A building information model component-level updating reconstruction method based on point cloud data

Abstract

The application discloses a building information model component level updating reconstruction method based on point cloud data, and comprises the following steps: firstly, creating a directional bounding box for each component element of an original building information model, sorting the component elements based on the volume of the directional bounding box, and distributing corresponding point cloud data to each component element according to the directional bounding box and the distance from a calculation point to a nearest voxel grid of the model; then, extracting model shape features based on a convolutional neural network and iteratively updating the selected component element model; finally, realizing the refinement of the selected component element model through a triangulation and remeshing algorithm, and updating and reconstructing the building information model fitted with the point cloud data at a high resolution. The application can realize high-resolution updating reconstruction of the building information model based on three-dimensional point cloud data scanned on site, can realize high-resolution geometric updating of the selected component model without changing the semantic information of the original model, and has important application value.

Description

A Component-Level Update and Reconstruction Method for Building Information Modeling Based on Point Cloud Data Technical Field

[0001] This invention relates to the field of three-dimensional updating and reconstruction technology of building information models, and in particular to a component-level updating and reconstruction method for building information models based on point cloud data. Background Technology

[0002] Building Information Modeling (BIM) is a core data carrier in modern engineering. By integrating geometric, semantic, and mechanical performance information, BIM models provide fundamental support for the digital representation and intelligent analysis of building engineering objects. Their applications permeate various research areas of building engineering, including completion acceptance, structural health monitoring, and structural finite element analysis. Although BIM models create the information needed to help plan, manage, and coordinate construction during the project design phase, the BIM at the design stage often deviates significantly from the final condition due to construction errors, material deformation, or environmental influences. Existing 3D update and reconstruction methods are mostly based on feature extraction or surface reconstruction, failing to consider both the geometric details and semantic information of the model. Therefore, there is an urgent need for a high-resolution update and reconstruction method for BIM components based on point cloud data, which can achieve high-resolution geometric updates of selected component models without altering the semantic information of the original model, possessing significant application value. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for updating and reconstructing building information models at the component level based on point cloud data.

[0004] The objective of this invention is achieved through the following technical solution: a method for component-level updating and reconstruction of a building information model based on point cloud data, comprising the following steps:

[0005] (1) Obtain real-world point cloud data and original building information model data, and preprocess the point cloud data;

[0006] (2) Using the geometric spatial features of the component elements of the original building information model, create a corresponding oriented bounding box for each component element of the original building information model;

[0007] (3) Sort the component elements according to the volume of the oriented bounding box, assign corresponding point cloud data to each component element based on the oriented bounding box and the nearest voxel mesh distance from the point cloud to the building information model, and create a point cloud-component element matching group list;

[0008] (4) Select the target component group from the point cloud-component element matching group list, use a convolutional neural network to extract the model shape features and iteratively update the selected component element model to create a preliminary updated model of the selected target component.

[0009] (5) Based on the initial update model of the target component, the component element model of the target component is refined through triangulation and remeshing algorithms, and a high-resolution update model that fits the target component with the point cloud data is created.

[0010] (6) For the selected target component element, replace the original geometric model in the original building information model with the high-resolution update model to achieve component-level update of the building information model.

[0011] Furthermore, the preprocessing of the point cloud data specifically includes:

[0012] First, voxel downsampling is performed on the point cloud data. Then, noise is removed based on Gaussian filtering and radius filtering. Finally, coordinate system registration is performed with the building information model using the iterative nearest point algorithm to obtain the preprocessed point cloud data.

[0013] Furthermore, step (2) specifically includes:

[0014] Based on the geometric spatial features of the component elements in the original building information model, principal component analysis is used to traverse the component elements in the original building information model to obtain the main direction and the three-dimensional spatial coordinates of the center point of the geometric model of each component element. Based on the space in which each component element is located, the extension length of the oriented bounding box along the xyz axis in the main direction spatial coordinate system is calculated to create an oriented bounding box for the component element.

[0015] Furthermore, step (3) specifically includes:

[0016] First, sort all component elements according to the volume of the oriented bounding box; then, traverse the oriented bounding boxes of the component elements, set a distance threshold, and map the point cloud data enclosed by the oriented bounding box or the point cloud data whose distance to the nearest voxel mesh of the building information model is less than or equal to the distance threshold to the component element; finally, create a point cloud-component element matching group list.

[0017] Furthermore, in step (4), the convolutional neural network is built based on the U-Net encoder-decoder model architecture, specifically including a grid convolutional layer, a grid pooling layer, and a grid depooling layer; the input data of the convolutional neural network includes the corresponding building information model data and a constant random noise vector, wherein the building information model data is transmitted in the convolutional neural network in the form of a three-dimensional tensor, and the constant random noise vector is used to randomly initialize the model weights in each iteration; the output of the convolutional neural network is an array list consisting of a series of relative displacement edges.

[0018] Further, in step (4), when iteratively updating the selected component element model, the relative displacement of the vertex is obtained after reconstruction based on the array list of a series of relatively displaced edges output by the convolutional neural network. The new component element model is reconstructed based on the relative displacement of the vertex and the previous component element model. The newly built component element model is uniformly sampled to obtain a uniform sampling point set. The bidirectional chamfer distance is calculated as a loss function based on the uniform sampling point set and the real-world point cloud dataset. The optimization objective is to minimize the loss function value. The model parameters of the convolutional neural network are updated by backpropagation to iteratively update the component element model and obtain the preliminary updated model of the target component.

[0019] Furthermore, in step (5), the refinement of the component element model of the target component through triangulation and remeshing algorithms specifically includes:

[0020] The initial update model of the target component is downsampled from a uniform mesh to a sparse model. Then, the point cloud data and the component element model faces are grouped based on the triangular patch normal vectors and connectivity to obtain a face set. For each group, edge constraint triangulation reconstruction is performed to obtain a new face set. Finally, taubin-based smooth mesh reconstruction is performed to obtain the final high-resolution update model that fits the target component with the point cloud data.

[0021] The beneficial effects of this invention are as follows: This invention establishes a framework for high-resolution update and reconstruction of Building Information Models (BIM) components by integrating semantic preservation mechanisms and 3D model reconstruction technology. This method combines oriented bounding boxes with the calculation of the distance from a point to the nearest voxel mesh in the model, establishing a spatial matching mechanism between point cloud data and component elements, and associating the semantic information of the point cloud data with that of the original BIM. Then, based on a U-Net encoder-decoder 3D network architecture with dynamic noise suppression, model features are extracted, the selected component element models are iteratively updated, and finally, model detail reshaping is completed through methods such as triangulation and remeshing. Compared with traditional feature extraction or surface reconstruction methods, this invention can balance high geometric accuracy and complete semantic information during the BIM update process, solving the problem of lack of geometric details or semantic information in traditional model update and reconstruction. Its automated processing flow can be applied to engineering scenarios with strict requirements for model fidelity, such as final acceptance and structural monitoring, improving the engineering applicability of BIM. Attached Figure Description

[0022] Figure 1 is a flowchart of the component-level update and reconstruction method for building information model based on point cloud data according to the present invention;

[0023] Figure 2 is a case diagram of the application of the present invention in actual structural engineering (cable-stayed bridge tower);

[0024] Figure 3 is a schematic diagram of the present invention that maps point cloud data to component elements based on oriented bounding boxes and the distance from a point to the nearest voxel mesh in the model;

[0025] Figure 4 is a schematic diagram of the result of mapping and grouping point cloud data and component elements in this invention;

[0026] Figure 5 is a schematic diagram of a single iteration of the model update based on a convolutional neural network (CNN) according to the present invention;

[0027] Figure 6 is a schematic diagram of the three-dimensional model data structure of the input convolutional neural network (CNN) in this invention;

[0028] Figure 7 is a schematic diagram of the process of achieving model refinement based on triangulation and remeshing algorithms in this invention. Detailed Implementation

[0029] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims. It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not intended to limit this application.

[0030] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0031] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "in response to determination," or "includes." Moreover, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process or method. Without further limitations, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0032] The present invention will now be described in detail with reference to the accompanying drawings. Unless otherwise specified, the features of the following embodiments and implementations can be combined with each other.

[0033] Referring to Figure 1, the component-level update and reconstruction method for building information model based on point cloud data of the present invention specifically includes the following steps:

[0034] (1) Obtain real-world point cloud data and original building information model data, and preprocess the point cloud data.

[0035] In the current step (1), two essential key data for this invention need to be prepared. The first is point cloud data collected on-site from the engineering project, which can be obtained from video images using LiDAR or Structure of Motion (SFM). The second is the original building information model (BIM) of the engineering project. This BIM data can be obtained from the design model created during the engineering project design phase or from design drawings. The model data input supports the open standard format for BIM data exchange using the IFC (Industry Fundamentals Classes) standard, as well as the common CAD model open standard formats such as OBJ and STL.

[0036] Furthermore, point cloud data needs to be preprocessed according to the actual situation before use. The preprocessing of point cloud data includes: first, voxel downsampling of point cloud data, then removing noise based on Gaussian filtering and radius filtering, and then using the Iterative Closest Point (ICP) algorithm to perform coordinate system registration with the building information model.

[0037] It should be understood that the ICP algorithm is a classic algorithm for 3D point cloud registration, mainly used to align two or more sets of point cloud data to achieve the best spatial match. This ICP algorithm iteratively optimizes the calculation of the optimal rotation matrix (R) and translation vector (t) to minimize the error between the source and target point clouds.

[0038] Referring to the case shown in Figure 2, this embodiment uses point cloud data of the cable-stayed bridge tower obtained by lidar and the original building information model roughly established based on drawings to update and reconstruct the elements of the bridge tower components.

[0039] (2) Using the geometric spatial features of the component elements of the original building information model, create a corresponding oriented bounding box for each component element of the original building information model.

[0040] Specifically, based on the geometric spatial characteristics of the component elements in the original building information model, principal component analysis is used to traverse the component elements in the original building information model to obtain the main direction and the three-dimensional spatial coordinates of the center point of the geometric model of each component element. Based on the space in which each component element is located, the extension length of the oriented bounding box along the xyz axis in the main direction spatial coordinate system is calculated to create an oriented bounding box for the component element.

[0041] (3) Sort the component elements according to the volume of the oriented bounding box, and assign corresponding point cloud data to each component element based on the oriented bounding box and the nearest voxel mesh distance from the point cloud to the building information model, and create a point cloud-component element matching group list.

[0042] Specifically, step (3) can be seen in Figure 3. First, sort all component elements according to the volume of the oriented bounding box. Then, traverse the oriented bounding boxes of the component elements, set a distance threshold according to the actual situation of the project, and map the point cloud data enclosed by the oriented bounding box or the point cloud data whose distance to the nearest voxel grid of the building information model is less than or equal to the distance threshold to the component element. Finally, create a point cloud-component element matching group list, and its visualization result is shown in Figure 4.

[0043] (4) Select the target component group from the point cloud-component element matching group list, use a convolutional neural network (CNN) to extract the model shape features and iteratively update the selected component element model to create a preliminary updated model of the selected target component.

[0044] Furthermore, in step (4), the CNN is built using a U-Net encoder-decoder model architecture, specifically including special operation layers such as mesh convolution (MeshConv) layer, mesh pooling (MeshPool) layer and mesh unpooling (MeshUnpool) layer; the input data of the CNN includes the corresponding building information model data and a constant random noise vector, wherein the building information model data is passed in the CNN in the form of a three-dimensional tensor, and the constant random noise vector is used to randomly initialize the model weights in each iteration, thus effectively avoiding bias and overfitting; the output of the CNN is an array list consisting of a series of relatively displacement edges.

[0045] Furthermore, in step (4), when iteratively updating the selected component element model, the relative vertex displacement is obtained after reconstruction based on the array list of a series of relatively displaced edges output by the CNN. The relative vertex displacement is then compared with the previous component element model. Rebuild a new component element model And the newly created component element model Uniform sampling is performed to obtain a uniform sampling point set P. i Based on the uniform sampling point set P i and real-world cloud dataset P t The bidirectional chamfer distance is calculated as the loss function, and the optimization objective is to minimize the loss function value. The CNN model parameters are updated by backpropagation, and the component element model is iteratively updated to obtain the initial updated model of the target component.

[0046] In this embodiment, a CNN is first used to extract the shape features of the model and iteratively update the selected component element model, as shown in Figure 5. The data structure of the Building Information Model (BIM) data is shown in Figure 6. Since the BIM data is transmitted in the CNN in the form of a three-dimensional tensor, for different BIM data formats, open-source frameworks such as ifcopenshell or trimesh are used to read the BIM data and convert it into a triangular mesh (V, F, E), where V, F, and E are used to record the vertex, face, and edge data of the component element model, respectively. Subsequently, the currently iterated BIM data, converted into a triangular mesh, is... and constant random noise vector C i-1 The input is fed into a CNN built on a U-Net encoder-decoder model architecture. The constant random noise vector C i-1 This will be used to randomly initialize model weights in each iteration, avoiding bias and overfitting. The CNN will then output an array list consisting of a series of relatively shifted edges. The rebuild module will then proceed to the rebuild module. The relative displacement of a vertex is obtained by averaging the predicted vertex displacements of each edge. This process can be represented as follows:

[0047]

[0048] In the formula, v represents the relative displacement of the vertex in the i-th iteration. i This represents the vertex of the i-th iteration. Let represent the predicted vertex displacement of the j-th edge, and n represent the total number of edges. This is based on the relative vertex displacements and the original model. Reconstruct a new model And obtain the point set P by uniform sampling i Combined with the newly created uniform sampling point set P i and real-world cloud dataset P t The bidirectional chamfer distance is calculated as a loss function, and minimizing this loss function value is the optimization objective. Backpropagation is then used to update the CNN model parameters. The bidirectional chamfer distance can be expressed as:

[0049]

[0050] In the formula, Loss:d(P i ,P t ) represents the bidirectional chamfer distance, and also represents the loss function; p i Let P represent the uniform sampling point set in the i-th iteration process. i Point cloud data in p t Represents the real-world cloud dataset P t Point cloud data in; n i n represents the total number of point clouds in a uniformly sampled set. t This represents the total number of point clouds in the real-world point cloud dataset. The total number of iterations can be set according to the actual situation. Furthermore, this invention also supports setting a fixed number of iterations and then uniformly upsampling the number of current model surfaces to achieve higher resolution. For details, please refer to step (5) below.

[0051] (5) Based on the preliminary update model of the target component obtained in step (4), the component element model of the target component is refined by triangulation and remeshing algorithm, and a high-resolution update model that fits the target component with the point cloud data is created.

[0052] Furthermore, the component element model of the target component is refined through triangulation and remeshing algorithms. Specifically, this includes: downsampling the initial update model of the target component into a sparse model using a uniform mesh; grouping the point cloud data and component element model faces based on the triangular facet normal vectors and connectivity to obtain a face set; performing edge-constrained triangulation reconstruction for each group to obtain a new face set; and finally performing taubin-based smooth mesh reconstruction to obtain a high-resolution update model that fits the target component with the point cloud data.

[0053] Specifically, the refinement of the component element model is performed after the component element model has been updated and iterated, and the process is shown in Figure 7. The high-resolution model after iteration is then... Uniform grid downsampling to sparse model Then based on the triangular patch normal vector and connectivity A ij The point cloud data and component element model faces are grouped to obtain a face set. The grouping criteria can be expressed as:

[0054]

[0055] In the formula, Dominant Axis represents the grouping basis; A ij The connectivity A represents the connectivity between point cloud i and point cloud j if they are adjacent. ij If n is 1, then n is 0; otherwise n is 0. x n y n z Representing normal vectors respectively The components along the x, y, and z coordinate axes. Then, the point cloud P within a certain distance threshold range... i Mapping to face groups and performing triangulation reconstruction based on the original edge constraints yields a new set of faces. Finally, by integrating the surface set and performing taubin-based smooth mesh reconstruction, a high-resolution updated model that fits the target component with the point cloud data is obtained.

[0056] (6) For the selected target component element, replace the original geometric model in the original building information model with the high-resolution update model obtained in step (5) to realize the component-level update of the building information model.

[0057] In this invention, the updated and reconstructed building information model does not change the semantic information of the original building information model, while achieving high-resolution geometric updates of the selected target component element model.

[0058] In summary, this invention integrates a semantic preservation mechanism for Building Information Modeling (BIM) with 3D model reconstruction technology to establish a framework for high-resolution update and reconstruction of BIM components. Based on the original BIM model and point cloud data acquired from the field, this invention can balance high geometric accuracy with complete semantic information during BIM updates, effectively solving the problem of traditional model updates and reconstructions lacking geometric details or semantic information. Its automated processing flow can be applied to engineering scenarios with strict requirements for model fidelity, such as final acceptance and structural monitoring, thereby improving the engineering applicability of BIM models.

[0059] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for component-level updating and reconstruction of building information model based on point cloud data, characterized in that, Includes the following steps: (1) Obtain real-world point cloud data and original building information model data, and preprocess the point cloud data; (2) Using the geometric spatial features of the component elements of the original building information model, create a corresponding oriented bounding box for each component element of the original building information model; (3) Sort the component elements according to the volume of the oriented bounding box, and assign corresponding point cloud data to each component element based on the oriented bounding box and the nearest voxel mesh distance from the point cloud to the building information model, and create a point cloud-component element matching group list; (4) Select the target component group from the point cloud-component element matching group list, use a convolutional neural network to extract the model shape features and iteratively update the selected component element model, and create a preliminary update model of the selected target component; In step (4), the convolutional neural network is built based on the U-Net encoder-decoder model architecture, specifically including a grid convolutional layer, a grid pooling layer and a grid depooling layer; The input data of the convolutional neural network includes the corresponding building information model data and a constant random noise vector, wherein the building information model data is transmitted in the convolutional neural network in the form of a three-dimensional tensor, and the constant random noise vector is used to randomly initialize the model weights in each iteration; The output of the convolutional neural network is an array list composed of a series of relative displacement edges; In step (4), when iteratively updating the selected component element model, the relative displacement of the vertex is obtained after reconstruction based on the array list of a series of relative displacement edges output by the convolutional neural network. The new component element model is reconstructed based on the relative displacement of the vertex and the previous component element model. The newly created component element model is uniformly sampled to obtain a uniform sampling point set. The bidirectional chamfer distance is calculated based on the uniform sampling point set and the real point cloud dataset as the loss function. The optimization objective is to minimize the loss function value. The model parameters of the convolutional neural network are updated by backpropagation to iteratively update the component element model and obtain the preliminary update model of the target component. (5) Based on the preliminary update model of the target component, the component element model of the target component is refined through triangulation and remeshing algorithms to create a high-resolution update model that fits the target component with the point cloud data. (6) For the selected target component element, the original geometric model in the original building information model is replaced by the high-resolution update model to realize the component-level update of the building information model.

2. The method for component-level updating and reconstruction of building information model based on point cloud data according to claim 1, characterized in that, The preprocessing of point cloud data specifically includes: first, voxel downsampling of the point cloud data; then, removing noise based on Gaussian filtering and radius filtering; and finally, using the iterative nearest point algorithm to perform coordinate system registration with the building information model to obtain the preprocessed point cloud data.

3. The method for component-level updating and reconstruction of building information model based on point cloud data according to claim 1, characterized in that, The specific steps (2) include: based on the geometric spatial features of the component elements in the original building information model, using principal component analysis to traverse the component elements in the original building information model, obtaining the main direction and center point spatial three-dimensional coordinates of the geometric model direction of each component element, and based on the space where each component element is located, calculating the extension length of the oriented bounding box along the xyz axis in the main direction spatial coordinate system to create an oriented bounding box for the component element.

4. The method for component-level updating and reconstruction of building information model based on point cloud data according to claim 1, characterized in that, The specific steps (3) include: first, sorting all component elements according to the volume of the oriented bounding box; then, traversing the oriented bounding boxes of the component elements, setting a distance threshold, and mapping the point cloud data enclosed by the oriented bounding box or the point cloud data whose distance to the nearest voxel grid of the building information model is less than or equal to the distance threshold to the component element; finally, creating a point cloud-component element matching group list.

5. The method for component-level updating and reconstruction of building information model based on point cloud data according to claim 1, characterized in that, In step (5), the refinement of the component element model of the target component through triangulation and remeshing algorithms specifically includes: downsampling the initial update model of the target component into a sparse model using a uniform mesh; grouping the point cloud data and component element model faces based on the triangular facet normal vectors and connectivity to obtain a face set; performing edge-constrained triangulation reconstruction for each group to obtain a new face set; and finally performing taubin-based smooth mesh reconstruction to obtain a high-resolution update model that fits the target component with the point cloud data.

Images