Three-dimensional reconstruction method and device for power facilities based on multi-source point cloud cooperation

Abstract

This invention discloses a method and equipment for 3D reconstruction of power facilities based on multi-source point cloud collaboration. The method includes: multi-source data acquisition; data preprocessing; registration and fusion of multi-source point clouds; classification of the fused point clouds based on features and target extraction; individual modeling of power tower point clouds; construction of a 3D power line curve model through power line point cloud extraction and adaptive threshold curve fitting; determination of fitting results by comparing the fitted curve with the original point cloud using a comprehensive threshold and accuracy threshold; automatic adjustment of fitting parameters and refitting if the results are not met; fusion of the power tower model and power line curve model under a unified coordinate system if the results are met; and 3D visualization and / or performance analysis. This invention enables centimeter-level 3D reconstruction of power facilities and efficient web-based display, providing technical support for smart grids, digital twin cities, and infrastructure monitoring.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional reconstruction technology, and in particular to a method and equipment for three-dimensional reconstruction of power facilities based on multi-source point cloud collaboration. Background Technology

[0002] With the increasing popularity of concepts such as smart cities, digital twins, and virtual reality, 3D geographic information technology has ushered in unprecedented development opportunities. Against this backdrop, web-based 3D visualization technology has become a research focus. Cesium, as a browser-based open-source 3D globe platform, has been widely used in the field of geospatial information due to its powerful 3D visualization capabilities, flexible data compatibility, and convenient development interfaces. It provides an efficient means for 3D reconstruction of the world, and the massive amounts of 3D model data generated can realistically reflect the details of surface topography and urban architecture.

[0003] However, for complex linear infrastructure such as power towers and power lines, the 3D models generated by traditional oblique photogrammetry suffer from problems such as local missing parts, insufficient accuracy, and texture distortion, making it difficult to meet the needs of refined management. As an important component of the power grid's basic power infrastructure, accurate 3D digital modeling of power towers and power lines is of great significance for power inspection, fault prediction, energy dispatch, and disaster emergency response. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a method and equipment for 3D reconstruction of power facilities based on multi-source point cloud collaboration. This method overcomes the limitations of traditional power facility modeling in terms of structural complexity, accuracy consistency, and information silos, significantly improving the accuracy of spatial information acquisition and model representation in power engineering. Relying on the advantages of point cloud acquisition through the fusion of laser scanning and oblique photography, this invention can achieve centimeter-level geometric reconstruction and panoramic visual display of power facilities (power towers, power lines, etc.).

[0005] To achieve the above-mentioned technical objectives and effects, the present invention is implemented through the following technical solution:

[0006] The method for 3D reconstruction of power facilities based on multi-source point cloud collaboration includes the following steps:

[0007] S1. Oblique photography technology is used to obtain multi-view images of the power facility area, and lidar technology is used to obtain three-dimensional point cloud data of the power facility area.

[0008] S2. Standardize the original image and 3D point cloud data;

[0009] S3. Register and fuse the point cloud generated by oblique photography with the point cloud of lidar, so that the point cloud data are placed in the same coordinate system;

[0010] S4. Classify the fused point cloud according to its features, including power towers, ground, vegetation, and power lines, and extract the target.

[0011] S5. Perform individual modeling of the power tower point cloud;

[0012] S6. Construct a three-dimensional curve model of electric lines by extracting electric line point clouds and fitting adaptive threshold curves: Cluster electric line point clouds using the adaptive region growing method to extract continuous line segments; use a bivariate quadratic polynomial to fit spatial curves and obtain its three-dimensional geometric structure.

[0013] S7. By comparing the comprehensive threshold between the fitted curve and the original point cloud. With precision threshold To determine whether the fitting result meets the standard: if Execute step S8; if Proceed to step S9;

[0014] S8. Automatically adjust the fitting parameters and return to step S6 to refit;

[0015] S9. Integrate the power tower model and the power line curve model under a unified coordinate system;

[0016] S10. Perform 3D visualization and / or performance analysis.

[0017] Preferably, the point cloud registration method in step S3 specifically includes the following steps:

[0018] S301, Assuming the source coordinate system is around Axis rotation Angle, the coordinate transformation after rotation is:

[0019] ;

[0020] in, for Axis rotation matrix, ;

[0021] The coordinates are from the source coordinate system. For this point Coordinates after axis rotation transformation;

[0022] Assuming the source coordinate system is around Axis rotation Angle, the coordinate transformation after rotation is:

[0023] ;

[0024] in, for Axis rotation matrix, ;

[0025] The coordinates are from the source coordinate system. For this point Coordinates after axis rotation transformation;

[0026] Assuming the source coordinate system is around Axis rotation Angle, the coordinate transformation after rotation is:

[0027] ;

[0028] in, for Axis rotation matrix, ;

[0029] The coordinates are from the source coordinate system. For this point Coordinates after axis rotation transformation;

[0030] Then the rotation transformation matrix of the point cloud for:

[0031] ;

[0032] S302, Adjust the origin of the source coordinate system according to... The axes are translated towards the origin of the target coordinate system, with the three translation parameters being: The translation matrix is:

[0033] ;

[0034] in, respectively along Translation in three axes; For this point Coordinates after axis translation;

[0035] S303. The formula for rotating and translating the source coordinate system to the target coordinate system is:

[0036] ;

[0037] in, It is a translation vector. ; For this point Coordinates after axis rotation and translation transformation;

[0038] S304. Select three or more sets of feature points for matching and solve. and ;

[0039] S305. Let the set of points in the source point cloud participating in feature point matching calculation be... The set of points in the target point cloud that participate in the nearest point matching calculation is: Define the nearest point matching error between point clouds. for: , ; The number of points in the source point cloud that participated in the matching calculation. The number of points in the target point cloud that participate in the matching calculation;

[0040] S306. Set the initial rigid body transformation for the source point cloud in the initial iteration. ,in Let be the initial rotation transformation matrix. It is a special orthogonal group. Let be the initial translation vector. The set is the set of real numbers; when prior pose information exists... and Determined by prior information; when no prior pose information exists, let , The identity matrix; based on the initial rigid body transformation Perform coordinate transformation on the source point cloud to obtain the initial source point cloud. ; then in the In this iteration, the source point cloud... Each point in In the target point set Find the nearest neighbor , forming a set of point pairs ;

[0041] S307, in the In the next iteration, based on all matching point pairs Calculate the new rotation transformation matrix Translation vector ;

[0042] S308, Based on the new rigid body transformation Update the source point cloud to obtain the first point cloud. Second iteration point cloud ;

[0043] S309. Repeat steps S307 to S308. When the number of iterations reaches the set value or the transformation error is less than the threshold, the iteration is completed; otherwise, continue to the next iteration.

[0044] Preferably, step 307 specifically includes the following steps:

[0045] First, calculate the... The centroid of the source point cloud in the next iteration and the centroid of the target point cloud : , ;

[0046] Then, the peer-to-peer pairs are decentralized: ;

[0047] in, They represent the first In the next iteration, the decentralized points of the source point cloud and the decentralized points of the target point cloud are used to construct the covariance matrix. ; for The transpose; and for Perform singular value decomposition: Thus, a new rotation matrix is ​​obtained. , where the matrix Let be the orthogonal matrix, diagonal matrix, and orthogonal matrix obtained from singular value decomposition, where This represents the diagonal matrix construction operator, used to arrange a given scalar sequentially along the main diagonal of a matrix, with the remaining elements set to zero; The determinant of the matrix is ​​used to determine whether the rotation result includes mirror reflection; the new translation vector is... ;

[0048] The new rigid body transformation is .

[0049] Preferably, the power line point cloud extraction in step S6 specifically includes the following steps:

[0050] S601. Construct a KD-tree index;

[0051] S602. Randomly select a point from the original point cloud of power lines that has not yet been clustered. As the initial seed point;

[0052] S603, using seed point Centered on the KD-tree neighborhood, select a set number of nearest neighbors. Calculate its spatial distance :

[0053] ;

[0054] S604. Calculate the average distance of all points in the current neighborhood. and standard deviation And set an adaptive distance threshold. :

[0055] ;

[0056] in, These are adaptive fitting parameters, with values ​​ranging from 1 to 2;

[0057] S605, for conditions satisfying the distance from the seed point For the neighboring points, determine whether they have been clustered. If not, add the point to the current power line point cloud set and remove it from the original point cloud.

[0058] S606. Select a nearest neighbor from the newly added points as a new seed point, and repeat steps 603 to 605 until all expandable points in the current class have been processed.

[0059] S607. Select new seed points from the remaining unclustered points and execute steps 602 to 606 above until all points are divided, completing the separation of all power lines one by one.

[0060] Preferably, in step S6, when constructing the three-dimensional curve model of the power line, a bivariate quadratic polynomial is selected. Curve fitting is performed on the mathematical model, where These are the fitting coefficients. The horizontal coordinates of the electric field line point in the horizontal plane; The coordinates of the electric field line point are the longitudinal coordinates in the horizontal plane; and the least squares method is used to estimate them.

[0061] Preferably, in step S7, the comprehensive threshold is... The calculation formula is:

[0062] ;

[0063] ;

[0064] ;

[0065] ;

[0066] in, For the weighting coefficients, satisfying ; The root mean square error threshold is adaptively adjusted based on the quality indicators and noise level of the input point cloud data. The root mean square error between the model's predicted values ​​and the actual values; The number of point cloud data points that do not form continuous power line segments after power line point cloud clustering is counted. The coordinates of the original point cloud data. The coordinates are calculated for the fitted model; The Pearson correlation coefficient between the fitted curve and the original point cloud data; For the original data in The arithmetic mean in the direction; The point cloud data obtained from the fitting model calculation is in The arithmetic mean of directions; The coefficient of determination.

[0067] Preferably, the automatic adjustment step of the fitting parameters in step S8 includes:

[0068] S801, Set initial values ​​for fitting parameters The last iteration number was And the fitting parameters in the previous iteration were ;

[0069] S802, according to the... Subfit Adjusting the fitting parameters by changing : ; It's about adjusting the step size; Indicates the first Error of the next iteration With the The next iteration The amount of change in error ;

[0070] like Then increase the fitting parameters. ;

[0071] like Then reduce the fitting parameters .

[0072] Preferred, The adaptive adjustment steps are as follows:

[0073] A. Parameter Preprocessing: Based on the power line point cloud extraction results in step S6, select a complete point cloud dataset for a single power line. ,in This represents the total number of point clouds for the power line. For the point cloud data points that constitute the power line; from the set Three non-collinear feature points are randomly selected from the data. Ensure coverage of different sections of the power line, among which, , and These are the three-dimensional coordinates of each feature point;

[0074] B. Neighborhood Feature Calculation: For each feature point, based on an adaptive distance threshold... Filter the most recent There are 10 neighboring points, forming three sets of neighboring points:

[0075] ;

[0076] Calculate the distance between each feature point and its neighboring points:

[0077] ;

[0078] ;

[0079] ;

[0080] in, To represent feature points Centered on an adaptive distance threshold The first set of neighborhood points obtained under constraints; To represent feature points Centered on an adaptive distance threshold The second set of neighborhood points obtained under constraints; To represent feature points Centered on an adaptive distance threshold The third set of neighborhood points obtained under constraints; Represent the first (i) in each set of neighboring points. There are 1 neighboring points, among which ,and This represents the number of neighboring points; For feature points Its first Neighboring points The square of the Euclidean distance between them; For feature points Its first neighborhood points The square of the Euclidean distance between them; For feature points Its first Neighboring points The square of the Euclidean distance between them; These are the three-dimensional coordinates of the corresponding neighborhood points;

[0081] C. Point Cloud Quality Feature Calculation: Calculate the distance statistics of the three sets of neighboring point sets to quantify the point cloud quality and noise level. Calculate the average distance of the three sets of neighboring point sets as follows: The distances from the standard deviations are respectively The noise intensities are respectively The average of the three values ​​is taken as the overall noise intensity of the power line point cloud. ;

[0082] D. Adaptive threshold calculation: Based on the overall noise intensity of the point cloud, an adaptive root mean square error threshold is calculated. ;

[0083] in, The root mean square error threshold is set to 0.02m. This is the noise adjustment factor.

[0084] Correspondingly, a computer device includes a processor and a memory, wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the three-dimensional reconstruction method for power facilities based on multi-source point cloud collaboration as described above.

[0085] The beneficial effects of this invention are:

[0086] First, this invention comprehensively utilizes UAV oblique photogrammetry and airborne LiDAR data, and introduces an iterative nearest-point algorithm to achieve spatial registration of multi-source point clouds, obtaining a high-precision fused point cloud in a unified coordinate system; parametric modeling is performed based on the fused point cloud in the Revit environment to construct a single-unit BIM model including the tower body, tower base, and auxiliary components; spatial curve fitting is performed on the power line point cloud extracted from the airborne LiDAR data to restore the true spatial form of the power lines; and three-dimensional visualization loading of power facilities is realized on the Cesium platform, supporting model browsing, spatial measurement, and interactive analysis.

[0087] Secondly, this invention achieves high-precision 3D reconstruction and visualization of power towers and power lines without requiring extensive manual modeling. It has the advantages of high fusion accuracy, strong automation, and excellent visualization performance. It can realize centimeter-level 3D reconstruction of power facilities and efficient display on the Web, providing technical support for smart grids, digital twin cities, and infrastructure monitoring. It has significant application value for improving the intelligent management and operation and maintenance efficiency of power facilities.

[0088] Third, by introducing a rigid body transformation solution method based on singular value decomposition in the point cloud fine registration stage and combining it with determinant constraints to correct the rotation direction, this invention effectively avoids the problems of mirror reflection and non-rigid body transformation during the registration process, and improves the stability and robustness of multi-source point cloud registration. It is especially suitable for point cloud data with occlusion, noise and uneven point density in complex power facility scenarios.

[0089] Fourth, in the process of three-dimensional modeling of power lines, this invention introduces an adaptive distance threshold-driven region growth and error feedback-driven automatic parameter adjustment mechanism, which can dynamically optimize fitting parameters according to the point cloud noise level and data quality, reduce reliance on human experience, improve the convergence speed and reconstruction accuracy of curve fitting, and enhance the versatility and scalability of the method in different survey areas and under different acquisition conditions. Attached Figure Description

[0090] Figure 1 This is a flowchart of the three-dimensional reconstruction method for power facilities based on multi-source point cloud collaboration according to the present invention;

[0091] Figure 2 This is a photogrammetric point cloud model rendering of the power tower in an embodiment of the present invention;

[0092] Figure 3 This is a diagram showing the effect of multi-source point cloud data registration before implementation in an embodiment of the present invention;

[0093] Figure 4 This is a diagram showing the effect of multi-source point cloud data registration in an embodiment of the present invention;

[0094] Figure 5 This is a diagram showing the effect of multi-source point cloud data fusion in an embodiment of the present invention;

[0095] Figure 6 This is a schematic diagram of the tower body modeling process in an embodiment of the present invention;

[0096] Figure 7 This is a schematic diagram of the point cloud of four power lines extracted in an embodiment of the present invention;

[0097] Figure 8 It is a polynomial fitting curve in an embodiment of the present invention. Detailed Implementation

[0098] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand the present invention and implement it. However, the embodiments are not intended to limit the present invention.

[0099] A method for 3D reconstruction of power facilities based on multi-source point cloud collaboration, such as Figure 1 As shown, it includes the following steps:

[0100] S1. Oblique photography technology is used to obtain multi-view images of the power facility area, and lidar technology is used to obtain three-dimensional point cloud data of the power facility area.

[0101] Step S1 is used for multi-source data acquisition. Preferably, high-resolution multi-view images of the power facility area are obtained by using UAV oblique photography, while airborne LiDAR is used to obtain high-density three-dimensional point cloud data. Control points can be set up in a unified geodetic coordinate system such as CGCS2000 and the UAV platform can be positioned by RTK (real-time dynamic differential positioning) / PPK (post-processing differential positioning).

[0102] S2. Standardize the original image and 3D point cloud data.

[0103] Step S2 is used for data preprocessing, standardizing the original imagery and 3D point cloud data to ensure the accuracy and consistency of subsequent modeling. For example, aerial triangulation is performed on the oblique photogrammetric imagery to generate orthophotos and dense point clouds; noise removal, redundant point clipping, and hole filling are performed on the LiDAR data. Finally, the data is unified into a standardized format, and the resulting photogrammetric point cloud model of the power tower is shown below. Figure 2 As shown, multi-source data can provide more comprehensive and detailed geographic information, ensuring the accuracy of power facility modeling.

[0104] S3. Register and fuse the point cloud generated by oblique photography with the LiDAR point cloud to place the point cloud data in the same coordinate system and ensure accurate spatial alignment.

[0105] Since image data and LiDAR point cloud data are data types with different sources and properties, spatial registration cannot be performed. However, by performing photogrammetric processing on image data, dense point clouds can be generated. These point clouds are more similar to LiDAR point clouds in spatial distribution, thus creating conditions for registration of the two types of data. Therefore, this invention uses photogrammetric point clouds as a medium to achieve data fusion with airborne LiDAR point clouds in steps. The first step involves coarse spatial alignment between the photogrammetric point cloud and the airborne LiDAR point cloud, known as coarse registration. This step aims to eliminate significant initial position and pose differences between the two types of point clouds. The second step, based on the coarse registration result, uses the ICP algorithm for fine registration, ultimately achieving high-precision fusion of the two point cloud data.

[0106] Preferably, this invention employs a seven-parameter rigid body transformation (Bursa seven-parameter model) for initial registration to unify the coordinate reference, followed by fine registration using the Iterative Closest Point (ICP) algorithm. Multiple rounds of iterative matching and error calculation are performed until the registration residual is less than a threshold or the maximum number of iterations is reached. This is described in detail below.

[0107] The point cloud registration method in step S3 specifically includes the following steps:

[0108] S301, Assuming the source coordinate system is around Axis rotation Angle, the coordinate transformation after rotation is:

[0109] ;

[0110] in, for Axis rotation matrix, ;

[0111] The coordinates are from the source coordinate system. For this point Coordinates after axis rotation transformation;

[0112] Assuming the source coordinate system is around Axis rotation Angle, the coordinate transformation after rotation is:

[0113] ;

[0114] in, for Axis rotation matrix, ;

[0115] The coordinates are from the source coordinate system. For this point Coordinates after axis rotation transformation;

[0116] Assuming the source coordinate system is around Axis rotation Angle, the coordinate transformation after rotation is:

[0117] ;

[0118] in, for Axis rotation matrix, ;

[0119] The coordinates are from the source coordinate system. For this point Coordinates after axis rotation transformation;

[0120] Then the rotation transformation matrix of the point cloud for:

[0121] .

[0122] S302, Adjust the origin of the source coordinate system according to... The axes are translated towards the origin of the target coordinate system, with the three translation parameters being: The translation matrix is:

[0123] ;

[0124] in, respectively along Translation in three axes; For this point The coordinates after axis translation.

[0125] S303. The formula for rotating and translating the source coordinate system to the target coordinate system is:

[0126] ;

[0127] in, It is a translation vector. ; For this point Coordinates after axis rotation and translation transformation.

[0128] S304. Select three or more sets of feature points for matching and solve. and .

[0129] S305. Let the set of points in the source point cloud participating in feature point matching calculation be... The set of points in the target point cloud that participate in the nearest point matching calculation is: Define the nearest point matching error between point clouds. for: , ; The number of points in the source point cloud that participated in the matching calculation. The number of points in the target point cloud that participate in the matching calculation.

[0130] S306. Set the initial rigid body transformation for the source point cloud in the initial iteration. ,in Let be the initial rotation transformation matrix. It is a special orthogonal group. Let be the initial translation vector. The set is the set of real numbers; when prior pose information exists... and Determined by prior information; when no prior pose information exists, let , The identity matrix; based on the initial rigid body transformation Perform coordinate transformation on the source point cloud to obtain the initial source point cloud. ; then in the In this iteration, the source point cloud... Each point in In the target point set Find the nearest neighbor , forming a set of point pairs .

[0131] S307, in the In the next iteration, based on all matching point pairs Calculate the new rotation transformation matrix Translation vector Specifically:

[0132] First, calculate the... The centroid of the source point cloud in the next iteration and the centroid of the target point cloud : , ;

[0133] Then, the peer-to-peer pairs are decentralized: ;

[0134] in, They represent the first In the next iteration, the decentralized points of the source point cloud and the decentralized points of the target point cloud are used to construct the covariance matrix. ; for The transpose; and for Perform singular value decomposition: Thus, a new rotation matrix is ​​obtained. , where the matrix Let be the orthogonal matrix, diagonal matrix, and orthogonal matrix obtained from singular value decomposition, where This represents the diagonal matrix construction operator, used to arrange a given scalar sequentially along the main diagonal of a matrix, with the remaining elements set to zero; The determinant of a matrix is ​​used to determine whether the rotation result includes mirror reflection; this is achieved by introducing a matrix. The rotation direction is corrected to ensure that the final rotation matrix satisfies Rigid body rotation constraint; the new translation vector is ;

[0135] The new rigid body transformation is .

[0136] S308, Based on the new rigid body transformation Update the source point cloud to obtain the first point cloud. Second iteration point cloud .

[0137] S309. Repeat steps S307 to S308. When the number of iterations reaches the set value or the transformation error is less than the threshold, the iteration is completed; otherwise, continue to the next iteration.

[0138] After registration, the overlapping point clouds are fused to eliminate density unevenness and form a unified, high-quality point cloud dataset. The before-and-after registration results are shown in the image below. Figure 3 and Figure 4 As shown, after registration, the oblique photogrammetry point cloud and the airborne LiDAR point cloud are fused. Based on the advantages of different data sources, low-quality point clouds are removed, and only high-quality data is retained and merged to obtain complete power tower point cloud data, as shown. Figure 5 As shown, the fusion and fine registration of multi-source point clouds reduce the error of point cloud data and improve the overall accuracy.

[0139] S4. Classify the fused point cloud according to its features, including power towers, ground, vegetation, and power lines, and extract the target.

[0140] Preferably, based on the characteristics of point clouds such as elevation, density, geometry, and spatial distribution, clustering or rule extraction methods are used to classify point clouds into different types such as power towers, ground, vegetation, and power lines.

[0141] S5. Perform individual modeling of the power tower point cloud.

[0142] Preferably, the extracted power tower point cloud is imported into BIM (Building Information Modeling) software (such as Revit). Components such as the tower base, tower body, and crossarms are constructed using parametric modeling, and semantic attributes such as material, dimensions, and voltage level are assigned to these components. This results in a single-unit BIM model output with structural and visualization capabilities. The tower body modeling process is as follows: Figure 6 As shown.

[0143] S6. Construct a three-dimensional curve model of electric lines by extracting electric line point clouds and adaptive threshold curve fitting: Cluster the electric line point clouds using the adaptive region growing method to extract continuous line segments; use a bivariate quadratic polynomial to perform spatial curve fitting to obtain its three-dimensional geometric structure.

[0144] The aforementioned point cloud data includes not only point clouds of power lines and towers, but also point clouds of other ground features such as ground and vegetation. Because different types of point clouds are mixed, this invention achieves accurate separation of a single power line point cloud by setting an adaptive distance threshold. Preferably, the power line point cloud extraction specifically includes the following steps:

[0145] S601. Due to the large amount of power line point cloud data, this invention constructs a KD tree index to accelerate neighborhood search.

[0146] S602. Randomly select a point from the original point cloud of power lines that has not yet been clustered. As an initial seed point, it is used to trigger the region growth process;

[0147] S603, using seed point Centered on the KD-tree neighborhood, select a set number of nearest neighbors. Calculate its spatial distance :

[0148] ;

[0149] S604. Calculate the average distance of all points in the current neighborhood. and standard deviation And set an adaptive distance threshold. :

[0150] ;

[0151] in, The adaptive fitting parameters range from 1 to 2 to ensure the adaptive distance threshold. Less than the distance between different power lines;

[0152] S605, for conditions satisfying the distance from the seed point For the neighboring points, determine whether they have been clustered. If not, add the point to the current power line point cloud set and remove it from the original point cloud.

[0153] S606. Select a nearest neighbor from the newly added points as a new seed point, and repeat steps 603 to 605 until all expandable points in the current class have been processed.

[0154] S607. Select new seed points from the remaining unclustered points and execute steps 602 to 606 above until all points are divided, completing the separation of all power lines one by one.

[0155] Through this process, the seed point continuously moves within the point cloud data of the same power line, enabling automatic identification and extraction of the point cloud data of the same power line. The point cloud data of the four power lines between the two power towers extracted by this invention is as follows: Figure 7 As shown, the automated point cloud extraction process reduces manual intervention and improves efficiency by dynamically adjusting the adaptive distance threshold. It accurately separates power line point clouds from other point clouds, improving the accuracy of the extraction results.

[0156] Preferred, such as Figure 8 As shown, when constructing the three-dimensional curve model of the power line, a bivariate quadratic polynomial is selected. Curve fitting is performed on the mathematical model, where These are the fitting coefficients. The horizontal coordinates of the electric field line point in the horizontal plane; The vertical coordinates of the electric field line points in the horizontal plane are given; and the least squares method is used to estimate the polynomial coefficients to minimize the error between the observed data and the fitted curve, thereby improving the fitting accuracy.

[0157] S7. By comparing the comprehensive threshold between the fitted curve and the original point cloud. With precision threshold To determine whether the fitting result meets the standard: if Execute step S8; if Proceed to step S9.

[0158] Preferably, the present invention judges model performance based on three indicators: root mean square error (RMSE). Pearson correlation coefficient ( ) and coefficient of determination ( As an error assessment indicator, a comprehensive threshold is set. The calculation formula is:

[0159] ;

[0160] ;

[0161] ;

[0162] ;

[0163] in, For the weighting coefficients, satisfying ; The root mean square error threshold is adaptively adjusted based on the quality indicators and noise level of the input point cloud data; it is reduced when the point cloud density is high and the noise level is low. To improve the fitting accuracy constraint, and vice versa. ; The root mean square error between the model's predicted values ​​and the actual values; The number of point cloud data points that do not form continuous power line segments after power line point cloud clustering is counted. The coordinates of the original point cloud data. The coordinates are calculated for the fitted model; The Pearson correlation coefficient between the fitted curve and the original point cloud data; For the original data in The arithmetic mean in the direction; The point cloud data obtained from the fitting model calculation is in The arithmetic mean of directions; The coefficient of determination.

[0164] Set standard values ​​for the system under ideal conditions. =1, therefore, It is usually set to a convergence threshold that is slightly lower than the ideal value.

[0165] Preferred, The adaptive adjustment steps are as follows:

[0166] A. Parameter Preprocessing: Based on the power line point cloud extraction results in step S6, select a complete point cloud dataset for a single power line. ,in This represents the total number of point clouds for the power line. For the point cloud data points that constitute the power line; from the set Three non-collinear feature points are randomly selected from the data. Ensure coverage of different sections of the power line, among which, , and These are the three-dimensional coordinates of each feature point;

[0167] B. Neighborhood Feature Calculation: For each feature point, based on an adaptive distance threshold... Filter the most recent neighborhood points ( The value ranges from 10 to 14, with 12 being the preferred value, forming three sets of neighborhood points:

[0168] ;

[0169] Calculate the distance between each feature point and its neighboring points:

[0170] ;

[0171] ;

[0172] ;

[0173] in, To represent feature points Centered on an adaptive distance threshold The first set of neighborhood points obtained under constraints; To represent feature points Centered on an adaptive distance threshold The second set of neighborhood points obtained under constraints; To represent feature points Centered on an adaptive distance threshold The third set of neighborhood points obtained under constraints; Represent the first (i) in each set of neighboring points. There are 1 neighboring points, among which ,and This represents the number of neighboring points, with a value ranging from 10 to 14. For feature points Its first Neighboring points The square of the Euclidean distance between them; For feature points Its first Neighboring points The square of the Euclidean distance between them; For feature points Its first Neighboring points The square of the Euclidean distance between them; These are the three-dimensional coordinates of the corresponding neighborhood points;

[0174] C. Point Cloud Quality Feature Calculation: Calculate the distance statistics of the three sets of neighboring point sets to quantify the point cloud quality and noise level. Calculate the average distance of the three sets of neighboring point sets as follows: The distances from the standard deviations are respectively The noise intensities are respectively The average of the three values ​​is taken as the overall noise intensity of the power line point cloud. ;

[0175] D. Adaptive threshold calculation: Based on the overall noise intensity of the point cloud, an adaptive root mean square error threshold is calculated. ;

[0176] in, The root mean square error threshold is set to 0.02m. The noise adjustment coefficient is preferably selected as follows: .

[0177] This invention selected four sets of experimental data, all from the same region, covering the power lines between two power towers. The original point cloud data were all collected using the CGCS2000 coordinate system, and the terrain was a plain. In these four sets of experimental data, the minimum horizontal sampling interval was 0.277 meters, and the maximum was 0.358 meters. The basic information for the four sets of experimental data is shown in Table 1 below.

[0178] Table 1. Experimental data and basic information of four power lines

[0179]

[0180] Based on the actual scene and power line distribution in the survey area, this invention reconstructs power lines. The accuracy statistics for 3D power line reconstruction based on airborne LiDAR point clouds are as follows: root mean square error ranges from 2.8467 to 3.4666 cm; correlation coefficient ranges from 0.975 to 0.987; coefficient of determination ranges from 0.950 to 0.974; and computation time ranges from 1.1249 to 1.4558 s. Comprehensive analysis shows that this method can stably and accurately reconstruct the power line morphology, maintaining high consistency with the original data, reflecting strong linear correlation, and exhibiting good fitting performance. The performance evaluation of each experimental data is shown in Table 2.

[0181] Table 2 Evaluation results of the power line reconstruction model

[0182]

[0183] S8. Automatically adjust the fitting parameters and return to step S6 to refit.

[0184] Preferably, a dynamic adjustment mechanism based on error feedback is adopted, through automatic adjustment. To optimize the fitting process, the automatic adjustment steps for fitting parameters include:

[0185] S801, Set initial values ​​for fitting parameters The last iteration number was And the fitting parameters in the previous iteration were ;

[0186] S802. After each iteration, according to the... Subfit Adjusting the fitting parameters by changing : ; The step size determines the relative magnitude of each adjustment, thus optimizing the selection. ; Indicates the first Error of the next iteration With the The next iteration The amount of change in error ;

[0187] like Then increase the fitting parameters. This accelerates convergence;

[0188] like Then reduce the fitting parameters This slows down the fitting process.

[0189] S9. Integrate the power tower model and the power line curve model under a unified coordinate system.

[0190] S10. Perform 3D visualization and / or performance analysis, such as importing the fused model into the Cesium platform for Web 3D display.

[0191] Correspondingly, a computer device includes a processor and a memory, the memory storing at least one instruction, which is loaded and executed by the processor to perform the three-dimensional reconstruction method for power facilities based on multi-source point cloud collaboration as described in any of the above.

[0192] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for three-dimensional reconstruction of power facilities based on multi-source point cloud collaboration, characterized in that, Includes the following steps: S1. Oblique photography technology is used to obtain multi-view images of the power facility area, and lidar technology is used to obtain three-dimensional point cloud data of the power facility area. S2. Standardize the original image and 3D point cloud data; S3. Register and fuse the point cloud generated by oblique photography with the point cloud of lidar, so that the point cloud data are placed in the same coordinate system; S4. Classify the fused point cloud according to its features, including power towers, ground, vegetation, and power lines, and extract the target. S5. Perform individual modeling of the power tower point cloud; S6. Construct a three-dimensional curve model of electric lines by extracting electric line point clouds and fitting adaptive threshold curves: Cluster the electric line point clouds using the adaptive region growing method to extract continuous line segments; The three-dimensional geometric structure is obtained by fitting a spatial curve using a bivariate quadratic polynomial. S7. By comparing the comprehensive threshold between the fitted curve and the original point cloud. With precision threshold To determine whether the fitting result meets the standard: if Execute step S8; if Proceed to step S9; S8. Automatically adjust the fitting parameters and return to step S6 to refit; S9. Integrate the power tower model and the power line curve model under a unified coordinate system; S10. Perform 3D visualization and / or performance analysis; Step S6, power line point cloud extraction, specifically includes the following steps: S601. Construct a KD-tree index; S602. Randomly select a point from the original point cloud of power lines that has not yet been clustered. As the initial seed point; S603, using seed point Centered on the KD-tree neighborhood, select a set number of nearest neighbors. Calculate its spatial distance : ； S604. Calculate the average distance of all points in the current neighborhood. and standard deviation And set an adaptive distance threshold. : ； in, These are adaptive fitting parameters, with values ​​ranging from 1 to 2; S605, for conditions satisfying the distance from the seed point For the neighboring points, determine whether they have been clustered. If not, add the point to the current power line point cloud set and remove it from the original point cloud. S606. Select a nearest neighbor from the newly added points as a new seed point, and repeat steps 603 to 605 until all expandable points in the current class have been processed. S607. Select new seed points from the remaining unclustered points and execute steps 602 to 606 above until all points are divided, completing the separation of all power lines one by one.

2. The method for three-dimensional reconstruction of power facilities based on multi-source point cloud collaboration according to claim 1, characterized in that, The point cloud registration method in step S3 specifically includes the following steps: S301, Assuming the source coordinate system is around Axis rotation Angle, the coordinate transformation after rotation is: ； in, for Axis rotation matrix, ; The coordinates are from the source coordinate system. For this point Coordinates after axis rotation transformation; Assuming the source coordinate system is around Axis rotation Angle, the coordinate transformation after rotation is: ； in, for Axis rotation matrix, ; The coordinates are from the source coordinate system. For this point Coordinates after axis rotation transformation; Assuming the source coordinate system is around Axis rotation Angle, the coordinate transformation after rotation is: ； in, for Axis rotation matrix, ; The coordinates are from the source coordinate system. For this point Coordinates after axis rotation transformation; Then the rotation transformation matrix of the point cloud for: ； S302, Adjust the origin of the source coordinate system according to... The axes are translated towards the origin of the target coordinate system, with the three translation parameters being: The translation matrix is: ； in, respectively along Translation in three axes; For this point Coordinates after axis translation; S303. The formula for rotating and translating the source coordinate system to the target coordinate system is: ； in, It is a translation vector. ; For this point Coordinates after axis rotation and translation transformation; S304. Select three or more sets of feature points for matching and solve. and ; S305. Let the set of points in the source point cloud participating in feature point matching calculation be... The set of points in the target point cloud that participate in the nearest point matching calculation is: Define the nearest point matching error between point clouds. for: , ; The number of points in the source point cloud that participated in the matching calculation. The number of points in the target point cloud that participate in the matching calculation; S306. Set the initial rigid body transformation for the source point cloud in the initial iteration. ,in Let be the initial rotation transformation matrix. It is a special orthogonal group. Let be the initial translation vector. The set is the set of real numbers; when prior pose information exists... and Determined by prior information; when no prior pose information exists, let , The identity matrix; based on the initial rigid body transformation Perform coordinate transformation on the source point cloud to obtain the initial source point cloud. ; then in the In this iteration, the source point cloud... Each point in In the target point set Find the nearest neighbor , forming a set of point pairs ; S307, in the In the next iteration, based on all matching point pairs Calculate the new rotation transformation matrix Translation vector ; S308, Based on the new rigid body transformation Update the source point cloud to obtain the first point cloud. Second iteration point cloud ; S309. Repeat steps S307 to S308. When the number of iterations reaches the set value or the transformation error is less than the threshold, the iteration is completed; otherwise, continue to the next iteration.

3. The method for three-dimensional reconstruction of power facilities based on multi-source point cloud collaboration according to claim 2, characterized in that, Step 307 specifically includes the following steps: First, calculate the... The centroid of the source point cloud in the next iteration and the centroid of the target point cloud : , ; Then, the peer-to-peer pairs are decentralized: ; in, They represent the first In the next iteration, the decentralized points of the source point cloud and the decentralized points of the target point cloud are used to construct the covariance matrix. ; for The transpose; and for Perform singular value decomposition: Thus, a new rotation matrix is ​​obtained. , where the matrix Let be the orthogonal matrix, diagonal matrix, and orthogonal matrix obtained from singular value decomposition, where This represents the diagonal matrix construction operator, used to arrange a given scalar sequentially along the main diagonal of a matrix, with the remaining elements set to zero; The determinant of the matrix is ​​used to determine whether the rotation result includes mirror reflection; the new translation vector is... ; The new rigid body transformation is .

4. The method for three-dimensional reconstruction of power facilities based on multi-source point cloud collaboration according to claim 1, characterized in that, In step S6, when constructing the three-dimensional curve model of the power line, a bivariate quadratic polynomial is selected. Curve fitting is performed on the mathematical model, where These are the fitting coefficients. The horizontal coordinates of the electric field line point in the horizontal plane; The coordinates of the electric field line point are the longitudinal coordinates in the horizontal plane; and the least squares method is used to estimate them.

5. The method for three-dimensional reconstruction of power facilities based on multi-source point cloud collaboration according to claim 4, characterized in that, In step S7, the overall threshold is calculated. The calculation formula is: ； ； ； ； in, For the weighting coefficients, satisfying ; The root mean square error threshold is adaptively adjusted based on the quality indicators and noise level of the input point cloud data. The root mean square error between the model's predicted values ​​and the actual values; The number of point cloud data points that do not form continuous power line segments after power line point cloud clustering is counted. The coordinates of the original point cloud data. The coordinates are calculated for the fitted model; The Pearson correlation coefficient between the fitted curve and the original point cloud data; For the original data in The arithmetic mean in the direction; The point cloud data obtained from the fitting model calculation is in The arithmetic mean of directions; The coefficient of determination.

6. The method for three-dimensional reconstruction of power facilities based on multi-source point cloud collaboration according to claim 5, characterized in that, The automatic adjustment steps for the fitting parameters in step S8 include: S801, Set initial values ​​for fitting parameters The last iteration number was And the fitting parameters in the previous iteration were ; S802, according to the... Subfit Adjusting the fitting parameters by changing : ; It's about adjusting the step size; Indicates the first Error of the next iteration With the The next iteration The amount of change in error ; like Then increase the fitting parameters. ; like Then reduce the fitting parameters .

7. The method for three-dimensional reconstruction of power facilities based on multi-source point cloud collaboration according to claim 5, characterized in that, The adaptive adjustment steps are as follows: A. Parameter Preprocessing: Based on the power line point cloud extraction results in step S6, select a complete point cloud dataset for a single power line. ,in This represents the total number of point clouds for the power line. For the point cloud data points that constitute the power line; from the set Three non-collinear feature points are randomly selected from the data. Ensure coverage of different sections of the power line, among which, , and These are the three-dimensional coordinates of each feature point; B. Neighborhood Feature Calculation: For each feature point, based on an adaptive distance threshold... Filter the most recent There are 10 neighboring points, forming three sets of neighboring points: ; Calculate the distance between each feature point and its neighboring points: ； ； ； in, To represent feature points Centered on an adaptive distance threshold The first set of neighborhood points obtained under constraints; To represent feature points Centered on an adaptive distance threshold The second set of neighborhood points obtained under constraints; To represent feature points Centered on an adaptive distance threshold The third set of neighborhood points obtained under constraints; Represent the first (i) in each set of neighboring points. There are 1 neighboring points, among which ,and This represents the number of neighboring points; For feature points Its first Neighboring points The square of the Euclidean distance between them; For feature points Its first Neighboring points The square of the Euclidean distance between them; For feature points Its first Neighboring points The square of the Euclidean distance between them; These are the three-dimensional coordinates of the corresponding neighborhood points; C. Point Cloud Quality Feature Calculation: Calculate the distance statistics of the three sets of neighboring point sets to quantify the point cloud quality and noise level. Calculate the average distance of the three sets of neighboring point sets as follows: The distances from the standard deviations are respectively The noise intensities are respectively The average of the three values ​​is taken as the overall noise intensity of the power line point cloud. ; D. Adaptive threshold calculation: Based on the overall noise intensity of the point cloud, an adaptive root mean square error threshold is calculated. ; in, The root mean square error threshold is set to 0.02m. This is the noise adjustment factor.

8. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing at least one instruction, which is loaded and executed by the processor to implement the three-dimensional reconstruction method for power facilities based on multi-source point cloud collaboration as described in any one of claims 1-7.

Images