Method for automatically generating BIM model based on point cloud data

Abstract

The invention discloses a method for automatically generating a BIM model based on point cloud data. The method comprises the following steps: step 1, preprocessing original point cloud data; 2, carrying out the semantic segmentation of the point cloud of the building construction; step 3, segmenting point cloud instances of the building components; 4, performing IFC construction reverse Revit modeling based on the point cloud data and the point cloud instance segmentation result, and converting the point cloud data into a three-dimensional BIM model of the building component; 5, after reverse modeling of the building components is completed, all the building components are recombined and integrated based on the coordinate data, a complete three-dimensional BIM building model is formed, accurate positions and relations of all the building components in the global coordinate system are ensured, and overall optimization of the BIM building model is conducted. The invention relates to the technical field of information, and can solve the problems that in the prior art, modeling based on point cloud data depends on manual stroking and mold turning, the efficiency is low, and errors are large.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for automatically generating a BIM model based on point cloud data. Background Art

[0002] With the acceleration of urbanization and the aging of urban buildings, urban renewal projects are becoming increasingly important. These projects involve not only the repair and renovation of buildings, but also the replanning and utilization of urban space. In this process, accurate and detailed building information models (BIM) are essential for effective planning, design, and construction. Traditional manual measurement and modeling methods are time-consuming, labor-intensive, and prone to errors, and cannot meet the needs of modern urban renewal projects for high precision and efficiency. Therefore, point cloud-based 3D modeling methods have become the mainstream method of reverse modeling in urban renewal projects.

[0003] Point cloud data is three-dimensional spatial data obtained through laser scanning or photogrammetry technology, which can accurately record the geometric shape and spatial position of buildings. Laser scanners (such as LiDAR) and photogrammetry technology (such as drone aerial photography) can quickly and comprehensively capture detailed information about buildings and urban spaces and generate high-density point cloud data. These point cloud data provide rich basic information for building high-precision three-dimensional models.

[0004] In urban renewal projects, the point cloud-based 3D modeling method has significant advantages. First, it can significantly improve the efficiency of data collection and modeling. Traditional measurement methods often require a lot of time for on-site measurement and recording, while laser scanning and photogrammetry technology can obtain large-scale, high-precision point cloud data in a relatively short time, reducing on-site work time and labor costs. Secondly, the high density and high precision of point cloud data can ensure that the generated 3D model has a high level of detail and accuracy, meeting the needs of refined design and construction. In addition, the point cloud-based 3D modeling method can also effectively handle complex architectural forms and urban environments. In urban renewal projects, the forms of buildings are often complex and changeable, and traditional modeling methods are difficult to accurately describe. Point cloud data can fully capture the complex geometric forms of buildings, and through advanced point cloud processing and modeling algorithms, highly detailed and accurate 3D models can be generated. At the same time, point cloud data can also reflect the real state of buildings, including damage and aging, providing a reliable basis for the planning and decision-making of urban renewal projects.

[0005] However, currently, modeling based on point cloud data mainly relies on manual tracing and reshaping, which is inefficient and has large errors. Therefore, it is necessary to provide a method for automatically generating BIM models based on point cloud data, which can solve the problem of low efficiency and large errors in the prior art of modeling based on point cloud data relying on manual tracing and reshaping. Summary of the invention

[0006] The purpose of the present invention is to provide a method for automatically generating a BIM model based on point cloud data, which can solve the problem of low efficiency and large error in the prior art of modeling based on point cloud data relying on manual outlining and mold turning.

[0007] The present invention is achieved in that:

[0008] A method for automatically generating a BIM model based on point cloud data comprises the following steps:

[0009] Step 1: Preprocess the original point cloud data;

[0010] Step 2: semantic segmentation of building point cloud;

[0011] Step 3: Point cloud instance segmentation of building components;

[0012] Step 4: Perform IFC construction reverse Revit modeling based on the point cloud data and point cloud instance segmentation results to convert the point cloud data into a three-dimensional BIM model of the building component;

[0013] Step 5: After the reverse modeling of the building components is completed, all building components are reorganized and integrated based on the coordinate data to form a complete three-dimensional BIM building model, ensuring the accurate position and relationship of all building components in the global coordinate system, and performing overall optimization of the BIM building model.

[0014] The step 1 comprises the following sub-steps:

[0015] Step 1.1: De-noise the original point cloud data;

[0016] Step 1.2: Downsample the denoised point cloud data;

[0017] Step 1.3: Align and register the point cloud data so that the point cloud data from different perspectives are aligned in the same coordinate system;

[0018] Step 1.4: Crop and segment the point cloud data.

[0019] In the step 1.1, the denoising method includes statistical filtering, median filtering, and mean filtering.

[0020] In the step 1.2, the downsampling process includes voxel grid downsampling and random downsampling.

[0021] The step 2 comprises the following sub-steps:

[0022] Step 2.1: Label the building component categories of the pre-processed point cloud data, and use the labeled point cloud data component training data set;

[0023] Step 2.2: Select the PointNet++ network model to train the annotated point cloud data;

[0024] Step 2.3: After sufficient training, the PointNet++ network model automatically classifies the unlabeled point cloud data and classifies each point cloud data into the corresponding building component category.

[0025] In the step 2.1, the building component category annotations of the point cloud data include walls, doors and windows, and columns.

[0026] In the step 2.2, the PointNet++ network model directly processes the original coordinates of the point cloud data, uses symmetric functions to extract global features, and performs classification and segmentation; through the hierarchical feature extraction mechanism, the features are refined layer by layer to improve the expression ability of the PointNet++ network model;

[0027] During the training process, a loss function is used to measure the difference between the predicted results of the PointNet++ network model and the actual annotations, and the model parameters are continuously optimized through the back-propagation algorithm. Data enhancement techniques, including random rotation, translation, and scaling, are used to transform the training data and increase data diversity.

[0028] The step 3 comprises the following sub-steps:

[0029] Step 3.1: Obtain preliminary labels of building component categories through point cloud semantic segmentation in step 2;

[0030] Step 3.2: Use the instance segmentation model to cluster similar points and identify each independent instance;

[0031] Step 3.3: Add a segmentation branch based on target detection through the Mask R-CNN algorithm, and combine the geometric information of the point cloud data to achieve fine segmentation of the instance;

[0032] Step 3.4: Based on the CRF post-processing method, the relationship between points is introduced to optimize the segmentation results, eliminate noise points and erroneous segmentation, and further refine the instance boundaries.

[0033] The step 4 comprises the following sub-steps:

[0034] Step 4.1: Import the point cloud instance segmentation result in step 3 into Revit software;

[0035] Step 4.2: Based on the IFC standard, reverse modeling is performed using Revit software to create three-dimensional BIM models of building components one by one.

[0036] The step 5 comprises the following sub-steps:

[0037] Step 5.1: Convert and align the coordinate data of all independent building components so that all independent building components are correctly spliced ​​in the same coordinate system;

[0038] Step 5.2: Using the ICP global registration algorithm, the rigid transformation matrix between the two point clouds or building component models is iteratively optimized to minimize the registration error and integrate the building component models to form a complete BIM building model;

[0039] Step 5.3: Analyze the geometry and topology of the 3D BIM building model, identify and remove duplicate points, lines and surfaces, and optimize the 3D BIM building model; the optimization of the 3D BIM building model includes deleting redundant data, repairing damaged areas, and smoothing geometric surfaces.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. The 3D modeling technology based on point cloud data of the present invention can quickly and efficiently obtain the precise geometric information of the building and generate a high-precision 3D BIM building model without manual outlining and mold turning. Therefore, compared with the traditional method, the present invention can greatly improve the modeling efficiency, reduce the on-site verification and measurement time, and reduce the labor cost.

[0042] 2. The present invention automatically generates a three-dimensional BIM building model based on point cloud data. The high-density characteristics of point cloud data ensure the details and accuracy of the model, and can truly reflect the current status of the building. The refinement of the three-dimensional BIM building model is greatly improved through alignment, registration, and deep learning convolutional neural networks. The refined modeling capability provides reliable data support for planning, design, and construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of a method for automatically generating a BIM model based on point cloud data of the present invention. DETAILED DESCRIPTION

[0044] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0045] Please see attached Figure 1 , a method for automatically generating a BIM model based on point cloud data, comprising the following steps:

[0046] Step 1: Preprocess the original point cloud data.

[0047] The original point cloud data can be obtained by conventional laser scanning, photogrammetry and other technologies, which will not be described here. After obtaining the original point cloud data, point cloud preprocessing is performed to obtain high-quality and clearly structured point cloud data, thereby improving the quality and availability of point cloud data and providing a reliable data foundation for subsequent building construction.

[0048] The step 1 comprises the following sub-steps:

[0049] Step 1.1: De-noise the original point cloud data.

[0050] The original point cloud data usually contains a large number of noise points, which may come from sensor errors, environmental interference, etc. Preferably, the denoising method includes but is not limited to statistical filtering, median filtering, mean filtering, etc., which can effectively remove isolated points and abnormal points in the original point cloud data.

[0051] Among them, statistical filtering identifies and removes noise points by statistically analyzing the distribution of points in the neighborhood around each point. Median filtering calculates the median of the points in the neighborhood around each point and replaces the value of the current point with the median to eliminate sharp noise. Mean filtering smoothes point cloud data by calculating the average value of points in the neighborhood.

[0052] Step 1.2: Downsample the denoised point cloud data.

[0053] The amount of point cloud data is usually very large. In order to reduce the computational complexity and storage requirements, downsampling is used to reduce the point cloud data beam. Downsampling can be done by using voxel grid downsampling, random downsampling and other methods.

[0054] Among them, in voxel grid downsampling, the point cloud is divided into a series of voxels, and the points in each voxel are replaced by the voxel center point or the average value of the points in the voxel, thereby reducing the point cloud density. In random downsampling, the point cloud density is reduced by randomly selecting a certain proportion of points. Voxel grid-based downsampling is superior in preserving geometric morphology.

[0055] Step 1.3: Align and register the point cloud data.

[0056] Point cloud data contains coordinate data. For point cloud data collected from multiple perspectives, alignment and registration are required to align point cloud data from different perspectives in the same coordinate system. Preferably, the rigid transformation matrix between the two point cloud data can be iteratively optimized based on the ICP (Iterative Closest Point) algorithm to minimize the registration error. In order to speed up the registration process, the global SIFT (Scale-invariant feature transform) and SURF (Speeded Up Robust Features) feature matching methods are combined to perform rough alignment first, and then fine alignment is performed through ICP.

[0057] Step 1.4: Crop and segment the point cloud data.

[0058] Specifically, the required area is extracted according to specific needs, and irrelevant background point cloud data is eliminated. Starting from the seed point based on the region growing algorithm, it is expanded to the neighboring points according to the similarity criterion to extract the region of interest.

[0059] Step 2: Semantic segmentation of building point cloud.

[0060] Point cloud semantic segmentation is to assign each point in the point cloud data to a predefined category to achieve preliminary recognition and classification of building components. Preferably, semantic segmentation of point cloud data can be performed based on deep learning convolutional neural network (CNN) and graph convolutional network (GCN) models.

[0061] The step 2 comprises the following sub-steps:

[0062] Step 2.1: Label the building component categories of the preprocessed point cloud data and use the labeled point cloud data component training data set.

[0063] The building component category annotations of the point cloud data include but are not limited to walls, doors and windows, columns, etc.

[0064] Step 2.2: Select the PointNet++ network model to train the annotated point cloud data, which can effectively process the three-dimensional spatial characteristics of the point cloud.

[0065] The PointNet++ network model directly processes the original coordinates of the point cloud data, uses symmetric functions (such as maximum pooling) to extract global features, and performs classification and segmentation. And through the hierarchical feature extraction mechanism, the features are refined layer by layer to improve the expression ability of the PointNet++ network model. The PointNet++ network model is a commonly used point cloud data training model in this field, and its working process and working principle will not be repeated here:

[0066] In step 2.2, during the training process, a loss function is used to measure the difference between the prediction results of the PointNet++ network model and the actual annotations, and the model parameters are continuously optimized through the back propagation algorithm. In order to enhance the generalization ability of the model, data enhancement techniques are used, including random rotation, translation, scaling, etc., to transform the training data and increase data diversity.

[0067] Step 2.3: After sufficient training, the PointNet++ network model can automatically classify unlabeled point cloud data and classify each point cloud data into the corresponding building component category.

[0068] In order to further improve the accuracy of semantic segmentation, we can combine technologies such as multi-scale feature extraction and context information fusion. Among them, multi-scale feature extraction captures local and global features of point clouds by extracting features at different scales, improves segmentation accuracy, and enhances the ability of the PointNet++ network model to understand complex scenes. The semantic segmentation results provide a basis for subsequent instance segmentation and modeling, ensuring that different types of building components can be correctly identified and processed.

[0069] Step 3: Point cloud instance segmentation of building components.

[0070] Point cloud instance segmentation further distinguishes different instances of the same category based on point cloud semantic segmentation. Multiple columns in the same building are classified as the same category in semantic segmentation, and they can be distinguished as independent individual components in instance segmentation. That is, the task of instance segmentation is to identify and segment each independent building component instance. By learning the spatial distribution and geometric features of point cloud data, different building component instances can be accurately segmented.

[0071] The step 3 comprises the following sub-steps:

[0072] Step 3.1: Obtain preliminary labels of building component categories through point cloud semantic segmentation in step 2.

[0073] Step 3.2: Use the instance segmentation model to cluster similar points and identify each independent instance.

[0074] Step 3.3: Through the Mask R-CNN (Mask Region-based Convolutional Neural Network) algorithm, a segmentation branch is added on the basis of target detection to achieve fine segmentation of instances. At the same time, in order to improve the accuracy of instance segmentation, the geometric information of point cloud data, including normal vectors, curvature, etc., is combined to perform feature enhancement and optimization.

[0075] Among them, the normal vector is the direction of the normal line of the point cloud surface, reflecting the local geometric characteristics of the point cloud. The curvature reflects the degree of curvature of the point cloud surface. By combining these geometric features, the accuracy of instance segmentation can be improved.

[0076] The Mask R-CNN algorithm is a commonly used instance segmentation algorithm in this field. It is a deep learning model for target detection and instance segmentation used for "target detection", "target instance segmentation", "target key point detection", etc. Its working process and working principle will not be described here.

[0077] Step 3.4: Based on the CRF (Conditional Random Field) post-processing method, the relationship between points is introduced to optimize the segmentation results, eliminate noise points and erroneous segmentation, and further refine the instance boundaries.

[0078] The point cloud instance segmentation result in step 3 can not only provide more detailed building component information, but also provide accurate building component boundaries and positions for reverse modeling. Based on the point cloud instance segmentation, further refinement can be performed, such as 3D reconstruction of the point cloud instance to generate a more refined geometric model.

[0079] Step 4: Based on the point cloud data and point cloud instance segmentation results, IFC (Industry Foundation Classes) is used to build reverse Revit modeling to convert the point cloud data into a three-dimensional BIM model of the building components.

[0080] The step 4 comprises the following sub-steps:

[0081] Step 4.1: Import the point cloud instance segmentation results in step 3 into Revit software.

[0082] Step 4.2: Based on the IFC standard, reverse modeling is performed using Revit software to create three-dimensional BIM models of building components one by one.

[0083] The IFC standard is an international standard in the BIM field that defines the geometric and semantic information of building components.

[0084] In the process of reverse modeling, components such as walls, columns, and floor slabs need to be manually drawn and adjusted according to the geometric features of the point cloud data. By using the modeling tools of Revit software, the geometric form of building components can be accurately drawn, and the semantic and instance information in the point cloud data can be converted into the attribute information of the building components. For example: when drawing a wall, the wall is manually drawn according to the wall outline in the point cloud data, and the corresponding material, thickness and other attribute information are given. The generated BIM model of the building component not only has detailed geometric information, but also contains rich attribute information.

[0085] Step 5: After the reverse modeling of the building components is completed, all building components are reorganized and integrated based on the coordinate data to form a complete three-dimensional BIM building model, ensuring the accurate position and relationship of all building components in the global coordinate system, and performing overall optimization of the BIM building model.

[0086] The step 5 comprises the following sub-steps:

[0087] Step 5.1: Convert and align the coordinate data of all independent building components so that all independent building components can be correctly spliced ​​in the same coordinate system.

[0088] Step 5.2: Using the ICP global registration algorithm, the rigid transformation matrix between the two point clouds or building component models is iteratively optimized to minimize the registration error and integrate the building component models to form a complete BIM building model.

[0089] In order to speed up the registration process, the SIFT global feature matching method is combined to perform rough alignment first, and then fine alignment is performed through ICP.

[0090] Step 5.3: Analyze the geometry and topology of the 3D BIM building model, identify and remove duplicate points, lines and surfaces, and optimize the 3D BIM building model to improve the overall quality of the 3D BIM building model.

[0091] The optimization processing of the three-dimensional BIM building model includes deleting redundant data, repairing damaged areas, smoothing geometric surfaces, etc. The required optimization processing method can be selected according to actual needs to achieve the purpose of optimizing the three-dimensional BIM building model.

[0092] As urban renewal projects require more efficient and accurate modeling, point cloud-based 3D modeling methods have become an ideal choice for reverse modeling due to their high efficiency, high precision, and strong adaptability. By making full use of point cloud data to automatically generate and optimize 3D BIM building models, urban renewal projects can achieve a comprehensive and accurate description of existing buildings and urban spaces, thereby providing solid data support for urban planning, design, and construction, and promoting the smooth development of urban renewal work.

[0093] The above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the invention. Therefore, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for automatically generating a BIM model based on point cloud data, characterized by: The following steps are involved: Step 1: Preprocess the original point cloud data; Step 2: semantic segmentation of building point cloud; Step 3: Point cloud instance segmentation of building components; Step 4: Perform IFC construction reverse Revit modeling based on the point cloud data and point cloud instance segmentation results to convert the point cloud data into a three-dimensional BIM model of the building component; Step 5: After the reverse modeling of the building components is completed, all building components are reorganized and integrated based on the coordinate data to form a complete three-dimensional BIM building model, ensuring the accurate position and relationship of all building components in the global coordinate system, and performing overall optimization of the BIM building model.

2. The method for automatically generating a BIM model based on point cloud data according to claim 1, characterized in that: The step 1 comprises the following sub-steps: Step 1.1: De-noise the original point cloud data; Step 1.2: Downsample the denoised point cloud data; Step 1.3: Align and register the point cloud data so that the point cloud data from different perspectives are aligned in the same coordinate system; Step 1.4: Crop and segment the point cloud data.

3. The method for automatically generating a BIM model based on point cloud data according to claim 2, characterized in that: In the step 1.1, the denoising method includes statistical filtering, median filtering, and mean filtering.

4. The method for automatically generating a BIM model based on point cloud data according to claim 2, characterized in that: In the step 1.2, the downsampling process includes voxel grid downsampling and random downsampling.

5. The method for automatically generating a BIM model based on point cloud data according to claim 1, characterized in that: The step 2 comprises the following sub-steps: Step 2.1: Label the building component categories of the pre-processed point cloud data, and use the labeled point cloud data component training data set; Step 2.2: Select the PointNet++ network model to train the annotated point cloud data; Step 2.3: After sufficient training, the PointNet++ network model automatically classifies the unlabeled point cloud data and classifies each point cloud data into the corresponding building component category.

6. The method for automatically generating a BIM model based on point cloud data according to claim 5, characterized in that: In the step 2.1, the building component category annotations of the point cloud data include walls, doors and windows, and columns.

7. The method for automatically generating a BIM model based on point cloud data according to claim 5, characterized in that: In the step 2.2, the PointNet++ network model directly processes the original coordinates of the point cloud data, uses symmetric functions to extract global features, and performs classification and segmentation; through the hierarchical feature extraction mechanism, the features are refined layer by layer to improve the expression ability of the PointNet++ network model; During the training process, a loss function is used to measure the difference between the predicted results of the PointNet++ network model and the actual annotations, and the model parameters are continuously optimized through the back-propagation algorithm. Data enhancement techniques, including random rotation, translation, and scaling, are used to transform the training data and increase data diversity.

8. The method for automatically generating a BIM model based on point cloud data according to claim 1, characterized in that: The step 3 comprises the following sub-steps: Step 3.1: Obtain preliminary labels of building component categories through point cloud semantic segmentation in step 2; Step 3.2: Use the instance segmentation model to cluster similar points and identify each independent instance; Step 3.3: Add a segmentation branch based on target detection through the Mask R-CNN algorithm, and combine the geometric information of the point cloud data to achieve fine segmentation of the instance; Step 3.4: Based on the CRF post-processing method, the relationship between points is introduced to optimize the segmentation results, eliminate noise points and erroneous segmentation, and further refine the instance boundaries.

9. The method for automatically generating a BIM model based on point cloud data according to claim 1, characterized in that: The step 4 comprises the following sub-steps: Step 4.1: Import the point cloud instance segmentation result in step 3 into Revit software; Step 4.2: Based on the IFC standard, reverse modeling is performed using Revit software to create three-dimensional BIM models of building components one by one.

10. The method for automatically generating a BIM model based on point cloud data according to claim 1, characterized in that: The step 5 comprises the following sub-steps: Step 5.1: Convert and align the coordinate data of all independent building components so that all independent building components are correctly spliced ​​in the same coordinate system; Step 5.2: Using the ICP global registration algorithm, the rigid transformation matrix between the two point clouds or building component models is iteratively optimized to minimize the registration error and integrate the building component models to form a complete BIM building model; Step 5.3: Analyze the geometry and topology of the 3D BIM building model, identify and remove duplicate points, lines and surfaces, and optimize the 3D BIM building model; The optimization of 3D BIM building models includes deleting redundant data, repairing damaged areas, and smoothing geometric surfaces.

Images