Component positioning method based on bim and laser scanning

Abstract

The application discloses a component positioning method based on BIM and laser scanning, and belongs to the technical field of computer vision, and comprises the following steps: collecting three-dimensional data of a prefabricated assembly type building site by means of a laser scanner, acquiring point cloud data of actual positions of components in a building structure, and preliminarily registering the point cloud data with design values of BIM to obtain a deviation distribution diagram after preliminary alignment; extracting a nested level relationship feature of a region exceeding a preset threshold value of the deviation distribution diagram, and obtaining a hierarchical sequence of the nested level by means of topological traversal; calculating differences between actual values and design values of space constraint conditions by means of the hierarchical sequence of the nested level and the point cloud data, and obtaining a quantitative index of dynamic matching deviation; and processing the quantitative index of the dynamic matching deviation to obtain a final component positioning scheme.

Description

Technical Field

[0001] This invention belongs to the field of computer vision technology, and in particular relates to a component positioning method based on BIM and laser scanning. Background Technology

[0002] Existing component positioning methods often reveal significant shortcomings when dealing with complex building structures. Many traditional techniques rely on single measurement methods or manual verification, making it difficult to handle the complex spatial relationships between components and the dynamically changing construction environment. For example, in actual construction, the actual installation position of a component may deviate from the design position due to site conditions, transportation deformation, or other factors. Existing methods often lack sufficient adaptability and accuracy when dealing with multi-component, multi-layered spatial relationships. This limitation makes the identification and adjustment of complex relationships between components inefficient during construction, especially when multi-layered nested components are involved.

[0003] One of the core technical challenges in this field lies in accurately identifying the nested hierarchical relationships between components. Nested hierarchical relationships refer to the interdependence and connection methods of components in space, such as the multi-level assembly relationships of beams, columns, and slabs at complex nodes. Because these relationships exist in the BIM model in an idealized geometric and topological form, the actual installation state of components during construction may deviate from the design due to site deviations, making it difficult to directly locate them using the BIM model. Furthermore, the complexity of these nested relationships amplifies another technical challenge: the dynamic matching of spatial constraints. Spatial constraints include the relative positions, angles, and geometric requirements of contact surfaces between components, which are dynamically adjusted due to environmental changes during actual construction. For example, during the installation of a precast concrete node, the nested connection between beams and columns may experience misalignment of the contact surfaces due to minor deviations, thus affecting the overall structural stability.

[0004] Therefore, in complex building structures, how to accurately identify and locate the deviations in the nesting relationships between components based on the idealized design of the BIM model and the actual construction status obtained by laser scanning has become a key issue. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a component positioning method based on BIM and laser scanning, thereby resolving the issues present in the prior art.

[0006] To achieve the above objectives, the present invention provides a component positioning method based on BIM and laser scanning, comprising the following steps:

[0007] A laser scanner is used to collect three-dimensional data on the site of the prefabricated building to obtain point cloud data of the actual positions of the components in the building structure, and the geometric representation of the components is obtained based on the point cloud data.

[0008] The iterative nearest point algorithm is used to perform preliminary registration of the point cloud data of the actual location of the component with the BIM design value to obtain a deviation distribution map after preliminary alignment;

[0009] Extract the nested hierarchical relationship features of the region in the deviation distribution map that exceeds a preset threshold, and obtain the hierarchical sequence of the nested hierarchy by recursively traversing the component connection topology;

[0010] By using the hierarchical sequence of the nested levels and point cloud data, the difference between the actual value and the design value of the spatial constraints is calculated to obtain a quantitative index of the dynamic matching deviation.

[0011] The type of deviation is determined based on the quantitative index of the dynamic matching deviation, and the adjusted spatial attitude parameters are determined based on the type of deviation.

[0012] The geometric representation of the components in the point cloud data is updated based on the adjusted spatial attitude parameters to generate a corrected 3D model and obtain positioning guidance data consistent with BIM.

[0013] Based on the positioning guidance data, the installation process is simulated to obtain the final component positioning scheme.

[0014] Optionally, the process of obtaining the geometric representation of the component includes:

[0015] A point cloud feature extraction method based on an improved graph attention network is used to extract features from the point cloud data to obtain multi-scale features;

[0016] The registered point cloud is obtained by feature matching of the multi-scale features based on the Lie algebra optimization method.

[0017] Boundary point set is obtained by performing boundary extraction on the registered point cloud;

[0018] The geometric representation of the component is obtained by three-dimensional reconstruction using the NURBS surface fitting method based on the boundary point set.

[0019] Optionally, the process of using the iterative nearest-point algorithm to perform preliminary registration of the point cloud data of the actual location of the component and the BIM design value to obtain a preliminary aligned deviation distribution map includes:

[0020] A multi-scale feature pyramid network based on an attention mechanism is constructed, and the point cloud data and BIM design values ​​are input into the multi-scale feature pyramid network to obtain a deep feature descriptor.

[0021] A point cloud-model registration mathematical model based on optimal transmission theory is established, and the depth feature descriptor is input into the point cloud-model registration mathematical model to obtain the optimal registration result;

[0022] Based on the optimal registration result, a deviation field modeling method based on Gaussian process regression is adopted to construct a spatially continuous deviation distribution function. The deviation distribution function generates a deviation distribution map through Kriging interpolation.

[0023] Optionally, the expression for the point cloud-model registration mathematical model is:

[0024]

[0025] In the formula, For the transfer matrix, T ij For point cloud p i With design value q i The matching probability between them, R∈S0(3) is a three-dimensional rotation matrix. Let H(T) be the three-dimensional translation, H(T) be the entropy regularization term, ε be the coefficient of the entropy regularization term, and C be the coefficient of the three-dimensional translation. ij μ is an element of the cost matrix. i Let ν be the probability distribution of the source point cloud. j Let N be the probability distribution of the source point cloud, and M be the number of points in the target point cloud.

[0026] Optionally, the process of extracting the nested hierarchical relationship features of the regions in the deviation distribution map that exceed a preset threshold, and obtaining the hierarchical sequence of nested levels by recursively traversing the component connection topology, includes:

[0027] Based on the deviation values ​​of each grid node in the deviation distribution map, an adaptive threshold detection algorithm and a spatial clustering method are used to process the data to obtain a set of abnormal regions.

[0028] Based on the component topological connectivity in the set of abnormal regions, a graph attention network is used to extract the deep feature representation of the component nodes to obtain an enhanced topological feature map.

[0029] Based on the enhanced topological feature map, a hidden Markov model is used to model the component hierarchical state, and the optimal state transition path is solved by the Viterbi algorithm to obtain the preliminary hierarchical sequence.

[0030] Based on the initial hierarchical sequence, a geometric constraint and semantic consistency optimization method is used to fuse hierarchical relationships and generate a complete nested hierarchical sequence.

[0031] Optionally, an adaptive threshold detection algorithm is used to obtain the expression for exceeding a preset threshold:

[0032] Γ(x)=μ(x)+k·σ(x)+λ·G(x)

[0033] In the formula, G(x) is the Gaussian weighted bias, σ(x) is the local standard deviation, μ(x) is the local mean, k is the sensitivity coefficient, λ is the spatial weight coefficient, and Γ(x) is the threshold function.

[0034] Optionally, the process of calculating a quantitative indicator of dynamic matching deviation includes:

[0035] Based on the geometric correspondence between the hierarchical sequence and the point cloud data, a multi-scale feature alignment method is used for geometric matching to obtain a preliminary position deviation field.

[0036] Based on the preliminary position deviation field, the relative pose of the component is parametrically modeled using a Lie group and Lie algebra framework, and the pose difference between the actual value and the design value is solved by an optimization algorithm to obtain the rigid body transformation parameters.

[0037] Based on the rigid body transformation parameters, the translation and rotation components are decomposed using screw theory, and the position deviation vector and attitude deviation angle are calculated respectively to obtain the separated spatial constraint deviation.

[0038] Based on the spatial constraint deviation, an adaptive threshold judgment mechanism is used to detect abnormal deviation regions, and the topological mapping relationship is dynamically adjusted through a graph structure optimization method to obtain the corrected deviation distribution.

[0039] Based on the corrected deviation distribution, the Kalman filter method is used to fuse multi-source dynamic deviation indices, and the final comprehensive quantitative deviation index is generated through weighted Bayesian ensemble.

[0040] Optionally, the process of processing the quantitative index of the dynamic matching deviation to obtain the final component positioning scheme includes:

[0041] Based on the quantitative index of the dynamic matching deviation, the geometric representation of the component in the point cloud data is updated using the attitude parameter adjustment method to obtain the corrected component data;

[0042] Based on the corrected component data, a corrected 3D model is generated using a 3D reconstruction method, and the corresponding model representation is obtained through BIM consistency comparison.

[0043] Based on the model representation, a feature extraction method is used to obtain positioning guidance data, and positioning guidance data consistent with BIM is obtained through building information matching.

[0044] Based on the positioning guidance data, a mesh model is constructed using the finite element analysis method, and the deformation response is calculated by applying simulated installation loads to obtain the structural stress distribution;

[0045] Based on the stress distribution of the structure, a safety value is calculated using a stability assessment method, and the position deviation parameter is adjusted based on the safety value to obtain the corrected simulation results;

[0046] Based on the corrected simulation results, the final component positioning scheme is determined.

[0047] The present invention also provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned component positioning method based on BIM and laser scanning.

[0048] The present invention also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned component positioning method based on BIM and laser scanning.

[0049] Compared with the prior art, the present invention has the following advantages and technical effects:

[0050] This invention achieves precise positioning and construction quality control of prefabricated building components through deep integration of laser scanning and BIM models. First, high-precision point cloud data of the components on site is acquired using laser scanning. This data is then registered with the BIM design model using an iterative nearest-point algorithm to quickly identify the distribution of construction deviations. For areas with out-of-tolerance conditions, nested hierarchical features are extracted by recursively traversing the component topology to establish a hierarchical sequence relationship. Based on spatial constraints, a dynamic matching deviation quantification index is calculated, and the least squares method is used to optimize the spatial attitude parameters of the components, generating a corrected 3D model. Finally, finite element analysis is used to verify structural stability and output an optimized positioning scheme. This method significantly improves the construction accuracy of complex building structures, reduces manual adjustment time and rework costs, ensures assembly quality and structural safety, and provides digital technical support for intelligent construction. Attached Figure Description

[0051] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0052] Figure 1 This is a flowchart of an embodiment of the present invention. Detailed Implementation

[0053] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0054] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0055] Example 1

[0056] like Figure 1 As shown, this embodiment provides a component positioning method based on BIM and laser scanning, including the following steps:

[0057] S101. Use a laser scanner to collect three-dimensional data on the site of the prefabricated building, obtain point cloud data of the actual position of the components in the building structure, and obtain the geometric representation of the components based on the point cloud data.

[0058] Furthermore, a laser scanner is used to collect 3D data from the prefabricated building site, obtaining point cloud data containing the actual locations of components in the complex building structure. The process of obtaining the geometric representation of the components includes: extracting features from the point cloud data using a point cloud feature extraction method based on an improved graph attention network to obtain multi-scale features; performing feature matching on the multi-scale features using a Lie algebra optimization method to obtain a registered point cloud; extracting the boundaries of the registered point cloud to obtain a boundary point set; and performing 3D reconstruction based on the boundary point set using a NURBS surface fitting method to obtain the geometric representation of the component.

[0059] Furthermore, as a specific implementation of this embodiment, when using a laser scanner for 3D data acquisition at a prefabricated building site, the scanning can be performed on the beam-column structural area of ​​a multi-story residential building. Scanners such as the FaroFocus series emit laser pulses from the ground and multiple height positions, capturing surface reflection signals from building components such as concrete beams and steel columns, thereby generating point cloud data containing millions of points. These point clouds not only record the actual position of the components but also include information on slight tilting or displacement due to construction deviations, ensuring that the data reflects the true state of the site. In this way, the acquisition process can cover hidden corners in complex building structures, such as welded areas at beam-column junctions, avoiding blind spots of traditional measurement methods.

[0060] Furthermore, based on the point cloud data obtained from multi-view scanning, an improved graph attention network is used to construct a feature extraction model to obtain multi-scale depth features. The feature extraction model is as follows:

[0061]

[0062] In the formula, Let σ be the feature vector of the i-th node in layer l+1, σ be the sigmoid activation function, N(i) be the set of neighboring nodes of node i, and α be the feature vector of the i-th node in layer l+1. ij W represents the attention weight of node i to node j. (l) The weight matrix is ​​a learnable matrix. Let be the feature vector of the j-th node in layer l.

[0063] Multi-scale feature fusion is: F fused =σ(W g [F 16 ||F 32 ||F 64 ]⊙tanh(W f [F 16 ||F 32 ||F 64 ])

[0064] In the formula, F fused W represents the multi-scale features of the fusion. g W is the gated weight matrix. f Let F be the feature transformation weight matrix. 16 F 32 F 64 This indicates multi-scale features extracted using graph convolution kernels of different scales. The numbers in these subscripts (16, 32, 64) represent the neighborhood range (receptive field size) or feature dimension of the graph convolution at that scale.

[0065] Furthermore, as a specific implementation of this embodiment, the multi-scale feature registration process based on Lie algebra adopts a coarse-to-fine optimization strategy, first at the largest scale (F... 64 Initialize the transformation parameters on a large receptive field (F), perform coarse registration using large receptive field features, and quickly converge to the roughly aligned region; then at a medium scale (F... 32 Optimize transformation parameters at the minimum scale (F), and precisely adjust pose relationships through component-level features; finally, optimize transformation parameters at the minimum scale (F). 16 Fine registration is performed on the model, utilizing fine-grained features to achieve final fine alignment. An iterative optimization approach is employed within each scale, calculating the feature matching residuals and Jacobian matrix to solve the incremental equations and update the Lie algebra parameters until convergence before proceeding to the next scale for optimization. Finally, the obtained optimal transformation matrix is ​​applied to generate the registered point cloud data.

[0066] Furthermore, as a specific implementation of this embodiment, when extracting the boundaries of the registered point cloud, a region growing algorithm based on normal vectors is first used to identify continuous surface regions in the point cloud. Then, an edge detection algorithm is used to extract the boundary points of each surface region, forming a preliminary boundary point set. Next, a curvature-based filtering method is used to remove noise points and pseudo-boundary points, ultimately obtaining an accurate set of boundary feature points. When performing 3D reconstruction based on the boundary point set, the boundary points are first parameterized to determine the node vectors of the NURBS surface. Then, the control point coordinates and weight factors are calculated using the least squares method to generate an initial NURBS surface model. Next, an iterative optimization algorithm is used to adjust the control point positions so that the surface fits the boundary point set to the maximum extent. Finally, the reconstructed surface is smoothed and its quality verified to obtain an accurate geometric representation of the components.

[0067] S102. The iterative nearest point algorithm is used to perform preliminary registration of the point cloud data of the actual location of the component with the design value of BIM to obtain the deviation distribution map after preliminary alignment.

[0068] Furthermore, the process of using the iterative nearest-point algorithm to perform preliminary registration of the point cloud data of the actual location of the component and the design values ​​of the BIM model to obtain a preliminary aligned deviation distribution map includes: constructing a multi-scale feature pyramid network based on an attention mechanism, inputting the point cloud data and the design values ​​of the BIM model into the multi-scale feature pyramid network to obtain a deep feature descriptor; establishing a point cloud-model registration mathematical model based on optimal transport theory, inputting the deep feature descriptor into the point cloud-model registration mathematical model to obtain the optimal registration result; and, for the optimal registration result, using a deviation field modeling method based on Gaussian process regression to construct a spatially continuous deviation distribution function, and generating a deviation distribution map through the Kriging interpolation method.

[0069] Furthermore, as a specific implementation of this embodiment, a deep feature extraction model is constructed based on point cloud data and BIM model data using a multi-scale feature pyramid network: F = FPN(P, Q; θ), where P is the point cloud data, Q is the BIM model sampling data, and θ is the network parameter. This is achieved through an attention mechanism β. ij Deep feature descriptors with rotation and translation invariance are extracted; based on these deep feature descriptors, a mathematical model for point cloud-model registration is constructed using optimal transport theory. The expression for the mathematical model of point cloud-model registration is:

[0070]

[0071] In the formula, For the transfer matrix, T ij For point cloud p i With design value q i The matching probability between them, R∈S0(3) is a three-dimensional rotation matrix. Let H(T) be the three-dimensional translation, H(T) be the entropy regularization term, ε be the coefficient of the entropy regularization term, and C be the coefficient of the three-dimensional translation. ij μ is an element of the cost matrix. i Let ν be the probability distribution of the source point cloud. j Let N be the probability distribution of the source point cloud, and M be the number of points in the target point cloud.

[0072] The optimal transfer matrix is ​​solved using the Sinkhorn iterative algorithm. Based on the optimal transfer matrix, a bias field model is constructed using Gaussian process regression: f(x) ~ gP(m(x), k(x,x')). The Matern kernel is used as the kernel function, and a spatially continuous bias distribution function is obtained through posterior prediction distribution. Based on the bias distribution function, a visualization model is constructed using Kriging interpolation. The weight coefficients are obtained by solving a system of linear equations using a variogram model to obtain a high-precision bias distribution map. Finally, three-dimensional visualization of the bias field is achieved through grid mapping.

[0073] The visualization model constructed using the Kriging interpolation method is as follows:

[0074] The variogram model is as follows:

[0075] The system of linear equations to be solved is:

[0076] In the formula, Let z(x) be the predicted deviation at the interpolation point. i ) represents the location of the measurement point with a known deviation value, w i denoted by , where h is the spatial distance between point pairs, c0 represents measurement error and microvariability, c is the upper limit of overall variability, a is the range of influence of spatial correlation, and γ is the variogram matrix between observation points.

[0077] S103. Extract the nested hierarchical relationship features of the region in the deviation distribution map that exceeds the preset threshold, and obtain the hierarchical sequence of the nested hierarchy by recursively traversing the component connection topology.

[0078] Furthermore, the process of extracting nested hierarchical relationship features from regions exceeding a preset threshold in the deviation distribution map and obtaining a hierarchical sequence of nested levels by recursively traversing the component connection topology includes: processing the deviation values ​​of each grid node in the deviation distribution map using an adaptive threshold detection algorithm and spatial clustering method to obtain an abnormal region set; extracting the deep feature representation of component nodes using a graph attention network based on the component topology connection relationships in the abnormal region set to obtain an enhanced topology feature map; modeling the component hierarchical state using a hidden Markov model based on the enhanced topology feature map and solving for the optimal state transition path using the Viterbi algorithm to obtain a preliminary hierarchical sequence; and fusing hierarchical relationships using a geometric constraint and semantic consistency optimization method based on the preliminary hierarchical sequence to generate a complete nested hierarchical sequence.

[0079] Furthermore, as a specific implementation of this embodiment, the expression for exceeding a preset threshold is obtained using an adaptive threshold detection algorithm:

[0080] Γ(x)=μ(x)+k·σ(x)+λ·G(x)

[0081] In the formula, G(x) is the Gaussian weighted bias, σ(x) is the local standard deviation, μ(x) is the local mean, k is the sensitivity coefficient, λ is the spatial weight coefficient, and Γ(x) is the threshold function.

[0082] Furthermore, as a specific implementation method of this embodiment, based on the enhanced topological feature map, a Hidden Markov Model (HMM) is first used to probabilistically model the hierarchical relationships of components: the position of each component in different levels is defined as a hidden state, and the connection relationships and geometric attributes in the topological feature map are used as observation variables. Statistical learning is used to obtain the state transition probability and observation probability matrix. On this basis, the Viterbi algorithm is used to search for the global optimal path. This algorithm calculates the maximum probability path for each state step by step through dynamic programming and retains path backtracking information, finally outputting the state sequence with the highest probability, i.e., the preliminary hierarchical sequence. Subsequently, multi-constraint optimization is performed based on the preliminary hierarchical sequence: first, geometric features (such as component spatial coordinates, orientation angles, etc.) are extracted from the sequence to establish a geometric constraint model. Spatial position conflicts are corrected through nearest neighbor distance detection and angle tolerance analysis; simultaneously, a semantic consistency rule base is constructed by combining BIM semantic information (such as component type, functional classification, assembly logic, etc.) to semantically verify the hierarchical relationships. An iterative optimization strategy is adopted to process geometric and semantic conflicts sequentially. Abnormal connections are eliminated by combining local adjustments and global rearrangements, ultimately generating a complete nested hierarchical sequence that simultaneously satisfies topological connectivity, geometric rationality, and semantic consistency.

[0083] S104. By using the nested hierarchical sequence and point cloud data, calculate the difference between the actual value and the design value of the spatial constraints to obtain a quantitative index of the dynamic matching deviation.

[0084] Furthermore, the process of calculating the quantitative index of dynamic matching deviation includes: based on the geometric correspondence between the hierarchical sequence and point cloud data, a multi-scale feature alignment method is used for geometric matching to obtain a preliminary position deviation field; based on the preliminary position deviation field, a Lie group and Lie algebra framework are used to parametrically model the relative pose of the components, and the pose difference between the actual value and the design value is solved through an optimization algorithm to obtain rigid body transformation parameters; based on the rigid body transformation parameters, the spinor theory is used to decompose the translation and rotation components, and the position deviation vector and attitude deviation angle are calculated respectively to obtain the separated spatial constraint deviation; based on the spatial constraint deviation, an adaptive threshold judgment mechanism is used to detect abnormal deviation regions, and the topological mapping relationship is dynamically adjusted through a graph structure optimization method to obtain the corrected deviation distribution; based on the corrected deviation distribution, a Kalman filter method is used to fuse multi-source dynamic deviation indices, and a weighted Bayesian ensemble is used to generate the final comprehensive quantitative deviation index.

[0085] Furthermore, as a specific implementation of this embodiment, based on the geometric correspondence between the hierarchical sequence and the point cloud data, a multi-scale feature alignment method is used to construct a geometric matching objective function to obtain the preliminary position deviation field. The geometric matching objective function is:

[0086]

[0087] Based on the theory of Lie groups and Lie algebras, the rigid body transformation parameterizes into Lie algebra vectors using the exponential mapping method, and the optimal pose parameters are solved by a weighted robust optimization algorithm to obtain an accurate representation of the rigid body transformation.

[0088] Based on the screw theory framework, a logarithmic mapping method is used to decompose pose deviation into translation and rotation components, and the separated spatial constraint deviation is obtained through norm calculation. Based on preset threshold parameters, a logical judgment function is used to detect abnormal deviation regions, and a graph structure optimization method is used to dynamically adjust the topological mapping relationship to obtain the corrected deviation distribution. Based on Kalman filtering theory, a state-space modeling method is used to fuse multi-source dynamic deviation indices, and a final comprehensive quantitative deviation index is generated through Bayesian inverse probability weighting.

[0089] S105. Determine the type of deviation based on the quantitative index of dynamic matching deviation, and determine the adjusted spatial attitude parameters based on the type of deviation.

[0090] Furthermore, the deviation type is obtained based on the quantitative indicators. If the deviation type is angular deviation, the component attitude is processed using the least squares method. The least squares method takes the observed and designed values ​​of the component attitude as inputs and outputs the attitude estimate that minimizes the residuals, thus obtaining optimized attitude data. Spatial attitude parameters are extracted from the optimized attitude data, and adjustment parameter values ​​are identified for angular deviations to determine preliminary adjustment parameters. The adjustment results are then fused using the preliminary adjustment parameters to obtain the final spatial attitude parameters.

[0091] Furthermore, as a specific implementation of this embodiment, in the business scenario of building component assembly, when the system identifies the type of deviation based on previously calculated quantitative indicators such as deviation value sequences, if the detected deviation is an angular deviation, this usually stems from the rotational difference between the actual installation posture of the component in the point cloud scan data and the design model. For example, the tilt angle of a steel beam is actually 5 degrees but designed to be 0 degrees. In this case, the least squares method needs to be activated to handle the component posture. The least squares method is an optimization algorithm whose principle is to solve for the best fitting parameters by minimizing the sum of squares of the residuals between the observed and predicted values. Here, the inputs include the component posture observation values ​​extracted from the point cloud data, such as the actual Euler angle coordinates, and the design values, such as the ideal rotation matrix. Then, the parameters are adjusted through iterative calculation to minimize the residuals, and the optimized posture estimation data is output, thereby helping to correct the installation accuracy of beams and columns on the assembly line.

[0092] Specifically, the process of extracting spatial attitude parameters from optimized attitude data can be understood as parsing key parameters such as yaw angle, pitch angle, and roll angle. For example, in the installation of bridge components, the optimized data may show that the actual deviation of the yaw angle is 3 degrees. At this time, the angle deviation is identified, and the value is adjusted by comparing the actual parameter with a threshold such as 2 degrees. The initial adjustment parameter is determined, such as correcting the yaw angle to 1 degree to approach the design requirements.

[0093] In one embodiment, the final spatial attitude parameters are obtained by fusing the results of preliminary parameter adjustments. This involves weighted integration of multiple parameters such as angle and position deviation. For example, in the assembly of wind turbine tower components, the preliminary parameters may include a 2-degree adjustment of the roll angle and a 1-degree correction of the pitch angle. Then, during fusion, an averaging strategy is used to generate the final parameters such as the integrated attitude vector.

[0094] S106. Update the component geometric representation in the point cloud data based on the adjusted spatial attitude parameters, generate the corrected 3D model, and obtain positioning guidance data consistent with the BIM model.

[0095] Furthermore, by adjusting the fusion spatial bias through attitude parameters, the component representations in the point cloud data are updated to obtain corrected component data. A 3D model is generated from the corrected component data, and a model representation corresponding to the component data is obtained by combining model consistency comparison. Positioning guidance data is extracted from the model representation, and positioning guidance data consistent with the building information model is obtained through building information matching.

[0096] Furthermore, as a specific implementation of this embodiment, in the process of adjusting the posture of building components, it is first necessary to fuse spatial deviations using posture parameters. Specifically, this adjustment involves applying posture parameters such as rotation angles and displacement vectors to point cloud data. For example, for a steel beam component of a bridge, assuming that the initial point cloud data shows that the beam has a rotational deviation of 5 degrees and a translational deviation of 2 meters on the x-axis, the posture parameter adjustment algorithm inputs these deviation parameters into the point cloud transformation matrix, performs coordinate transformation on each point cloud point, thereby updating the representation of the component in three-dimensional space, and finally obtaining the corrected component data.

[0097] In one embodiment, when generating a 3D model from the corrected component data, a meshing method can be used to convert the point cloud into a surface model. For the corrected point cloud data of the bridge steel beams mentioned above, a preliminary 3D model is first generated using a triangular meshing algorithm. Then, a model consistency comparison is performed, that is, the generated model is geometrically aligned and compared with the design BIM model, and the Hausdorff distance is calculated to evaluate consistency, thereby obtaining an accurate model representation corresponding to the component data.

[0098] For example, extracting positioning guidance data from a model representation typically involves extracting key feature points from the model, such as the center point, edge lines, and connection nodes of components. For instance, in a bridge construction scenario, the coordinates of the anchorage points at the beam ends are extracted from the steel beam model representation as positioning guidance data. Then, through building information matching, these data are matched with corresponding elements in the building information model. For example, the ICP algorithm is used to iteratively align the two sets of points to obtain consistent positioning guidance data. This matching process can effectively integrate on-site data and design data to achieve accurate component installation guidance.

[0099] S107. Based on the positioning guidance data, the installation process is simulated to obtain the final component positioning scheme.

[0100] Furthermore, the installation process is simulated based on the positioning guidance data. Finite element analysis is used to verify the overall structural stability and determine whether the simulation results meet the preset safety threshold, thus obtaining the final component positioning scheme. Initial component positions are obtained from the positioning guidance data. A mesh model is constructed by dividing the data into mesh elements and defining material properties using finite element analysis. The deformation response is calculated by applying simulated installation loads to the mesh nodes, yielding the structural stress distribution. The overall stability is evaluated based on the stress distribution. If the preset safety threshold is exceeded, the position deviation parameters are adjusted, and the component position is iteratively optimized to minimize the stress peak, obtaining a corrected simulation result. Based on the corrected simulation result, it is determined whether the threshold requirement is met. An optimized adjustment scheme is extracted from this result to determine the final component positioning scheme.

[0101] This embodiment also provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a component positioning method based on BIM and laser scanning.

[0102] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a component positioning method based on BIM and laser scanning.

[0103] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A component positioning method based on BIM and laser scanning, characterized in that, Includes the following steps: A laser scanner is used to collect three-dimensional data on the site of the prefabricated building to obtain point cloud data of the actual positions of the components in the building structure, and the geometric representation of the components is obtained based on the point cloud data. The iterative nearest point algorithm is used to perform preliminary registration of the point cloud data of the actual location of the component with the BIM design value to obtain a deviation distribution map after preliminary alignment; The nested hierarchical relationship features of regions exceeding a preset threshold in the deviation distribution map are extracted, and a hierarchical sequence of nested levels is obtained by recursively traversing the component connection topology. The process of obtaining the hierarchical sequence of nested levels includes: processing the deviation values ​​of each grid node in the deviation distribution map using an adaptive threshold detection algorithm and a spatial clustering method to obtain an abnormal region set; extracting the deep feature representation of component nodes using a graph attention network based on the component topology connection relationships in the abnormal region set to obtain an enhanced topology feature map; modeling the component hierarchical state using a hidden Markov model based on the enhanced topology feature map, and solving for the optimal state transition path using the Viterbi algorithm to obtain a preliminary hierarchical sequence; and fusing hierarchical relationships using a geometric constraint and semantic consistency optimization method based on the preliminary hierarchical sequence to generate a complete nested hierarchical sequence. Using the nested hierarchical sequence and point cloud data, the difference between the actual and design values ​​of spatial constraints is calculated to obtain a quantitative index of dynamic matching deviation. The process of calculating the quantitative index of dynamic matching deviation includes: based on the geometric correspondence between the hierarchical sequence and point cloud data, a multi-scale feature alignment method is used for geometric matching to obtain a preliminary position deviation field; based on the preliminary position deviation field, a Lie group and Lie algebra framework are used to parametrically model the relative pose of the components, and an optimization algorithm is used to solve for the pose difference between the actual and design values ​​to obtain rigid body transformation parameters; based on the rigid body transformation parameters, spinor theory is used to decompose translation and rotation components, and the position deviation vector and attitude deviation angle are calculated respectively to obtain the separated spatial constraint deviation; based on the spatial constraint deviation, an adaptive threshold judgment mechanism is used to detect abnormal deviation regions, and a graph structure optimization method is used to dynamically adjust the topological mapping relationship to obtain a corrected deviation distribution; based on the corrected deviation distribution, a Kalman filter method is used to fuse multi-source dynamic deviation indices, and a weighted Bayesian ensemble is used to generate the final comprehensive quantitative deviation index. The type of deviation is determined based on the quantitative index of the dynamic matching deviation, and the adjusted spatial attitude parameters are determined based on the type of deviation. The component geometric representation in the point cloud data is updated based on the adjusted spatial attitude parameters to generate a corrected 3D model and obtain positioning guidance data consistent with the BIM model. Based on the positioning guidance data, the installation process is simulated to obtain the final component positioning scheme.

2. The component positioning method based on BIM and laser scanning according to claim 1, characterized in that, The process of obtaining the geometric representation of a component includes: A point cloud feature extraction method based on an improved graph attention network is used to extract features from the point cloud data to obtain multi-scale features; The registered point cloud is obtained by feature matching of the multi-scale features based on the Lie algebra optimization method. Boundary point set is obtained by performing boundary extraction on the registered point cloud; The geometric representation of the component is obtained by three-dimensional reconstruction using the NURBS surface fitting method based on the boundary point set.

3. The component positioning method based on BIM and laser scanning according to claim 1, characterized in that, The process of using the iterative nearest-point algorithm to perform preliminary registration of the point cloud data of the actual location of the component with the design values ​​of the BIM model to obtain a preliminary aligned deviation distribution map includes: A multi-scale feature pyramid network based on an attention mechanism is constructed, and the point cloud data and the design values ​​of the BIM model are input into the multi-scale feature pyramid network to obtain deep feature descriptors. A point cloud-model registration mathematical model based on optimal transmission theory is established, and the depth feature descriptor is input into the point cloud-model registration mathematical model to obtain the optimal registration result; Based on the optimal registration result, a deviation field modeling method based on Gaussian process regression is adopted to construct a spatially continuous deviation distribution function. The deviation distribution function generates a deviation distribution map through Kriging interpolation.

4. The component positioning method based on BIM and laser scanning according to claim 3, characterized in that, The expression for the point cloud-model registration mathematical model is as follows: In the formula, For the transfer matrix, T ij For point cloud p i With design value q i The probability of matching between them It is a three-dimensional rotation matrix. Let H(T) be the three-dimensional translation, H(T) be the entropy regularization term, ε be the coefficient of the entropy regularization term, and C be the coefficient of the three-dimensional translation. ij For cost matrix elements, Let be the probability distribution of the source point cloud. Let N be the probability distribution of the source point cloud, and M be the number of points in the target point cloud.

5. The component positioning method based on BIM and laser scanning according to claim 1, characterized in that, The expression for obtaining the threshold value exceeding the preset threshold using the adaptive threshold detection algorithm is as follows: In the formula, G(x) is the Gaussian weighted bias. For local standard deviation, Let λ be the local mean, k be the sensitivity coefficient, and λ be the spatial weighting coefficient. This is a threshold function.

6. The component positioning method based on BIM and laser scanning according to claim 1, characterized in that, The process of processing the quantitative index of the dynamic matching deviation to obtain the final component positioning scheme includes: Based on the quantitative index of the dynamic matching deviation, the geometric representation of the component in the point cloud data is updated using the attitude parameter adjustment method to obtain the corrected component data; Based on the corrected component data, a corrected 3D model is generated using a 3D reconstruction method, and the corresponding model representation is obtained through BIM consistency comparison. Based on the model representation, a feature extraction method is used to obtain positioning guidance data, and positioning guidance data consistent with BIM is obtained through building information matching. Based on the positioning guidance data, a mesh model is constructed using the finite element analysis method, and the deformation response is calculated by applying simulated installation loads to obtain the structural stress distribution; Based on the stress distribution of the structure, a safety value is calculated using a stability assessment method, and the position deviation parameter is adjusted based on the safety value to obtain the corrected simulation results; Based on the corrected simulation results, the final component positioning scheme is determined.

7. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the component positioning method based on BIM and laser scanning as described in claim 1.

8. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the component positioning method based on BIM and laser scanning as described in claim 1.

Images