Method for reversely establishing electromechanical pipeline BIM model based on three-dimensional laser point cloud

Abstract

The invention discloses a method for reversely establishing an electromechanical pipeline BIM model based on a three-dimensional laser point cloud, and the method employs a three-dimensional laser scanning technology to obtain an on-site electromechanical pipeline point cloud, and improves the efficiency and precision of on-site data collection. A point cloud processing algorithm is combined, a point cloud reverse modeling scheme is designed, CV-Filter, DR-RANSAC, MC-RANSAC, SW-DBSCAN and a pipeline axis optimization algorithm based on feature iteration are provided, and electromechanical pipeline reverse modeling is carried out. The electromechanical pipeline reverse modeling method has higher automation level, reduces the workload, improves the accuracy of electromechanical pipeline reverse modeling, and solves the problems of slow speed, large workload, low model accuracy and the like of a traditional manual electromechanical pipeline reverse modeling mode.

Description

Technical Field

[0001] The present invention belongs to the field of pipeline reverse modeling, and specifically relates to a method for reversely establishing a BIM model of an electromechanical pipeline based on a three-dimensional laser point cloud. Background Art

[0002] Building Information Modeling (BIM) is a tool that uses digital technology to express a building's geometric, physical, and functional information. During the construction phase, an as-built BIM model is established based on on-site construction information, allowing for rapid analysis of construction deviations and corrective actions. During the operation and maintenance phase, an operation and maintenance platform is established based on the as-built BIM model, making the building visible, manageable, and controllable. The establishment of as-built BIM models is limited by reverse modeling technology, which refers to the three-dimensional reconstruction of a completed building through manual modification or automated means. Traditional reverse modeling methods involve manually modifying BIM models in BIM modeling software (such as Revit). This is slow, has a low degree of automation, and suffers from low model accuracy. Furthermore, some buildings even lack floor plans, significantly hindering BIM reverse modeling. Three-dimensional laser scanning technology has garnered widespread attention in recent years due to its ability to achieve rapid, high-precision, and highly automated three-dimensional reconstruction of buildings.

[0003] 3D laser scanning technology, a high-precision measurement method, can rapidly capture millimeter-level 3D point cloud data using terrestrial laser scanners. These data points are typically represented by X, Y, and Z coordinates and can also include information about color and surface reflectance. By processing the 3D data points on an object's surface, a highly accurate 3D building model can be reconstructed. This technology offers advantages such as fast measurement speed, high accuracy, high point cloud density, non-contact operation, strong real-time performance, and independence from lighting conditions, effectively eliminating human error and operational risks. Therefore, 3D laser scanning technology holds broad application prospects in architectural surveying, urban planning, cultural heritage preservation, and other fields.

[0004] At present, research on architectural reverse modeling based on 3D laser technology has made certain progress, with improved accuracy and automation compared to manual modeling methods based on on-site photos. However, research has not yet been able to efficiently convert point clouds into semantically rich BIM models. Previous studies required standard BIM models of components as a prerequisite and relied heavily on accurate local features (such as normal vectors and curvature), making them difficult to apply in environments with severe point cloud noise and occlusion. In addition, these studies only established the main pipeline model and did not propose a specific method for pipeline node modeling. Therefore, it is necessary to establish a complete and practical solution for automatically extracting component information and performing parametric modeling. Summary of the Invention

[0005] The purpose of the present invention is to address the deficiencies of the existing technology and propose a method for reverse engineering the electromechanical pipeline BIM model based on three-dimensional laser point cloud.

[0006] The technical solution adopted by the present invention is as follows: a method for reversely establishing a BIM model of electromechanical pipelines based on a three-dimensional laser point cloud, characterized by comprising the following steps:

[0007] Step 1. On-site point cloud acquisition: Use a 3D laser scanner to scan the electromechanical pipelines at the construction site to obtain the original point cloud of the on-site electromechanical pipelines;

[0008] Step 2. Point cloud data preprocessing: The original point cloud is registered based on the registration algorithm, the registered point cloud is sampled to create a station cloud, and the point cloud is denoised and streamlined. The pipeline point cloud is segmented by combining Euclidean clustering and region growing algorithms.

[0009] Step 3. Point cloud orientation identification and slicing: Voxelize the pipeline, fit the voxel points, and use the line with the most inliers and the shortest distance from the inliers to the line as the axis direction of the pipeline point cloud. Rotate the axis direction of the pipeline point cloud to the horizontal direction, and slice the 3D pipeline point cloud into several 2D point clouds along the x-axis.

[0010] Step 4. Point cloud slice fitting: Add constraints to the RANSAC fitting algorithm and design the MC-RANSAC algorithm to fit the sliced point cloud; output the center coordinates and radius of the circular pipe point cloud slice, and output the center point coordinates, width, and height of the rectangular section of the rectangular pipe point cloud slice.

[0011] Step 5. Point cloud axis fitting: Design the SW-DBSCAN algorithm, consider the direction weights between points and the corresponding cross-sectional radius, obtain the weighted distance between points, cluster the center points of the slice cross sections obtained in step 4, and use the least squares method to fit each cluster to obtain the coordinates of the starting and ending points of the pipeline axis, as well as the pipeline radius or width and height;

[0012] Step 6. Fine-tune the point cloud axis: Design a pipeline axis optimization algorithm based on feature iteration, select the endpoints of the pipeline axis as the initial seed points, set the initial direction vector and step size, iteratively update the point cloud features near each seed point until there are no available seed points, compare the pipeline axis obtained in step 5 with the original point cloud, and fine-tune the starting and ending positions of the pipeline axis according to the correction vector obtained by minimizing the eigenvalue and the dynamically adjusted step size.

[0013] Step 7. Connection mode detection: Based on the coordinates and cross-sectional information of the single-segment pipeline point cloud obtained in step 6, the connection mode between each segment of the pipeline point cloud (such as elbow and tee connection) is obtained according to the spatial geometry algorithm.

[0014] Step 8. Parametric modeling: Perform parametric modeling based on the pipeline information obtained in steps 6 and 7, and automatically generate pipeline and pipeline accessory models.

[0015] Furthermore, in step 2, point cloud denoising and simplification are performed based on radius filtering and CV-Filter. The CV-Filter algorithm is:

[0016]

[0017] Among them, P rep represents the representative point within the voxel, P i ,P j ,P k They represent the i-th, j-th and k-th points in a voxel respectively, and n represents the number of points in the voxel.

[0018] Furthermore, the point cloud direction recognition and slicing process in step 3 includes the following specific steps:

[0019] Step 3.1 voxelize the pipeline point cloud by taking the average or weighted average of all points in a fixed cubic area to obtain voxel points, which replace all points in the original cubic area. In this way, all pipeline point clouds are voxelized;

[0020] Step 3.2: Use the RANSAC algorithm to iteratively fit the voxel points to obtain several candidate lines;

[0021] Step 3.3 sets a threshold and classifies voxel points whose vertical straight-line distance is less than the threshold as adjacent points;

[0022] Step 3.4: Take the direction of the line with the largest number of adjacent points as the main direction of the pipeline;

[0023] Step 3.5: Take the direction of the line with the second largest number of adjacent points as the secondary direction of the pipeline;

[0024] Step 3.6: Rotate the pipeline point cloud so that the primary direction is horizontal and the secondary direction is vertical.

[0025] Step 3.7: Slice the point cloud at a certain interval along the x-axis, take the average of the x-coordinates of all points in the same slice, and obtain several two-dimensional point cloud slices.

[0026] Furthermore, the point cloud slice fitting process in step 4 includes the following specific steps:

[0027] Step 4.1 Design the MC-RANSAC algorithm to fit the sliced point cloud. Add axial direction, density, size, dynamic distance threshold, and normal vector constraints to the RANSAC fitting algorithm. For circular tubes, the algorithm can directly fit a circle and obtain the radius and center coordinates. For square tubes, the algorithm can fit the edges of a rectangle and obtain the width, height, and center coordinates of the rectangle.

[0028] Step 4.2 classifies the slices of the same pipe according to radius, width, or height, thereby identifying the reducer and dividing it into multiple pipe sections. Filter conditions are set to delete the slices with inconsistent radius, width, or height. Furthermore, the point cloud axis fitting process in step 5 is as follows:

[0029] Step 5.1 Consider the direction weights between points and the corresponding cross-section radius to obtain the weighted distances between points, and design the SW-DBSCAN algorithm to cluster the center points of the slice sections obtained in step 4;

[0030] Step 5.2: Use the least squares method to fit each cluster to obtain the starting and ending coordinates of the pipeline axis and the radius or width and height;

[0031] In step 5.3, during the point cloud segmentation and slicing classification process, the same pipe may be mistakenly divided into multiple pipes. By comparing the axis coordinate range and the pipe radius or width and height, the partially fitted axes are merged. The filtering conditions are set to delete the pipes whose length, radius, width, and height do not match.

[0032] Furthermore, the point cloud axis fine-tuning process in step 6 includes the following specific steps:

[0033] Step 6.1: Use the starting and ending points of the pipeline axis as the initial seed points, and set the direction vector and step size;

[0034] Step 6.2: Calculate the point cloud feature values near the seed point and determine whether they meet the threshold requirements. If they meet the requirements, find a new seed point based on the current seed point and update the direction vector and step size.

[0035] Step 6.3 Iterate the search for seed points until there are no available seed points, and fine-tune the start and end positions of the pipeline axis according to the correction vector and step size.

[0036] Furthermore, the parametric modeling process in step 8 includes the following specific steps:

[0037] Step 8.1 Export the pipeline geometry information and pipeline connection methods obtained in steps 6 and 7 to an Excel file, and set up a node in Dynamo to read the Excel file;

[0038] Step 8.2 Write the Dynamo script and design the process generation pipeline and attachments.

[0039] The present invention has the following beneficial effects: Traditional manual reverse modeling of electromechanical pipelines suffers from slow speed, high workload, and low model accuracy. This invention utilizes 3D laser scanning technology to acquire on-site electromechanical pipeline point clouds, improving the efficiency and accuracy of on-site data collection. Incorporating point cloud processing algorithms, a point cloud reverse modeling solution is designed, achieving a higher level of automation and reducing workload. Furthermore, the proposed CV-Filter, DR-RANSAC, MC-RANSAC, SW-DBSCAN, and feature-based iteration-based pipeline axis optimization algorithms enhance the accuracy of reverse modeling of electromechanical pipelines. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 This is a flow chart of the electromechanical pipeline reverse modeling method of the present invention;

[0042] Figure 2 This is the flow chart of the electromechanical pipeline point cloud direction recognition and slicing process of the present invention. DETAILED DESCRIPTION

[0043] The present invention will be described in detail below based on the accompanying drawings and preferred embodiments. The purpose and effects of the present invention will become more apparent. The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention.

[0044] like Figure 1 As shown, the present invention provides a method for reversely establishing a BIM model of electromechanical pipelines based on a three-dimensional laser point cloud, comprising the following steps:

[0045] Step 1. On-site point cloud acquisition: Develop a scanning plan based on the actual construction site conditions, determine the layout of each measuring station, and use a 3D laser scanner to scan the electromechanical pipelines to obtain the original point cloud of the on-site electromechanical pipelines;

[0046] Step 2. Point cloud data preprocessing: The original point cloud is registered using the ICP registration algorithm. A station cloud is created using the step-size sampling method. A radius filter is used to denoise the point cloud and remove outliers. An improved voxel filter (Centroid-Voting Voxel Filter, CV-Filter) algorithm is designed to streamline the point cloud. The point cloud is segmented using a combination of Euclidean clustering and region growing algorithms. The CV-Filter algorithm is:

[0047]

[0048] P rep represents the representative point within the voxel, P i ,P j ,P k They represent the i-th, j-th and k-th points in a voxel respectively, and n represents the number of points in the voxel.

[0049] Step 3. Point cloud direction identification and slicing: Figure 2 As shown, the pipeline is voxelized, and the voxels along the length of the pipeline can be approximately connected into a straight line. The DR-RANSAC algorithm (Direction Recognition RANSAC) is designed to fit the voxel points. The line with the most fitting points and the shortest distance from the point to the line is selected. This line is parallel to the axis of the longest pipeline in the point cloud and is recorded as the primary direction. Then, the line with the second largest number of fitting points is selected, which is parallel to the axis of the second longest pipeline in the point cloud and is recorded as the secondary direction. The DR-RANSAC algorithm is:

[0050]

[0051]

[0052]

[0053] I k ={v i ∈V|δ ik <∈}

[0054]

[0055] V represents the voxel center point set, v i Represents the voxel center coordinates, L k represents the k-th fitting straight line, (α k ,β k ,γ k ) represents the direction vector of the line, (x0, y0, z0) represents a point on the line, δ ik Represents point v i To line L k distance, Represents the origin to the straight line L k The direction vector of the previous point p, Represents the straight line L k The direction vector, ∈ represents the distance threshold, I k represents the interior point set, k opt Indicates the selected straight line. Represents the main direction vector.

[0056] According to the main direction and the secondary direction, the point cloud is rotated to the xoy plane, with the main direction along the x-axis, and the corresponding rotation matrix is recorded.

[0057] Step 4. Point cloud slice fitting: Design the MC-RANSAC (Multi-Constrained RANSAC) algorithm to fit the sliced point cloud. Add constraints such as axial direction, density, size, dynamic distance threshold, and normal vector to the original RANSAC fitting algorithm to enhance the fitting effect. Output the center coordinates and radius of the circular pipe point cloud slice, and output the center point coordinates, width, and height of the rectangular cross section of the rectangular pipe point cloud slice. Taking the fitting of a circular pipe as an example, the MC-RANSAC algorithm is:

[0058]

[0059] S opt =min S {E(S)||S|≥m min}

[0060] S represents the circular cross section of the fitting, Represents the point coordinates, represents the coordinates of the center of the circle, r represents the radius of the circle, r e represents the expected radius, represents the normal vector of the circular plane, represents the pipeline axial vector, m represents the number of fitting points, ρ represents the expected point density, d ij Indicates the distance between adjacent points, w1, w2, w3, w4, w5 represent weight coefficients, m min Indicates the minimum point threshold.

[0061] Step 5. Point cloud axis fitting: In actual point clouds, incompleteness is common along the length of the pipe, while perpendicular to the length, multiple pipes may be closely arranged, and the pipe cross-sectional dimensions may change multiple times. Therefore, the SW-DBSCAN (Size-Weighted DBSCAN) algorithm is designed to cluster the center points of the slice cross-sections obtained in step 4. Each cluster is fitted using the least squares method to obtain the coordinates of the starting and ending points of the pipe axis and the pipe radius or width and height. The SW-DBSCAN algorithm is:

[0062]

[0063] λ(θ ij )=w l cos 2 (θ ij )+w t sin 2 (θ ij )

[0064]

[0065] D ij represents the weighted distance between point i and point j, Δx ij ,Δy ij ,Δz ij Represents the coordinate difference, r i ,r j represents the cross-sectional radius corresponding to point i and point j, represents the pipeline axial vector, Represents the vector from point i to point j, θ ij Represents a vector Angle with the axial vector, w l represents the axial weight, w t represents the lateral weight, μ represents the radius difference weight, λ(θ ij ) represents the direction-dependent weight function.

[0066] Step 6. Pipeline axis fine-tuning: Design a pipeline axis optimization algorithm based on feature iteration, select the pipeline axis endpoint as the initial seed point, and calculate the initial direction vector Set the initial step size δ. Calculate the point cloud features near each seed point, specifically:

[0067]

[0068]

[0069]

[0070]

[0071] in, represents the coordinates of the seed point, represents the neighborhood point cloud, r represents the pipeline radius, n k represents the number of points in sector k, A k represents the area of sector k, represents the average point density, r j Indicates the distance from the point to the axis. represents the average radius, l represents the total number of sectors, and n represents the number of points in the neighborhood.

[0072] Determine whether the eigenvalue meets the threshold requirement. If so, find a new seed point based on the current seed point and update the direction vector and step size.

[0073]

[0074] in is the new direction vector, is the initial direction vector, is the direction vector corresponding to each step, is the correction vector obtained by minimizing the eigenvalue, δ i is the step size of dynamic adjustment.

[0075] The above process is iterated until there are no available seed points. The starting and ending positions of the pipeline axis are fine-tuned according to the correction vector and step size to solve the problem of incomplete axis fitting caused by incompleteness and occlusion.

[0076] Step 7. Connection method detection: Based on the coordinates and cross-sectional information of the single-segment pipeline point cloud obtained in step 6, determine the connection method between the pipeline point clouds (such as elbows, tee connections, short pipes that are not parallel to the main pipeline, tees that are not parallel to the main pipeline, short pipes that are parallel to the main pipeline, etc.).

[0077] Step 8. Parametric Modeling: Export the pipeline information obtained in steps 6 and 7 as an Execl file and import it into Dynamo for parametric modeling. Automatically generate pipeline and pipeline accessory models in Revit. The specific operation is as follows: Use the "List.GetItemAtIndex" node to read each data item separately, and use the "Point.Bycoordinates" and "Line.ByStartPointEndPoint" nodes to generate line segments in Dynamo's 3D view. Use the node package "MEPover" to automatically generate pipelines and their accessories. Use the "Pipe.ByLines" node to generate circular pipes; use the "DuctRectangular.ByLines" node to generate square pipes; use the "Tee.By2MEPCurves" node to generate tees; use the "Elbow.ByMEPCurves" or "MEPFitting.ByCurves" nodes to generate elbows.

[0078] The above embodiments are used to illustrate the present invention rather than to limit the present invention. Any modifications and changes made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A method for reverse engineering a BIM model of electromechanical pipelines based on 3D laser point clouds, characterized in that: The steps include: Step 1. On-site point cloud acquisition: Use a 3D laser scanner to scan the electromechanical pipelines at the construction site to obtain the original point cloud of the on-site electromechanical pipelines; Step 2. Point cloud data preprocessing: The original point cloud is registered based on the registration algorithm, the registered point cloud is sampled to create a station cloud, and the point cloud is denoised and streamlined. The pipeline point cloud is segmented by combining Euclidean clustering and region growing algorithms. Step 3. Point cloud orientation identification and slicing: Voxelize the pipeline, fit the voxel points, and use the line with the most inliers and the shortest distance from the inliers to the line as the axis direction of the pipeline point cloud. Rotate the axis direction of the pipeline point cloud to the horizontal direction, and slice the 3D pipeline point cloud into several 2D point clouds along the x-axis. Step 4. Point cloud slice fitting: Add constraints to the RANSAC fitting algorithm and design the MC-RANSAC algorithm to fit the sliced point cloud; output the center coordinates and radius of the circular pipe point cloud slice, and output the center point coordinates, width, and height of the rectangular section of the rectangular pipe point cloud slice; Step 5. Point cloud axis fitting: Design the SW-DBSCAN algorithm, consider the direction weights between points and the corresponding cross-sectional radius, obtain the weighted distance between points, cluster the center points of the slice cross sections obtained in step 4, and use the least squares method to fit each cluster to obtain the coordinates of the starting and ending points of the pipeline axis, as well as the pipeline radius or width and height; Step 6. Fine-tune the point cloud axis: Design a pipeline axis optimization algorithm based on feature iteration, select the endpoints of the pipeline axis as the initial seed points, set the initial direction vector and step size, iteratively update the point cloud features near each seed point until there are no more available seed points, compare the pipeline axis obtained in Step 5 with the original point cloud, and fine-tune the starting and ending positions of the pipeline axis based on the correction vector obtained by minimizing the eigenvalue and the dynamically adjusted step size; Step 7. Connection mode detection: Based on the coordinates and cross-sectional information of the single-segment pipeline point cloud obtained in step 6, the connection mode between the point clouds of each segment of the pipeline is obtained using a spatial geometry algorithm; Step 8. Parametric modeling: Perform parametric modeling based on the pipeline information obtained in steps 6 and 7, and automatically generate pipeline and pipeline accessory models.

2. The method for reverse engineering a BIM model of electromechanical pipelines based on a three-dimensional laser point cloud according to claim 1, characterized in that: In step 2, point cloud denoising and simplification are performed based on radius filtering and CV-Filter. The CV-Filter algorithm is: Among them, P rep represents the representative point within the voxel, P i ,P j ,P k They represent the i-th, j-th and k-th points in a voxel respectively, and n represents the number of points in the voxel.

3. The method for reverse engineering a BIM model of electromechanical pipelines based on a three-dimensional laser point cloud according to claim 1, characterized in that: The point cloud direction recognition and slicing process in step 3 is as follows: Step 3.1 voxelize the pipeline point cloud by taking the average or weighted average of all points in a fixed cubic area to obtain voxel points, which replace all points in the original cubic area. In this way, all pipeline point clouds are voxelized; Step 3.2: Use the RANSAC algorithm to iteratively fit the voxel points to obtain several candidate lines; Step 3.3 sets a threshold and classifies voxel points whose vertical straight-line distance is less than the threshold as adjacent points; Step 3.4: Take the direction of the line with the largest number of adjacent points as the main direction of the pipeline; Step 3.5: Take the direction of the line with the second largest number of adjacent points as the secondary direction of the pipeline; Step 3.6: Rotate the pipeline point cloud so that the primary direction is horizontal and the secondary direction is vertical. Step 3.7: Slice the point cloud at a certain interval along the x-axis, take the average of the x-coordinates of all points in the same slice, and obtain several two-dimensional point cloud slices.

4. The method for reverse engineering a BIM model of electromechanical pipelines based on a three-dimensional laser point cloud according to claim 1, characterized in that: The point cloud slice fitting process in step 4 is as follows: Step 4.1 Design the MC-RANSAC algorithm to fit the sliced point cloud. Add axial direction, density, size, dynamic distance threshold, and normal vector constraints to the RANSAC fitting algorithm. For circular tubes, the algorithm can directly fit a circle and obtain the radius and center coordinates. For square tubes, the algorithm can fit the edges of a rectangle and obtain the width, height, and center coordinates of the rectangle. Step 4.2 classifies several slices of the same pipe according to radius or width and height, thereby identifying the reducer and dividing it into multiple pipe sections. Set filtering conditions to delete slices with inconsistent radius or width and height.

5. The method for reverse engineering a BIM model of electromechanical pipelines based on a three-dimensional laser point cloud according to claim 1, characterized in that: The point cloud axis fitting process in step 5 is as follows: Step 5.1 Consider the direction weights between points and the corresponding cross-section radius to obtain the weighted distances between points, and design the SW-DBSCAN algorithm to cluster the center points of the slice sections obtained in step 4; Step 5.2: Use the least squares method to fit each cluster to obtain the starting and ending coordinates of the pipeline axis and the radius or width and height; In step 5.3, during the point cloud segmentation and slicing classification process, the same pipe may be mistakenly divided into multiple pipes. By comparing the axis coordinate range and the pipe radius or width and height, the partially fitted axes are merged. The filtering conditions are set to delete the pipes whose length, radius, width, and height do not match.

6. The method for reverse engineering a BIM model of electromechanical pipelines based on a three-dimensional laser point cloud according to claim 1, characterized in that: The point cloud axis fine-tuning process in step 6 is as follows: Step 6.1: Use the starting and ending points of the pipeline axis as the initial seed points, and set the direction vector and step size; Step 6.2: Calculate the point cloud feature values near the seed point and determine whether they meet the threshold requirements. If they meet the requirements, find a new seed point based on the current seed point and update the direction vector and step size. Step 6.3 Iterate the search for seed points until there are no available seed points, and fine-tune the start and end positions of the pipeline axis according to the correction vector and step size.

7. The method for reverse engineering a BIM model of electromechanical pipelines based on a three-dimensional laser point cloud according to claim 1, characterized in that: The parametric modeling process in step 8 includes the following specific steps: Step 8.1 Export the pipeline geometry information and pipeline connection methods obtained in steps 6 and 7 to an Excel file, and set up a node in Dynamo to read the Excel file; Step 8.2 Write the Dynamo script and design the process generation pipeline and attachments.

Images