Method and system for reconstructing manhattan building axes with high precision using indoor and outdoor point clouds

Abstract

This invention provides a method and system for high-precision reconstruction of Manhattan building axes based on indoor and outdoor point clouds, including: acquiring high-precision indoor and outdoor point cloud data of buildings; extracting point cloud data of target floors of buildings; projecting the point cloud data of target floors onto the XOY plane and removing indoor non-wall structure point cloud data to obtain wall structure point cloud data; extracting the inner and outer contours of the wall point cloud based on the wall structure point cloud data; calculating the thickness of each wall segment based on the inner and outer contours of the wall point cloud; performing straight-line detection on the outer boundary of the wall based on the inner and outer contours of the wall point cloud; determining the collinearity or orthogonality of the boundary based on the geometric relationship inference of the wall boundary based on the inner and outer contours of the wall point cloud; optimizing the overall parameters of the outer boundary of the wall based on orthogonal and symmetric constraints; calculating the coordinates of the corner points of the wall axis and the length of each axis segment based on the wall boundary position information, wall thickness, and length geometric dimension information, and establishing a BIM model in conjunction with the floor height information.

Description

Technical Field

[0001] This invention relates to the field of point cloud data processing technology, specifically to a method and system for high-precision reconstruction of Manhattan building axes using indoor and outdoor point clouds, primarily aimed at high-precision BIM modeling of buildings that conform to the Manhattan hypothesis. Background Technology

[0002] Most buildings consist of parallel and orthogonal walls, beams, and columns, with spatial relationships between components conforming to the Manhattan hypothesis. Existing reconstruction techniques largely employ methods that extract point cloud boundaries and then perform linear or planar fitting. However, due to errors in the point cloud and incomplete data collection, these methods suffer from several problems: the geometric parallelism and orthogonality of axes cannot be guaranteed; and the fitted boundaries include the thickness of wall decorations, making comparison with design data difficult and failing to meet the standards for accurate architectural drawings. This technology addresses the BIM modeling problem of buildings conforming to the Manhattan hypothesis by directly restoring the building's axes, ultimately achieving high-precision BIM modeling and application.

[0003] Patent document CN110443875A (application number: 201910679891.0) discloses an automatic axis drawing system based on building point cloud data. The system includes: a data acquisition module for acquiring raw ordered point cloud data; a feature value calculation module for performing normal calculations on the raw ordered point cloud data to obtain normal feature values; a building element feature extraction module for calculating the verticality score of the raw ordered point cloud data and extracting axis grid deviation feature values ​​based on the verticality score; and an axis grid drawing module for drawing an axis grid based on the normal feature values ​​and the axis grid deviation feature values. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for high-precision reconstruction of the Manhattan building axis using indoor and outdoor point clouds.

[0005] A method for high-precision reconstruction of the Manhattan building axis based on indoor and outdoor point clouds according to the present invention includes:

[0006] Step S1: Use 3D laser scanning technology to collect high-precision indoor and outdoor point cloud data of buildings;

[0007] Step S2: Extract point cloud data of the target floor of the building;

[0008] Step S3: Project the point cloud data of the target floor onto the XOY plane and use a filtering algorithm to remove the point cloud data of the non-wall structure indoors to obtain the point cloud data of the wall structure.

[0009] Step S4: Extract the inner and outer contours of the wall point cloud using the Alpha-Shape algorithm based on the wall structure point cloud data;

[0010] Step S5: Calculate the thickness of each wall segment based on the inner and outer contours of the wall point cloud;

[0011] Step S6: Based on the inner and outer contours of the wall point cloud, use an improved Hough transform to detect straight lines on the outer boundary of the wall.

[0012] Step S7: Based on the inner and outer contours of the wall point cloud, perform geometric relationship reasoning on the wall boundary to determine whether the boundaries are collinear or orthogonal;

[0013] Step S8: Based on orthogonal and symmetric constraints, the LS-TLS method is used to optimize the overall parameters of the outer boundary of the wall.

[0014] Step S9: Based on the wall boundary location information, wall thickness and length geometric dimensions, calculate the coordinates of the corner points of the wall axis and the length of each axis segment, and establish a BIM model by combining the floor height information.

[0015] Preferably, step S2 involves: preprocessing the target floor of the building using Cyclone software based on the floor height, and extracting the three-dimensional point cloud data of the target floor of the building.

[0016] Preferably, step S3 involves: projecting the extracted single-layer point cloud data of the building onto the XOY plane and using a pass-through filtering algorithm to remove the point cloud data of non-wall structures in the interior, thereby obtaining the point cloud data of the wall structure, and then using SOR statistical filtering to remove noise.

[0017] Preferably, step S4 employs the following methods:

[0018] Step S4.1: Select any point D1 from the wall structure point cloud data P, and form a point set P' from points less than 2α away from D1, where α represents the given radius of the circle;

[0019] Step S4.2: Take any point D2 from P' and find the center O of the circle passing through points D1 and D2;

[0020] Step S4.3: Calculate the distance h from the center O of the circle to the other points in P' besides D1 and D2;

[0021] Step S4.4: When h≥α, points D1 and D2 are boundary points and stored in point set S; when h<α, another point is selected from P', and steps S4.2 to S4.4 are repeated until all points in P' have been judged.

[0022] Step S4.5: Select another point from P, and repeat steps S4.1 to S4.5 until all points in P are judged, and obtain the wall boundary contour point set S.

[0023] Preferably, step S5 involves taking no less than n points for each segment based on the inner and outer contours of the wall point cloud, calculating the difference between their X or Y coordinates, and taking the average value as the wall thickness parameter for each segment.

[0024] Preferably, step S6 employs the following methods:

[0025] Step S6.1: Establish a two-dimensional array in the parameter (ρ, θ) space, and treat the current two-dimensional array as a ballot box; where ρ represents the polar radius of a point in the parameter space, and θ represents the polar angle;

[0026] Step S6.2: Sequentially search for target points, find the corresponding position of each target point in the parameter space, increment the corresponding position in the corresponding voting box by 1, and record the current target point;

[0027] Step S6.3: Set the ballot box threshold; ballot boxes with parameters greater than the current threshold will have linear parameters.

[0028] Step S6.4: Determine the corresponding line parameters and the target point of the line in the two-dimensional space through the position (ρ′, θ′) in the parameter space; where ρ′ represents the polar radius of a point in the parameter space of the line, and θ′ represents the polar angle.

[0029] Step S6.5: Select a line as the initial line, traverse the remaining lines, and if the angle between the initial line and the remaining lines is less than θ... lim And the straight-line distance is less than d lim If the lines are merged into a single line, the new line is used as the initial line. Steps S6.1 to S6.5 are repeated to complete the line merging and obtain the line parameters and point set.

[0030] Step S6.6: Remove merged lines from the original set of lines, and repeat step S6.5 until all lines are merged;

[0031] Step S6.7: Output the parameters of each merged line.

[0032] Preferably, step S7 involves: calculating the inner product of the direction vectors of adjacent lines based on the detected lines, and setting thresholds Thresholding1 and Thresholding2. When the inner product of the direction vectors of adjacent lines is less than Thresholding1, the two lines are considered collinear; when the inner product of the direction vectors of adjacent lines is greater than Thresholding2, the two lines are considered orthogonal. Thresholding1 is greater than 0 and approaches 0; Thresholding2 is less than 1 and approaches 1.

[0033] Preferably, step S8 employs:

[0034] Step S8.1: The global least squares solution function model for the boundary problem consisting of n mutually perpendicular or parallel lines simplifies to:

[0035] min:e1 T e1

[0036] st:(A1-E1)ξ1+A2ξ2=0

[0037] ξ1 T ξ1=1

[0038] Where A1∈R m×2 It is a random matrix containing error E1; A2∈R m×n It is a fixed matrix; e1 = vec(E1); ξ1 ∈ R 2 and ξ2∈R n It is a parameter vector; T represents the matrix transpose; m and n represent the number of rows and columns of the matrix, respectively;

[0039]

[0040] ξ1=(α c α s ) T ξ2=(ρ1…ρ n ) T

[0041] Where, α c =cosα,α s =sinα, where α is the coordinate azimuth angle of the line, and ρ is the distance from the origin to the nth line;

[0042] Step S8.2: Given the partitioned point cloud dataset, construct the design matrices A1 and A2;

[0043] Step S8.3: Calculation The eigenvector corresponding to the smallest eigenvalue is used to obtain the estimated value of ξ1;

[0044] Step S8.4: Substitute the estimated value of ξ1 into... In this process, we obtain an estimate of ξ2.

[0045] Preferably, step S9 employs the following methods:

[0046] Step S9.1: Calculate the coordinates of the corner points of the wall axis and the length of each segment of the axis based on the wall thickness, wall boundary position information and length geometry, and draw the precise axis of the wall using Revit software;

[0047] Step S9.2: Accurate wall axis lines. Using Revit software, a BIM model is created in conjunction with floor height information.

[0048] A system for high-precision reconstruction of the Manhattan building axis based on indoor and outdoor point clouds, according to the present invention, includes:

[0049] Module M1: Uses 3D laser scanning technology to acquire high-precision indoor and outdoor point cloud data of buildings;

[0050] Module M2: Extracts point cloud data of the target floor of the building;

[0051] Module M3: Projects the point cloud data of the target floor onto the XOY plane and uses a filtering algorithm to remove the point cloud data of the non-wall structure indoors, thus obtaining the point cloud data of the wall structure.

[0052] Module M4: Extracts the inner and outer contours of the wall point cloud using the Alpha-Shape algorithm based on the wall structure point cloud data;

[0053] Module M5: Calculates the thickness of each wall segment based on the inner and outer contours of the wall point cloud;

[0054] Module M6: Based on the inner and outer contours of the wall point cloud, an improved Hough transform is used to detect straight lines on the outer boundary of the wall.

[0055] Module M7: Based on the inner and outer contours of the wall point cloud, perform geometric relationship reasoning of the wall boundary to determine the collinear or orthogonal relationship of the boundary;

[0056] Module M8: Based on orthogonal and symmetric constraints, the LS-TLS method is used to optimize the overall parameters of the outer boundary of the wall.

[0057] Module M9: Based on the wall boundary location information, wall thickness and length geometric dimensions, calculate the coordinates of the corner points of the wall axis and the length of each axis segment, and build a BIM model by combining the floor height information.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. This invention improves the automation level of axis reconstruction. Existing software is based on rapid modeling from design drawings to BIM (CAD2BIM). However, the modeling problem from unstructured point clouds scanned by laser to BIM (Scan2BIM) requires multiple software programs and a lot of manual processing. The reconstruction method of this invention helps to improve the automation level of axis reconstruction.

[0060] 2. This invention is beneficial for improving the geometric accuracy of automated reconstruction. The method of this invention takes into account the parallel and orthogonal characteristics between axes and adopts the idea of ​​overall optimization to improve the geometric accuracy of reconstruction.

[0061] 3. This invention determines the precise position and size information of the axis by using the outer edge contour, which can avoid the influence of the building wall surface decoration and facilitate the comparison of existing buildings and design data;

[0062] 4. The method of the present invention starts from the original point cloud and can be directly applied to commercial software to improve the automation and geometric accuracy of point cloud-based building reconstruction. Attached Figure Description

[0063] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0064] Figure 1 A flowchart of a method for high-precision reconstruction of Manhattan building axes based on indoor and outdoor point clouds.

[0065] Figure 2 To extract the original point cloud data map of a single floor of a building.

[0066] Figure 3 This is a point cloud data image after noise reduction using direct-pass filtering and SOR statistical filtering.

[0067] Figure 4 This is a point cloud distribution map after projecting 3D point cloud data onto a horizontal plane.

[0068] Figure 5 This is the inner contour map of the wall point cloud extracted using the Alpha-Shape algorithm.

[0069] Figure 6 This is the outer contour map of the wall point cloud extracted using the Alpha-Shape algorithm.

[0070] Figure 7 This is a diagram for detecting straight lines at wall boundaries using an improved Hough transform.

[0071] Figure 8 To create a regularized wall boundary map using the LS-TLS method.

[0072] Figure 9 For accurate wall axis diagrams.

[0073] Figure 10 This is for the final, accurate BIM model. Detailed Implementation

[0074] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0075] Example 1

[0076] According to the present invention, a method for high-precision reconstruction of the Manhattan building axis based on indoor and outdoor point clouds is provided, such as... Figure 1 As shown, it includes:

[0077] Step S1: Use 3D laser scanning technology to collect high-precision indoor and outdoor point cloud data of buildings;

[0078] The scanning instrument is a LEICA P40, with a distance scanning accuracy of 1.2mm + 10ppm, a scanning interval of 0.02m, and a data volume of no less than 200,000 data points.

[0079] Step S2: Extract point cloud data of the target floor of the building;

[0080] Specifically, when extracting building point clouds, the building's floor height should be considered. While preserving as much of the building's outline information as possible, Cyclone software should be used to preprocess the target floor. Attention should be paid to maintaining the integrity of the point cloud, and the 3D point cloud data of the target floor should be extracted, such as... Figure 2 As shown.

[0081] Step S3: Project the point cloud data of the target floor onto the XOY plane and use a filtering algorithm to remove the point cloud data of the non-wall structure indoors to obtain the point cloud data of the wall structure.

[0082] Specifically, step S3 involves: performing a three-dimensional laser scan of the building to obtain point clouds, which, in addition to wall point clouds, also include a large number of point clouds such as beams, columns, and furniture;

[0083] At this point, the extracted single-story point cloud data of the building is projected onto the XOY plane and a pass-through filtering algorithm is used to remove the point cloud data of the non-wall structures in the interior, so as to obtain the point cloud data of the wall structure, and SOR statistical filtering is used to remove noise.

[0084] More specifically, 1) Specify a dimension and its value range; 2) Traverse each point in the point cloud, determining whether its value in the specified dimension is within the value range, and deleting points whose values ​​are outside the range; 3) After the traversal, the remaining points constitute the filtered point cloud. Statistical filtering removes outliers as follows: 1) Given a point cloud density threshold, if a point cloud has a density less than this threshold, the point cloud is invalid; 2) Calculate the average distance from each point to its k nearest neighbors, assuming the result follows a Gaussian distribution; 3) Given the mean and variance, remove points with a variance outside 3σ. Point cloud data denoised using direct-pass filtering and SOR statistical filtering is shown below. Figure 3 As shown.

[0085] Step S4: Extract the inner and outer contours of the wall point cloud using the Alpha-Shape algorithm based on the wall structure point cloud data;

[0086] Specifically, step S4 involves using the Alpha-Shape algorithm to extract and segment the boundary of the wall point cloud. Attention should be paid to the value of the radius α. It is recommended to use a trial-and-error algorithm and select a reasonable value of α based on the actual results.

[0087] Step S4.1: Select any point D1 from the wall structure point cloud data P, and form a point set P' from points less than 2α away from D1, where α represents the given radius of the circle;

[0088] Step S4.2: Take any point D2 from P' and find the center O of the circle passing through points D1 and D2;

[0089] Step S4.3: Calculate the distance h from the center O of the circle to the other points in P' besides D1 and D2;

[0090] Step S4.4: When h≥α, points D1 and D2 are boundary points and stored in point set S; when h<α, another point is selected from P', and steps S4.2 to S4.4 are repeated until all points in P' have been judged.

[0091] Step S4.5: Select another point from P, and repeat steps S4.1 to S4.5 until all points in P are judged, and obtain the wall boundary contour point set S. Figure 5 , Figure 6 These are the inner and outer contour images of the wall point cloud extracted using the Alpha-Shape algorithm.

[0092] Step S5: Calculate the thickness of each wall segment based on the inner and outer contours of the wall point cloud;

[0093] Specifically, step S5 involves: selecting no fewer than 500 points for each segment based on the inner and outer contours of the wall point cloud, ensuring the uniformity of the selected points. The difference between the X and Y coordinates is calculated, and the average value is used as the wall thickness parameter for each segment. More specifically, the X and Y coordinates of the points extracted from the inner and outer contours are determined, considering whether they are parallel or perpendicular to the coordinate axes, and the difference between the X and Y coordinates is calculated. Finally, the average value of these 500 sets of coordinate differences is taken as the wall thickness for that segment.

[0094] Step S6: Based on the inner and outer contours of the wall point cloud, use the improved Hough transform to perform straight line detection on the outer boundary of the wall;

[0095] Specifically, step S6 involves: when using the improved Hought transform for line detection, the point cloud data should first undergo statistical filtering to remove outliers, and then Gaussian filtering should be used for smoothing.

[0096] Step S6.1: Establish a two-dimensional array in the parameter (ρ, θ) space, and treat the current two-dimensional array as a ballot box; where ρ represents the polar radius of a point in the parameter space, and θ represents the polar angle;

[0097] Step S6.2: Sequentially search for target points, find the corresponding position of each target point in the parameter space, increment the corresponding position in the corresponding voting box by 1, and record the current target point;

[0098] Step S6.3: Set the ballot box threshold; ballot boxes with parameters greater than the current threshold will have linear parameters.

[0099] Step S6.4: Determine the corresponding line parameters and the target point of the line in the two-dimensional space through the position (ρ′, θ′) in the parameter space; where ρ′ represents the polar radius of a point in the parameter space of the line, and θ′ represents the polar angle.

[0100] Step S6.5: Select a line as the initial line, traverse the remaining lines, and if the angle between the initial line and the remaining lines is less than θ... lim And the straight-line distance is less than d lim If the lines are merged into a single line, the new line is used as the initial line. Steps S6.1 to S6.5 are repeated to complete the line merging and obtain the line parameters and point set.

[0101] Step S6.6: Remove merged lines from the original set of lines, and repeat step S6.5 until all lines are merged;

[0102] Step S6.7: Output the parameters of each merged line. The final result of the wall boundary line detection using the improved Hough transform is as follows: Figure 7 As shown.

[0103] Step S7: Based on the inner and outer contours of the wall point cloud, perform geometric relationship reasoning on the wall boundary to determine whether the boundaries are collinear or orthogonal;

[0104] Specifically, step S7 involves: calculating the inner product of the direction vectors of adjacent lines based on the detected lines, and setting thresholds Thresholding1 and Thresholding2. When the inner product of the direction vectors of adjacent lines is less than Thresholding1, the two lines are considered collinear; when the inner product of the direction vectors of adjacent lines is greater than Thresholding2, the two lines are considered orthogonal. Thresholding1 is greater than 0 and approaches 0; Thresholding2 is less than 1 and approaches 1.

[0105] For the values ​​of Thresholding1 and Thresholding2, a trial-and-error method can be used, with the values ​​set in advance as [0.1, 0.9], and the threshold parameters adjusted according to the results.

[0106] Step S8: Based on orthogonal and symmetric constraints, the LS-TLS method is used to optimize the overall parameters of the outer boundary of the wall; the precise dimensions of the wall are determined through overall parameter optimization, so as to carry out subsequent modeling.

[0107] Specifically, since the point cloud data obtained from 3D laser scanning all come from the same instrument, assuming that the point cloud data follows an independent distribution of equal precision, then... In the formula I 2m Let σ0 be a 2m×2m identity matrix, where σ0 represents the standard deviation of Gaussian noise in the point cloud data, N represents a normal distribution, and e2 is the standard deviation of the noise. ij The Gaussian noise in the point cloud data is represented by the matrix. Considering that the design matrix can be divided into random and fixed matrices, the overall least squares solution function model for the boundary problem composed of n mutually perpendicular or parallel lines can be simplified to:

[0108] min:e1 T e1

[0109] st:(A1-E1)ξ1+A2ξ2=0

[0110] ξ1 T ξ1=1

[0111] Where A1∈R m×2 It is a random matrix containing error E1, A2∈R m×n It is a fixed matrix, e1 = vec(E1). ξ1 ∈ R 2 and ξ2∈R n It is a parameter vector.

[0112]

[0113] ξ1=(α c α s ) T ξ2=(ρ1…ρ n ) T

[0114] Where, α c =cosα,α s =sinα, where α is the coordinate azimuth angle of the line, and ρ is the distance from the origin to the nth line (the line is represented in Hessian normal form);

[0115] The specific steps of the hybrid LS-TLS method are as follows:

[0116] (1) Given a partitioned point cloud dataset, construct design matrices A1 and A2; where the given partitioned point cloud dataset is a processed point cloud dataset used for overall regularization.

[0117] (2) Calculation The eigenvector corresponding to the smallest eigenvalue is the estimated value of ξ1.

[0118] (3) Substitute the estimated value of ξ1 into In this process, we obtain an estimate of ξ2.

[0119] By solving the above mathematical model, the wall boundary can be regularized, and thus accurate dimensional information can be obtained. Figure 8 The wall boundary map after wall regularization is obtained by using a hybrid LS-TLS method based on spatial moments.

[0120] Step S9: Based on the wall boundary location information, wall thickness and length geometric dimensions, calculate the coordinates of the corner points of the wall axis and the length of each axis segment, and establish a BIM model by combining the floor height information.

[0121] Specifically, step S9 employs the following:

[0122] Step S9.1: Calculate the coordinates of the corner points of the wall axis and the length of each segment of the axis based on the wall thickness, wall boundary position information and length geometry, and draw the precise axis of the wall using Revit software;

[0123] Step S9.2: Accurate wall axis lines. Using Revit software, a BIM model is created in conjunction with floor height information.

[0124] Extracting the coordinates of the wall boundary corner points: Boundary(x i ,y i ), (i=1,2,3,4), wall thickness t i(i = 1, 2, 3, 4), calculate the precise coordinates of the corner points of the wall axis. i (x i ,y i ), (i=1,2,3,4), and the length l of each axis segment. i (i = 1, 2, 3, 4), and use Revit software to draw the precise axis lines of the walls. Figure 9 It is a precise wall axis diagram.

[0125] A precise BIM model is built by combining floor height information. The specific method is as follows: accurately obtain and extract the wall thickness t. i (i = 1, 2, 3, 4), and the length l of each axis segment. i After obtaining the precise axis lines of the walls (i = 1, 2, 3, 4), a precise BIM model is created using Revit software, combined with floor height information, as follows: Figure 10 As shown.

[0126] A system for high-precision reconstruction of the Manhattan building axis based on indoor and outdoor point clouds, according to the present invention, includes:

[0127] Module M1: Uses 3D laser scanning technology to acquire high-precision indoor and outdoor point cloud data of buildings;

[0128] The scanning instrument is a LEICA P40, with a distance scanning accuracy of 1.2mm + 10ppm, a scanning interval of 0.02m, and a data volume of no less than 200,000 data points.

[0129] Module M2: Extracts point cloud data of the target floor of the building;

[0130] Specifically, when extracting building point clouds, the building's floor height should be considered. While preserving as much of the building's outline information as possible, Cyclone software should be used to preprocess the target floor. Attention should be paid to maintaining the integrity of the point cloud, and the 3D point cloud data of the target floor should be extracted, such as... Figure 2 As shown.

[0131] Module M3: Projects the point cloud data of the target floor onto the XOY plane and uses a filtering algorithm to remove the point cloud data of the non-wall structure indoors, thus obtaining the point cloud data of the wall structure.

[0132] Specifically, module M3 uses point clouds obtained by performing three-dimensional laser scanning on buildings, including not only wall point clouds but also a large number of point clouds such as beams, columns, and furniture.

[0133] At this point, the extracted single-story point cloud data of the building is projected onto the XOY plane and a pass-through filtering algorithm is used to remove the point cloud data of the non-wall structures in the interior, so as to obtain the point cloud data of the wall structure, and SOR statistical filtering is used to remove noise.

[0134] More specifically, 1) Specify a dimension and its value range; 2) Traverse each point in the point cloud, determining whether its value in the specified dimension is within the value range, and deleting points whose values ​​are outside the range; 3) After the traversal, the remaining points constitute the filtered point cloud. Statistical filtering removes outliers as follows: 1) Given a point cloud density threshold, if a point cloud has a density less than this threshold, the point cloud is invalid; 2) Calculate the average distance from each point to its k nearest neighbors, assuming the result follows a Gaussian distribution; 3) Given the mean and variance, remove points with a variance outside 3σ. Point cloud data denoised using direct-pass filtering and SOR statistical filtering is shown below. Figure 3 As shown.

[0135] Module M4: Extracts the inner and outer contours of the wall point cloud using the Alpha-Shape algorithm based on the wall structure point cloud data;

[0136] Specifically, module M4 employs the Alpha-Shape algorithm to extract and segment the boundary of the wall point cloud. Attention should be paid to the value of the radius α. It is recommended to use a trial-and-error algorithm and select a reasonable value of α based on the actual results.

[0137] Module M4.1: Select any point D1 from the wall structure point cloud data P, and form a point set P' from points less than 2α away from D1, where α represents the given radius of the circle;

[0138] Module M4.2: Take any point D2 from P' and find the center O of the circle passing through points D1 and D2;

[0139] Module M4.3: Calculate the distance h from the center O of the circle to all points in P' except D1 and D2;

[0140] Module M4.4: When h≥α, points D1 and D2 are boundary points and stored in point set S; when h<α, another point is selected from P', and modules M4.2 to M4.4 are repeatedly triggered until all points in P' have been judged and the process ends.

[0141] Module M4.5: Select another point from P, and repeatedly trigger modules M4.1 to M4.5 until all points in P are judged, and obtain the wall boundary contour point set S. Figure 5 , Figure 6 These are the inner and outer contour images of the wall point cloud extracted using the Alpha-Shape algorithm.

[0142] Module M5: Calculates the thickness of each wall segment based on the inner and outer contours of the wall point cloud;

[0143] Specifically, module M5 employs the following method: Based on the inner and outer contours of the wall point cloud, at least 500 points are selected for each segment, ensuring uniformity in point selection. The difference between the X and Y coordinates is calculated, and the average value is used as the wall thickness parameter for each segment. More specifically, the X and Y coordinates of the points extracted from the inner and outer contours are determined, considering whether they are parallel or perpendicular to the coordinate axes, and the difference between the X and Y coordinates is calculated. Finally, the average value of these 500 sets of coordinate differences is taken as the wall thickness for that segment.

[0144] Module M6: Based on the inner and outer contours of the wall point cloud, an improved Hough transform is used to detect straight lines on the outer boundary of the wall.

[0145] Specifically, module M6 employs the following approach: when using the improved Hought transform for line detection, the point cloud data should first undergo statistical filtering to remove outliers, and then Gaussian filtering should be used for smoothing.

[0146] Module M6.1: Establishes a two-dimensional array in the parameter (ρ, θ) space, where the current two-dimensional array is regarded as a ballot box; where ρ represents the polar radius of a point in the parameter space, and θ represents the polar angle;

[0147] Module M6.2: Sequentially search for target points, find the corresponding position of each target point in the parameter space, increment the corresponding position in the corresponding ballot box by 1, and record the current target point;

[0148] Module M6.3: Sets the ballot box threshold; ballot boxes with parameters greater than the current threshold will have linear parameters.

[0149] Module M6.4: Determines the corresponding line parameters and the target point of the line in two-dimensional space by using the position (ρ′, θ′) in the parameter space; where ρ′ represents the polar radius of a point in the parameter space of the line, and θ′ represents the polar angle.

[0150] Module M6.5: Select a line as the initial line, traverse the remaining lines, and if the angle between the initial line and the remaining lines is less than θ... lim And the straight-line distance is less than d lim If the lines are merged into a single line, the new line will be used as the initial line. Modules M6.1 to M6.5 will be repeatedly triggered to complete the line merging, resulting in the line parameters and point set.

[0151] Module M6.6: Remove merged lines from the original set of lines, and repeat step Module M6.5 until all lines are merged;

[0152] Module M6.7: Outputs parameters for each merged line. The final result of wall boundary line detection using the improved Hough transform is as follows: Figure 7 As shown.

[0153] Module M7: Based on the inner and outer contours of the wall point cloud, perform geometric relationship reasoning of the wall boundary to determine the collinear or orthogonal relationship of the boundary;

[0154] Specifically, module M7 employs the following method: calculating the inner product of the direction vectors of adjacent lines based on the detected lines, and setting thresholds Thresholding1 and Thresholding2. When the inner product of the direction vectors of adjacent lines is less than Thresholding1, the two lines are considered collinear; when the inner product of the direction vectors of adjacent lines is greater than Thresholding2, the two lines are considered orthogonal. Thresholding1 is greater than 0 and approaches 0; Thresholding2 is less than 1 and approaches 1.

[0155] For the values ​​of Thresholding1 and Thresholding2, a trial-and-error method can be used, with the values ​​set in advance as [0.1, 0.9], and the threshold parameters adjusted according to the results.

[0156] Module M8: Based on orthogonal and symmetric constraints, the LS-TLS method is used to optimize the overall parameters of the outer boundary of the wall; the precise dimensions of the wall are determined through overall parameter optimization, so as to carry out subsequent modeling.

[0157] Specifically, since the point cloud data obtained from 3D laser scanning all come from the same instrument, assuming that the point cloud data follows an independent distribution of equal precision, then... In the formula I 2m Let σ0 be a 2m×2m identity matrix, where σ0 represents the standard deviation of Gaussian noise in the point cloud data, N represents a normal distribution, and e2 is the standard deviation of the noise. ij The Gaussian noise in the point cloud data is represented by the matrix. Considering that the design matrix can be divided into random and fixed matrices, the overall least squares solution function model for the boundary problem composed of n mutually perpendicular or parallel lines can be simplified to:

[0158] min:e1 T e1

[0159] st:(A1-E1)ξ1+A2ξ2=0

[0160] ξ1 T ξ1=1

[0161] Where A1∈R m×2 It is a random matrix containing error E1, A2∈R m×n It is a fixed matrix, e1 = vec(E1). ξ1 ∈ R 2 and ξ2∈R n It is a parameter vector.

[0162]

[0163] ξ1=(α c α s ) T ξ2=(ρ1…ρ n ) T

[0164] Where, α c =cosα,α s =sinα, where α is the coordinate azimuth angle of the line, and ρ is the distance from the origin to the nth line (the line is represented in Hessian normal form);

[0165] The specific steps of the hybrid LS-TLS method are as follows:

[0166] (1) Given a partitioned point cloud dataset, construct design matrices A1 and A2; where the given partitioned point cloud dataset is a processed point cloud dataset used for overall regularization.

[0167] (2) Calculation The eigenvector corresponding to the smallest eigenvalue is the estimated value of ξ1.

[0168] (3) Substitute the estimated value of ξ1 into In this process, we obtain an estimate of ξ2.

[0169] By solving the above mathematical model, the wall boundary can be regularized, and thus accurate dimensional information can be obtained. Figure 8 The wall boundary map after wall regularization is obtained by using a hybrid LS-TLS method based on spatial moments.

[0170] Module M9: Based on the wall boundary location information, wall thickness and length geometric dimensions, calculate the coordinates of the corner points of the wall axis and the length of each axis segment, and build a BIM model by combining the floor height information.

[0171] Specifically, module M9 adopts:

[0172] Module M9.1: Calculates the coordinates of the corner points of the wall axis and the length of each axis segment based on the wall thickness, wall boundary position information, and length geometry, and draws the precise axis of the wall using Revit software;

[0173] Module M9.2: Precise wall axis lines are established using Revit software, combined with floor height information, to create a BIM model.

[0174] Extracting the coordinates of the wall boundary corner points: Boundary(x i ,y i ), (i=1,2,3,4), wall thickness t i(i = 1, 2, 3, 4), calculate the precise coordinates of the corner points of the wall axis. i (x i ,y i ), (i=1,2,3,4), and the length l of each axis segment. i (i = 1, 2, 3, 4), and use Revit software to draw the precise axis lines of the walls. Figure 9 It is a precise wall axis diagram.

[0175] A precise BIM model is built by combining floor height information. The specific method is as follows: accurately obtain and extract the wall thickness t. i (i = 1, 2, 3, 4), and the length l of each axis segment. i After obtaining the precise axis lines of the walls (i = 1, 2, 3, 4), a precise BIM model is created using Revit software, combined with floor height information, as follows: Figure 10 As shown.

[0176] This invention fully considers the characteristics of Manhattan architecture, employing orthogonal projection to further reduce data complexity while preserving as much original point cloud data information as possible. It uses the Alpha-Shape algorithm to accurately extract the inner and outer contours of the wall point cloud to determine wall thickness; an improved Hough transform is used to detect straight lines on the outer boundary of the wall; and considering orthogonal and symmetry constraints, a hybrid LS-TLS method based on spatial moments is used to optimize the overall parameters of the boundary, thereby obtaining accurate wall axis dimensions. This method allows for the rapid and accurate extraction of building axes and the creation of high-precision BIM models.

[0177] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0178] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for high-precision reconstruction of the Manhattan building axis based on indoor and outdoor point clouds, characterized in that, include: Step S1: Use 3D laser scanning technology to collect high-precision indoor and outdoor point cloud data of buildings; Step S2: Extract point cloud data of the target floor of the building; Step S3: Project the point cloud data of the target floor onto the XOY plane and use a filtering algorithm to remove the point cloud data of the non-wall structure indoors to obtain the point cloud data of the wall structure. Step S4: Extract the inner and outer contours of the wall point cloud using the Alpha-Shape algorithm based on the wall structure point cloud data; Step S5: Calculate the thickness of each wall segment based on the inner and outer contours of the wall point cloud; Step S6: Based on the inner and outer contours of the wall point cloud, use the improved Hough transform to perform straight line detection on the outer boundary of the wall; Step S7: Based on the inner and outer contours of the wall point cloud, perform geometric relationship reasoning on the wall boundary to determine whether the boundaries are collinear or orthogonal; Step S8: Based on orthogonal and symmetric constraints, the LS-TLS method is used to optimize the overall parameters of the outer boundary of the wall. Step S9: Based on the wall boundary location information, wall thickness and length geometric dimensions, calculate the coordinates of the corner points of the wall axis and the length of each axis segment, and establish a BIM model by combining the floor height information; Step S4 employs the following: Step S4.1: Select any point D1 from the wall structure point cloud data P, where the distance from D1 is less than 2. The points form a point set P', where, Indicates a given rounding radius; Step S4.2: Take any point D2 from P' and find the center O of the circle passing through points D1 and D2; Step S4.3: Calculate the distance h from the center O of the circle to all points in P' except D1 and D2; Step S4.4: When Then points D1 and D2 are boundary points and are stored in point set S; when h < If so, select another point from P' and repeat steps S4.2 to S4.4 until all points in P' have been judged and the process ends. Step S4.5: Select another point from P, and repeat steps S4.1 to S4.5 until all points in P are judged, and obtain the wall boundary contour point set S.

2. The method for high-precision reconstruction of the Manhattan building axis based on indoor and outdoor point clouds according to claim 1, characterized in that, Step S2 involves: preprocessing the target floor of the building using Cyclone software based on the floor height, and extracting the three-dimensional point cloud data of the target floor of the building.

3. The method for high-precision reconstruction of the Manhattan building axis based on indoor and outdoor point clouds according to claim 1, characterized in that, Step S3 involves: projecting the extracted single-layer point cloud data of the building onto the XOY plane and using a pass-through filtering algorithm to remove the point cloud data of non-wall structures in the interior, thereby obtaining the point cloud data of the wall structure, and then using SOR statistical filtering to remove noise.

4. The method for high-precision reconstruction of the Manhattan building axis based on indoor and outdoor point clouds according to claim 1, characterized in that, Step S5 involves taking no fewer than n points for each segment based on the inner and outer contours of the wall point cloud, calculating the difference between their X or Y coordinates, and taking the average value as the wall thickness parameter for each segment.

5. The method for high-precision reconstruction of the Manhattan building axis based on indoor and outdoor point clouds according to claim 1, characterized in that, Step S7 involves: calculating the inner product of the direction vectors of adjacent lines based on the detected lines, and setting thresholds Thresholding1 and Thresholding2. When the inner product of the direction vectors of adjacent lines is less than Thresholding1, the two lines are considered collinear; when the inner product of the direction vectors of adjacent lines is greater than Thresholding2, the two lines are considered orthogonal. Thresholding1 is greater than 0 and approaches 0; Thresholding2 is less than 1 and approaches 1.

6. The method for high-precision reconstruction of the Manhattan building axis based on indoor and outdoor point clouds according to claim 1, characterized in that, Step S9 employs the following: Step S9.1: Calculate the coordinates of the corner points of the wall axis and the length of each segment of the axis based on the wall thickness, wall boundary position information and length geometry, and draw the precise axis of the wall using Revit software; Step S9.2: Accurate wall axis lines. Using Revit software, a BIM model is created in conjunction with floor height information.

7. A system for high-precision reconstruction of the Manhattan building axis based on indoor and outdoor point clouds, characterized in that, include: Module M1: Uses 3D laser scanning technology to acquire high-precision indoor and outdoor point cloud data of buildings; Module M2: Extracts point cloud data of the target floor of the building; Module M3: Projects the point cloud data of the target floor onto the XOY plane and uses a filtering algorithm to remove the point cloud data of the non-wall structure indoors, thus obtaining the point cloud data of the wall structure. Module M4: Extracts the inner and outer contours of the wall point cloud using the Alpha-Shape algorithm based on the wall structure point cloud data; Module M5: Calculates the thickness of each wall segment based on the inner and outer contours of the wall point cloud; Module M6: Based on the inner and outer contours of the wall point cloud, an improved Hough transform is used to detect straight lines on the outer boundary of the wall. Module M7: Based on the inner and outer contours of the wall point cloud, perform geometric relationship reasoning of the wall boundary to determine the collinear or orthogonal relationship of the boundary; Module M8: Based on orthogonal and symmetric constraints, the LS-TLS method is used to optimize the overall parameters of the outer boundary of the wall. Module M9: Based on the wall boundary location information, wall thickness and length geometric dimensions, calculate the coordinates of the corner points of the wall axis and the length of each axis segment, and build a BIM model by combining the floor height information; The module M4 includes: Module M4.1: Select any point D1 from the wall structure point cloud data P, where the distance from D1 is less than 2. The points form a point set P', where, Indicates a given rounding radius; Module M4.2: Take any point D2 from P' and find the center O of the circle passing through points D1 and D2; Module M4.3: Calculate the distance h from the center O of the circle to all points in P' except D1 and D2; Module M4.4: When Then points D1 and D2 are boundary points and are stored in point set S; when h < If so, select another point from P' and repeatedly trigger modules M4.2 to M4.4 until all points in P' have been judged and the process ends; Module M4.5: Select another point from P, and repeatedly trigger modules M4.1 to M4.5 until all points in P are judged, and obtain the wall boundary contour point set S.

Images