BIM comparison-based fabricated building construction progress determination method and system

Abstract

The application provides a kind of based on BIM comparison's fabricated building construction progress determination method, system, including acquisition construction site fabricated building three-dimensional point cloud data, through cycle iteration completes component point cloud segmentation, each round selects the point with minimum local geometric entropy and normalized neighborhood density variance weighted sum as seed point, relies on seed point to establish growth area and iteration expansion, combines weighted covariance matrix, Mahalanobis distance and anisotropic growth inhibition factor setting determination condition, completes the independent segmentation of various component point clouds, extracts the centroid, main direction, circumscribed body size and other geometric characteristics of segmented component, compares and matches with BIM model preset parameter, identifies actual assembly component on site, according to the proportion of the number of matched components on site and the total number of components of BIM model, accurately estimates the actual construction progress of fabricated building on site, which helps to improve the segmentation accuracy and the reliability of progress determination result, adapts to the progress accounting needs of fabricated construction site.

Description

Technical Field

[0001] This application belongs to the field of building construction, and in particular relates to a method and system for determining the construction progress of prefabricated buildings based on BIM comparison. Background Technology

[0002] The key to prefabricated building construction lies in the efficient integration of factory-prefabricated components and on-site assembly operations. Timely and accurate monitoring of on-site construction progress is crucial for ensuring project schedules and improving project management. Traditional construction progress monitoring mainly relies on manual inspections or two-dimensional image comparisons, which suffers from high manpower requirements, significant subjective judgment, and susceptibility to interference from camera angles and site obstructions. With the widespread application of 3D laser scanning technology and Building Information Modeling (BIM), acquiring 3D point cloud data of the construction site using scanning equipment and comparing it with the established BIM model has become an important technical approach for construction progress monitoring. However, the complex environment of construction sites, coupled with uneven density, equipment obstruction, interference from temporary facilities, and significant environmental noise, makes it challenging to accurately extract the geometry of specific prefabricated components from the overall point cloud. If the component point cloud segmentation results are inaccurate, subsequent matching and identification with BIM model parameters will be affected, leading to inaccurate construction progress assessments.

[0003] Region growing algorithms typically select an initial seed point and gradually expand to the neighborhood based on geometric features such as normal vectors and curvature until the preset growth conditions are no longer met. In the seed point selection stage, traditional methods often use random selection or simply rely on the principle of minimum curvature, failing to fully consider local geometric entropy and neighborhood density distribution characteristics. This makes them prone to misselecting boundary points, seam points, or outliers as seed points under noise interference, leading to subsequent region shifts. In the iterative expansion stage, traditional region growing methods usually use Euclidean distance and fixed angle thresholds as criteria, making it difficult to weight and update the covariance matrix based on the distance difference from the point to the region centroid, and failing to fully consider the abrupt changes in the normal field gradient at component boundaries. Therefore, when processing seams, sharp edges, or complex corner areas of prefabricated components, over-segmentation or under-segmentation is likely to occur, reducing the extraction accuracy of geometric features such as the centroid, principal direction, and circumscribed body dimensions of the component point cloud, and affecting the matching and recognition effect with the BIM model, thus making it difficult to obtain reliable calculation results for prefabricated building construction progress. Summary of the Invention

[0004] To improve the accuracy of monitoring the construction progress of prefabricated buildings, this invention provides a method and system for determining the construction progress of prefabricated buildings based on BIM comparison.

[0005] According to one aspect of the present invention, a method for determining the construction progress of prefabricated buildings based on BIM comparison is provided, comprising the following steps: Acquire 3D point cloud data of prefabricated buildings; The component segmentation is performed iteratively until the target component point cloud segmentation is completed, or the remaining unsegmented points meet the preset criteria for noise points, occlusion residue points, or points to be verified. In each iteration, the eigenvalues ​​of the three-dimensional coordinate covariance matrix in the neighborhood of each point in the unsegmented point cloud are calculated to obtain the local geometric entropy and the normalized neighborhood density variance. The point with the smallest weighted sum of the local geometric entropy and the normalized neighborhood density variance is used as the initial seed point for the current iteration. The growth region is constructed using the initial seed point and iteratively expanded. In each iteration, the weighted covariance matrix of all points in the current growth region is updated, where the weight of each point in the current growth region is inversely proportional to its distance to the centroid of the current growth region. The Mahalanobis distance of the point to be determined on the boundary of the growth region is calculated. The anisotropic growth suppression factor of the boundary normal field gradient is calculated. When the Mahalanobis distance is less than the distance threshold and the angle between the normal vector of the point to be determined and the average normal vector of the region is less than the angle threshold corrected by the anisotropic growth suppression factor, the point to be determined is incorporated into the growth region until the segmentation of a single segmentation component point cloud is completed. Extract the centroid position, principal direction, and circumscribed body size of the point cloud of each segmented component as geometric features; match the geometric features with the preset parameters of the building information model to identify the corresponding components; determine the construction progress based on the ratio of the total number of matched components to the total number of components in the building information model.

[0006] Optionally, acquiring the three-dimensional point cloud data of the prefabricated building includes: The construction site was scanned using a 3D LiDAR scanner to obtain raw point cloud data; Three-dimensional pass-through filtering is used to isolate the raw point cloud data from irrelevant external regions; A voxel grid is constructed to downsample the point cloud, and the average coordinates of points within each voxel are calculated as representative points to obtain a simplified point set. Calculate the average distance between each point in the simplified point set and its neighboring point sets, and remove outliers whose absolute difference between the average distance and the global average distance is greater than a multiple of the standard deviation threshold to obtain the three-dimensional point cloud data.

[0007] Optionally, in each iteration, the eigenvalues ​​of the 3D coordinate covariance matrix in the neighborhood of each point in the unsegmented point cloud are calculated to obtain the local geometric entropy and the normalized neighborhood density variance, including: A three-dimensional coordinate covariance matrix is ​​constructed for the neighborhood of each point in the unsegmented point cloud and eigenvalue decomposition is performed. Three eigenvalues ​​are extracted and normalized to obtain three normalized eigenvalues. Calculate the negative of the product of each normalized eigenvalue and its natural logarithm, and add the three negative values ​​together to obtain the local geometric entropy; A spatial neighborhood sphere is constructed with each point in the unsegmented point cloud as its center and a preset detection radius. The local neighborhood density is obtained by dividing the number of points in the spatial neighborhood sphere by the volume of the spatial neighborhood sphere. Calculate the overall average density of the point cloud, and divide the square of the difference between the local neighborhood density and the overall average density by the square of the overall average density to obtain the normalized neighborhood density variance.

[0008] Optionally, updating the weighted covariance matrix of all points within the current growth region includes: Extract the three-dimensional coordinates of all points within the current growth region, calculate the average value of each coordinate axis, and obtain the centroid coordinates of the current growth region; Calculate the distance from each point to the centroid coordinates, take the reciprocal of the distance and normalize it to obtain the weight parameters of each point; Based on the difference between the coordinates of each point in the current growth region and the coordinates of the centroid, column vectors and row vectors are generated respectively, and the column vectors and row vectors are multiplied to obtain the initial covariance matrix of each point; The initial covariance matrix of each point is multiplied by the corresponding weight parameter, and the weighted covariance matrix of all points is summed to obtain the weighted covariance matrix.

[0009] Optionally, the calculation of the Mahalanobis distance of the point to be determined on the boundary of the growth region includes: The pseudo-inverse of the weighted covariance matrix is ​​obtained by taking the pseudo-inverse of the weighted covariance matrix; Subtract the centroid coordinates of the current growth region from the three-dimensional coordinates of the point to be determined to obtain the deviation column vector; The transpose of the deviation column vector is multiplied on the left by the weighted covariance pseudo-inverse matrix, then multiplied on the right by the deviation column vector, and the square root is taken to obtain the Mahalanobis distance.

[0010] Optionally, the anisotropic growth inhibition factor for calculating the boundary normal field gradient includes: Obtain the K nearest neighbors of the point to be determined, and calculate the difference vector between the normal vector of each nearest neighbor and the normal vector of the point to be determined; Calculate the squared L2 norm of each difference vector and sum them to obtain the gradient index of the boundary normal field; The anisotropic growth inhibition factor is calculated using the natural constant e as the base and the negative value of the boundary normal field gradient index as the exponent.

[0011] Optionally, the extraction of the centroid position, principal direction, and circumscribed volume dimensions of the point cloud of each segmented component as geometric features includes: Calculate the average coordinates of the point cloud of each segmented component on the three coordinate axes, and use this as the centroid position; Calculate the covariance matrix of the point cloud of each segmented component and perform principal component analysis to extract three orthogonal eigenvectors corresponding to the largest eigenvalue, the second largest eigenvalue, and the smallest eigenvalue. Use the eigenvector corresponding to the largest eigenvalue as the principal direction. A local coordinate system is established with the centroid position as the origin and the three orthogonal feature vectors as coordinate axes. The point cloud of the segmented component is projected onto each coordinate axis of the local coordinate system. The difference between the maximum and minimum values ​​of the projection on each coordinate axis is calculated and used as the length, width, and height features of the circumscribed body.

[0012] According to another aspect of the present invention, a system for determining the construction progress of prefabricated buildings based on BIM comparison is provided, comprising the following modules: The acquisition module is used to acquire 3D point cloud data of prefabricated buildings; The calculation module is used to perform component segmentation in a loop until the target component point cloud segmentation is completed, or the remaining unsegmented points meet the preset criteria for noise points, occlusion residual points, or points to be verified. In each loop, the eigenvalues ​​of the three-dimensional coordinate covariance matrix in the neighborhood of each point in the unsegmented point cloud are calculated to obtain the local geometric entropy and the normalized neighborhood density variance. The point with the smallest weighted sum of the local geometric entropy and the normalized neighborhood density variance is used as the initial seed point for the current loop. The segmentation module is used to construct a growth region using the initial seed point and iteratively expand it. In each iteration, it updates the weighted covariance matrix of all points in the current growth region, where the weight of each point in the current growth region is inversely proportional to its distance to the centroid of the current growth region. It calculates the Mahalanobis distance of the point to be determined on the boundary of the growth region; it calculates the anisotropic growth suppression factor of the boundary normal field gradient; when the Mahalanobis distance is less than a distance threshold and the angle between the normal vector of the point to be determined and the average normal vector of the region is less than the angle threshold corrected by the anisotropic growth suppression factor, the point to be determined is incorporated into the growth region until the segmentation of a single segmentation component point cloud is completed. The determination module is used to extract the centroid position, principal direction, and circumscribed body size of the point cloud of each segmented component as geometric features; match the geometric features with preset parameters of the building information model to identify the corresponding components; and determine the construction progress based on the ratio of the total number of matched components to the total number of components in the building information model.

[0013] Furthermore, acquiring the 3D point cloud data of the prefabricated building includes: The construction site was scanned using a 3D LiDAR scanner to obtain raw point cloud data; Three-dimensional pass-through filtering is used to isolate the raw point cloud data from irrelevant external regions; A voxel grid is constructed to downsample the point cloud, and the average coordinates of points within each voxel are calculated as representative points to obtain a simplified point set. Calculate the average distance between each point in the simplified point set and its neighboring point sets, and remove outliers whose absolute difference between the average distance and the global average distance is greater than a multiple of the standard deviation threshold to obtain the three-dimensional point cloud data.

[0014] Furthermore, in each iteration, the eigenvalues ​​of the 3D coordinate covariance matrix within the neighborhood of each point in the unsegmented point cloud are calculated to obtain the local geometric entropy and the normalized neighborhood density variance, including: A three-dimensional coordinate covariance matrix is ​​constructed for the neighborhood of each point in the unsegmented point cloud and eigenvalue decomposition is performed. Three eigenvalues ​​are extracted and normalized to obtain three normalized eigenvalues. Calculate the negative of the product of each normalized eigenvalue and its natural logarithm, and add the three negative values ​​together to obtain the local geometric entropy; A spatial neighborhood sphere is constructed with each point in the unsegmented point cloud as its center and a preset detection radius. The local neighborhood density is obtained by dividing the number of points in the spatial neighborhood sphere by the volume of the spatial neighborhood sphere. Calculate the overall average density of the point cloud, and divide the square of the difference between the local neighborhood density and the overall average density by the square of the overall average density to obtain the normalized neighborhood density variance.

[0015] Further, updating the weighted covariance matrix of all points within the current growth region includes: Extract the three-dimensional coordinates of all points within the current growth region, calculate the average value of each coordinate axis, and obtain the centroid coordinates of the current growth region; Calculate the distance from each point to the centroid coordinates, take the reciprocal of the distance and normalize it to obtain the weight parameters of each point; Based on the difference between the coordinates of each point in the current growth region and the coordinates of the centroid, column vectors and row vectors are generated respectively, and the column vectors and row vectors are multiplied to obtain the initial covariance matrix of each point; The initial covariance matrix of each point is multiplied by the corresponding weight parameter, and the weighted covariance matrix of all points is summed to obtain the weighted covariance matrix.

[0016] Further, the calculation of the Mahalanobis distance of the point to be determined on the boundary of the growth region includes: The pseudo-inverse of the weighted covariance matrix is ​​obtained by taking the pseudo-inverse of the weighted covariance matrix; Subtract the centroid coordinates of the current growth region from the three-dimensional coordinates of the point to be determined to obtain the deviation column vector; The transpose of the deviation column vector is multiplied on the left by the weighted covariance pseudo-inverse matrix, then multiplied on the right by the deviation column vector, and the square root is taken to obtain the Mahalanobis distance.

[0017] Furthermore, the anisotropic growth inhibition factor for calculating the boundary normal field gradient includes: Obtain the K nearest neighbors of the point to be determined, and calculate the difference vector between the normal vector of each nearest neighbor and the normal vector of the point to be determined; Calculate the squared L2 norm of each difference vector and sum them to obtain the gradient index of the boundary normal field; The anisotropic growth inhibition factor is calculated using the natural constant e as the base and the negative value of the boundary normal field gradient index as the exponent.

[0018] Furthermore, the extraction of the centroid position, principal direction, and circumscribed volume dimensions of the point cloud of each segmented component as geometric features includes: Calculate the average coordinates of the point cloud of each segmented component on the three coordinate axes, and use this as the centroid position; Calculate the covariance matrix of the point cloud of each segmented component and perform principal component analysis to extract three orthogonal eigenvectors corresponding to the largest eigenvalue, the second largest eigenvalue, and the smallest eigenvalue. Use the eigenvector corresponding to the largest eigenvalue as the principal direction. A local coordinate system is established with the centroid position as the origin and the three orthogonal feature vectors as coordinate axes. The point cloud of the segmented component is projected onto each coordinate axis of the local coordinate system. The difference between the maximum and minimum values ​​of the projection on each coordinate axis is calculated and used as the length, width, and height features of the circumscribed body.

[0019] This invention selects initial seed points by comprehensively calculating local geometric entropy and normalized neighborhood density variance, and introduces planarity candidate constraints in its implementation. This reduces the possibility of linear edge points or seam points being mistakenly selected as seed points due to low single entropy values, thus improving the stability and reliability of the initial position of the region growth. During the region expansion process, weights are determined and the weighted covariance matrix is ​​updated based on the reciprocal of the distance from the point to the region centroid. This reduces the interference of outliers or boundary points far from the centroid on the growth direction estimation, enhancing the noise resistance of the point cloud segmentation process. Furthermore, by combining Mahalanobis distance, which reflects the spatial distribution characteristics of the data, and an anisotropic growth suppression factor calculated based on the boundary normal field gradient, the angle threshold is adaptively corrected. This helps suppress erroneous cross-domain growth at the boundaries of different components, improving the accuracy and completeness of component point cloud segmentation. Based on this, by extracting the geometric features of the segmented components and matching them with the building information model, the installed components and the overall construction progress are determined, thereby improving the automation level and management efficiency of prefabricated building construction progress accounting. Attached Figure Description

[0020] Figure 1 A flowchart of a method for determining the construction schedule of prefabricated buildings based on BIM comparison; Figure 2A schematic diagram of seed point selection for structural geometry; Figure 3 This is a schematic diagram illustrating the ablation effect comparison. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] Example 1 In Embodiment 1 of the present invention, as Figure 1 As shown, a method for determining the construction schedule of prefabricated buildings based on BIM comparison includes the following steps: S1, acquire the 3D point cloud data of the prefabricated building.

[0023] The construction site is scanned using a terrestrial 3D laser scanner or a drone equipped with LiDAR to acquire raw 3D point cloud data containing the building structure and construction environment. A pass-through filtering algorithm is used to remove background noise points that exceed the boundaries of the set construction area. The VoxelGrid algorithm in the PCL point cloud library is then used to perform voxel downsampling on the filtered raw 3D point cloud data. Appropriate voxel side lengths are set to reduce the data volume while preserving the building's geometric features. The StatisticalOutlierRemoval algorithm is then used to calculate the distance between each point and a specified number of neighboring points, removing outliers outside the global distance distribution. The processed point cloud data is stored in a KDTree data structure for subsequent rapid neighborhood search and calculation.

[0024] In some embodiments, acquiring the three-dimensional point cloud data of the prefabricated building includes: The construction site was scanned using a 3D LiDAR scanner to obtain raw point cloud data; Three-dimensional pass-through filtering is used to isolate the raw point cloud data from irrelevant external regions; A voxel grid is constructed to downsample the point cloud, and the average coordinates of points within each voxel are calculated as representative points to obtain a simplified point set. Calculate the average distance between each point in the simplified point set and its neighboring point sets, and remove outliers whose absolute difference between the average distance and the global average distance is greater than a multiple of the standard deviation threshold to obtain the three-dimensional point cloud data.

[0025] The original site point cloud can be acquired using a 3D LiDAR with a 360° horizontal field of view, a 100° vertical field of view, and millimeter-level ranging accuracy. Spatial filtering boundary parameters for 3D pass-through filtering are set, such as limiting the X and Y axes to within the building foundation pit outline and the Z axis to -2m to 50m, to reduce redundant data caused by the sky, underground, and surrounding distant views. In the simplification stage, a cubic voxel mesh structure with a spatial side length of 0.03m is generated. The arithmetic mean of the X, Y, and Z dimensions is calculated for points falling within the same voxel, and the generated mean coordinate point is used as the representative point to replace the original point cloud. This downsampling process preserves the main geometric outline of the original building while reducing the data volume. The nearest neighbor search parameter K for outlier statistical filtering is configured to be 50, and the mean Euclidean distance from each point to its 50 nearest neighbors is calculated. The global mean of the average distance of the entire point set is then calculated. And the global standard deviation. The standard deviation threshold is set to 2.0; any local average distance whose absolute value differs from the global mean by more than 2.0 is considered a valid standard deviation. All points are identified as outliers and removed, resulting in a spatially uniform 3D point cloud dataset.

[0026] S2, perform component segmentation repeatedly until the target component point cloud segmentation is completed, or the remaining unsegmented points meet the preset criteria for noise points, occlusion residue points, or points to be verified. In each round of the loop, calculate the eigenvalues ​​of the three-dimensional coordinate covariance matrix in the neighborhood of each point in the unsegmented point cloud to obtain the local geometric entropy and normalized neighborhood density variance. The point with the smallest weighted sum of the local geometric entropy and normalized neighborhood density variance is used as the initial seed point for the current round.

[0027] Initialize an undivided point cloud set and construct a KDTree structure within it. Iterate through each point in the set and call the K-nearest neighbor search algorithm to obtain a preset number of nearest neighbors for the current point, or use the radius search algorithm to obtain the nearest neighbors within a fixed search radius for the current point. Calculate the covariance matrix of the three-dimensional coordinates in the neighborhood and use the singular value decomposition (SVD) algorithm to obtain three eigenvalues ​​arranged in descending order. The ratios of the three eigenvalues ​​to their sum are calculated to obtain three normalized eigenvalues. Based on the Shannon entropy formula, the three normalized eigenvalues ​​are added to the negative of their natural logarithm product to calculate the local geometric entropy. This local geometric entropy characterizes the dispersion of the neighborhood's geometric distribution. To avoid low entropy values ​​in linear seams or edge regions due to dominance of a single dominant direction, a planarity candidate constraint is further set during seed point selection. This constraint requires that the ratio of the second largest eigenvalue to the largest eigenvalue is not less than a preset linear exclusion threshold, and the ratio of the smallest eigenvalue to the sum of the three eigenvalues ​​is not greater than a preset thickness threshold. Points that do not meet this planarity candidate constraint are not considered final seed point candidates, or a preset penalty value is added to their fusion index. For each point, a local spatial neighborhood sphere is constructed with that point as the center and a fixed detection radius. The number of points in the local spatial neighborhood sphere is counted and divided by the volume of the neighborhood sphere to obtain the local neighborhood density corresponding to that point. Then, the overall average density of the point cloud is calculated. The square of the difference between the local neighborhood density of that point and the overall average density is divided by the square of the overall average density to obtain the normalized neighborhood density variance corresponding to that point. Preset weight constants are assigned to the local geometric entropy and the normalized neighborhood density variance, and they are added together to obtain the fusion index. The current unsegmented set is traversed, and the point with the smallest fusion index value is found among the points that satisfy the planarity candidate constraint. This point is marked as the center of a region with high flatness and relatively uniform distribution, and is added to the current growth region point set as the initial seed point for the current round of region growth.

[0028] In some embodiments, in each iteration, the eigenvalues ​​of the 3D coordinate covariance matrix in the neighborhood of each point in the unsegmented point cloud are calculated to obtain the local geometric entropy and the normalized neighborhood density variance, including: A three-dimensional coordinate covariance matrix is ​​constructed for the neighborhood of each point in the unsegmented point cloud and eigenvalue decomposition is performed. Three eigenvalues ​​are extracted and normalized to obtain three normalized eigenvalues. Calculate the negative of the product of each normalized eigenvalue and its natural logarithm, and add the three negative values ​​together to obtain the local geometric entropy; A spatial neighborhood sphere is constructed with each point in the unsegmented point cloud as its center and a preset detection radius. The local neighborhood density is obtained by dividing the number of points in the spatial neighborhood sphere by the volume of the spatial neighborhood sphere. Calculate the overall average density of the point cloud, and divide the square of the difference between the local neighborhood density and the overall average density by the square of the overall average density to obtain the normalized neighborhood density variance.

[0029] For the unsegmented point set, extract the nearest neighbors by setting a neighborhood size K=30 for each point and generating a 3×3 covariance matrix. Obtain the eigenvalues ​​in descending order through eigenvalue decomposition. , and The normalized eigenvalue is obtained by dividing each individual eigenvalue by the sum of the three eigenvalues. , and According to the formula The local geometric entropy E is calculated. For flat component areas such as prefabricated wall panels, it typically manifests as follows: and All have a certain value and Smaller; typically appears as smaller at seam edges or linear structures. Dominant , The local geometric entropy is relatively small. Therefore, this embodiment does not solely rely on the smaller local geometric entropy as the criterion for judging the planar region, but instead first utilizes... / ≥ and / ()≤ Constitutes planar candidate constraints, where A value of 0.2 to 0.6 is acceptable. The value can be between 0.01 and 0.1; among the candidate points that satisfy the planarity constraint, a weighted selection is performed by combining local geometric entropy and normalized neighborhood density variance. A local spatial neighborhood sphere is constructed by setting the preset detection radius to r = 0.2m. The actual number of points contained within the sphere is counted and divided by the volume of the sphere. The local neighborhood density of each point is obtained. The global average density is calculated based on the overall bounding box volume of the point cloud mesh and the total number of points, expressed as (local neighborhood density - global average density). 2 / (Global average density) 2 The normalized variance formula calculates the distribution dispersion corresponding to each point. The above index can locate the center point of the core region in the 3D point cloud that satisfies the planarity condition and has relatively uniform surface density, using this as the initial seed point, such as... Figure 2 As shown, the horizontal axis represents the index of the geometric center seed point, and the vertical axis represents the volume density spatial variance, indicating the spatial density distribution characteristics of different candidate seed points. Points with lower variance correspond to regions with more uniform surface density, which is consistent with the characteristics of the optimal seed point selected by the fusion index.

[0030] S3, construct a growth region using the initial seed point and iteratively expand it. In each iteration, update the weighted covariance matrix of all points in the current growth region, where the weight of each point in the current growth region is inversely proportional to its distance to the centroid of the current growth region; calculate the Mahalanobis distance of the point to be determined on the boundary of the growth region; calculate the anisotropic growth suppression factor of the boundary normal field gradient; when the Mahalanobis distance is less than the distance threshold and the angle between the normal vector of the point to be determined and the average normal vector of the region is less than the angle threshold corrected by the anisotropic growth suppression factor, the point to be determined is incorporated into the growth region until the segmentation of a single segmentation component point cloud is completed.

[0031] A breadth-first search (BFS) queue is established, and the initial seed point is pushed into the queue as the initial growth region. To avoid covariance matrix degradation caused by a single seed point, nearest neighbors are searched with the initial seed point as the center. Initial support points are selected based on Euclidean distance thresholds, normal angle thresholds, and planarity constraints. The initial seed point and its support points together form the initial growth region. If the number of support points is less than the preset minimum number of points, the search radius is widened, or the seed point is temporarily suspended and the next candidate seed point is selected. The preset minimum number of points is preferably not less than 5. During the iterative expansion process, the centroid of the current growth region is calculated, and the normalized weights are determined based on the reciprocal of the distance from each point in the region to the centroid. The weighted covariance matrix and its pseudo-inverse matrix with singular value lower bound constraints are updated. The head point of the queue is used as the boundary point. Unsegmented neighboring points around it are searched using KDTree as the decision points. Mahalanobis distance is calculated based on the deviation vector between the decision point and the centroid of the region and the weighted covariance pseudo-inverse matrix. Simultaneously, after unifying the orientation of the normal vectors, the angle between the normal vector of the point to be determined and the average normal vector of the current region is calculated, and the anisotropic growth inhibition factor is calculated based on the boundary normal field gradient index G. The initial angle threshold is multiplied by the suppression factor to obtain the corrected angle threshold. When the Mahalanobis distance of the point to be determined is less than the preset distance threshold and its normal angle is less than the corrected angle threshold, the point to be determined is determined to belong to the same prefabricated component surface and is added to the current growth region and BFS queue, while being deleted from the unsegmented point cloud set. The above dequeue, search, determination and enqueue process is repeated until the queue is empty, and the candidate component point cloud corresponding to the current growth region is output. When the same prefabricated component is segmented into multiple adjacent candidate patches due to sharp corners, seams or multiple visible surfaces, the candidate patches belonging to the same prefabricated component are merged according to the spatial adjacency relationship, boundary distance, main direction angle and size constraint relationship with the same BIM candidate component, and the merged point cloud is used as the segmented component point cloud for subsequent geometric feature extraction. During the cyclic segmentation process, if the number of remaining unsegmented points is less than the preset minimum number of component points, or if the remaining points cannot meet the requirements of planarity candidate constraints, the number of initial support points, and the minimum cluster size, they will be marked as noise points, occlusion residual points, or points to be reviewed, and the current segmentation cycle will end.

[0032] In some embodiments, updating the weighted covariance matrix of all points within the current growth region includes: Extract the three-dimensional coordinates of all points within the current growth region, calculate the average value of each coordinate axis, and obtain the centroid coordinates of the current growth region; Calculate the distance from each point to the centroid coordinates, take the reciprocal of the distance and normalize it to obtain the weight parameters of each point; Based on the difference between the coordinates of each point in the current growth region and the coordinates of the centroid, column vectors and row vectors are generated respectively, and the column vectors and row vectors are multiplied to obtain the initial covariance matrix of each point; The initial covariance matrix of each point is multiplied by the corresponding weight parameter, and the weighted covariance matrix of all points is summed to obtain the weighted covariance matrix.

[0033] As the region growth process progresses, all points that have been incorporated into the current set are extracted in real time. For example, if there are currently 500 interior points, the arithmetic mean of the X, Y, and Z coordinate sequences is calculated to generate updated centroid coordinates. Calculate the points within each set. Euclidean distance to the center of mass C To avoid division by zero errors, a very small tolerance parameter is added to the denominator. ,Pick As the basic weight values, the basic weight of each point is divided by the sum of the basic weights of all points to complete the normalization operation, so that the extracted weight parameters are distributed in the interval (0,1] and the sum is equal to 1. In space, the closer a point is to the geometric centroid, the higher its weight is to the current plane. Based on the three-dimensional coordinates of each interior point, the centroid coordinates are subtracted from each interior point to extract a 3×1 dimensional deviation column vector v and a 1×3 dimensional row vector formed by its transpose. Perform vector multiplication. Output the initial 3×3 covariance matrix components for each point. Multiply the individual covariance matrix of each point by the corresponding assigned normalized weight parameter, and obtain the overall weighted covariance matrix of the region by linear summation. This matrix reduces the deflection effect on the overall plane fitting caused by boundary interference at newly formed peripheral edge points, and stabilizes the extraction of the principal plane normal characteristics.

[0034] In some embodiments, calculating the Mahalanobis distance of the point to be determined on the boundary of the growth region includes: The pseudo-inverse of the weighted covariance matrix is ​​obtained by taking the pseudo-inverse of the weighted covariance matrix; Subtract the centroid coordinates of the current growth region from the three-dimensional coordinates of the point to be determined to obtain the deviation column vector; The transpose of the deviation column vector is multiplied on the left by the weighted covariance pseudo-inverse matrix, then multiplied on the right by the deviation column vector, and the square root is taken to obtain the Mahalanobis distance.

[0035] For prefabricated floor slabs and wall panels, which exhibit flat 3D point cloud distributions, the aforementioned weighted covariance matrix often displays singular matrix characteristics, making matrix inversion prone to divergence. Therefore, singular value decomposition is used to obtain the diagonal matrix, and a singular value tolerance threshold is set. When constructing the weighted covariance pseudo-inverse matrix, singular values ​​above the singular value lower bound parameter τ are inversely calculated based on their actual values; singular values ​​below the singular value lower bound parameter τ are not truncated to zero, but are used as the lower bound for the inverse operation, or equivalently, a small regularization term of the form τI is added to the weighted covariance matrix before inversion. This avoids completely ignoring small eigenvalues ​​in the thickness direction, allowing the thickness direction to retain the distance penalty effect. At this point, the outer extension point of the boundary to be measured is input, and a 3×1 deviation column vector v is established from the centroid of the region to the point to be determined through three-dimensional coordinate difference calculation. The measurement is completed using the Mahalanobis distance quadratic form calculation formula, that is, the 1×3 transposed deviation row vector is multiplied left by the constructed 3×3 pseudo-inverse matrix, and the resulting 1×3 row vector is then multiplied right by the 3×1 deviation column vector v to obtain a scalar result. The arithmetic square root operation is performed on the scalar result to obtain the Mahalanobis distance. This calculation process can relax the geometric distance determination constraint along the extension direction of the principal components of the preform point cloud, and retain the necessary constraint strength in the thickness direction through singular value lower bound regularization. By combining typical thresholds such as... =3.0, filtering out background noise points that have no topological relationship with the target component plane.

[0036] In some embodiments, the calculation of the anisotropic growth inhibition factor of the boundary normal field gradient includes: Obtain the K nearest neighbors of the point to be determined, and calculate the difference vector between the normal vector of each nearest neighbor and the normal vector of the point to be determined; Calculate the squared L2 norm of each difference vector and sum them to obtain the gradient index of the boundary normal field; The anisotropic growth inhibition factor is calculated using the natural constant e as the base and the negative value of the boundary normal field gradient index as the exponent.

[0037] Using the constructed KDTree data structure, a local neighborhood search is performed on the point to be judged, extracting K=20 nearest neighbors and obtaining the 3D normal vectors of each neighbor and the point to be judged. Before calculating the normal vector difference, the orientation of each normal vector is unified according to the scanning viewpoint direction or the average normal vector of the current region; or a method is used... As a normal difference measure, the influence of normal vector sign reversal on the calculation results is eliminated. Subsequently, the difference vector between the normal vectors of each neighboring point and the normal vector of the point to be decided is calculated, and the squared L2 norms of each difference vector are accumulated to obtain the boundary normal field gradient index G. This parameter represents the degree of curvature change of the microsurface centered at the decision point. Taking the negative of the gradient index G and substituting it into the exponent of the natural base e, the suppression factor is calculated. Within a specific building environment point cloud, the G-value on a flat inner surface is typically small, and the calculated inhibition factor f ranges from 0.95 to 1.0. Once the point cloud spreads to the L-shaped interface or joint between prefabricated column and wall components, the G-value increases due to the angle between the normals of adjacent surfaces, and the inhibition factor f decays to approximately 0.1. Multiplying this decay factor by a set growth angle threshold can reduce the angle judgment threshold when approaching the prefabricated joint, thereby interrupting the growth cycle.

[0038] S4. Extract the centroid position, principal direction, and circumscribed body size of the point cloud of each segmented component as geometric features; match the geometric features with the preset parameters of the building information model to identify the corresponding components; determine the construction progress based on the ratio of the total number of matched components to the total number of components in the building information model.

[0039] Before performing geometric feature matching, the 3D point cloud data is uniformly transformed to the coordinate system of the Building Information Model (BIM) based on construction site control points, station coordinates, or ICP registration algorithms, and the corresponding rigid body transformation matrix is ​​recorded. Then, the Euclidean distance between the centroids, the cosine similarity of the principal directions, and the relative dimensional errors are calculated in the unified coordinate system. For each independently segmented component point cloud output from the above process, the arithmetic mean of the 3D coordinates of all points is calculated as the centroid position coordinates. The PCA algorithm from the PCL library is used to extract the feature vector with the largest variance in the point cloud data distribution as the principal direction of the component. The maximum and minimum coordinate values ​​are calculated along this principal direction and two other mutually orthogonal axes to generate the minimum circumscribed cuboid that encloses the point cloud using the Directed Bounding Box (OBB) algorithm. The absolute values ​​of the length, width, and height of this circumscribed cuboid are obtained as the geometric features of the circumscribed body dimensions. The IFC Plus parsing engine reads the pre-imported standard file of the prefabricated building information BIM model, parses and extracts the theoretical 3D coordinate orientation quaternions and length, width and height dimensions of each component in the BIM model, and constructs a feature cost matrix in memory, with the centroid Euclidean distance, principal direction cosine similarity and relative dimensional error as independent variables. The Kuhn-Munkres matching algorithm from graph theory is used to perform a global bipartite graph optimal matching solution between the geometric feature set of the actual segmented component point cloud and the preset parameter set of the BIM model. Matching pairs with matching costs below a set threshold are marked as successfully identified and installed components in the database. The total number of successfully identified and actually installed components is counted, and this total number is divided by the total number of components to be installed according to the corresponding construction stage in the BIM model to obtain the completion percentage. This percentage is formatted and output to the terminal display interface to determine the construction progress.

[0040] In some embodiments, extracting the centroid position, principal direction, and circumscribed volume dimensions of each segmented component point cloud as geometric features includes: Calculate the average coordinates of the point cloud of each segmented component on the three coordinate axes, and use this as the centroid position; Calculate the covariance matrix of the point cloud of each segmented component and perform principal component analysis to extract three orthogonal eigenvectors corresponding to the largest eigenvalue, the second largest eigenvalue, and the smallest eigenvalue. Use the eigenvector corresponding to the largest eigenvalue as the principal direction. A local coordinate system is established with the centroid position as the origin and the three orthogonal feature vectors as coordinate axes. The point cloud of the segmented component is projected onto each coordinate axis of the local coordinate system. The difference between the maximum and minimum values ​​of the projection on each coordinate axis is calculated and used as the length, width, and height features of the circumscribed body.

[0041] After the point cloud of an independent prefabricated component in the prefabricated structure is segmented, the coordinates of this point set are traversed, and the coordinate values ​​on the X, Y, and Z axes are accumulated sequentially and divided by the total number of points in the set. The centroid of the target in three-dimensional space is extracted as the centroid position of the component. The total covariance matrix is ​​calculated through global point cloud centering, and principal component analysis is used to decompose and output the largest, second largest, and smallest eigenvalues ​​and three corresponding mutually perpendicular three-dimensional unit vectors. Since the prefabricated component has a long and narrow geometric proportion, the eigenvector matching the largest eigenvalue is taken as the principal direction in space. Using the solved centroid as the origin coordinate, a rotation matrix is ​​assembled using the three mutually orthogonal unit eigenvectors extracted in the previous step to establish a directed bounding box local spatial coordinate system that coincides with the tilted pose of the target prefabricated component. The extracted local point set is subjected to a vector inner product operation on the three directional basis axes of the newly established coordinate system, and the point cloud is mapped and unfolded into a projected coordinate sequence. By comparing the maximum and minimum values ​​of the boundary coordinates at both ends of each projection sequence, an absolute difference operation is performed to calculate the axial extension dimension of the entity. This outputs three feature parameters, such as length 3.6m, width 0.2m, and height 2.8m. These values ​​can be used as semantic comparison criteria to align with the geometric features defined in the Building Information Modeling (BIM) database.

[0042] This experiment uses a building point cloud dataset obtained from a prefabricated construction site scan as the test object. The dataset contains five million points, covering various typical prefabricated components such as precast floor slabs and wall panels. The baseline is set as a region growing algorithm using traditional Euclidean distance and a fixed angle threshold. Based on this, three ablation experiment variants are constructed. Variant 1 introduces a weighted covariance matrix and Mahalanobis distance boundary determination module on top of the baseline. Variant 2 introduces an anisotropic growth suppression factor module for the boundary normal field gradient on top of the baseline. The complete scheme includes all the above optimization modules. The evaluation metrics are point cloud segmentation accuracy, recall, and F1 score.

[0043] In the segmentation test of point cloud data from a construction site, the baseline algorithm achieved a point cloud segmentation accuracy of 81.0%, a recall of 78.0%, and an F1 score of 79.5%. Variant 1 showed an improved accuracy of 86.5%, a recall of 84.5%, and an F1 score of 85.5%. Variant 2 achieved an accuracy of 85.0%, a recall of 82.0%, and an F1 score of 83.5%. Under the experimental settings, the complete solution yielded relatively better results, achieving a point cloud segmentation accuracy of 94.5%, a recall of 91.5%, and a calculated F1 score of 93.0%. Figure 3 As shown in the figure, the performance differences of the baseline algorithm, variant A, variant B and the complete solution in terms of segmentation accuracy, target recall and overall F1 score are displayed in the form of bar charts, which reflects the role of each optimization module in improving the segmentation effect.

[0044] Experimental data comparison shows that the complete solution improves upon the benchmark algorithm in all metrics, with a total F1 score improvement of 13.5 percentage points. By inputting the weighted covariance matrix and Mahalanobis distance, the deflection interference caused by newly formed peripheral edge points on the overall plane fitting is reduced. Geometric constraints are relaxed along the extension direction of the principal components of the prefabricated point cloud, and necessary distance constraints in the thickness direction are preserved through singular value lower bound regularization, improving the recall rate of boundary points of flat components. Combined with the anisotropic growth suppression factor calculated from the boundary normal field gradient, the algorithm rapidly reduces the angle judgment threshold the instant the point cloud growth touches the interface or seam of the prefabricated component, helping to reduce erroneous cross-domain growth and enhancing the accuracy and stability of point cloud segmentation in multi-component scenarios.

[0045] Example 2 Embodiment 2 of the present invention proposes a prefabricated building construction progress determination system based on BIM comparison, comprising the following modules: The acquisition module is used to acquire 3D point cloud data of prefabricated buildings; The calculation module is used to perform component segmentation in a loop until the target component point cloud segmentation is completed, or the remaining unsegmented points meet the preset criteria for noise points, occlusion residual points, or points to be verified. In each loop, the eigenvalues ​​of the three-dimensional coordinate covariance matrix in the neighborhood of each point in the unsegmented point cloud are calculated to obtain the local geometric entropy and the normalized neighborhood density variance. The point with the smallest weighted sum of the local geometric entropy and the normalized neighborhood density variance is used as the initial seed point for the current loop. The segmentation module is used to construct a growth region using the initial seed point and iteratively expand it. In each iteration, it updates the weighted covariance matrix of all points in the current growth region, where the weight of each point in the current growth region is inversely proportional to its distance to the centroid of the current growth region. It calculates the Mahalanobis distance of the point to be determined on the boundary of the growth region; it calculates the anisotropic growth suppression factor of the boundary normal field gradient; when the Mahalanobis distance is less than a distance threshold and the angle between the normal vector of the point to be determined and the average normal vector of the region is less than the angle threshold corrected by the anisotropic growth suppression factor, the point to be determined is incorporated into the growth region until the segmentation of a single segmentation component point cloud is completed. The determination module is used to extract the centroid position, principal direction, and circumscribed body size of the point cloud of each segmented component as geometric features; match the geometric features with preset parameters of the building information model to identify the corresponding components; and determine the construction progress based on the ratio of the total number of matched components to the total number of components in the building information model.

[0046] In this specification, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise limited, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. In this document, "a," "an," "the," "the," and "its" may also include plural forms unless the context clearly indicates otherwise. "Multiple" refers to at least two, such as 2, 3, 5, or 8, etc. "And / or" includes any and all combinations of the associated listed items.

[0047] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0048] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for determining the construction progress of prefabricated buildings based on BIM comparison, characterized in that, Includes the following steps: Acquire 3D point cloud data of prefabricated buildings; The component segmentation is performed iteratively until the target component point cloud segmentation is completed, or the remaining unsegmented points meet the preset criteria for noise points, occlusion residue points, or points to be verified. In each iteration, the eigenvalues ​​of the three-dimensional coordinate covariance matrix in the neighborhood of each point in the unsegmented point cloud are calculated to obtain the local geometric entropy and the normalized neighborhood density variance. The point with the smallest weighted sum of the local geometric entropy and the normalized neighborhood density variance is used as the initial seed point for the current iteration. The growth region is constructed using the initial seed point and iteratively expanded. In each iteration, the weighted covariance matrix of all points in the current growth region is updated, where the weight of each point in the current growth region is inversely proportional to its distance to the centroid of the current growth region. The Mahalanobis distance of the point to be determined on the boundary of the growth region is calculated. The anisotropic growth suppression factor of the boundary normal field gradient is calculated. When the Mahalanobis distance is less than the distance threshold and the angle between the normal vector of the point to be determined and the average normal vector of the region is less than the angle threshold corrected by the anisotropic growth suppression factor, the point to be determined is incorporated into the growth region until the segmentation of a single segmentation component point cloud is completed. Extract the centroid position, principal direction, and circumscribed body size of the point cloud of each segmented component as geometric features; match the geometric features with the preset parameters of the building information model to identify the corresponding components; determine the construction progress based on the ratio of the total number of matched components to the total number of components in the building information model.

2. The method according to claim 1, characterized in that, The acquisition of 3D point cloud data of prefabricated buildings includes: The construction site was scanned using a 3D LiDAR scanner to obtain raw point cloud data; The original point cloud data is isolated by three-dimensional pass-through filtering to isolate irrelevant external areas; A voxel grid is constructed to downsample the point cloud, and the average coordinates of points within each voxel are calculated as representative points to obtain a simplified point set. Calculate the average distance between each point in the simplified point set and its neighboring point sets, and remove outliers whose absolute difference between the average distance and the global average distance is greater than a multiple of the standard deviation threshold to obtain the three-dimensional point cloud data.

3. The method according to claim 1, characterized in that, In each iteration, the eigenvalues ​​of the 3D coordinate covariance matrix within the neighborhood of each point in the unsegmented point cloud are calculated to obtain the local geometric entropy and the normalized neighborhood density variance, including: A three-dimensional coordinate covariance matrix is ​​constructed for the neighborhood of each point in the unsegmented point cloud and eigenvalue decomposition is performed. Three eigenvalues ​​are extracted and normalized to obtain three normalized eigenvalues. Calculate the negative of the product of each normalized eigenvalue and its natural logarithm, and add the three negative values ​​together to obtain the local geometric entropy; A spatial neighborhood sphere is constructed with each point in the unsegmented point cloud as its center and a preset detection radius. The local neighborhood density is obtained by dividing the number of points in the spatial neighborhood sphere by the volume of the spatial neighborhood sphere. Calculate the overall average density of the point cloud, and divide the square of the difference between the local neighborhood density and the overall average density by the square of the overall average density to obtain the normalized neighborhood density variance.

4. The method according to claim 2, characterized in that, The step of updating the weighted covariance matrix of all points within the current growth region includes: Extract the three-dimensional coordinates of all points within the current growth region, calculate the average value of each coordinate axis, and obtain the centroid coordinates of the current growth region; Calculate the distance from each point to the centroid coordinates, take the reciprocal of the distance and normalize it to obtain the weight parameters of each point; Based on the difference between the coordinates of each point in the current growth region and the coordinates of the centroid, column vectors and row vectors are generated respectively, and the column vectors and row vectors are multiplied to obtain the initial covariance matrix of each point; The initial covariance matrix of each point is multiplied by the corresponding weight parameter, and the weighted covariance matrix of all points is summed to obtain the weighted covariance matrix.

5. The method according to claim 1 or 3, characterized in that, The calculation of the Mahalanobis distance of the point to be determined on the boundary of the growth region includes: The pseudo-inverse of the weighted covariance matrix is ​​obtained by taking the pseudo-inverse of the weighted covariance matrix; Subtract the centroid coordinates of the current growth region from the three-dimensional coordinates of the point to be determined to obtain the deviation column vector; The transpose of the deviation column vector is multiplied on the left by the weighted covariance pseudo-inverse matrix, then multiplied on the right by the deviation column vector, and the square root is taken to obtain the Mahalanobis distance.

6. The method according to claim 1, characterized in that, The anisotropic growth inhibition factor for calculating the boundary normal field gradient includes: Obtain the K nearest neighbors of the point to be determined, and calculate the difference vector between the normal vector of each nearest neighbor and the normal vector of the point to be determined; Calculate the squared L2 norm of each difference vector and sum them to obtain the gradient index of the boundary normal field; The anisotropic growth inhibition factor is calculated using the natural constant e as the base and the negative value of the boundary normal field gradient index as the exponent.

7. The method according to claim 1, characterized in that, The extraction of the centroid position, principal direction, and circumscribed volume dimensions of the point cloud of each segmented component as geometric features includes: Calculate the average coordinates of the point cloud of each segmented component on the three coordinate axes, and use this as the centroid position; Calculate the covariance matrix of the point cloud of each segmented component and perform principal component analysis to extract three orthogonal eigenvectors corresponding to the largest eigenvalue, the second largest eigenvalue, and the smallest eigenvalue. Use the eigenvector corresponding to the largest eigenvalue as the principal direction. A local coordinate system is established with the centroid position as the origin and the three orthogonal feature vectors as coordinate axes. The point cloud of the segmented component is projected onto each coordinate axis of the local coordinate system. The difference between the maximum and minimum values ​​of the projection on each coordinate axis is calculated and used as the length, width, and height features of the circumscribed body.

8. A system for determining the construction progress of prefabricated buildings based on BIM comparison, characterized in that, Includes the following modules: The acquisition module is used to acquire 3D point cloud data of prefabricated buildings; The calculation module is used to perform component segmentation in a loop until the target component point cloud segmentation is completed, or the remaining unsegmented points meet the preset criteria for noise points, occlusion residual points, or points to be verified. In each loop, the eigenvalues ​​of the three-dimensional coordinate covariance matrix in the neighborhood of each point in the unsegmented point cloud are calculated to obtain the local geometric entropy and the normalized neighborhood density variance. The point with the smallest weighted sum of the local geometric entropy and the normalized neighborhood density variance is used as the initial seed point for the current loop. The segmentation module is used to construct a growth region using the initial seed point and iteratively expand it. In each iteration, it updates the weighted covariance matrix of all points in the current growth region, where the weight of each point in the current growth region is inversely proportional to its distance to the centroid of the current growth region. It calculates the Mahalanobis distance of the point to be determined on the boundary of the growth region; it calculates the anisotropic growth suppression factor of the boundary normal field gradient; when the Mahalanobis distance is less than a distance threshold and the angle between the normal vector of the point to be determined and the average normal vector of the region is less than the angle threshold corrected by the anisotropic growth suppression factor, the point to be determined is incorporated into the growth region until the segmentation of a single segmentation component point cloud is completed. The determination module is used to extract the centroid position, principal direction, and circumscribed body size of the point cloud of each segmented component as geometric features; match the geometric features with preset parameters of the building information model to identify the corresponding components; and determine the construction progress based on the ratio of the total number of matched components to the total number of components in the building information model.

9. The system according to claim 8, characterized in that, The acquisition of 3D point cloud data of prefabricated buildings includes: The construction site was scanned using a 3D LiDAR scanner to obtain raw point cloud data; The original point cloud data is isolated by three-dimensional pass-through filtering to isolate irrelevant external areas; A voxel grid is constructed to downsample the point cloud, and the average coordinates of points within each voxel are calculated as representative points to obtain a simplified point set. Calculate the average distance between each point in the simplified point set and its neighboring point sets, and remove outliers whose absolute difference between the average distance and the global average distance is greater than a multiple of the standard deviation threshold to obtain the three-dimensional point cloud data.

10. The system according to claim 8, characterized in that, In each iteration, the eigenvalues ​​of the 3D coordinate covariance matrix within the neighborhood of each point in the unsegmented point cloud are calculated to obtain the local geometric entropy and the normalized neighborhood density variance, including: A three-dimensional coordinate covariance matrix is ​​constructed for the neighborhood of each point in the unsegmented point cloud and eigenvalue decomposition is performed. Three eigenvalues ​​are extracted and normalized to obtain three normalized eigenvalues. Calculate the negative of the product of each normalized eigenvalue and its natural logarithm, and add the three negative values ​​together to obtain the local geometric entropy; A spatial neighborhood sphere is constructed with each point in the unsegmented point cloud as its center and a preset detection radius. The local neighborhood density is obtained by dividing the number of points in the spatial neighborhood sphere by the volume of the spatial neighborhood sphere. Calculate the overall average density of the point cloud, and divide the square of the difference between the local neighborhood density and the overall average density by the square of the overall average density to obtain the normalized neighborhood density variance.

Images