probability that the sampling point of the corresponding feature value belongs to the wall structure itself.

[0071] For example, the probability density function satisfies the following formula: in, Let be the probability density function. The input values ​​are the feature values ​​corresponding to each sampling point in the sparse point cloud data. For the observation sample set The number of eigenvalues ​​in the data. The bandwidth parameter controls the smoothness of the probability density curve. A smaller bandwidth parameter results in a more refined fit to the sample data, while a larger bandwidth parameter produces a smoother curve. The Silverman rule of thumb can be used to calculate automatically, that is... ,in The standard deviation of the sample is 1. The interquartile range. Alternatively, when computational resources are limited, it can be set to a fixed empirical value (e.g., This ensures that the probability distribution curve has a moderate smoothness. For the feature values ​​in the observed sample set, The kernel function is used to weight the influence range of each sample. In this embodiment, the standard Gaussian kernel function is selected. .in These are the input variables for the kernel function. It is a natural constant. Pi is the mathematical constant of a circle.

[0072] Step 303: Determine the probability density of wall attribution for each sampling point in the sparse point cloud data based on the probability density function.

[0073] This application can input the feature value corresponding to each sampling point in sparse point cloud data into the probability density function to calculate the wall belonging probability density corresponding to each sampling point. The wall belonging probability density can reflect the probability that the corresponding sampling point belongs to the wall structure itself.

[0074] Step 304: Determine the wall affiliation weight of each sampling point in the registered point cloud data based on the wall affiliation probability density of each sampling point in the sparse point cloud data.

[0075] In one possible implementation, this application can normalize the wall attribution probability density of each sampling point in the sparse point cloud data to obtain the sparse weights of each sampling point in the sparse point cloud data. Then, based on the mapping relationship between the sparse point cloud data and the sampling points in the registered point cloud data, the sparse weights of each sampling point in the sparse point cloud data are mapped to each sampling point in the registered point cloud data to obtain the wall attribution weights of each sampling point in the registered point cloud data.

[0076] For example, sparse weights satisfy the following formula: in, For the first Sparse weights for each sampling point For the first The feature values ​​of each sampling point For the first The probability density of wall attribution for each sampling point. For the first The feature values ​​of each sampling point For the first The probability density of wall attribution for each sampling point. This represents the maximum value of the probability density of wall ownership at each sampling point.

[0077] The above formula divides the wall attribution probability density of each sampling point by the maximum value of the wall attribution probability density, performs maximum value normalization processing, and generates sparse weights.

[0078] Since sparse point cloud data is generated from registered point cloud data through voxel downsampling, each sparse point corresponds to a voxel grid, which contains multiple sampled points from the registered point cloud data. Therefore, for each sampled point in the registered point cloud data, the sparse weights corresponding to its respective voxel grid can be directly inherited to obtain the wall belonging weights of the registered point cloud data. Finally, the wall belonging weights, corresponding one-to-one with the sampled points in the registered point cloud data, are output as the core input for subsequent weighted fitting.

[0079] Based on the above technical solution, this application constructs an observation sample set by extracting feature values ​​corresponding to the reference surface sample data from the connectivity feature vector, obtaining high-confidence wall structure distribution samples in the feature space, then performing kernel density estimation based on the observation sample set to construct a probability density function, realizing the quantitative calculation of the probability of attribution for any feature value in the feature space, then determining the wall attribution probability density of each sampling point in the sparse point cloud data according to the probability density function, quantifying the probability that each sampling point belongs to a wall structure, and finally determining the wall attribution weight of each sampling point in the registered point cloud data based on the wall attribution probability density of each sampling point in the sparse point cloud data. This solution can use the reference surface sample data as a supervision signal to achieve stable and reliable probability estimation, generate continuous wall attribution weights, avoid the randomness of unsupervised clustering algorithms, ensure the stability and robustness of the algorithm under different construction scenarios, and solve the edge jaggedness problem caused by hard threshold segmentation, further improving the accuracy of model generation.

[0080] As a possible embodiment of this application, the above-mentioned "weighted parameter fitting of the registered point cloud data based on the wall ownership weight of each sampling point in the registered point cloud data to generate a building information model" can be achieved through the following steps: Step 401: Based on the wall affiliation weights of each sampling point in the registered point cloud data, perform weighted least squares fitting on the registered point cloud data to determine the wall position correction parameters.

[0081] Among them, the wall position correction parameter is used to characterize the intercept of the fitting plane in the normal direction of the design plane. Combined with the normal vector of the design plane, the precise position of the wall in space of the reconstructed building information model can be uniquely determined.

[0082] In one possible implementation, this application can construct a weighted least squares objective function.

[0083] The weighted least squares objective function is used to characterize the weighted sum of squared distances from each sampling point in the registered point cloud data to the fitting plane, and the normal vector constraint of the fitting plane is the design plane normal vector.

[0084] In order to meet the standard requirements for evaluating deviations based on design axes in building construction acceptance, this application can introduce the design plane normal vector as a strong constraint to keep the normal vector of the fitted plane consistent with the design plane normal vector.

[0085] For example, the weighted least squares objective function satisfies the following formula: in, The objective function is the weighted least squares function. The parameters for correcting the wall position are the intercepts of the fitted plane in the normal direction of the design plane. This represents the number of sampling points in the registered point cloud data. For the registered point cloud data The weight of the wall belonging degree of each sampling point. For the first The three-dimensional coordinates of each sampling point To design the plane normal vector.

[0086] The above formula iterates through all sampling points, calculating the distance from each sampling point to the fitting plane. Then, a weighted summation is performed, incorporating wall affiliation weights. This weighting automatically adjusts the contribution of each sampling point to the objective function, enhancing the contribution of high-weighted sampling points and suppressing the contribution of low-weighted sampling points, thus achieving robust handling of interference data. The final weighted least squares objective function value is used to evaluate the overall fit between the fitted plane and the weighted point cloud data under given wall position correction parameters; a smaller objective function value indicates a better fit.

[0087] Then, the weighted least squares objective function is optimized to determine the wall position correction parameters.

[0088] Among them, the weighted least squares objective function Regarding parameters By taking the derivative and setting it to zero, the optimal wall position correction parameters can be obtained. .

[0089] For example, the wall position correction parameters satisfy the following formula: in, Adjust parameters for wall position. This represents the number of sampling points in the registered point cloud data. For the registered point cloud data The weight of the wall belonging degree of each sampling point. For the first The three-dimensional coordinates of each sampling point To design the plane normal vector. This is a safety parameter used to correct for fractions with a denominator of 0. It ensures the stability of numerical calculations and avoids calculation anomalies caused by a zero denominator. It can be set to... .

[0090] In the above formula This represents the projection moment of the weighted point cloud onto the normal of the design plane. This represents the sum of weights plus a smoothing term, which is used to calculate the weighted centroid projection position of the full-field point cloud in the design normal direction after removing interference, through weighted averaging.

[0091] Step 402: Determine the fitting plane equation based on the wall location correction parameters, and generate a building information model by combining the preset wall thickness parameters.

[0092] Among them, the fitted plane equation represents the wall plane determined based on the design plane normal vector and the wall position correction parameters, and the preset wall thickness parameters refer to the standard wall thickness parameters specified in the architectural design model.

[0093] This application can use the fitted plane as one side of the outer surface of the wall, and stretch the distance corresponding to the preset wall thickness parameter in the opposite direction of the design plane normal vector to generate a closed three-dimensional solid model of the wall, i.e., a building information model.

[0094] For example, the generated building information model can be exported as an IfcWall object conforming to the Industry Foundation Classes (IFC) standard, meeting the requirements for digital delivery of building engineering projects.

[0095] Based on the above technical solution, this application can perform weighted least squares fitting on the registered point cloud data according to the wall belonging weight of each sampling point in the registered point cloud data, determine the wall position correction parameters, and suppress the influence of interference data on the fitting results through weight constraints to obtain the optimal geometric parameters that can truly reflect the position of the wall body. Then, the fitting plane equation is determined according to the wall position correction parameters, and a building information model is generated by combining it with preset wall thickness parameters, realizing the automatic conversion from point cloud data to a standardized building information model (BIM) entity model. The above solution can ensure that the generated building information model conforms to the real concrete base surface in terms of geometric accuracy and meets the architectural design benchmarks in terms of spatial orientation, further improving the accuracy and standardization of model generation and meeting the application requirements of digital acceptance of construction quality.

[0096] As one possible embodiment of this application, the method further includes the following steps: Step 501: Obtain the position residual value of each sampling point in the registered point cloud data to the corresponding position in the building information model.

[0097] Among them, the location residual value refers to the vertical distance between the sampling point in the registered point cloud data and the wall plane of the generated building information model, which can intuitively reflect the concavity and convexity of the sampling point relative to the fitted wall plane.

[0098] For example, the location residual value satisfies the following formula: in, For the first The location residual value of each sampling point For the first The three-dimensional coordinates of each sampling point To design the plane normal vector, Correct parameters for the wall position.

[0099] Step 502: Detect quality defects based on the wall belonging weight and location residual value of each sampling point in the registered point cloud data.

[0100] In some embodiments, this application may perform quality defect detection by means of threshold determination.

[0101] If the wall belonging weight of a sampling point is greater than the weight threshold (indicating that the sampling point belongs to the wall connected region topologically) and the location residual value is greater than the flatness deviation threshold (indicating that the sampling point is geometrically significantly convex), then it is determined that there is a bulging quality defect in the area where the sampling point is located. This application can determine the residual value of the area. The mapping is converted into a color cloud map and written into the Pset_ConcreteElementGeneral attribute set of the BIM model as the basis for rectification.

[0102] If the wall affiliation weight of a sampling point is less than or equal to the weight threshold (indicating that the sampling point is topologically disconnected from the wall) and the location residual value is greater than the structural deviation threshold, then the area where the sampling point is located is determined to be an ancillary facility such as a distribution box or conduit. This application can generate marker information at the corresponding location in the model to prompt electromechanical professionals to review and avoid misjudging it as a structural deviation.

[0103] The weight threshold can range from 0 to 1. For example, a value of 0.5 indicates that the sampling point is closer to the center of the reference surface sample in the feature space, meaning that the probability of it belonging to the connected area of ​​the wall exceeds 50%. The flatness deviation threshold can be the allowable deviation of the surface flatness of the cast-in-place structure. Generally, 8mm is used for walls. The structural deviation threshold is used to identify the minimum significant protrusion height of electromechanical facilities (distribution boxes, conduits, etc.). To avoid misjudging minor surface unevenness (such as formwork joints) as facilities, the structural deviation threshold can be set to 5mm.

[0104] Based on the above technical solution, this application can obtain the positional residual values ​​of each sampling point in the registered point cloud data to its corresponding position in the building information model, quantify the deviation of each sampling point relative to the generated wall model, and then perform quality defect detection based on the wall belonging weight and positional residual values ​​of each sampling point in the registered point cloud data. Through joint analysis of topological weights and geometric deviations, it achieves automatic classification of construction defects and legitimate facilities, avoiding the problem of misclassifying electromechanical facilities as structural deviations. This solution can simultaneously complete intelligent diagnosis of construction quality on the basis of automatically generating a high-precision building information model, realizing a closed loop of the entire process from data acquisition to digital delivery and quality acceptance, further improving the automation level and accuracy of digital management and control of construction quality in building engineering.

[0105] This application embodiment also provides a hardware structure diagram of a high-precision automatic house model generation device for construction sites (referred to as high-precision automatic house model generation device 20 for construction sites), see [link to diagram]. Figure 2 The high-precision house model automatic generation device 20 for the construction site includes a processor 21, and optionally, a memory 22 connected to the processor 21.

[0106] In the first possible implementation, see Figure 2 The high-precision house model automatic generation device 20 for construction sites also includes a communication interface 23. The processor 21, memory 22, and communication interface 23 are connected via a bus. The communication interface 23 is used to communicate with other devices or communication networks. Optionally, the communication interface 23 may include a transmitter and a receiver. The device in the communication interface 23 used to implement the receiving function can be considered as a receiver, which is used to perform the receiving steps in the embodiments of this application. The device in the communication interface 23 used to implement the transmitting function can be considered as a transmitter, which is used to perform the transmitting steps in the embodiments of this application.

[0107] Based on the first possible implementation method Figure 2 The structural diagram shown can be used to illustrate the structure of the high-precision house model automatic generation device for construction sites involved in the above embodiments.

[0108] in, Figure 2The diagram can also illustrate the system chip in the high-precision house model automatic generation device at the construction site. In this case, the actions performed by the aforementioned high-precision house model automatic generation device at the construction site can be implemented by this system chip. The specific actions performed are described above and will not be repeated here.

[0109] In implementation, each step of the method provided in this embodiment can be completed by integrated logic circuits in the processor or by instructions in software form. The steps of the method disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or being executed by a combination of hardware and software modules in the processor.

Claims

1. A method for automatically generating high-precision house models at construction sites, characterized in that, include: Obtain registered point cloud data from the construction site; Based on the registered point cloud data, reference surface sample data is extracted, and feature analysis is performed on the registered point cloud data in combination with the reference surface sample data to determine the wall belonging weight of each sampling point in the registered point cloud data. The reference surface sample data is used to characterize the set of sampling points on the original concrete forming surface; the wall belonging weight is used to characterize the probability that the corresponding sampling point belongs to the wall structure. Based on the wall affiliation weights of each sampling point in the registered point cloud data, a weighted parameter fitting is performed on the registered point cloud data to generate a building information model.

2. The method for automatically generating high-precision house models at construction sites according to claim 1, characterized in that, Obtain the registered point cloud data of the construction site, including: Obtain the original point cloud data and design parameters of the construction site, including the coordinates of the wall design center point and the design plane normal vector; The original point cloud data is transformed according to the design parameters to generate the registered point cloud data.

3. The method for automatically generating high-precision house models at construction sites according to claim 1, characterized in that, Based on the registered point cloud data, reference surface sample data is extracted, and feature analysis is performed on the registered point cloud data in conjunction with the reference surface sample data to determine the wall attribution weight of each sampling point in the registered point cloud data, including: The registered point cloud data is subjected to voxel downsampling to generate sparse point cloud data; Statistical analysis is performed on the vertical deviation distance of each sampling point in the sparse point cloud data to extract the reference surface sample data from the sparse point cloud data; the vertical deviation distance is used to characterize the signed vertical distance of the sampling point relative to the design reference surface; A surface connectivity graph is constructed based on the sparse point cloud data, and spectral decomposition is performed on the surface connectivity graph to generate a connectivity feature vector; the connectivity feature vector is used to characterize the numerical distribution of each sampling point in the sparse point cloud data in the topological feature space. Supervised probability estimation is performed based on the connectivity feature vector and the reference surface sample data to determine the wall affiliation weight of each sampling point in the registered point cloud data.

4. The method for automatically generating high-precision house models at construction sites according to claim 3, characterized in that, Based on the statistical analysis of the vertical deviation distance of each sampling point in the sparse point cloud data, the reference surface sample data is extracted from the sparse point cloud data, including: Calculate the vertical deviation distance of each sampling point in the sparse point cloud data relative to the design reference plane; A frequency distribution histogram is constructed based on the vertical deviation distance of each sampling point, and local peak intervals are obtained from the frequency distribution histogram. The reference surface sample data is constructed based on the sampling points whose vertical deviation distance is located within the local peak interval.

5. The method for automatically generating high-precision house models at construction sites according to claim 3, characterized in that, A surface connectivity graph is constructed based on the sparse point cloud data, and spectral decomposition is performed on the surface connectivity graph to generate connectivity feature vectors, including: The local neighborhood point set of each sampling point in the sparse point cloud data is obtained, and local plane fitting is performed based on the local neighborhood point set of each sampling point to obtain the local normal vector of each sampling point. The surface connectivity graph is constructed based on the three-dimensional coordinates, vertical deviation distance, and local normal vector of each sampling point in the sparse point cloud data. The surface connectivity graph is subjected to spectral decomposition to generate the connectivity feature vector.

6. The method for automatically generating high-precision house models at construction sites according to claim 3, characterized in that, Based on the connectivity feature vector and the reference surface sample data, supervised probability estimation is performed to determine the wall affiliation weight of each sampling point in the registered point cloud data, including: Extract the feature values ​​corresponding to the reference surface sample data from the connectivity feature vector to construct the observation sample set; Kernel density estimation is performed based on the observed sample set to construct a probability density function; the probability density function is used to characterize the probability density of the distribution of feature values ​​of the sampling points in the feature space; The probability density of wall attribution for each sampling point in the sparse point cloud data is determined based on the probability density function. The wall affiliation weight of each sampling point in the registered point cloud data is determined based on the wall affiliation probability density of each sampling point in the sparse point cloud data.

7. The method for automatically generating high-precision house models at construction sites according to claim 6, characterized in that, The wall affiliation weight of each sampling point in the registered point cloud data is determined based on the wall affiliation probability density of each sampling point in the sparse point cloud data, including: The wall attribution probability density of each sampling point in the sparse point cloud data is normalized to obtain the sparse weight of each sampling point in the sparse point cloud data. Based on the mapping relationship between the sparse point cloud data and the sampling points in the registered point cloud data, the sparse weights of each sampling point in the sparse point cloud data are mapped to each sampling point in the registered point cloud data to obtain the wall belonging weights of each sampling point in the registered point cloud data.

8. The method for automatically generating high-precision house models at construction sites according to claim 1, characterized in that, Based on the wall ownership weights of each sampling point in the registered point cloud data, a weighted parameter fitting is performed on the registered point cloud data to generate a building information model, including: Based on the wall affiliation weights of each sampling point in the registered point cloud data, a weighted least squares fit is performed on the registered point cloud data to determine the wall position correction parameters; the wall position correction parameters are used to characterize the intercept of the fitting plane in the normal direction of the design plane. The fitted plane equation is determined based on the wall position correction parameters, and the building information model is generated by combining the preset wall thickness parameters.

9. The method for automatically generating high-precision house models at construction sites according to claim 8, characterized in that, Based on the wall affiliation weights of each sampling point in the registered point cloud data, a weighted least squares fit is performed on the registered point cloud data to determine the wall position correction parameters, including: A weighted least squares objective function is constructed, which is used to characterize the weighted sum of squared distances from each sampling point in the registered point cloud data to the fitting plane, and the normal vector constraint of the fitting plane is the design plane normal vector; The weighted least squares objective function is optimized and solved to determine the wall position correction parameters.

10. The method for automatically generating high-precision house models at construction sites according to claim 1, characterized in that, The method further includes: Obtain the position residual value of each sampling point in the registered point cloud data to the corresponding position in the building information model; Quality defect detection is performed based on the wall attribution weight and location residual value of each sampling point in the registered point cloud data.

Images