A purlin support plate point cloud automatic extraction method based on three-dimensional point cloud and BIM

Abstract

The application discloses a kind of based on three-dimensional point cloud and BIM purlin support plate point cloud automatic extraction method, comprising the following steps: S1, data acquisition and coordinate uniformity: obtaining completion point cloud data, registration is carried out to data, denoising and coordinate system uniformity processing, obtain BIM model and extract main structure component set and optional purlin support plate design component set, BIM model and point cloud model are registered to the same coordinate system;S2, based on the main frame point cloud of BIM is stripped;S3, based on the purlin support plate candidate point cloud cluster extraction of density clustering;S4, purlin support plate geometric constraint screening and fine denoising;S5, purlin support plate point cloud output and structured storage.The application can automatically identify and segment all independent purlin support plate point cloud clusters in the whole completion model containing hundreds of millions of point clouds without manual framing, and greatly reduces manual participation.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of civil engineering and computer vision, specifically a method for automatic extraction of purlin support plate point clouds based on 3D point clouds and BIM. Background Technology

[0002] With the widespread application of terrestrial 3D laser scanning technology in the engineering construction field, engineers can acquire large-scale, high-density as-built 3D point cloud data in a short time. This data can then be compared with BIM design models to conduct Scan-to-BIM and Scan-vs-BIM applications for quality inspection, deformation monitoring, and progress management. In complex steel structure projects, purlin brackets are typically embedded in the concrete frame or the exterior facade of the main structure, serving as crucial connecting components for the precise installation of subsequent steel roofs and maintenance systems. To accurately determine the actual spatial position and orientation of the purlin brackets before steel structure installation, it is necessary to extract the point cloud cluster corresponding to each purlin bracket from the overall as-built point cloud as input for subsequent fitting and deviation analysis.

[0003] In existing technologies, point cloud extraction for small-scale auxiliary components such as purlin brackets mainly suffers from the following problems and shortcomings: 1. Relies on manual selection, resulting in low efficiency and poor stability. In practice, the common practice is for operators to manually place multiple cross-sectional frames and clipping frames in point cloud processing software, selecting areas suspected to be purlin support plates one by one or in small batches. By manually judging the geometric shape and positional relationship, point clouds are repeatedly added and deleted to form a purlin support plate point cloud cluster for each purlin support plate. This method is highly dependent on the operator's experience, involves a huge workload, and is highly repetitive. In a large project containing hundreds or even thousands of purlin support plates, it often requires several days of manual operation, and the results obtained by different personnel are inconsistent.

[0004] 2. General geometric segmentation methods have limited effectiveness in recognizing small components. Existing point cloud geometric segmentation methods, such as region growing algorithms based on plane fitting and volume clustering algorithms based on voxels, are mostly used to extract large-volume beam, slab, and column components. For purlin support plates that are small in size and close to the surface of beams, slabs and columns, their point clouds are often closely adjacent to or even partially overlap with the point clouds of the main structure. Based solely on local geometric features, it is difficult to reliably distinguish the purlin support plate from components of similar size such as the main structure, scaffolding, and reserved parts. These methods usually do not utilize the accurate design information provided by the BIM model, resulting in the segmentation results being extremely sensitive to noise and occlusion.

[0005] 3. Pure point cloud clustering methods are susceptible to noise interference and are sensitive to parameters. Some studies have attempted to use density-based spatial clustering algorithms (such as DBSCAN) or spectral clustering and K-means-based methods to directly cluster the overall point cloud: When there are many interfering points such as support frames, temporary components, and debris at the construction site, pure clustering methods often produce a large number of meaningless small clusters, or merge multiple adjacent purlin support plates into one cluster; The key parameters of algorithms such as DBSCAN (neighborhood radius Eps, minimum number of core points MinPts) are usually set manually based on experience, which cannot take into account the actual situation of "high density of point cloud in the near area and low density of point cloud in the far area", and is prone to under-clustering or over-clustering.

[0006] 4. Insufficient utilization of BIM information and lack of "structural-level" guidance. In current Scan-vs-BIM scenarios, BIM models are often used to detect overall structural deviations and to compare the positions of large components (beams, slabs, columns, etc.). However, in the point cloud extraction stage of small components such as purlins, BIM is often only used as a visualization reference and has not been systematically incorporated into the point cloud segmentation algorithm. This fails to achieve the following: first, using the geometric information of the main structure from BIM to isolate large components and narrow the search space; and second, using geometric filtering and verification of the clustering results based on the theoretical arrangement rules of purlins in BIM.

[0007] In summary, existing technologies still lack an automatic extraction method for purlin support plate point clouds that fully utilizes BIM geometric prior information while balancing automation and robustness in practical applications on complex construction sites. This makes it impossible to efficiently and stably separate all purlin support plate point cloud clusters from massive point cloud data, creating a bottleneck for subsequent intelligent inspection of construction quality. Summary of the Invention

[0008] The purpose of this invention is to provide an automatic extraction method for purlin support plate point clouds based on 3D point clouds and BIM, so as to solve the problems in the prior art.

[0009] To achieve the above objectives, the present invention provides the following technical solution: an automatic extraction method for purlin support plate point clouds based on 3D point clouds and BIM, comprising the following steps: S1. Data Acquisition and Coordinate Unification: Acquire as-built point cloud data, perform registration, noise reduction, and coordinate system unification processing on the data; acquire the BIM model and extract the main component set and purlin support plate design component set, and register the BIM model and point cloud model to the same coordinate system; S2. BIM-based main frame point cloud stripping: Using BIM main structure information, calculate the minimum Euclidean distance from each point in the overall as-built point cloud model P to the surface of the main components. Based on the set distance tolerance threshold of the main components, strip the main structure point cloud and retain the remaining point cloud including the purlin bracket. S3. Extraction of candidate point cloud clusters for purlins based on density clustering: The DBSCAN density clustering algorithm is executed on the remaining point cloud. Based on the design size of the purlin, the average point distance of the point cloud and the surface area, the neighborhood radius Eps and the minimum number of core points MinPts are adaptively set to obtain a set of candidate point cloud clusters. S4. Purlin Support Geometric Constraint Screening and Refined Noise Reduction: Calculate the number of points, 3D bounding box size, and principal component analysis eigenvalues ​​of each candidate point cloud cluster, and perform initial screening based on the purlin support design size range and planarity threshold; screen out the purlin support point cloud clusters that truly correspond to the purlin support, and further eliminate remaining noise; perform secondary discrimination of candidate point cloud clusters based on BIM priors and geometric features within the purlin support point cloud clusters; S5. Purlin Support Point Cloud Output and Structured Storage: Assign a unique number to each purlin support point cloud cluster, record the point cloud set, geometric center coordinates, fitting principal plane normal vector and corresponding BIM component ID within the purlin support point cloud cluster, and perform structured storage.

[0010] Preferably, in step S1, data acquisition and coordinate unification specifically include the following steps: S11. Obtain as-built point cloud data: Use a ground-based 3D laser scanner to set up multiple stations around the structure and inside the building to perform a surround scan of the concrete frame and the embedded purlin support plate, and obtain the original point cloud dataset of multiple stations. Registration and denoising were performed on the point clouds at each survey point, and the point clouds were uniformly transformed to the overall coordinate system of the project, resulting in the overall as-built point cloud model P={p k}, where each point p k = (x k y k , z k ) ; S12. Obtain and preprocess the BIM model: Export component information relevant to this analysis from the BIM model in the design phase, including at least the BIM main component set B of beams, slabs, and columns. main and purlin support plate design component set B purlin ; By manually selecting common feature points, using automatic feature matching, or employing the target coordinate method, the BIM model and the point cloud model are registered to ensure that they are in the same coordinate system.

[0011] Preferably, in step S2, the point cloud stripping of the main framework based on BIM specifically includes the following steps: S21. Represent the BIM main component set as follows: B main ={S i |i=1,...,M} ; Where S i Let i be the three-dimensional geometric surface of the i-th BIM main component; S22. For any point p in the overall as-built point cloud model P, calculate the distance S from that point to the surface of each main component. i The minimum Euclidean distance d(p,S) i ): ; Where q is the surface S of the component i any point on; S23. Set the distance tolerance threshold ε for the main components. main When there exists some i such that: d(p,S) i )<ε main At that time, mark point p as the main structural point and classify it with S. i The corresponding component cluster P i ; S24. After processing all points, the overall as-built point cloud model P is represented as: ; Among them, P res The remaining point cloud set that has not been assigned to any main component mainly consists of purlin bracket point cloud, a small number of unmodeled component point clouds, and on-site noise points; S25, P res This serves as the input point cloud for subsequent extraction of the purlin support plate.

[0012] Preferred, ε main The average point distance δ of the scanned point cloud and the structural dimensions of the main component are set to a range of 2δ to 5δ.

[0013] Preferably, step S3, the extraction of candidate point cloud clusters for purlins based on density clustering, specifically includes the following steps: S31. Parameter setting: The neighborhood radius Eps is set according to the design size of the purlin plate and the average point distance δ of the scanned point cloud, so that any adjacent points on the same purlin plate surface can fall into each other's Eps neighborhood. S32, DBSCAN clustering process: For P res For each point in the array, search for the number of points within its Eps neighborhood; If the number of neighborhood points of a certain point is not less than MinPts, it is marked as a core point, and a candidate point cloud cluster C is recursively expanded using it as a seed. j ; After penetrating the entire system, we obtain a set of non-noise clusters C: C = C j |j=1,...,K And a set P of discrete points that are partially labeled as noise noise ; S33. Since the purlin support plate is an independent steel plate component, the purlin support plate will appear in set C in the form of one or a few close clusters.

[0014] Preferably, in step S4, the purlin support plate geometric constraint screening and refined noise reduction specifically includes the following steps: S41. Cluster-level geometric feature calculation: For each candidate point cloud cluster C j Calculate: Number of points N j , where N j Candidate point cloud cluster C j The number of point clouds in the data; Minimum circumscribed cuboid size in three dimensions (L) j W j ,T j ), where L j ≥W j ≥T j L j W j T j These represent the dimensions in the length, width, and thickness directions, respectively. S42. Initial screening based on size and shape: combining the design size range of the purlin support plate (L) min ,L max ), (W min W max (T) min ,T max Set filter criteria: ; in, The planarity threshold is used; candidate point cloud clusters that meet the above conditions are retained as a set of suspected purlin support point cloud clusters. C (1) ={C j ∈C∣C j Satisfying dimensional and planarity constraints; in, Candidate point cloud clusters C j The first and third eigenvalues ​​obtained from principal component analysis of point clouds, Used to characterize the planarity of candidate point cloud clusters.

[0015] Preferably, in step S4, the purlin geometric constraint screening and refined noise reduction further includes: S43, morphological refinement based on two-dimensional density projection: For each C j ∈ C (1) ; Candidate point cloud cluster C jThe point cloud is projected along its principal normal direction onto the fitting principal plane to obtain the two-dimensional point set Q. j ={(u k ,v k )}; A grid of a certain resolution is constructed on the UV plane, and the number of points in each grid is mapped to pixel grayscale to generate a two-dimensional density image I of the purlin support plate. j (u,v); to I j Perform denoising operations such as Gaussian smoothing and shape opening / closing operations to eliminate isolated small regions; The contour and central axis of the main connected component are extracted from the processed image, and further judgment is made on whether it conforms to the morphological characteristics of the purlin support plate based on its aspect ratio, rectangularity, edge straightness and other indicators. Candidate point cloud clusters conforming to the above morphological constraints constitute the final set of purlin clusters: C (purlin) ={C j |C j Simultaneously satisfying constraints on size, planarity, and two-dimensional morphological characteristics.

[0016] Preferably, in step S4, the purlin support plate geometric constraint screening and refined noise reduction further includes: S44. Spatial consistency verification with the theoretical position of the BIM purlin support plate: If the BIM model contains a set of purlin support plate design components B purlin ={B m Then each candidate point cloud cluster C can be... j The geometric center g j With the corresponding BIM purlin design center Comparison: ; When it is less than the position tolerance ε pos At that time, the candidate point cloud cluster C j With design component B m Establish a one-to-one correspondence for subsequent deviation analysis and missing component identification; if no purlin support plate point cloud cluster that meets the conditions is found near a certain design component, it can be determined as "suspected missing" or "serious deviation".

[0017] Preferably, in step S5, the purlin support plate point cloud output and structured storage, the structured storage adopts the form of files or databases, and the stored information is used for subsequent purlin support plate attitude calculation, Scan-vs-BIM deviation analysis, and quality assessment.

[0018] Compared with the prior art, the beneficial effects of the present invention are: 1. Significantly improves the automation and efficiency of purlin bracket point cloud extraction: Through the technical route of "BIM main structure guided stripping + DBSCAN density clustering + multi-level geometric screening", a large number of manual selection and manual cutting steps in traditional methods are eliminated; in a large-scale engineering example containing hundreds of millions of point clouds, hundreds or even thousands of purlin bracket point cloud clusters can be automatically extracted in tens of minutes, which can save a lot of manpower and time costs compared with completely manual operation.

[0019] 2. Excellent robustness and accuracy in complex construction site environments: By using the BIM main structure to pre-extract the point clouds of beams, slabs, and columns, the range of remaining point clouds is effectively reduced, allowing DBSCAN clustering to focus on local small components and noise, significantly reducing the probability of false or missed clustering; based on the joint screening of size, planarity, and two-dimensional morphological features, it effectively filters out non-purlin support plate clusters such as scaffolding and support components, and has good adaptability to purlin support plates of different sizes and installation heights; in actual engineering applications, it can completely retain the main part of the purlin support plate point cloud, providing sufficient point support for subsequent high-precision fitting and deviation analysis.

[0020] 3. Fully utilize prior BIM information to form an integrable Scan-vs-BIM pre-module: This invention no longer treats BIM merely as a visualization background, but directly embeds its component geometry (especially the main frame) into the point cloud segmentation algorithm, enabling deep integration of point cloud computing and model data; through optional consistency verification with BIM purlin support components, the location of "should be but missing" embedded plates can be automatically identified, providing a basis for marking missing parts and issuing out-of-tolerance warnings in the quality report; this method, as a "pre-data processing module" in the Scan-vs-BIM quality inspection process, has good modularity and reusability.

[0021] 4. Provides high-quality data support for intelligent inspection of steel structure construction quality: After processing by this invention, each purlin support plate point cloud cluster is stored in a structured manner and can be directly input into the subsequent RANSAC plane fitting, three-dimensional pose solving and three-axis deviation calculation modules; through accurate and stable point cloud extraction, the reliability of deviation analysis results is guaranteed, making indicators such as Z-axis height deviation and normal angle deviation more credible, thereby providing a solid data foundation for the precise installation of steel roof and external enclosure system. Attached Figure Description

[0022] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention.

[0024] Please see Figure 1 In this embodiment of the invention, an automatic extraction method for purlin support plate point clouds based on 3D point clouds and BIM includes the following steps: S1. Data Acquisition and Coordinate Unification: Acquire as-built point cloud data, perform registration, noise reduction and coordinate system unification processing on the data, acquire the BIM model and extract the main component set and optional purlin design component set, and register the BIM model and point cloud model to the same coordinate system; In step S1, the data acquisition and coordinate unification specifically include the following steps: S11. Obtain as-built point cloud data: Use a ground-based 3D laser scanner to set up multiple stations around the structure and inside the building to perform a surround scan of the concrete frame and the embedded purlin support plate, and obtain the original point cloud dataset of multiple stations. Registration and denoising were performed on the point clouds at each survey point, and the point clouds were uniformly transformed to the overall coordinate system of the project, resulting in the overall as-built point cloud model P={p k}, where each point p k =|(x k y k , z k ) ; S12. Obtain and preprocess the BIM model: Export component information relevant to this analysis from the BIM model in the design phase, including at least the BIM main component set B of beams, slabs, and columns. main and purlin support plate design component set B purlin ; By manually selecting common feature points, using automatic feature matching, or employing the target coordinate method, the BIM model and the point cloud model are registered to ensure that they are in the same coordinate system.

[0025] S2. BIM-based main frame point cloud stripping: Using BIM main structure information, calculate the minimum Euclidean distance from each point in the overall point cloud to the surface of the main component. Based on the set distance tolerance threshold of the main component, strip the main structure point cloud and retain the remaining point cloud including the purlin bracket. In step S2, the point cloud stripping of the main framework based on BIM specifically includes the following steps: S21. The collection of BIM main components (such as a beam, a floor slab, or a column) is represented as: B main ={S i |i=1,...,M}; Where S i Let i be the three-dimensional geometric surface of the i-th BIM main component; S22. For any point p in the overall as-built point cloud model P, calculate the distance S from that point to the surface of each main component. i The minimum Euclidean distance d(p,S) i ): ; Where q is the surface S of the component i any point on; S23. Set the distance tolerance threshold ε for the main components. main When there exists some i such that: d(p,S) i )<ε main At that time, mark point p as the main structural point and classify it with S. i The corresponding component cluster p i ; S24. After processing all points, the overall as-built point cloud model P is represented as: ; Among them, P res The remaining point cloud set that has not been assigned to any main component mainly consists of purlin bracket point cloud, a small number of unmodeled component point clouds, and on-site noise points; S25, P res This serves as the input point cloud for subsequent extraction of the purlin support plate.

[0026] Preferred, ε main The average point distance δ of the scanned point cloud and the structural dimensions of the main component are set within the range of 2δ to 5δ to balance error robustness and segmentation accuracy.

[0027] S3. Extraction of candidate point cloud clusters for purlins based on density clustering: The DBSCAN density clustering algorithm is executed on the remaining point cloud. Based on the design size of the purlin, the average point distance of the point cloud and the surface area, the neighborhood radius Eps and the minimum number of core points MinPts are adaptively set to obtain a set of candidate point cloud clusters. In step S3, the extraction of candidate point cloud clusters for purlin support plates based on density clustering specifically includes the following steps: S31. Parameter setting: The neighborhood radius Eps is set according to the design dimensions of the purlin plate (such as length and width) and the average point distance δ of the scanned point cloud, so that any adjacent points on the same purlin plate surface can fall into each other's Eps neighborhood. In one embodiment, Eps can be taken as k·δ; Where k is 3 to 5; in specific engineering cases, Eps is set to 5mm.

[0028] Minimum number of core points (MinPts): based on the purlin support plate surface area A p Based on the estimated number of points per unit area η, ensure that a complete purlin support plate contains at least one set of stable core points. A suitable value is: MinPts = □η·A p "; In a typical embodiment, MinPts is set to 500 to filter out tiny clusters generated by scattering noise.

[0029] S32, DBSCAN clustering process: For P res For each point in the array, search for the number of points within its Eps neighborhood; If the number of neighborhood points of a certain point is not less than MinPts, it is marked as a core point, and a candidate point cloud cluster C is recursively expanded using it as a seed. j ; After penetrating the entire system, we obtain several sets of non-noise clusters C: C=C j |j=1,...,K And a set P of discrete points that are partially labeled as noise noise ; S33. Since the purlin support plate is an independent steel plate component, its point cloud exhibits the characteristics of "high cohesion and obvious interval with surrounding points" in space. Therefore, most purlin support plates will appear in set C in the form of one or a few tight clusters.

[0030] S4. Purlin Support Geometric Constraint Screening and Refined Noise Reduction: Calculate the number of points, 3D bounding box size, and principal component analysis eigenvalues ​​of each candidate point cloud cluster, and perform initial screening based on the purlin support design size range and planarity threshold; screen out the purlin support point cloud clusters that truly correspond to the purlin support, and further eliminate remaining noise; perform secondary discrimination of candidate point cloud clusters based on BIM priors and geometric features within the purlin support point cloud clusters; In step S4, the purlin support plate geometric constraint screening and refined noise reduction specifically includes the following steps: S41. Cluster-level geometric feature calculation: For each candidate point cloud cluster C j Calculate: Number of points N j ; where N j Candidate point cloud cluster Cj The number of point clouds in the data; Minimum circumscribed cuboid size in three dimensions (L) j W j ,T j ), where L j ≥W j ≥T j L j W j T j These represent the dimensions in the length, width, and thickness directions, respectively. S42. Initial screening based on size and shape: combining the design size range of the purlin support plate (L) min ,L max ), (W min W max (T) min ,T max Set filter criteria: ; in, The planarity threshold is used; candidate point cloud clusters that meet the above conditions are retained as a set of suspected purlin clusters. C (1) ={C j ∈C∣C j Satisfying dimensional and planarity constraints; in, Candidate point cloud clusters C j The first and third eigenvalues ​​obtained from principal component analysis of point clouds, Used to characterize the planarity of candidate point cloud clusters.

[0031] S43. Morphological refinement based on two-dimensional density projection: For each C j ∈ C (1) : Candidate point cloud cluster C j The point cloud is projected along its principal normal direction onto the fitting principal plane to obtain the two-dimensional point set Q. j ={(u k ,v k )}; A grid of a certain resolution is constructed on the UV plane, and the number of points in each grid is mapped to pixel grayscale to generate a two-dimensional density image I of the purlin support plate. j (u,v); to I j Perform denoising operations such as Gaussian smoothing and shape opening / closing operations to eliminate isolated small regions; The contour and central axis of the main connected component are extracted from the processed image, and further judgment is made on whether it conforms to the morphological characteristics of the purlin support plate based on its aspect ratio, rectangularity, edge straightness and other indicators. Clusters conforming to the above morphological constraints constitute the final set of purlin clusters: C (purlin) ={C j |C j Simultaneously satisfying constraints on size, planarity, and two-dimensional morphological characteristics.

[0032] S44. Spatial consistency verification with the theoretical position of the BIM purlin support plate: If the BIM model contains a set of purlin support plate design components B purlin ={B m Then each candidate point cloud cluster C can be... j The geometric center g j With the corresponding BIM purlin design center Comparison: ; When it is less than the position tolerance ε pos At that time, the candidate point cloud cluster C j With design component B m Establish a one-to-one correspondence for subsequent deviation analysis and missing component identification; if no cluster meeting the conditions is found near a certain design component, it can be determined as "suspected missing" or "serious deviation".

[0033] S5. Purlin Support Point Cloud Output and Structured Storage: A unique number is assigned to each purlin support point cloud cluster. The point cloud set, geometric center coordinates, fitted principal plane normal vector, and corresponding BIM component ID within the purlin support point cloud cluster are recorded and structured. This structured storage is in file or database format, and the stored information is used for subsequent purlin support attitude calculation, Scan-vs-BIM deviation analysis, and quality assessment.

[0034] Through the above multi-level geometric constraints and optional BIM consistency verification, this invention can effectively eliminate point clouds of non-target components such as scaffold uprights and small reserved parts, and retain only the point cloud clusters that truly correspond to the purlin support plate.

[0035] Example: Point cloud extraction of purlin support plates in a concrete frame + steel roof project 1. Project and Data Overview The actual project is a multi-story concrete frame structure with a reserved large-span steel roof at the top. Hundreds of embedded purlin brackets are arranged on the exterior facade of the columns and the beams of the exterior walls. The project team used a terrestrial 3D laser scanner to set up a total of 64 stations inside and outside the structure, with a point cloud density of 3mm@10m. The scanning time per station was approximately 5 minutes, and the total data acquisition time was approximately 8 hours.

[0036] After registration and denoising, the average accuracy of the overall point cloud stitching is about 3.2 mm, with a maximum error of no more than 6.8 mm, forming an as-built point cloud model under a unified coordinate system.

[0037] 2. Registration of the main frame BIM model and point cloud model Export geometric models of main structural components such as concrete beams, slabs, and columns from BIM software, including the design components of purlin support plates; Several easily identifiable corner points and intersection points of construction edges are selected in the BIM model and point cloud respectively as corresponding feature points. Rigid body transformation is used to solve the registration matrix from model to point cloud, and the BIM model is transformed into the point cloud coordinate system. Visual inspection confirmed that the outlines of the main beams, slabs, and columns basically match the point cloud.

[0038] 3. Subject Point Cloud Stripping Based on Geometric Distance All beam, slab, and column components are denoted as the main component set B. main For each point p in the overall point cloud P, calculate the values ​​down to each S. j ∈B main Minimum distance between surfaces; Empirical statistics show that the average point spacing of the point cloud in this project is δ≈3 mm. Considering on-site construction errors and scanning noise, ε is selected. main =15mm is used as the subject distance threshold; Satisfying d(p,S) i Points less than 15mm are assigned to the corresponding component cluster P. i In the middle, the remaining points are added to the remaining point set P. res .

[0039] According to statistics, after the main body is stripped, P res The number of midpoints is approximately 15% of the original point cloud, including purlin support point cloud, a small amount of unmodeled component point cloud, and noise.

[0040] 4. DBSCAN Clustering and Candidate Cluster Generation By analyzing P res Sampling analysis of the point cloud density in local areas yielded an average point spacing δ'≈4mm; combined with the design dimensions of the purlin support plate (length approximately 300–600 mm, width approximately 150–300 mm), Eps=5mm was selected; Based on the surface area of ​​the purlin support plate and the desired point density, MinPts is set to 500; In P res The DBSCAN algorithm is then run to cluster spatially densely connected points into several candidate point cloud clusters, resulting in a set of non-noise clusters C and a set of noisy discrete points P. noise .

[0041] Clustering results show that the point cloud near the purlin support plate is aggregated into multiple medium-sized clusters, while some scaffolding members, rebar ends, etc. form slender or scattered small clusters.

[0042] 5. Cluster-level geometric feature calculation and initial size screening For each cluster C j； Calculate the number of points N j ; The principal orientation is obtained using PCA, and the 3D bounding box size L is calculated. j W j ,T j and eigenvalue ratio ; Based on the design size range of the purlin support plate, the following conditions are set: ; Candidate point cloud clusters that meet the above conditions are considered as suspected purlin support point cloud clusters and enter set C. (1) .

[0043] 6. Two-dimensional density projection and morphological refinement (optional) For each C j ∈ C (1) : Project its point cloud onto a two-dimensional plane along the principal plane to construct a density image with a pixel resolution of 2mm×2mm; Gaussian filtering and opening operations are performed on the density image to eliminate isolated small spots; Use connected component analysis to extract the main region with the largest area, and calculate its aspect ratio, rectangularity, and boundary straightness; Only clusters with aspect ratios close to the design value, high rectangularity, and near-straight edges are retained, while irregularly shaped temporary components are excluded.

[0044] After this step, the final set of point cloud clusters C for the purlin support plate is obtained. (purlin) In this engineering example, the size of the set is basically consistent with the number of purlin brackets designed.

[0045] 7. Correspondence and output with BIM purlin support plate For each candidate point cloud cluster C j ∈ C (purlin) Calculate the geometric center g j And query the nearest purlin support plate design component center in the BIM model. ; when At that time, the candidate point cloud cluster C j Establish a one-to-one correspondence with the corresponding BIM components, and record the cluster ID, BIM ID, center coordinates, and principal normal information; If no matching purlin support plate point cloud clusters are found around a certain BIM purlin support plate component, it will be marked as "suspected missing or serious deviation" for subsequent quality reports to focus on.

[0046] Ultimately, each purlin plate was associated with an independent point cloud cluster, providing standardized input for the next steps of RANSAC plane fitting, triaxial deviation calculation, and generation of 3D visualization deviation cloud maps.

[0047] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for automatic extraction of purlin support plate point clouds based on 3D point clouds and BIM, characterized in that, Includes the following steps: S1. Data Acquisition and Coordinate Unification: Acquire as-built point cloud data, perform registration, noise reduction, and coordinate system unification processing on the data; acquire the BIM model and extract the main component set and purlin support plate design component set, and register the BIM model and point cloud model to the same coordinate system; S2. BIM-based main frame point cloud stripping: Using BIM main structure information, calculate the minimum Euclidean distance from each point in the overall as-built point cloud model P to the surface of the main components. Based on the set distance tolerance threshold of the main components, strip the main structure point cloud and retain the remaining point cloud including the purlin bracket. S3. Extraction of candidate point cloud clusters for purlins based on density clustering: The DBSCAN density clustering algorithm is executed on the remaining point cloud. Based on the design size of the purlin, the average point distance of the point cloud and the surface area, the neighborhood radius Eps and the minimum number of core points MinPts are adaptively set to obtain a set of candidate point cloud clusters. S4. Purlin Support Geometric Constraint Screening and Refined Noise Reduction: Calculate the number of points, 3D bounding box size, and principal component analysis eigenvalues ​​of each candidate point cloud cluster, and perform initial screening based on the purlin support design size range and planarity threshold; screen out the purlin support point cloud clusters that truly correspond to the purlin support, and further eliminate remaining noise; perform secondary discrimination of candidate point cloud clusters based on BIM priors and geometric features within the purlin support point cloud clusters; S5. Purlin Support Point Cloud Output and Structured Storage: Assign a unique number to each purlin support point cloud cluster, record the point cloud set, geometric center coordinates, fitting principal plane normal vector and corresponding BIM component ID within the purlin support point cloud cluster, and perform structured storage.

2. The method for automatic extraction of purlin support plate point cloud based on 3D point cloud and BIM according to claim 1, characterized in that, The step S1, data acquisition and coordinate unification, specifically includes the following steps: S11. Obtain as-built point cloud data: Use a ground-based 3D laser scanner to set up multiple stations around the structure and inside the building to perform a surround scan of the concrete frame and the embedded purlin support plate, and obtain the original point cloud dataset of multiple stations. Registration and denoising were performed on the point clouds at each survey point, and the point clouds were uniformly transformed to the overall coordinate system of the project, resulting in the overall as-built point cloud model P={p k }, where each point p k = (x k y k , z k） ; S12. Obtain and preprocess the BIM model: Export component information relevant to this analysis from the BIM model in the design phase, including at least the BIM main component set B of beams, slabs, and columns. main and purlin support plate design component set B purlin ; By manually selecting common feature points, using automatic feature matching, or employing the target coordinate method, the BIM model and the point cloud model are registered to ensure that they are in the same coordinate system.

3. The method for automatic extraction of purlin support plate point cloud based on 3D point cloud and BIM according to claim 2, characterized in that, Step S2, the BIM-based main framework point cloud stripping, specifically includes the following steps: S21. Represent the BIM main component set as follows: B main ={S i ∣i=1,...,M} Where S i Let i be the three-dimensional geometric surface of the i-th BIM main component; S22. For any point p in the overall as-built point cloud model P, calculate the distance S from that point to the surface of each main component. i The minimum Euclidean distance d(p,S) i ): ； Where q is the surface S of the component i any point on; S23. Set the distance tolerance threshold ε for the main components. main When there exists some i such that: d(p,S) i )<ε main At that time, mark point p as the main structural point and classify it with S. i The corresponding component cluster P i ; S24. After processing all points, the overall as-built point cloud model P is represented as: ; Among them, P res The remaining point cloud set that has not been assigned to any main component mainly consists of purlin bracket point cloud, a small number of unmodeled component point clouds, and on-site noise points; S25, P res This serves as the input point cloud for subsequent extraction of the purlin support plate.

4. The method for automatic extraction of purlin support plate point cloud based on 3D point cloud and BIM according to claim 3, characterized in that, ε main The average point distance δ of the scanned point cloud and the structural dimensions of the main component are set to a range of 2δ to 5δ.

5. The method for automatic extraction of purlin support plate point cloud based on 3D point cloud and BIM according to claim 3, characterized in that, Step S3, the extraction of candidate point cloud clusters for purlins based on density clustering, specifically includes the following steps: S31. Parameter setting: The neighborhood radius Eps is set according to the design size of the purlin plate and the average point distance δ of the scanned point cloud, so that any adjacent points on the same purlin plate surface can fall into each other's Eps neighborhood. S32, DBSCAN clustering process: For P res For each point in the array, search for the number of points within its Eps neighborhood; If the number of neighborhood points of a certain point is not less than MinPts, it is marked as a core point, and a candidate point cloud cluster C is recursively expanded using it as a seed. j ; After penetrating the entire system, we obtain a set of non-noise clusters C: C = C j |j=1,...,K And a set P of discrete points that are partially labeled as noise noise ; S33. Since the purlin support plate is an independent steel plate component, the purlin support plate will appear in set C in the form of one or a few close clusters.

6. The method for automatic extraction of purlin support plate point cloud based on 3D point cloud and BIM according to claim 1, characterized in that, The step S4, purlin support plate geometric constraint screening and refined noise reduction, includes the following steps: S41. Cluster-level geometric feature calculation: For each candidate point cloud cluster C j Calculate: Number of points N j , where N j Candidate point cloud cluster C j The number of point clouds in the data; Minimum circumscribed cuboid size in three dimensions (L) j W j ,T j ), where L j ≥W j ≥T j L j W j T j These represent the dimensions in the length, width, and thickness directions, respectively. S42. Initial screening based on size and shape: combining the design size range of the purlin support plate (L) min ,L max ), (W min W max (T) min ,T max Set filter criteria: ; in, The planarity threshold is used; candidate point cloud clusters that meet the above conditions are retained as a set of suspected purlin support point cloud clusters. C (1) ={C j ∈C∣C j Satisfying dimensional and planarity constraints; in, Candidate point cloud clusters C j The first and third eigenvalues ​​obtained from principal component analysis of point clouds, Used to characterize the planarity of candidate point cloud clusters.

7. The method for automatic extraction of purlin support plate point cloud based on 3D point cloud and BIM according to claim 6, characterized in that, The step S4, purlin support plate geometric constraint screening and refined noise reduction, also includes the following steps: S43. Morphological refinement based on two-dimensional density projection: For each C j ∈ C (1) ; Candidate point cloud cluster C j The point cloud is projected along its principal normal direction onto the fitting principal plane to obtain the two-dimensional point set Q. j ={(u k ,v k )}; A grid of a certain resolution is constructed on the UV plane, and the number of points in each grid is mapped to pixel grayscale to generate a two-dimensional density image I of the purlin support plate. j (u,v); to I j Perform denoising operations such as Gaussian smoothing and shape opening / closing operations to eliminate isolated small regions; The contour and central axis of the main connected component are extracted from the processed image, and further judgment is made on whether it conforms to the morphological characteristics of the purlin support plate based on its aspect ratio, rectangularity, edge straightness and other indicators. Candidate point cloud clusters conforming to the above morphological constraints constitute the final set of purlin clusters: C (purlin) ={C j |C j Simultaneously satisfying constraints on size, planarity, and two-dimensional morphological characteristics.

8. The method for automatic extraction of purlin support plate point cloud based on 3D point cloud and BIM according to claim 7, characterized in that, The step S4, purlin support plate geometric constraint screening and refined noise reduction, also includes the following steps: S44. Spatial consistency verification with the theoretical position of the BIM purlin support plate: If the BIM model contains a set of purlin support plate design components B purlin ={B m }, then each candidate point cloud cluster C j The geometric center g j With the corresponding BIM purlin design center Comparison: ; When it is less than the position tolerance ε pos At that time, the candidate point cloud cluster C j With design component B m Establish a one-to-one correspondence for subsequent deviation analysis and missing component identification; if no purlin support plate point cloud cluster that meets the conditions is found near a certain design component, it is judged as "suspected missing" or "serious deviation".

9. The method for automatic extraction of purlin support plate point cloud based on 3D point cloud and BIM according to claim 1, characterized in that, In step S5, the purlin support plate point cloud output and structured storage, the structured storage adopts the form of files or databases, and the stored information is used for subsequent purlin support plate attitude calculation, Scan-vs-BIM deviation analysis and quality assessment.

Images