Building structure intelligent reverse modeling and analysis system based on BIM

Abstract

The invention relates to the technical field of building information data analysis, in particular to a building structure intelligent reverse modeling and analysis system based on BIM, and the system achieves the denoising of the point cloud data of a building structure through a data collection module, a downsampling module, a structural similarity classification module, a data denoising module and a model construction module. And completing the construction of the building structure BIM model. The system analyzes the minimum point cloud data selected in the RANSAC fitting process by using the reproduction structure in the building structure based on the reproducibility of the building structure, so as to determine the weight of each point cloud data belonging to a priori interior point in the point cloud set selected by each fitting, and distribute a corresponding error threshold to the weight, thereby screening out noise points, and improving the robustness of the system. The accuracy of a reverse modeling structure can be effectively improved, and the robustness of the model to noise is further improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building information data analysis, in particular to a building structure intelligent reverse modeling and analysis system based on BIM. BACKGROUND

[0002] BIM (Building Information Modeling) is a comprehensive information platform based on digital three-dimensional model, integrating building geometric information, physical information, construction progress, cost, operation and maintenance, etc. whole life cycle data. At present, BIM technology is widely used to assist building design, construction and operation management at home and abroad. RANSAC (Random Sample Consensus) algorithm is an iterative algorithm for estimating mathematical model parameters from a set of observation data, especially suitable for dealing with a large number of noise and outliers.

[0003] The spatial grid structure of building structure is assembled by a large number of steel structures on the construction site, and is applied in building engineering, which has beautiful shape, light weight, stable structure and easy modification. The BIM model of the building is obtained by reverse modeling based on the point cloud data obtained by three-dimensional scanning of the steel structure that has been built, which can carry out structure performance evaluation, modification optimization and whole life cycle management. However, due to the limited space for arranging the scanner on the construction site, and the occlusion between the components of the steel structure when collecting data, the obtained point cloud data may have missing and noise data, which will affect the estimation of eigenvalue and curvature value of the point cloud data, and therefore will seriously affect the performance of the point cloud segmentation algorithm based on eigenvalue and curvature value calculation, and further affect the building modeling result based on BIM technology.

[0004] The RANSAC algorithm can improve the robustness of the model to noise by iteratively estimating the point cloud data. The RANSAC algorithm selects a minimum number of point cloud sets randomly within a certain range, analyzes the trend of the point cloud data, and iteratively selects inliers, and removes interference points, thereby improving the robustness of BIM technology to noise. RANSAC algorithm considers that normal point cloud data has a certain trend, or is a plane, or is a straight line, so the points deviating from the trend of the point cloud set are noise points. However, in this process, since the selection method is random, the selected minimum number of point cloud sets may not only contain inliers, causing the fitted model to be disturbed by noise, and further causing some inliers to be misjudged as noise points, thereby affecting the reverse modeling result. SUMMARY

[0005] In order to solve the above technical problems, the purpose of the present application is to provide a building structure intelligent reverse modeling and analysis system based on BIM.

[0006] The BIM-based building structure intelligent reverse modeling and analysis system provided by the embodiment of the application adopts the technical solutions and specifically comprises a data acquisition module, a down-sampling module, a structure similarity classification module, a data denoising module and a model construction module. The data acquisition module is configured to acquire point cloud data of a building structure. The down-sampling module is configured to analyze the positional relationship of each point cloud data and the plane where the neighborhood point cloud data is located, perform a down-sampling operation on the point cloud data, and obtain structure point cloud data. The structure similarity classification module is configured to perform structure similarity classification on the structure point cloud data, and obtain a plurality of unit modules of different categories. The data denoising module is configured to perform local fitting on the point cloud data by using the RANSAC algorithm, analyze the reproducibility of the point cloud data in the selected minimum point cloud subset in the RANSAC fitting process in the unit module, evaluate the error confidence weight of each point cloud data and adjust the error threshold, and obtain denoised point cloud data. The model construction module is configured to perform three-point building modeling based on the denoised point cloud data, and obtain a BIM model of the building structure.

[0007] In some embodiments of the application, the down-sampling module comprises: A neighborhood plane construction unit is configured to analyze the distribution trend of the plane where the point cloud data is located, and construct a neighborhood plane of the point cloud data, wherein the neighborhood plane contains a plurality of projection points. A down-sampling priority analysis unit is configured to analyze the spatial positional relationship between the projection points within the neighborhood range of any point cloud data based on the neighborhood plane, and obtain the down-sampling priority of each point cloud data. A structure point cloud data acquisition unit is configured to perform a down-sampling operation on the point cloud data according to the down-sampling priority, and obtain structure point cloud data.

[0008] In some embodiments of the application, the neighborhood plane construction unit is configured to: For any point cloud data, the principal component of any point cloud data is obtained by using the PCA dimension reduction algorithm based on all point cloud data within the neighborhood range of the point cloud data. A plane is constructed perpendicular to the principal component and passing through the point cloud data. All point cloud data within the neighborhood range is projected onto the plane to obtain the neighborhood plane of the point cloud data.

[0009] In some embodiments of the application, the down-sampling priority analysis unit is configured to: analyze differences between each projection point and other projection points in the point cloud data neighborhood range corresponding to the point cloud data to the neighborhood plane distance, to obtain the noise degree of each projection point in the point cloud data neighborhood range; In the neighborhood plane, any two projection points are connected into a straight line, and the number of straight lines passing through each projection point is counted to obtain the redundancy degree of each projection point in the point cloud data neighborhood range. In combination with the noise degree and the redundancy degree, the number of times each point cloud data is traversed as a projection point is obtained to obtain the down-sampling priority of each point cloud data.

[0010] In some embodiments of the present application, the structure similarity classification module is configured to: input the structure point cloud data into the trained pointNet model, perform structure similarity classification on the structure point cloud data, and obtain a plurality of category unit modules.

[0011] In some embodiments of the present application, the data denoising module comprises: a RANSAC algorithm model fitting unit for fitting the point cloud data locally by RANSAC algorithm and obtaining a minimum point cloud subset selected in the RANSAC fitting process; a minimum point cloud subset analysis unit for analyzing, in each iteration, the reproducibility of the point cloud data in the unit module to which the point cloud data belongs in the minimum point cloud subset, and evaluating the error confidence weight of each point cloud data in each iteration; an error threshold adjustment unit for adjusting the error threshold of the current iteration according to the error confidence weight and in combination with an initial error threshold to obtain an adjusted error threshold; an error distance adjustment unit for obtaining an adjusted error distance of each point cloud data in the current iteration according to the error confidence weight and in combination with an initial error distance; a point cloud data denoising unit for judging whether the adjusted error distance is less than the adjusted error threshold; if yes, the point cloud data is retained; if no, the point cloud data is removed to obtain denoised point cloud data.

[0012] In some embodiments of the present application, the RANSAC algorithm model fitting unit is configured to: divide the point cloud data into a plurality of voxel blocks, select a minimum point cloud subset in each voxel block, fit a plane model by the minimum subset, and obtain a fitting model.

[0013] In some embodiments of the present application, the minimum point cloud subset analysis unit is configured to: In each iteration, the distance of the point cloud data in the minimum point cloud subset in the unit module to the nearest point and the second nearest point is analyzed, and the error credible weight of each point cloud data in each iteration is evaluated.

[0014] In some embodiments of the present application, the model construction module is configured to: Through multiple iterations, the number of denoised point cloud data obtained in each iteration is counted; The fitting model with the largest number of denoised point cloud data in the fitting model is selected as the effective fitting model of each voxel block; According to the effective fitting model and the adjustment error threshold, the adjustment error distance of the point cloud data in the voxel block is judged to obtain effective point cloud data; According to the effective point cloud data, three-point building modeling is performed to obtain a BIM model of the building structure.

[0015] In some embodiments of the present application, the data acquisition module comprises: A ground laser scanner is used to measure at a specific scanning point to obtain main point cloud data of the building structure; A portable handheld scanner is used to perform local detail supplementary scanning to obtain fine point cloud data of the building structure; A data integration unit is used to project the main point cloud data and the fine point cloud data into the same three-dimensional space to obtain point cloud data of the building structure.

[0016] Compared with the prior art, the BIM-based intelligent reverse modeling and analysis system for building structures provided by the present application has the following beneficial effects: The system of the present application realizes comprehensive acquisition of building structure point cloud data through the data acquisition module, providing a data basis for subsequent analysis and model construction; through the downsampling module, the influence of untrustworthy point cloud data on the building structure is eliminated, the main structure of the point cloud data is obtained, and interference caused by matching of untrustworthy point cloud data on the unit module in the building structure is avoided; through the structure similarity classification module, the structure point cloud data is divided into several categories of unit modules, and through the data denoising module, the point cloud data is locally fitted, and the reproducibility of the point cloud data in the minimum point cloud subset selected in the RANSAC fitting process in the unit module is analyzed, and finally denoised point cloud data is obtained; finally, the model construction module completes the construction of the BIM model of the building structure. Based on the reproducibility of the building structure, the present application analyzes the minimum point cloud data selected in the RANSAC fitting process by using the recurring structure in the building structure, thereby determining the weight of each point cloud data in the point cloud set selected in each fitting belonging to the prior inner point, and assigning a corresponding error threshold to the point cloud data, thereby screening out noise points, improving the accuracy of the reverse modeling structure, and further improving the robustness of the model to noise. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, and the advantages thereof, a brief introduction will be given to the drawings needed in the embodiments or the prior art description. Obviously, the drawings in the following description only show some embodiments of the present application, and for those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 A basic component schematic diagram of a BIM-based building structure intelligent reverse modeling and analysis system provided by an embodiment of the present application; Figure 2 A one-ring neighborhood construction schematic diagram provided by an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined purposes, the specific embodiments, structures, features and effects of the BIM-based building structure intelligent reverse modeling and analysis system according to the present application are described in detail below in combination with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms such as "comprise", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that the circuit structure, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such article or device. Without more limitations, the element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the article or device comprising the element.

[0021] The specific scheme of the BIM-based building structure intelligent reverse modeling and analysis system provided by the present application is specifically described below in combination with the drawings.

[0022] Please refer to Figure 1 which shows the basic components of the BIM-based building structure intelligent reverse modeling and analysis system provided by an embodiment of the present application.

[0023] As Figure 1As shown, the BIM-based building structure intelligent reverse modeling and analysis system provided by one embodiment of the application mainly comprises a data acquisition module 10, a down-sampling module 20, a structure similarity classification module 30, a data denoising module 40 and a model construction module 50, wherein: The data acquisition module 10 is configured to acquire point cloud data of the building structure.

[0024] When performing reverse modeling on the building structure, firstly, the complexity of the building structure and the acquisition accuracy and complexity of the three-dimensional point cloud data need to be determined, so as to acquire the three-dimensional point cloud data of the building structure by using a ground laser scanner 11 and a structure light scanner and other acquisition modes.

[0025] Therefore, in the embodiment of the application, the data acquisition module 10 is configured to acquire point cloud data of the building structure. Further, the data acquisition module 10 comprises the ground laser scanner 11, a portable handheld scanner 12 and a data integration unit 13, and specifically: The ground laser scanner 11 is configured to measure at a specific scanning point to obtain main point cloud data of the building structure. The specific operation method is as follows: for fine scanning indoors and outdoors of the building structure, the ground laser scanner 11 (TLS) is used to measure at a specific scanning point, a laser pulse is emitted and a reflection time is recorded, the distance of the object surface is calculated, dense point cloud is generated, and the main point cloud data of the building structure is obtained. The advantage of this method is high accuracy (±1-3mm) and strong anti-light interference, and the disadvantage is that it cannot comprehensively acquire all local details.

[0026] The portable handheld scanner 12 is configured to perform local detail supplementary scanning to obtain fine point cloud data of the building structure. The specific operation method is as follows: in order to solve the problem that the ground laser scanner 11 (TLS) cannot comprehensively acquire local details, local detail supplementary scanning is needed. Therefore, the portable handheld scanner 12 is used to project a coded light pattern onto the object surface, and a three-dimensional coordinate is calculated by a camera capturing a deformed pattern, which is used to perform supplementary scanning on the point cloud data acquired by the ground laser scanner 11 (TLS).

[0027] The data integration unit 13 is configured to project the main point cloud data acquired by the ground laser scanner 11 (TLS) and the fine point cloud data acquired by the portable handheld scanner 12 into the same three-dimensional space to acquire point cloud data of the building structure. The point cloud data contains a large number of same unit modules, and the building structure is obtained by splicing multiple unit modules.

[0028] The down-sampling module 20 is configured to analyze the positional relationship between each point cloud data and the plane where the neighborhood point cloud data is located, perform down-sampling operation on the point cloud data, and acquire structure point cloud data.

[0029] The application aims to denoise the collected point cloud data by using the RANSAC algorithm, and reduce the iteration number of the RANSAC algorithm in the process, and specifically, the similarity relationship between the point cloud data belonging to one unit module in the building structure is used to achieve the purpose. Before determining the unit module, because there are a large number of fine point cloud data and noise point cloud data in the collected point cloud data, the noise point cloud data will interfere with the matching of the unit module in the building structure when the building structure is matched, so it is necessary to first perform a downsampling operation on the collected point cloud data to obtain the main structure of the point cloud data.

[0030] Based on the above analysis, in the embodiment of the application, by setting the downsampling module 20, the position relationship of each point cloud data with the plane where the neighborhood point cloud data is located is analyzed, the point cloud data is subjected to a downsampling operation, and the structural point cloud data is obtained. Because the collected point cloud data is the surface information of the building structure, it has a strong plane trend, so the relationship of each point cloud data with the plane where the neighborhood point cloud data is located can be analyzed to determine whether the point cloud data is redundant, and the importance of each point cloud data is obtained. Therefore, further, the downsampling module 20 includes a neighborhood plane construction unit 21, a downsampling priority analysis unit 22 and a structural point cloud data acquisition unit 23, specifically: The neighborhood plane construction unit 21 is configured to analyze the distribution trend of the plane where the point cloud data is located, and construct the neighborhood plane of the point cloud data, wherein the neighborhood plane contains a plurality of projection points. The downsampling priority analysis unit 22 is configured to: first, for any one point cloud data, using the PCA dimension reduction algorithm, the principal component of any one point cloud data is obtained, which represents the trend of the plane where the point cloud data is located; then, a plane is constructed perpendicular to the principal component and passing through the point cloud data; finally, all point cloud data in the neighborhood range are projected onto the plane to obtain the neighborhood plane of the point cloud data, and there are a plurality of projection points in the neighborhood plane, as shown in Figure 2 .

[0031] The downsampling priority analysis unit 22 is used to analyze the spatial positional relationship between projection points within the neighborhood of any point cloud data based on the neighborhood plane, and obtain the downsampling priority of each point cloud data. In the neighborhood plane of the point cloud data, for each projection point, the greater the spatial positional difference between it and the projection points in its neighborhood, the greater the probability that the projection point belongs to an unreliable noise point (here, it is only to eliminate the influence of unreliable projection points on the structure, not to determine the noise point); if there is a projection point that can be represented by other projection points, then it can also be indicated that the projection point is a redundant data point. Therefore, the downsampling priority analysis unit 22 is configured to: first, analyze the difference between the distances from each projection point to the neighborhood plane and the corresponding point cloud data of other projection points within the neighborhood of the point cloud data, and obtain the noise level of each projection point within the neighborhood of the point cloud data; then, in the neighborhood plane, connect any two projection points with a straight line, and count the number of straight lines passing through each projection point, and obtain the redundancy level of each projection point within the neighborhood of the point cloud data; finally, combine the noise level and the redundancy level, traverse the number of times each point cloud data is used as a projection point, and obtain the downsampling priority of each point cloud data.

[0032] First, based on the noise level and redundancy of the projection points, the first... Within the neighborhood of the point cloud data, the first Downsampling priority of each projection point The calculation formula is: In the formula, Indicates the first Within the neighborhood of the point cloud data, the first Single downsampling priority for each projection point; Indicates the first Within the neighborhood of the point cloud data, the first The projection point is the first The neighborhood plane of the nth point cloud data has a height of 0, and the nth point cloud data has a height of 0. The magnitude of the projection of the principal component of the point cloud data along the direction of the i-th point cloud data (i.e., the magnitude of the i-th point cloud data projection along the direction of the principal component) (The distance between each projection point and its corresponding point cloud data along the principal component direction). Indicates the first Within the neighborhood of the point cloud data, the first The nearest K-neighbors of the i-th projection point (K=5), with the i-th projection point as the nearest neighbor K-neighbor (K=5). The neighborhood plane of the nth point cloud data has a height of 0, and the nth point cloud data has a height of 0. The mean modulus of the projection of the principal component of each cloud data; Indicates the first A straight line connecting every two projected points in the neighborhood plane of the point cloud data passes through the first... The number of straight lines at each projection point.

[0033] The reaction of the first If the spatial position of a projection point differs more significantly from that of its neighboring points, it indicates that the point cloud data corresponding to that projection point may be a noise point, and it will not be retained when acquiring the spatial structure. The larger the value, the more likely the projected point is to be an intermediate data point that can be represented by other data points. Therefore, the point cloud data corresponding to the projected point is more likely to be a redundant point. Thus, when acquiring spatial structure, the need for downsampling to eliminate redundant points is greater.

[0034] Similarly, by obtaining the downsampling priority of each projection point within the neighborhood of each point cloud data, each point cloud data will have a downsampling priority when used as a projection point.

[0035] Therefore, for the first For point cloud data, its overall downsampling priority The calculation method is as follows: In the formula, Indicates the first Downsampling priority for point cloud data; Indicates the first The number of times each point cloud data is used as a projection point; Indicates the first The first point cloud data The single downsampling priority when the next point is used as the projection point; This represents the linear normalization function.

[0036] The structural point cloud data acquisition unit 23 is used to perform downsampling operations on the point cloud data according to the downsampling priority to acquire structural point cloud data. Specifically, based on the downsampling requirements, point cloud data with higher downsampling priority are preferentially deleted to obtain the structural point cloud data of the building structure. As an example, this invention deletes all point cloud data with a downsampling priority greater than or equal to 0.68 to obtain the structural point cloud data of the building structure.

[0037] The structural similarity classification module 30 is used to classify structural point cloud data based on structural similarity to obtain several categories of unit modules.

[0038] After obtaining the structural point cloud data, the structural point cloud data is the main structure of the building, and in modern building structures, the entire building is usually spliced by point cloud data belonging to the same unit module, that is, the point cloud data in the same unit module is approximate, so the point cloud data of the entire building structure can be classified to select the same unit module.

[0039] Based on the above analysis, in the embodiments of the present application, the structural point cloud data is classified by the structural similarity classification module 30 to obtain several categories of unit modules. Specifically, the structural similarity classification module 30 is configured to input the structural point cloud data into the trained pointNet (Point Cloud Neural Networks) model to classify the structural point cloud data by structural similarity and obtain several categories of unit modules. The loss function used in the classification is the cross-entropy loss function.

[0040] The data denoising module 40 is used to locally fit the point cloud data by the RANSAC algorithm, analyze the reproducibility of the point cloud data in the selected minimum point cloud subset in the RANSAC fitting process in the unit module, evaluate the error confidence weight of each point cloud data and adjust the error threshold to obtain the denoised point cloud data. Further, the data denoising module 40 includes a RANSAC algorithm model fitting unit 41, a minimum point cloud subset analysis unit 42, an error threshold adjustment unit 43, an error distance adjustment unit 44 and a point cloud data denoising unit 45.

[0041] Since the collected point cloud data contains noise data, these noise data interfere with the reverse modeling, so the initial point cloud data can be locally fitted by the RANSAC algorithm, and then the inliers and outliers are selected, where the inliers are the effective data points to be retained, and the outliers are the noise points to be removed.

[0042] Therefore, first, the RANSAC algorithm model fitting unit 41 is set up to locally fit the point cloud data by the RANSAC algorithm and obtain the minimum point cloud subset selected in the RANSAC fitting process. Specifically, the RANSAC algorithm model fitting unit 41 is configured to: first, divide the point cloud data into several voxel blocks; since the point cloud data of the building structure is large, and the RANSAC algorithm is used to eliminate noise points, the number of points fitted by the RANSAC algorithm needs to be reduced, so after the point cloud data is divided into voxel blocks, the minimum point cloud subset is selected in each voxel block (the number of point cloud data in the minimum point cloud subset in each voxel block is 40% of the total amount of point cloud data in the voxel block); finally, the plane model is fitted by the minimum subset to obtain the fitting model.

[0043] After constructing the fitting model, existing techniques filter valid point cloud data by analyzing the error between each point cloud data point and the fitting model. Specifically, if the error exceeds a set threshold, the point cloud data point is considered an outlier (noise point); if the error is less than or equal to the threshold, it is considered an inlier (valid point cloud data). After multiple iterations, the fitting model with the most inliers is selected as the valid fitting model for that voxel block. However, in this process, the traditional method of selecting the minimum subset involves both inliers and outliers in the selected point cloud data. Outliers can affect the shape of the fitting model, causing it to deviate from the normal building structure and ultimately leading to outliers being misclassified as inliers.

[0044] Therefore, in some embodiments of the present invention, a minimum point cloud subset analysis unit 42 is set up to analyze the reproducibility performance of point cloud data in the minimum point cloud subset within its respective unit module in each iteration, and to evaluate the error confidence weight of each point cloud data in each iteration. Specifically, the minimum point cloud subset analysis unit 42 is configured to: analyze the distances of point cloud data in the minimum point cloud subset to the nearest and second nearest points within its respective unit module in each iteration, and to evaluate the error confidence weight of each point cloud data in each iteration. Then, the first iteration performed within a voxel block... The iteration in which the selected minimum subset contains the th iteration. The nth point cloud data, then the nth Error confidence weight of point cloud data The calculation method is as follows: In the formula, This indicates the first time that occurs within a voxel block. The minimum subset selected in the nth iteration contains the nth Error confidence weights for each point cloud data; Indicates the first The number of unit modules in the category to which each point cloud data belongs (a category of unit modules may contain multiple unit modules). Indicates the first The point cloud data in the first Spatial Euclidean distance between each unit module and its nearest point cloud data in space; For the first The point cloud data in the first Spatial Euclidean distance between each unit module and its spatially nearest point cloud data; This represents the linear normalization function.

[0045] If the first If the same / similar point cloud data exists at the same position of each cell module, and the distance between the same / similar point cloud data and the nearest point cloud data is smaller, and the same / similar point cloud data does not exist in the periphery (the distance between the same / similar point cloud data and the second nearest point cloud data is larger), then the point cloud data is more likely to be an inlier. The point cloud data is fixed in the building structure of the same building system, and is more likely to be an inlier, so that the error confidence weight of the point cloud data is larger.

[0046] Since the selected minimum subset contains inliers and outliers, and the number is inconsistent, the error threshold needs to be adaptively adjusted according to the performance of the inliers of the selected point cloud data, so as to avoid misjudgment of the inliers as outliers due to the constant error threshold.

[0047] Therefore, in some embodiments of the present application, an error threshold adjustment unit 43 is arranged to adjust the error threshold of the current iteration according to the error confidence weight and the initial error threshold, to obtain an adjusted error threshold. Specifically, the adjusted error threshold of the n th iteration is calculated as follows: In the formula, E n represents the adjusted error threshold of the n th iteration; E n represents the adjusted error threshold of the n th iteration; E 0 represents the initial error threshold (the error threshold set in the RANSAC algorithm in the prior art); E n represents the number of the selected point cloud data in the minimum subset in the n th iteration in a voxel block; E n represents the number of the selected point cloud data in the minimum subset in the n th iteration in a voxel block; E n represents the error confidence weight of the selected point cloud data in the minimum subset in the n th iteration in a voxel block. Since the error confidence weight of the point cloud data is valued at 0-1, E n is also valued at 0-1, and the result is between 0.5-1.5 by adding 0.5, so that the initial error threshold is adjusted. The greater the average value of the error confidence weight of all point cloud data in the iteration (greater than 0.5), the more inliers in the selected point cloud data, and the adjustment of the error threshold needs to be increased, so that the inliers and outliers can be more effectively screened; on the contrary, the smaller the error confidence weight of the point cloud data in the iteration (less than 0.5), the fewer inliers in the selected point cloud data, and the adjustment of the error threshold needs to be reduced, so that the inliers and outliers can be more effectively screened.

[0048] Since the error confidence weight of the point cloud data is valued at 0-1, E n is also valued at 0-1, and the result is between 0.5-1.5 by adding 0.5, so that the initial error threshold is adjusted.

[0049] The greater the average value of the error confidence weight of all point cloud data in the iteration (greater than 0.5), the more inliers in the selected point cloud data, and the adjustment of the error threshold needs to be increased, so that the inliers and outliers can be more effectively screened; on the contrary, the smaller the error confidence weight of the point cloud data in the iteration (less than 0.5), the fewer inliers in the selected point cloud data, and the adjustment of the error threshold needs to be reduced, so that the inliers and outliers can be more effectively screened.

[0050] ​​​​The error distance adjustment unit 44 is configured to obtain the error distance of each point cloud data in the current iteration according to the error confidence weight and the initial error distance. Specifically, the error distance of each point cloud data in the current iteration is obtained according to the normal distance of each point cloud data to the plane generated in each iteration is the initial error threshold, which is used as the basis for judging whether the point cloud data is an inner point. The normal distance of the point cloud data to the plane is a known technology.

[0051] The adjustment error distance of the first point cloud data selected in the first iteration is calculated as follows: In the formula, d represents the adjustment error distance of the first point cloud data contained in the minimum subset selected in the first iteration within a voxel block; represents the initial error distance of the first point cloud data contained in the minimum subset selected in the first iteration within a voxel block; represents the initial error distance of the first point cloud data contained in the minimum subset selected in the first iteration within a voxel block; represents the error confidence weight of the first

[0052] The point cloud data denoising unit 45 is configured to judge whether the adjustment error distance is less than the adjustment error threshold. If yes, the point cloud data corresponding to the adjustment error distance is recorded as an inner point in the iteration, and the point cloud data is retained. If no, the point cloud data corresponding to the adjustment error distance is recorded as an outer point in the iteration, and the point cloud data is removed to obtain denoised point cloud data.

[0053] The model construction module 50 is configured to perform three-point building modeling based on the denoised point cloud data to obtain a BIM model of the building structure. Specifically, the model construction module 50 is configured to: first, count the number of denoised point cloud data obtained in each iteration through multiple iterations; then, select a fitting model with the largest number of denoised point cloud data in the fitting model as an effective fitting model of each voxel block; then, judge the adjustment error distance of all point cloud data in the voxel block according to the effective fitting model and the adjustment error threshold, remove all outer points, retain all inner points, and obtain effective point cloud data; finally, perform three-point building modeling according to all effective point cloud data (inner points) to obtain a BIM model of the building structure.

[0054] It is to be noted that the sequential order of the above-described embodiments of the present application only for the purpose of description, but not the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0055] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments.

Claims

1. A BIM-based intelligent reverse modeling and analysis system for building structures, characterized in that, The system comprises: a data acquisition module for acquiring point cloud data of a building structure; a down-sampling module for analyzing the positional relationship of each point cloud data with the plane in which the neighborhood point cloud data is located, performing a down-sampling operation on the point cloud data, and obtaining structural point cloud data; a structural similarity classification module for classifying the structural point cloud data according to structural similarity, and obtaining a plurality of categories of unit modules; a data denoising module for locally fitting the point cloud data by using a RANSAC algorithm, analyzing the reproducibility of the point cloud data in the selected minimum point cloud subset in the RANSAC fitting process, evaluating the error confidence weight of each point cloud data and adjusting the error threshold, and obtaining denoised point cloud data; a model construction module for constructing a three-point building model based on the denoised point cloud data, and obtaining a BIM model of the building structure.

2. The BIM-based building structure intelligent reverse modeling and analysis system according to claim 1, wherein, The down-sampling module comprises: a neighborhood plane construction unit for analyzing the distribution trend of the plane in which the point cloud data is located, and constructing a neighborhood plane of the point cloud data, wherein the neighborhood plane contains a plurality of projection points; a down-sampling priority analysis unit for analyzing the spatial positional relationship between the projection points in the neighborhood range of any point cloud data based on the neighborhood plane, and obtaining the down-sampling priority of each point cloud data; a structural point cloud data acquisition unit for performing a down-sampling operation on the point cloud data according to the down-sampling priority, and obtaining structural point cloud data.

3. The BIM-based building structure intelligent reverse modeling and analysis system according to claim 2, wherein, The neighborhood plane construction unit is configured to: for any point cloud data, obtain the principal component of any point cloud data by using a PCA dimension reduction algorithm with all point cloud data in its neighborhood range; construct a plane perpendicular to the principal component and passing through the point cloud data; project all point cloud data in the neighborhood range onto the plane to obtain the neighborhood plane of the point cloud data.

4. The BIM-based building structure intelligent reverse modeling and analysis system according to claim 2, wherein, The down-sampling priority analysis unit is configured to: analyze the difference between the distance of each projection point in the neighborhood range of the point cloud data and the corresponding point cloud data of other projection points to the neighborhood plane, and obtain the noise degree of each projection point in the neighborhood range of the point cloud data; in the neighborhood plane, connect any two projection points into a straight line, count the number of straight lines passing through each projection point, and obtain the redundancy degree of each projection point in the neighborhood range of the point cloud data; combine the noise degree and the redundancy degree, traverse the number of times each point cloud data is taken as a projection point, and obtain the down-sampling priority of each point cloud data.

5. The BIM-based building structure intelligent reverse modeling and analysis system according to claim 1, wherein, The structural similarity classification module is configured to: input the structural point cloud data into a trained pointNet model, classify the structural point cloud data according to structural similarity, and obtain a plurality of categories of unit modules.

6. The BIM-based building structure intelligent reverse modeling and analysis system according to claim 1, wherein, The data denoising module comprises: a RANSAC algorithm model fitting unit for locally fitting the point cloud data by using a RANSAC algorithm, and obtaining a minimum point cloud subset selected in the RANSAC fitting process; a minimum point cloud subset analysis unit configured to analyze, in each iteration, the reproducible performance of the point cloud data in the minimum point cloud subset in the unit module to which the point cloud data belongs, and evaluate the error credible weight of each point cloud data in each iteration; an error threshold adjustment unit configured to adjust the error threshold of the current iteration according to the error credible weight and in combination with an initial error threshold, to obtain an adjusted error threshold; an error distance adjustment unit configured to obtain the adjusted error distance of each point cloud data in the current iteration according to the error credible weight and in combination with an initial error distance; a point cloud data denoising unit configured to determine whether the adjusted error distance is less than the adjusted error threshold; if yes, the point cloud data is retained; if no, the point cloud data is removed to obtain denoised point cloud data.

7. The BIM-based building structure intelligent reverse modeling and analysis system according to claim 6, wherein, The RANSAC algorithm model fitting unit is configured to: divide the point cloud data into a plurality of voxel blocks, and select a minimum point cloud subset in each voxel block, and fit a plane model through the minimum subset to obtain a fitting model.

8. The BIM-based building structure intelligent reverse modeling and analysis system according to claim 7, wherein, The minimum point cloud subset analysis unit is configured to: analyze, in each iteration, the distance of the point cloud data in the minimum point cloud subset from the nearest point and the second nearest point in the unit module to which the point cloud data belongs, and evaluate the error credible weight of each point cloud data in each iteration.

9. The BIM-based building structure intelligent reverse modeling and analysis system according to claim 7, wherein, The model construction module is configured to: statistically count the number of denoised point cloud data obtained in each iteration through multiple iterations; select the fitting model with the largest number of denoised point cloud data in the fitting model as the effective fitting model of each voxel block; determine the adjusted error distance of the point cloud data in the voxel block according to the effective fitting model and the adjusted error threshold, to obtain effective point cloud data; perform three-point building modeling according to the effective point cloud data to obtain a BIM model of the building structure.

10. The BIM-based building structure intelligent reverse modeling and analysis system according to claim 1, wherein, The data acquisition module includes: a ground laser scanner configured to measure main point cloud data of the building structure at a specific scanning point; a portable handheld scanner configured to perform local detail supplementary scanning to obtain fine point cloud data of the building structure; a data integration unit configured to project the main point cloud data and the fine point cloud data to the same three-dimensional space to obtain point cloud data of the building structure.

Images