Method for adjusting as-built curtain wall model based on high-precision parameterization of point cloud
By using a high-precision parametric adjustment method based on point clouds, combined with 3D laser scanning and ICP registration algorithms, the accuracy and efficiency issues in curtain wall inspection are solved, enabling efficient and accurate adjustment of the curtain wall model and supporting subsequent maintenance work.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHEAST UNIV
- Filing Date
- 2023-03-02
- Publication Date
- 2026-06-26
Smart Images
Figure CN116385703B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of parametric adjustment of as-built curtain wall models, and specifically to a method for high-precision parametric adjustment of as-built curtain wall models based on point clouds. Background Technology
[0002] In practical engineering, glass curtain wall panels are relatively independent entities, numerous in number, and distributed around the building's perimeter, making close-up observation and testing difficult. Therefore, researching rapid inspection methods for curtain wall structures has been a hot research topic both domestically and internationally.
[0003] 3D laser scanning technology, following GPS, is another surveying technology and has become one of the important means of spatial data acquisition. Developed in the 1990s, 3D laser scanning technology, also known as "real-scene replication technology," can quickly acquire massive amounts of spatial coordinates of an object's surface. As an emerging non-contact measurement method, 3D laser scanning technology has unique advantages such as high speed, high accuracy, and digitalization. It can perform measurements in hazardous environments inaccessible to technicians without the need for reflective prisms. It is the most advanced method in the modern surveying industry, storing the spatial coordinates and surface information of the target object through a high-density data grid, allowing for a more detailed description of the object. Based on the data acquired by 3D scanning, a 3D solid model can be reconstructed from the scanned structure. 3D laser scanning technology has already been applied in fields such as topographic surveying, cultural relic protection and restoration, structural deformation monitoring, geological disaster assessment and post-disaster reconstruction, digital twins, and autonomous driving, and has achieved some progress. The application research prospects of this technology are bright, and its development potential is enormous. Although there is high attention paid to the inspection, monitoring, and maintenance of curtain walls both domestically and internationally, there are still many unknown areas regarding curtain wall inspection technology, evaluation standards, and maintenance strategies.
[0004] The flatness and alignment of building curtain walls describe the unevenness of the structural surface and the quality of the overall assembly and connection, and are one of the important standards for the quality acceptance of building projects. Current measurement methods mainly rely on simple methods such as straightedges and feeler gauges. Due to the randomness of the selection of test points, these methods suffer from low accuracy, slow speed, and low efficiency. Furthermore, scaffolding is required when inspecting high-rise buildings, posing safety hazards to workers. In the context of a rapidly developing digital society, traditional measurement methods are insufficient to meet the evolving needs of curtain wall inspection. Summary of the Invention
[0005] This invention provides a method for high-precision parametric adjustment of an as-built curtain wall model based on point cloud, avoiding feature point positioning deviations caused by subjective human intervention. It offers advantages of automation and high efficiency. The adjusted as-built curtain wall model is highly consistent with the actual installed curtain wall space, facilitating subsequent curtain wall maintenance and inspection. To achieve the above objectives, this invention adopts the following technical solution:
[0006] A method for high-precision parametric adjustment of an as-built curtain wall model based on point cloud includes:
[0007] S1. Collect the 3D point cloud information of the completed building structure. C1.
[0008] S2, 3D point cloud information C1 preprocessing;
[0009] S3. Detect and delete large planar point clouds unrelated to the curtain wall from the three-dimensional point cloud information C1, and obtain three-dimensional point cloud information C2 containing all point clouds on each curtain wall.
[0010] S4. For any one of the curtain walls, in the XY plane projection of the three-dimensional point cloud information C2, search for the point cloud within the search radius R with the midpoint of the curtain wall as the radius, and finally obtain the three-dimensional point cloud information C3 used to find the curtain wall platform.
[0011] S5. Locate the curtain wall countertop:
[0012] First, the point density histogram of the three-dimensional point cloud information C3 in step S4 is projected along the Z-axis to obtain the Z-axis coordinates of the two extreme points, which are located on different curtain wall surfaces.
[0013] Then, taking each extreme point as the midpoint and the height range H1 as the projection range, the XY plane projection of part of the three-dimensional point cloud information C3 in the projection range is obtained.
[0014] S6. Obtain feature points of the curtain wall point cloud:
[0015] The improved random sampling consensus algorithm is applied to the XY plane projection map in step S5, and the resulting intersection points are the feature points of the curtain wall point cloud; all the feature points of the curtain wall point cloud form a matrix A;
[0016] S7. Obtain relevant technical documents for the curtain wall and create a BIM model of the curtain wall based on the technical documents;
[0017] S8. Extract the design feature points on the BIM model from step S7. The design feature points correspond to matrix A in step S6. The design feature points form matrix B.
[0018] S9. First, substitute matrix B into the ICP registration algorithm to solve for the rotation matrix;
[0019] The BIM model of the curtain wall is then adjusted using a rotation matrix to achieve high-precision parametric adjustment of the curtain wall BIM model using point clouds.
[0020] Preferably, step S9 specifically includes:
[0021] (1) Extract the feature points of the curtain wall point cloud and the design feature points to form a matrix set A. i and B i And it is necessary to constrain min||B i -A i ||, calculate the translation and rotation parameters R and t;
[0022] (2) Obtain a new point set A' by rotation and translation. i ={A' i =RB i +t,B i ∈B}, calculate the feature points of the new curtain wall point cloud and the design feature points to form a matrix set A. i and B i The average distance d;
[0023] If the average distance d meets the given threshold, the calculation stops, and the new point sets in step (2) are combined to form the rotation matrix in S9; otherwise, the calculation continues until the value of d converges.
[0024] Preferably, the search radius R = L / 2 + 0.2m, where L is the width of the curtain wall.
[0025] Preferably, the height range is H1 = ±0.1m.
[0026] Compared with the prior art, the advantages of the present invention are:
[0027] (1) The as-built curtain wall model after high-precision parameterization adjustment based on point cloud is highly consistent with the actual installation space of the curtain wall, which is beneficial to the later curtain wall maintenance and inspection work.
[0028] (2) Compared with the general human-computer interactive point cloud processing, this method adopts a large number of automated algorithms, thus having the advantages of automation and high efficiency, and is not affected by human subjective influence, which may lead to feature point positioning deviation. Attached Figure Description
[0029] Figure 1 Delete the image from the large planar point cloud;
[0030] Figure 2 The neighborhood algorithm is used to search the point cloud of each curtain wall;
[0031] Figure 3 Projection plot of the point cloud density histogram for the curtain wall;
[0032] Figure 4 Point cloud map of important sections of the curtain wall;
[0033] Figure 5 To improve the random sampling consensus algorithm for obtaining the curtain wall intersection diagram;
[0034] Figure 6 This is a BIM model for the curtain wall.
[0035] Figure 7 This is a comparison image of the curtain wall after adjustment and the point cloud. Detailed Implementation
[0036] The present invention will now be described in more detail with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. It should be understood that those skilled in the art can modify the invention described herein while still achieving its advantageous effects. Therefore, the following description should be understood as being of general knowledge to those skilled in the art and is not intended to limit the invention.
[0037] like Figures 1-7 A method for adjusting the as-built curtain wall model using high-precision parameterization based on point clouds includes:
[0038] S1. Collect the 3D point cloud information of the completed building structure. C1.
[0039] After obtaining the relevant technical documents for the curtain wall, the installation route for the 3D laser scanner was determined through a site survey. The 3D laser scanner was then set up, and the scanning parameters were configured. The relevant technical documents include, but are not limited to, construction drawings.
[0040] Common ranging ranges (range settings) for 3D scanners are 120m, 270m, and 570m, among others, along with suitable target object resolutions such as 1.5mm / 10m, 3mm / 10m, and 6mm / 10m. Considering the characteristics of the indoor curtain wall, a distance parameter of 120m is sufficient to meet the requirements. Furthermore, the 3mm / 10m setting effectively improves data acquisition efficiency while maintaining point cloud quality; therefore, this setting is chosen.
[0041] S2. Preprocessing of 3D point cloud information C1 prepares the point cloud processing for step S3.
[0042] Step S2, the preprocessing of point cloud data, includes point cloud mosaicking, model-construction coordinate alignment, denoising, and downsampling. Specifically:
[0043] Point cloud data aggregation and model alignment with construction coordinates: Since the collected point cloud data will be consistent with the construction coordinate system, and the data will be aggregated between different points, a black and white checkerboard is set up for coordinate transformation.
[0044] The collected point cloud data from each site is imported into a professional point cloud processing platform, where the center coordinates of the checkerboard pattern for each site are extracted. Based on this data, the point clouds from multiple sites are merged to obtain a complete model, which is then aligned with the construction coordinates.
[0045] The requirements for setting up the checkerboard pattern are as follows: adjacent stations should have at least three common checkerboard patterns; the incident angle of the scanner should be as great as possible (greater than 60°) to ensure the data quality of the point cloud on the checkerboard pattern.
[0046] Denoising and Downsampling: Preprocessing of the 3D point cloud model includes point cloud denoising and downsampling. Due to environmental factors such as construction, some noise is unavoidable in the point cloud data, which is addressed using appropriate denoising algorithms. Additionally, due to the small indoor distances, some point cloud data contains excessively high numbers of points, necessitating downsampling to reduce point cloud density.
[0047] Given the susceptibility of on-site curtain walls to vibration, bilateral filtering is commonly used for point cloud noise reduction. Bilateral filtering is widely applied in point cloud noise reduction, and it is highly effective at preserving the edges of point cloud data.
[0048] S3. Detect and delete large planar point clouds unrelated to the curtain wall from the 3D point cloud information C1, and obtain 3D point cloud information C2 containing all point clouds on each curtain wall.
[0049] Specifically, algorithms are used to detect and delete large planar point clouds unrelated to the curtain wall from the point cloud data. For example, if the Mestimator Sample Consensus (MSAC) algorithm is used to detect large planar surfaces unrelated to the curtain wall, including ground and ceiling point clouds, these will be automatically deleted. See [link to documentation]. Figure 1 .
[0050] like Figure 1 The figures (a) to (b) represent: three-dimensional point cloud information C1, a large plane unrelated to the curtain wall, and three-dimensional point cloud information C2, respectively.
[0051] S4. For any one of the curtain walls, in the XY plane projection of the 3D point cloud information C2, a neighborhood algorithm is used to search for the point cloud within a search radius R with the midpoint of the curtain wall as the radius, and finally obtain the 3D point cloud information C3 used to find the curtain wall platform. See Figure 2 The search radius R = L / 2 + 0.2m, where L is the width of the curtain wall.
[0052] like Figure 2 As shown, since the search is performed in the XY plane projection of the three-dimensional point cloud information C2, and the search radius R is the plane radius, the three-dimensional point cloud information C3 contains the entire Z-axis coordinate of the curtain wall.
[0053] S5. Locate the top and bottom surfaces of the curtain wall.
[0054] First, the 3D point cloud information C3 from step S4 is projected along the Z-axis using a point density histogram to obtain the Z-axis coordinates of two extreme points (the second extreme point and the third extreme point, respectively), which are located on different curtain wall surfaces. Specifically, the second extreme point is located on the upper surface of the curtain wall, and the third extreme point is located on the lower surface of the curtain wall.
[0055] Because the point density of the curtain wall point cloud (3D point cloud information C3) exhibits a clear and regular variation along the Z-axis, the point density of the vertical component surface is relatively low in the Z-axis direction, but the point density of the horizontal component surface (especially the upper and lower platforms, such as...) is much higher. Figure 3 The curtain wall platform exhibits a sudden increase in size. Therefore, by finding the extreme points of the curtain wall point cloud along the Z-axis, the curtain wall platform represented by each extreme point can be found.
[0056] like Figure 3 As shown, the upper, lower, and top surfaces of the curtain wall have the highest point density along the Z-axis, with the top surface having the highest density. The point cloud distribution at the top of the curtain wall differs from that on the upper and lower surfaces and their vicinity; there is no point cloud around the top that can be fitted with a straight line. Therefore, the top is not selected. The distance between the upper and lower surfaces is known from the curtain wall design information, and the distance between the second and third extreme points is close to this distance. Therefore, Figure 3 In this study, an extreme point on the upper and lower surfaces of the curtain wall is selected as the second and third extreme points of the point density along the Z-axis. The second and third extreme points provide Z-values for the feature points of the curtain wall point cloud.
[0057] Then, using the second and third extreme points as midpoints and the height range H1 as the projection range, the XY plane projections of the point cloud near the upper platform and the point cloud near the lower platform of the 3D point cloud information C3 are obtained. The height range H1 = ±0.1m.
[0058] like Figure 4 As shown, the left image is the point cloud of a single curtain wall (any curtain wall), i.e., the 3D point cloud information C3, and the right image is the projection of the point cloud near the lower platform of the curtain wall extracted from the left image onto the XY plane.
[0059] Similarly, the projection of the point cloud near the tabletop on the curtain wall onto the XY plane (not shown in the figure) is also obtained in this step.
[0060] S6. Obtain feature points of the curtain wall point cloud.
[0061] The improved random sampling consensus algorithm is applied to the XY plane projections of the point clouds near the upper and lower platforms in step S5. The intersection points obtained are the feature points of the curtain wall point cloud. Figure 5 As shown; all feature points of the curtain wall point cloud form matrix A.
[0062] The feature points of the curtain wall point cloud represent the intersection points of each fitted straight line on the selected projection surface of the curtain wall. These intersection points include curtain wall nodes that are visible to the naked eye, and may also be virtual intersection points.
[0063] The improved Random Sampling Consensus (RANSAC) algorithm specifically includes the following:
[0064] The RANSAC algorithm (existing technology) takes a specified dataset as input, which includes inliers, outliers, and a parametric model (here, a linear model) that can be used to interpret the dataset. The RANSAC algorithm iteratively selects a subset of the dataset each time, which is then used as the parameters of the parametric model; the selected subset represents the inliers.
[0065] 1) Selected interior points are applicable to the linear model, meaning that other unknown parameters in the linear model can be calculated using these interior points;
[0066] 2) Fit the other data into the model obtained in step ①. If some points are suitable for the model, then these points that meet the conditions are also considered as inliers.
[0067] 3) If a large number of subsets are identified as inliers, then the linear model is reasonable;
[0068] 4) Refit the linear model using all inliers as input, since the linear model has only been fitted with the initially selected inliers.
[0069] 5) Finally, the model is improved by estimating the number of local points obtained and the error of the parametric model.
[0070] 6) Obtain the XY plane projection of the two important segments using the histogram-extreme method, remove all inliers, and perform the next line fitting.
[0071] S7. Obtain relevant technical documents for the curtain wall and create a BIM model of the curtain wall based on the technical documents.
[0072] The BIM model of the curtain wall is generated by 3D modeling from the design CAD drawings, such as... Figure 6 As shown, commonly used BIM software, such as Revit, can be used.
[0073] S8. Extract the design feature points from the BIM model in step S7. These design feature points correspond to matrix A in step S6; the design feature points form matrix B. The correspondence process is as follows:
[0074]
[0075] S9. First, substitute matrix B into the ICP registration algorithm to obtain the rotation matrix; then, use the rotation matrix to parametrically adjust the curtain wall's BIM model, thus completing the high-precision parametric adjustment of the curtain wall BIM model using point clouds, and obtaining the as-built curtain wall model, such as... Figure 7 As shown.
[0076] The ICP registration algorithm is an optimal matching algorithm based on the least squares method. Its basic principle is to find nearest points and register matrices A and B according to given constraints, resulting in the following error function:
[0077]
[0078]
[0079] In the above formula, R and t are the rotation parameter and translation parameter after iterative calculation, respectively, and A i and B i Let A be the point corresponding to matrix A and matrix B. i ∈A,B i ∈B, where d is the average distance between the two sets of data points.
[0080] The ICP registration algorithm can maximize the alignment and matching of feature points in the curtain wall BIM model with feature points in the curtain wall point cloud, so this algorithm is adopted.
[0081] Steps S8 to S9 are as follows:
[0082] (1) Curtain wall point cloud feature points A, extract the corresponding points in BIM model and S6 as matrix B;
[0083] (2) Extract the feature points of the curtain wall point cloud and the design feature points to form a matrix set A. i and B i And it is necessary to constrain min||B i -A i ||, calculate the translation and rotation parameters R and t;
[0084] (3) Obtain a new point set A' by rotation and translation. i ={A' i =RB i +t,B i ∈B}, calculate the feature points of the new curtain wall point cloud and the design feature points to form a matrix set A. i and Bi The average distance d;
[0085] (4) If the average distance d meets the given threshold, stop the calculation. At this time, the new point sets in step (3) are combined to form the rotation matrix in S9; otherwise, return to continue iterative calculation until the value of d converges.
[0086] The above are merely preferred embodiments of the present invention and do not constitute any limitation on the present invention. Any equivalent substitutions or modifications made by those skilled in the art to the technical solutions and content disclosed in the present invention without departing from the scope of the present invention shall be deemed to have remained within the protection scope of the present invention.
Claims
1. A method for high-precision parametric adjustment of an as-built curtain wall model based on point cloud, characterized in that, include: S1. Collect the 3D point cloud information of the completed building structure. C1. S2, 3D point cloud information C1 preprocessing; S3. Detect and delete large planar point clouds unrelated to the curtain wall from the three-dimensional point cloud information C1, and obtain three-dimensional point cloud information C2 containing all point clouds on each curtain wall. S4. For any one of the curtain walls, in the XY plane projection of the three-dimensional point cloud information C2, search for the point cloud within the search radius R with the midpoint of the curtain wall as the radius, and finally obtain the three-dimensional point cloud information C3 used to find the curtain wall platform. S5. Locate the curtain wall countertop: First, the point density histogram of the three-dimensional point cloud information C3 in step S4 is projected along the Z-axis to obtain the Z-axis coordinates of the two extreme points, which are located on different curtain wall surfaces. Then, taking each extreme point as the midpoint and the height range H1 as the projection range, the XY plane projection of part of the three-dimensional point cloud information C3 in the projection range is obtained. S6. Obtain feature points of the curtain wall point cloud: The improved random sampling consensus algorithm is applied to the XY plane projection map in step S5, and the resulting intersection points are the feature points of the curtain wall point cloud; all the feature points of the curtain wall point cloud form a matrix A; S7. Obtain relevant technical documents for the curtain wall and create a BIM model of the curtain wall based on the technical documents; S8. Extract the design feature points on the BIM model from step S7. The design feature points correspond to matrix A in step S6. The design feature points form matrix B. S9. First, substitute matrix B into the ICP registration algorithm to solve for the rotation matrix; The BIM model of the curtain wall is then adjusted using a rotation matrix to achieve high-precision parametric adjustment of the curtain wall BIM model using point clouds.
2. The method for adjusting the as-built curtain wall model based on high-precision parameterization of point clouds according to claim 1, characterized in that, Step S9 is as follows: (1) Extract the feature points of the curtain wall point cloud and the design feature points to form a matrix set A. i and B i And it is necessary to constrain minB i -A i Calculate the translation and rotation parameters R and t; (2) Obtain a new point set A' by rotation and translation. i ={A' i =RB i +t,B i ∈B}, calculate the feature points of the new curtain wall point cloud and the design feature points to form a matrix set A. i and B i The average distance d; If the average distance d meets the given threshold, the calculation stops, and the new point sets in step (2) are combined to form the rotation matrix in S9; otherwise, the calculation continues until the value of d converges.
3. The method for adjusting the as-built curtain wall model based on high-precision parameterization of point clouds according to claim 1, characterized in that, The search radius R = L / 2 + 0.2m, where L is the width of the curtain wall.
4. The method for adjusting the as-built curtain wall model based on high-precision parameterization of point clouds according to claim 1, characterized in that, Height range H1 = ±0.1m.