Quick coarse registration method for precast beam reinforcement cage point cloud and BIM (Building Information Modeling)

Abstract

The invention discloses a rapid coarse registration method for precast beam reinforcement cage point cloud and BIM, and belongs to the technical field of point cloud data and BIM registration methods. The method comprises the steps that after actual measurement point clouds and BIM point clouds are input, dimensionality reduction is carried out on the actual measurement point clouds and the BIM point clouds to a two-dimensional plane through principal component analysis, one-dimensional'projection point number-rotation angle 'distribution vectors are constructed to serve as global feature descriptors for rapid initial matching, and an initial coarse registration transformation matrix is generated; and then an asymmetric key area predefined in the BIM is used as arbitration to verify and carefully select a plurality of initial coarse registration transformation matrixes, and finally, an optimal coarse registration transformation matrix is output. According to the method, initial registration is performed on the point cloud through the global descriptor, a strategy that the registration result is improved by using the asymmetric key area is used, prior knowledge in the engineering field is converted into constraint conditions, and the problem of automatic registration caused by high repetition and symmetry of the reinforcement cage is solved.

Description

Technical Field

[0001] This invention belongs to the technical field of point cloud data and BIM (Building Information Model) registration methods, specifically involving a rapid coarse registration method for point cloud of precast beam steel reinforcement skeleton and BIM. Background Technology

[0002] In civil engineering fields such as bridges and buildings, the application of precast concrete box girders is becoming increasingly widespread. The steel reinforcement cage of the precast beam is the core component in the fabrication of the precast concrete box girder. Its geometric accuracy (such as rebar spacing, positional offset, and binding deviation) directly affects the overall quality of the component and the reliability of subsequent assembly, requiring quality inspection after fabrication. Traditional inspection methods rely on manual measurement (such as measuring tapes and calipers), which is inefficient and cannot fully reflect complex three-dimensional morphology. In recent years, three-dimensional laser scanning technology (such as structured light scanners) has been gradually applied to the quality inspection of steel reinforcement cages due to its non-contact, high-precision, and high-density characteristics. It can quickly acquire surface point cloud data for reverse modeling and deviation analysis. In practical engineering applications, to achieve automated, digital, and intelligent quality inspection of steel reinforcement cages, the measured point cloud data of the steel reinforcement cage collected at the construction site is usually compared and analyzed with the pre-designed Building Information Model (BIM) of the steel reinforcement cage, thereby achieving high-precision deviation identification and quality evaluation. The primary and most crucial step is to accurately align (i.e., register) the measured point cloud with the different coordinate systems of the BIM.

[0003] Existing precise registration algorithms, such as the Iterative Closest Points (ICP) algorithm and its variants, heavily rely on a good initial pose for convergence and accuracy. If the initial pose deviation is too large, the algorithm is prone to getting trapped in local optima, leading to registration failure. Therefore, coarse registration methods are typically needed to provide good initial conditions. Although existing feature-based coarse registration methods, such as Fast Point Feature Histograms (FPFH), Sample Consensus Initial Alignment (SAC_IA), and Normal Distributions Transform (NDT), have been widely applied in practice, automatically finding unique matching features in three-dimensional space becomes extremely difficult for dense, repetitive, and symmetrical mesh-like structures like precast box girder steel reinforcement skeletons, which consist of a large number of steel bars (slender cylinders) with extremely similar geometric features. This results in low success rates and poor robustness of existing coarse registration methods in such scenarios.

[0004] Currently, in practical engineering projects, operators typically rely on manual translation and rotation operations within the software to achieve approximate alignment. This method is inefficient, lacks automation, heavily depends on operator experience, and struggles to guarantee consistency, becoming a bottleneck for automated inspection. Therefore, there is an urgent need for a technical solution that can automatically, quickly, and robustly perform coarse registration between the rebar skeleton point cloud and BIM.

[0005] Chinese patent CN105701820A (publication date 2016.06.22) discloses a point cloud registration method based on matching regions. This method utilizes key regions with discriminative power (such as local point clouds with significant differences in curvature and normal vectors) for registration, offering advantages such as independence from predefined models and strong versatility. However, in highly repetitive structures like steel reinforcement skeletons, the algorithm may fail to find effective salient local regions. Chinese patent CN113674332A (publication date 2021.11.19) discloses a point cloud registration method based on topological structure and multi-scale features. This method extracts basic geometric shapes (such as planes and cylinders) from the point cloud and constructs their spatial topological relationship graph for matching. It exhibits strong robustness and high accuracy in registering structures composed of regular geometric shapes (such as industrial parts). However, this method has high computational complexity, requiring geometric shape segmentation before constructing and matching the topological graph. For structures like steel reinforcement skeletons composed of numerous small cylinders, the topological relationships may be overly complex and similar. The paper “Automatic Registration of Urban Laser Point Clouds and Images Based on Line Features” (He Peipei, Wan Youchuan, Yang Wei, et al., Acta Optica Sinica, 2015(5):9) proposes a registration method based on macroscopic and stable line features (such as road markings and curbs) for point cloud registration in road scenes. Although the registration accuracy is high by utilizing domain knowledge, it is still not applicable to steel skeleton structures with high repeatability and symmetry. Summary of the Invention

[0006] The purpose of this invention is to provide a rapid coarse registration method for the point cloud of precast beam reinforcement skeleton and BIM, which solves the problems of low matching success rate and poor robustness of existing coarse registration methods for reinforcement skeleton point cloud and BIM for highly repeatable and symmetrical grid-like dense structures.

[0007] The technical solution adopted in this invention is: a rapid coarse registration method for point clouds of precast beam reinforcement skeletons and BIM. It inputs measured point clouds and BIM point clouds, reduces their dimensionality to a two-dimensional plane through principal component analysis, and constructs a one-dimensional "projection point count - rotation angle" distribution vector as a global feature descriptor for rapid initial matching, generating an initial coarse registration transformation matrix. Furthermore, it utilizes predefined asymmetric key regions in BIM. As an arbitrator, multiple initial coarse registration transformation matrices are used. After verification and selection, the optimal coarse registration transformation matrix is ​​finally output as the coarse registration result.

[0008] The invention is further characterized by:

[0009] Furthermore, the method for rapid coarse registration of the point cloud of precast beam reinforcement skeleton with BIM specifically includes the following steps: S1: Input the measured point cloud of the reinforcing steel cage. Building Information Model (BIM) and Design ; S2: Principal component analysis was used to analyze the measured point cloud. and by model The generated BIM point cloud Feature extraction and dimensionality reduction are performed to construct a distribution vector as a global feature descriptor; S3: Based on global feature descriptor matching, generate an initial coarse registration transformation matrix. ; S4: Based on predefined building information models Asymmetric key regions For at least one initial coarse registration transformation matrix The matching was verified, and the matrix with the smallest matching error was selected as the final coarse registration result. ; S5: Output the final coarse registration result .

[0010] Furthermore, step S1 is detailed as follows: Input the measured point cloud of the steel reinforcement cage scanned on site. Building Information Model (BIM) and Design For the model Perform grid-based sampling to generate the corresponding BIM point cloud. ; in, This represents a data point in the measured point cloud of the steel reinforcement cage. Representing three-dimensional real space, This represents point elements in point cloud data, belonging to a three-dimensional real number space. This represents the total number of points in the point cloud dataset; similarly... This represents the points in the point cloud data generated by BIM, where the total number of point cloud elements is... .

[0011] Furthermore, step S2 includes: from the measured point cloud and BIM point cloud The distribution vector of "projection point count - rotation angle" is extracted as a global feature descriptor, and the point cloud is subjected to an initial coarse registration transformation based on the correlation between the global feature descriptors.

[0012] Furthermore, the feature extraction and dimensionality reduction processing in step S2 includes: For measured point cloud and BIM point cloud Principal component analysis is performed to calculate the three principal vector directions, which are the eigenvectors corresponding to the top three eigenvalues ​​obtained by eigenvalue decomposition of the three-dimensional covariance matrix of the point cloud. , , ; and the measured point cloud and BIM point cloud Projected onto the first two principal vectors respectively , On Zhang Cheng's plane, a two-dimensional projection image is obtained. and The projection process is represented by formula (1):

[0013] in , These represent the centroid coordinates of the point cloud, respectively.

[0014] Furthermore, the construction of the global feature descriptor in step S2 is as follows: Two-dimensional projection diagram centroid Using the origin as the reference point, at a fixed angle Interval rotating rays, generating ray; for the first A ray, i.e., an angle of Count the number of points that fall on the projection map. Generate a one-dimensional "projection point count - rotation angle" distribution vector. That is, the measured point cloud Global feature descriptor ; Similarly, we can obtain BIM point clouds. Distribution vector BIM point cloud Global feature descriptor ; This vector is insensitive to the translation of the point cloud and has cyclic shift properties.

[0015] Furthermore, step S3 includes: By calculating the distribution vector of the measured point cloud Distribution vector of BIM point cloud Similarity under different cyclic shifts determines the initial rotation angle. Based on the displacement of the center of the projected image after rotation, the initial translation vector is determined. This generates the initial coarse registration transformation matrix. ; Specifically as follows: S3.1: Calculation and Normalized cross-correlation coefficient As shown in formula (2):

[0016] in, and These are descriptors and The mean, is the number of rotated rays generated above, and s is the number of rotations between the feature descriptors representing the vectors; Find Maximum number of rotations Then the initial rotation angle The values ​​are shown in formula (3):

[0017] in, For a fixed angle; S3.2: Project the measured point cloud in two dimensions Rotation After the angle, calculate the centroid of the two points. and Translation vectors between The third principal component obtained by combining PCA (principal component analysis) Centroid difference in direction Together they form the initial translation vector. As shown in formula (4):

[0018] in, Here it is assumed that the main directions of the two cloud points are aligned; S3.3: Combined with rotation angle Translation vector Generate an initial coarse registration transformation matrix. As shown in formula (5):

[0019] in, It is the plane normal determined by the first two principal vectors, i.e., the rotation angle of the direction of the third vector. The three-dimensional rotation matrix.

[0020] Furthermore, step S4 is detailed as follows: S4.1: In Building Information Modeling Selecting asymmetric key regions The initial coarse registration transformation matrix Application to measured point cloud The transformed point cloud is obtained. ; S4.2: Calculate the transformed point cloud In the asymmetric critical region Internal and BIM point cloud Matching error That is, for Falling in the area Each point inside ,exist Find its nearest point and utilize Normal vector at point The calculation error is shown in formula (6):

[0021] in, express The middle is located in a predefined asymmetric key region A subset of points within, This represents the number of points in the set; S4.3: At the initial rotation angle Within its neighborhood, that is Sampling is performed on intervals for the transformed point cloud. Repeat steps S3.2 to S4.2 to generate a series of initial coarse registration transformation matrices. And calculate the corresponding error. ; S4.4: Compare the matching errors of all candidate hypotheses Selected in asymmetric key regions Top Match The matrix with the smallest error is taken as the final coarse registration result. .

[0022] Furthermore, matching error This represents the average distance from a point to a surface.

[0023] Furthermore, asymmetric key regions It is at least one of the following: prestressed anchorage zone, specific embedded part cluster, or non-standard steel reinforcement arrangement zone.

[0024] The beneficial effects of this invention are: (1) Fully automated and highly efficient: The method of the present invention realizes full automation from feature extraction, matching to verification without human intervention, which significantly improves detection efficiency.

[0025] (2) Strong robustness: The method of the present invention effectively solves the problem of incorrect matching caused by the high repetition and symmetry of the steel reinforcement cage by introducing the "asymmetric key area" as a feature, and greatly improves the success rate of registration.

[0026] (3) High computational efficiency: The method of the present invention performs rapid initial screening by reducing the three-dimensional point cloud to a two-dimensional projection and a one-dimensional descriptor, avoiding time-consuming feature calculation and matching in three-dimensional space. The algorithm is fast and suitable for engineering applications.

[0027] (4) Improved registration accuracy: The final coarse registration result of this invention can provide a sufficiently accurate initial value for subsequent fine registration, ensuring the reliability of the entire registration process and the accuracy of the final result. Attached Figure Description

[0028] Figure 1 This is a flowchart of a method for rapid registration of point cloud and BIM of steel reinforcement skeleton of precast beams provided by the present invention. Figure 2 This is a schematic diagram of the point cloud dimensionality reduction process provided in Embodiment 7 of the present invention; Figure 3 This is a schematic diagram of the statistical projection scanning process provided in Embodiment 7 of the present invention; Figure 4 This is the feature vector relationship diagram of "projection points - rotation angle" provided in Embodiment 7 of the present invention; Figure 5 This is the "normalized cross-correlation coefficient - rotation angle" relationship diagram provided in Embodiment 7 of the present invention; Figure 6 This is an example diagram of asymmetrical key areas in the BIM of the precast structural steel reinforcement skeleton provided in Embodiment 7 of the present invention; Figure 7 This is the "matching error - rotation angle" relationship diagram provided in Embodiment 7 of the present invention. Detailed Implementation

[0029] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0030] This invention provides a rapid coarse registration method for point clouds of precast beam reinforcement cages with BIM, such as... Figure 1 As shown, the specific steps are as follows: S1: Data Input and Preprocessing: Input the measured point cloud of the steel reinforcement cage scanned on-site. Building Information Model (BIM) and Design For the model Perform grid-based sampling to generate the corresponding BIM point cloud. ; S2: Feature Extraction and Dimensionality Reduction S2.1: Analyze the measured point cloud data respectively. and BIM point cloud Principal Component Analysis (PCA) is performed to calculate the three principal vector directions, which are the eigenvectors corresponding to the top three eigenvalues ​​obtained by eigenvalue decomposition of the three-dimensional covariance matrix of the point cloud. , , The two point clouds are projected onto their respective first two principal vectors. , On Zhang Cheng's plane, a two-dimensional projection image is obtained. and The projection process is represented by formula (1):

[0031] in , These represent the centroid coordinates of the point cloud, respectively. S2.2: Constructing a global feature descriptor: using a two-dimensional projection map centroid Using the origin as the reference point, at a fixed angle Interval rotating rays, generating ray, for the first A ray, i.e., an angle of Count the number of points that fall on the projection map. Generate a one-dimensional "projection point count - rotation angle" distribution vector. That is, the measured point cloud Global feature descriptor Similarly, we can obtain BIM point clouds. Distribution vector BIM point cloud Global feature descriptor This vector is insensitive to the translation of the point cloud and has cyclic shift properties. S3: Fast matching and initial transformation matrix generation: S3.1: Through calculation and The normalized cross-correlation (NCC) is shown in formula (2):

[0032] in, and These are descriptors and The mean, is the number of rotated rays generated above, and s is the number of rotations between the feature descriptors representing the vectors; Find Maximum number of rotations Then the initial rotation angle The values ​​are shown in formula (3):

[0033] in, For a fixed angle; S3.2: Project the measured point cloud in two dimensions Rotation After the angle, calculate the centroid of the two points. and Translation vectors between The third principal component obtained by combining PCA Centroid difference in direction Together they form the initial translation vector. As shown in formula (4):

[0034] in, Here it is assumed that the main directions of the two cloud points are aligned; S3.3: Combined with rotation angle Translation vector Generate an initial coarse registration transformation matrix. As shown in formula (5):

[0035] in, It is the plane normal determined by the first two principal vectors, i.e., the rotation angle of the direction of the third vector. The three-dimensional rotation matrix; S4: Based on asymmetric key regions Coarse registration transformation optimization: S4.1: In Building Information Modeling Selecting asymmetric key regions (e.g., prestressed anchorage zones, specific embedded component clusters, or non-standard rebar arrangement zones); using the initial transformation matrix obtained in step S3. For the measured point cloud Perform the transformation to obtain the transformed point cloud. ; S4.2: Calculate the transformed point cloud In the asymmetric critical region Inside, with BIM point cloud Average distance matching error between points and surfaces That is, for Falling in the area Each point inside ,exist Find its nearest point and utilize Normal vector at point The calculation error is shown in formula (6):

[0036] in, express The middle is located in a predefined asymmetric key region A subset of points within, This represents the number of points in the set; S4.3: At the initial rotation angle Within its neighborhood, that is Sampling is performed on intervals for the transformed point cloud. Repeat steps S3.2 to S4.2 to generate a series of transformation matrices. And calculate the corresponding error. ; S4.4: From all transformation matrices, select the key region for asymmetric transformations. Upper Matching Error Minimal transformation matrix This serves as the final coarse registration result; S5: Output the final coarse registration transformation matrix .

[0037] The technical solution of the present invention will be further illustrated below through embodiments.

[0038] Example 1 A rapid coarse registration method for point clouds and BIM of precast beam reinforcement skeletons is used. The method inputs measured point clouds and BIM point clouds, reduces their dimensionality to a two-dimensional plane using principal component analysis, and constructs a one-dimensional "projection point count - rotation angle" distribution vector as a global feature descriptor for rapid initial matching, generating an initial coarse registration transformation matrix. Furthermore, it utilizes predefined asymmetric key regions in BIM. As an arbitrator, multiple initial coarse registration transformation matrices are used. After verification and selection, the optimal coarse registration transformation matrix is ​​finally output as the coarse registration result.

[0039] Example 2 Based on Example 1, the specific steps include: S1: Input the measured point cloud of the reinforcing steel cage. Building Information Model (BIM) and Design ; S2: Principal component analysis was used to analyze the measured point cloud. and by model The generated BIM point cloud Feature extraction and dimensionality reduction are performed to construct a distribution vector as a global feature descriptor; S3: Based on global feature descriptor matching, generate an initial coarse registration transformation matrix. ; S4: Based on predefined building information models Asymmetric key regions For at least one initial coarse registration transformation matrix The matching was verified, and the matrix with the smallest matching error was selected as the final coarse registration result. ; S5: Output the final coarse registration result .

[0040] Example 3 Based on Example 1, the specific steps include: S1: Input the measured point cloud of the reinforcing steel cage. Building Information Model (BIM) and Design ; Input the measured point cloud of the steel reinforcement cage scanned on site. Building Information Model (BIM) and Design For the model Perform grid-based sampling to generate the corresponding BIM point cloud. ; S2: Principal component analysis was used to analyze the measured point cloud. and by model The generated BIM point cloud Feature extraction and dimensionality reduction are performed to construct a distribution vector as a global feature descriptor; S3: Based on global feature descriptor matching, generate an initial coarse registration transformation matrix. ; S4: Based on predefined building information models Asymmetric key regions For at least one initial coarse registration transformation matrix The matching was verified, and the matrix with the smallest matching error was selected as the final coarse registration result. ; S5: Output the final coarse registration result .

[0041] Example 4 Based on Example 2, step S2 includes: from the measured point cloud and BIM point cloud The distribution vector of "projection point count - rotation angle" is extracted as a global feature descriptor, and the point cloud is subjected to an initial coarse registration transformation based on the correlation between the global feature descriptors.

[0042] The feature extraction and dimensionality reduction process in step S2 includes: For measured point cloud and BIM point cloud Principal component analysis is performed to calculate the three principal vector directions, which are the eigenvectors corresponding to the top three eigenvalues ​​obtained by eigenvalue decomposition of the three-dimensional covariance matrix of the point cloud. , , ; And measured point cloud and BIM point cloud Projected onto the first two principal vectors respectively , On Zhang Cheng's plane, a two-dimensional projection image is obtained. and The projection process is represented by formula (1):

[0043] in , These represent the centroid coordinates of the point cloud, respectively.

[0044] The construction of the global feature descriptor in step S2 is as follows: Two-dimensional projection diagram centroid Using the origin as the reference point, at a fixed angle Interval rotating rays, generating ray; for the first A ray, i.e., an angle of Count the number of points that fall on the projection map. Generate a one-dimensional "projection point count - rotation angle" distribution vector. That is, the measured point cloud Global feature descriptor ; Similarly, we can obtain BIM point clouds. Distribution vector BIM point cloud Global feature descriptor ; This vector is insensitive to the translation of the point cloud and has cyclic shift properties.

[0045] Example 5 Based on Example 4, step S3 includes: By calculating the distribution vector of the measured point cloud Distribution vector of BIM point cloud Similarity under different cyclic shifts determines the initial rotation angle. ; Based on the displacement of the center of the rotated projected image, the initial translation vector is determined. This generates the initial coarse registration transformation matrix. The details are as follows: S3.1: Calculation and Normalized cross-correlation coefficient As shown in formula (2):

[0046] in, and These are descriptors and The mean, is the number of rotated rays generated above, and s is the number of rotations between the feature descriptors representing the vectors; Find Maximum number of rotations Then the initial rotation angle The values ​​are shown in formula (3):

[0047] in, For a fixed angle; S3.2: Project the measured point cloud in two dimensions Rotation After the angle, calculate the centroid of the two points. and Translation vectors between The third principal component obtained by combining PCA Centroid difference in direction Together they form the initial translation vector. As shown in formula (4):

[0048] in, Here it is assumed that the main directions of the two cloud points are aligned; S3.3: Combined with rotation angle Translation vector Generate an initial coarse registration transformation matrix. As shown in formula (5):

[0049] in, It is the plane normal determined by the first two principal vectors, i.e., the rotation angle of the direction of the third vector. The three-dimensional rotation matrix.

[0050] Example 6 Based on Example 5, step S4 is as follows: S4.1: In Building Information Modeling Selecting asymmetric key regions The initial coarse registration transformation matrix Application to measured point cloud The transformed point cloud is obtained. ; Asymmetric key regions It is at least one of the following: prestressed anchorage zone, specific embedded part cluster, or non-standard steel reinforcement arrangement zone; S4.2: Calculate the transformed point cloud In the asymmetric critical region Internal and BIM point cloud Matching error That is, for Falling in the area Each point inside ,exist Find its nearest point and utilize Normal vector at point Calculation error; Specifically, as shown in formula (6):

[0051] in, express The middle is located in a predefined asymmetric key region A subset of points within, This represents the number of points in the set; S4.3: At the initial rotation angle Within its neighborhood, that is Sampling is performed on intervals for the transformed point cloud. Repeat steps S3.2 to S4.2 to generate a series of initial coarse registration transformation matrices. And calculate the corresponding error. ; Matching error The average distance from a point to a surface; S4.4: Compare the matching errors of all candidate hypotheses Selected in asymmetric key regions Top Match The matrix with the smallest error is taken as the final coarse registration result. .

[0052] Example 7 like Figure 1 As shown in the figure, this embodiment provides a method for rapid coarse registration of point clouds of steel reinforcement cages with BIM, which specifically includes the following steps: S1: Data Input and Preprocessing Input the actual point cloud of the box girder reinforcement skeleton obtained by a terrestrial 3D laser scanner. Simultaneously, the BIM design model of the box girder's steel reinforcement skeleton is retrieved from the storage system. ; For ease of processing, the Poisson disk sampling algorithm is used to sample the surface of the BIM triangular mesh to generate a BIM point cloud with uniform density. ; S2: Feature Extraction and Dimensionality Reduction First, the measured point clouds were analyzed separately. and BIM point cloud Principal component analysis was performed to obtain the eigenvectors of its covariance matrix, and the measured point cloud was then analyzed. and BIM point cloud Projecting these components onto a two-dimensional plane composed of the first principal component (PC1) and the second principal component (PC2) yields a two-dimensional projection image. and The problem of rotation in three-dimensional space is simplified to a problem of rotation in a two-dimensional plane. The implementation process is as follows: Figure 2 As shown; Then, calculate the two-dimensional projection. and centroid and With the centroid as the origin, one ray is emitted every 5 degrees, for a total of 72 rays. The scanning process is as follows: Figure 3 As shown; The number of points falling on each ray in the projection image is counted, forming a 72-dimensional feature vector. and That is, the distribution vector of "projected points - rotation angle", which represents the measured point cloud. Global feature descriptor and BIM point cloud Global feature descriptor ,like Figure 4 As shown; this vector is not sensitive to the translation of the point cloud, and its cyclic shift characteristic facilitates rotation angle matching; S3: Fast Matching and Hypothesis Generation Calculate the global feature descriptor of the measured point cloud With BIM point cloud global feature descriptor Normalized cross-correlation coefficients under different cyclic shifts; find the number of rotations that maximizes the cross-correlation coefficients. To obtain the initial rotation angle The result is as follows Figure 5 As shown; Calculate the centroid on the rotated and aligned 2D projection. and Displacement between and Combined with the centroid difference along the direction of the third principal component (PC3) obtained from PCA analysis. Together they form the initial translation vector. ; This generates an initial coarse registration transformation matrix. ; S4: Based on asymmetric key regions Hypothesis testing and selection First, in Building Information Modeling In this process, the area of ​​the reinforcing bar protective pad in a specific embedded part is identified as the asymmetric critical area. ,like Figure 6 As shown; Then, with the initial rotation assumption Centered on, in Within the scope, Sampling was performed at intervals, generating a total of 101 rotation angles. For each rotation angle... Repeat step S3.2 to calculate the translation vector and generate the transformation matrix. ; Next, each transformation matrix Application to measured point cloud The transformed point cloud is obtained. ;calculate In the asymmetric critical region Inside, with BIM point cloud The average distance from the points to the surface between them is used as the matching error. ; Finally, the matching errors of these 101 rotation angles were compared. Choose the transformation matrix with the smallest error. As the final coarse registration result, such as Figure 7 As shown; S5: Output Results The optimal coarse registration transformation matrix The output is sent to a text file and used as input for subsequent fine-grained registration algorithms (such as the Trimmed ICP algorithm) to complete the final accurate registration.

Claims

1. A rapid coarse registration method for point clouds of precast beam reinforcement skeletons with BIM, characterized in that, Input the measured point cloud and BIM point cloud, and reduce them to a two-dimensional plane through principal component analysis. Construct a one-dimensional "projection point count - rotation angle" distribution vector as a global feature descriptor for fast initial matching, and generate an initial coarse registration transformation matrix. Furthermore, it utilizes predefined asymmetric key regions in BIM. As an arbitrator, multiple initial coarse registration transformation matrices are used. After verification and selection, the optimal coarse registration transformation matrix is ​​finally output as the coarse registration result. Specifically, the following steps are included: S1: Input the measured point cloud of the reinforcing steel cage. Building Information Model (BIM) and Design ; S2: The measured point cloud was analyzed using principal component analysis. and by the model The generated BIM point cloud Feature extraction and dimensionality reduction are performed to construct a distribution vector as a global feature descriptor; S3: Based on the matching of the global feature descriptors, generate an initial coarse registration transformation matrix. ; S4: Based on predefined elements in the Building Information Model Asymmetric key regions For at least one of the initial coarse registration transformation matrices The matching was verified, and the matrix with the smallest matching error was selected as the final coarse registration result. ; S5: Output the final coarse registration result. .

2. The rapid coarse registration method for point cloud and BIM of precast beam reinforcement skeleton according to claim 1, characterized in that, The specific steps of S1 are as follows: Input the measured point cloud of the steel reinforcement cage scanned on site. Building Information Model (BIM) and Design For the model Perform grid-based sampling to generate the corresponding BIM point cloud. ; in, This represents a data point in the measured point cloud of the steel reinforcement cage. Representing three-dimensional real space, This represents point elements in point cloud data, belonging to a three-dimensional real number space. This represents the total number of points in the point cloud dataset; similarly... This represents the points in the point cloud data generated by BIM, where the total number of point cloud elements is... .

3. The rapid coarse registration method for point cloud and BIM of precast beam reinforcement skeleton according to claim 2, characterized in that, Step S2 includes: from the measured point cloud and BIM point cloud The distribution vector of "projection point count - rotation angle" is extracted as a global feature descriptor, and the point cloud is subjected to an initial coarse registration transformation based on the correlation between the global feature descriptors.

4. The rapid coarse registration method for point cloud and BIM of precast beam reinforcement skeleton according to claim 3, characterized in that, The feature extraction and dimensionality reduction process in step S2 includes: For measured point cloud and BIM point cloud Principal component analysis is performed to calculate the three principal vector directions, which are the eigenvectors corresponding to the top three eigenvalues ​​obtained by eigenvalue decomposition of the three-dimensional covariance matrix of the point cloud. , , ; and the measured point cloud and BIM point cloud Projected onto the first two principal vectors respectively , On Zhang Cheng's plane, a two-dimensional projection image is obtained. and The projection process is represented by formula (1): in , These represent the centroid coordinates of the point cloud, respectively.

5. The rapid coarse registration method for point cloud and BIM of precast beam reinforcement skeleton according to claim 4, characterized in that, The construction of the global feature descriptor in step S2 is as follows: Two-dimensional projection diagram centroid Using the origin as the reference point, at a fixed angle Interval rotating rays, generating ray; for the first A ray, i.e., an angle of Count the number of points that fall on the projection map. Generate a one-dimensional "projection point count - rotation angle" distribution vector. That is, the measured point cloud Global feature descriptor ; Similarly, we can obtain BIM point clouds. Distribution vector BIM point cloud Global feature descriptor ; This vector is insensitive to the translation of the point cloud and has cyclic shift properties.

6. The rapid coarse registration method for point cloud and BIM of precast beam reinforcement skeleton according to claim 5, characterized in that, Step S3 includes: By calculating the distribution vector of the measured point cloud The distribution vector of the BIM point cloud Similarity under different cyclic shifts determines the initial rotation angle. Based on the displacement of the center of the projected image after rotation, the initial translation vector is determined. This generates the initial coarse registration transformation matrix. ; Specifically as follows: S3.1: Calculation and Normalized cross-correlation coefficients As shown in formula (2): in, and These are descriptors and The mean, is the number of rotated rays generated above, and s is the number of rotations between the feature descriptors representing vectors; Find Maximum number of rotations Then the initial rotation angle The values ​​are shown in formula (3): in, For a fixed angle; S3.2: Project the measured point cloud in two dimensions Rotation After the angle, calculate the centroid of the two points. and Translation vectors between The third principal component obtained by combining PCA Centroid difference in direction Together they form the initial translation vector. As shown in formula (4): in, Here it is assumed that the main directions of the two cloud points are aligned; S3.3: Combined with rotation angle Translation vector Generate an initial coarse registration transformation matrix. As shown in formula (5): in, It is the plane normal determined by the first two principal vectors, i.e., the rotation angle of the direction of the third vector. The three-dimensional rotation matrix.

7. The rapid coarse registration method for point cloud and BIM of precast beam reinforcement skeleton according to claim 6, characterized in that, Step S4 is as follows: S4.1: In Building Information Modeling Selecting asymmetric key regions The initial coarse registration transformation matrix Acting on the measured point cloud The transformed point cloud is obtained. ; S4.2: Calculate the transformed point cloud In the asymmetric key region Internal and the BIM point cloud Matching error That is, for Falling in the area Each point inside ,exist Find its nearest point and utilize Normal vector at point The calculation error is shown in formula (6): in, express The middle is located in the predefined asymmetric key region A subset of points within, This represents the number of points in the set; S4.3: At the initial rotation angle Within its neighborhood, that is Sampling is performed on intervals for the transformed point cloud. Repeat steps S3.2 to S4.2 to generate a series of initial coarse registration transformation matrices. And calculate the corresponding error. ; S4.4: Compare the matching errors of all candidate hypotheses Selected in asymmetric key regions Top Match The matrix with the smallest error is taken as the final coarse registration result. .

8. The rapid coarse registration method for point cloud and BIM of precast beam reinforcement skeleton according to claim 7, characterized in that, The matching error This represents the average distance from a point to a surface.

9. The rapid coarse registration method for point cloud and BIM of precast beam reinforcement skeleton according to any one of claims 1-8, characterized in that, The asymmetric key region It is at least one of the following: prestressed anchorage zone, specific embedded part cluster, or non-standard steel reinforcement arrangement zone.

Images