A substation equipment point cloud replacement method and device based on feature point matching
By using a feature point matching method, local neighborhood derivative matrix and SVD optimization to align point clouds, the problems of high computational cost and low accuracy in substation site cloud models are solved, generating high-precision 3D models.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANDONG ELECTRIC POWER TRANSMISSION & SUBSTATION ENG CO
- Filing Date
- 2026-01-26
- Publication Date
- 2026-06-02
AI Technical Summary
Existing traditional feature point extraction methods are not effective in large point cloud models of substations due to their large computational load and the excessive number of iterations or premature termination of the ICP algorithm, resulting in poor point cloud completion.
A feature point matching-based method is adopted, which constructs a second-order derivative correlation matrix through the local neighborhood derivative matrix, combines eigenvalue weighting and response function calculation, introduces a 3D non-maximum suppression mechanism, uses dynamic covariance matrix singular value decomposition (SVD), designs dual convergence judgment conditions, and optimizes the point cloud alignment process.
It improves the ability to recognize geometric features of point clouds, reduces the amount of computation, improves the alignment accuracy and computational efficiency of point clouds, and generates high-precision complete 3D models.
Smart Images
Figure CN122134988A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of point cloud processing technology, and in particular to a method and device for point cloud replacement of substation equipment based on feature point matching. Background Technology
[0002] Point cloud completion refers to the process of processing incomplete point cloud data to restore it to a more complete state. When point cloud data exists from multiple perspectives, the point cloud is incomplete because each perspective only captures partial information about the object. First, the poses of the point clouds from different perspectives need to be adjusted to ensure they are consistent in space. Then, information from overlapping areas of the point clouds is used to fill in the missing parts. During the fusion process, by comparing the positions and attributes of points in the overlapping areas, it is determined how to merge point clouds from different perspectives to fill in the missing areas. This method has wide applications in 3D reconstruction. For example, when scanning a large electrical device, point cloud data from multiple perspectives is obtained from different directions and heights. By using a point cloud completion method based on multi-view fusion, these incomplete point cloud data can be completed, constructing a more complete 3D model of the electrical device.
[0003] However, existing traditional feature point extraction methods are not effective when the geometric features of the point cloud are not obvious. Furthermore, in large point cloud models of substations, traditional feature point extraction methods extract a large number of feature points, and the uneven distribution of these feature points greatly increases the computational load. At the same time, traditional ICP relies on only a single convergence condition, such as an error threshold or a fixed number of iterations, which can lead to excessive iterations, computational redundancy, premature termination, and failure to achieve optimal alignment. Finally, existing point cloud completion methods require multiple suitable perspectives to complete the completion well. If the number of perspectives is insufficient or the perspective selection is unreasonable, it will be impossible to obtain enough overlapping area information to complete the missing parts, resulting in poor completion results. Summary of the Invention
[0004] This application provides a method and device for replacing point clouds of substation equipment based on feature point matching, in order to solve the above-mentioned problems.
[0005] On the one hand, this application provides a method for replacing point clouds of substation equipment based on feature point matching, characterized in that the method includes the following steps: Step S1: Input the residual point cloud and the complete point cloud of the substation equipment, and use the local neighborhood derivative matrix, eigenvalue weighting and 3D-NMS to filter key feature points, construct the complete point cloud kd tree, match the nearest neighbor points and calculate the error; Step S2: Set the error threshold and the maximum number of iterations, calculate the covariance matrix based on the corresponding point pairs, and solve the rotation matrix and translation vector through SVD to form the transformation matrix; Step S3: Apply the transformation matrix to update the residual cloud attitude, iterate and check the error and number of iterations. If it does not converge, repeat the solution of the matrix until the convergence condition is met. Step S4: Output the alignment result and transformation matrix, and replace the incomplete point cloud with the complete point cloud from the database.
[0006] In one implementation of this application, step S1 specifically includes: Input a fragmented point cloud P and a complete point cloud Q, and you have a fragmented 3D point cloud dataset. Each point is represented as: ; Construct a local neighborhood and determine its radius; For each point Iterate through all points in the point cloud and calculate the Euclidean distance between any two points: , distance Collect the points less than or equal to form a point. local neighborhood point set ; Calculate the derivative correlation matrix H and analyze the eigenvalues; Calculate the response function and determine the threshold.
[0007] In one implementation of this application, calculating the derivative correlation matrix H specifically includes: The first derivative is calculated using the numerical difference method, specifically: for The first derivative of the direction, at point 1 local neighborhood Internal calculation: set up yes The point in the middle, ,in It is the number of neighboring points; The approximate value of the first derivative in the direction is: ,in It is the interval, estimated based on the sampling interval of the point cloud; direction and The first derivatives in the directions are denoted as follows: and ; Constructing the second derivative correlation matrix The format is as follows: .
[0008] In one implementation of this application, the analysis of eigenvalues specifically includes: Solve for the eigenvalues of matrix H, for each point Correlation matrix Calculate its eigenvalues ; Sort the eigenvalues in descending order, i.e. .
[0009] In one implementation of this application, calculating the response function and determining the threshold specifically includes: The eigenvalues are weighted using the following weighting function:
[0010] Choose an empirical constant k and calculate the response function. For each point The response function value is: ; Set a threshold and identify feature points.
[0011] In one implementation of this application, step S1, constructing a complete point cloud kd-tree, specifically involves: Determine each feature point in the residual cloud. Nearest neighbor in a complete point cloud Record the corresponding point pairs; Calculate the distance error between corresponding point pairs, let the corresponding point pairs be... Then the error Represents Euclidean distance; The sum of errors for all corresponding point pairs is calculated and used as an error index under the current attitude to evaluate the quality of the attitude.
[0012] In one implementation of this application, step S2, the process of forming the transformation matrix, specifically includes: Let the point matrix of the residual defect cloud be... ,in The number of points corresponds to the point matrix of the complete point cloud. ; Calculate the covariance matrix of the point cloud ,in and They are respectively and For a 3D point cloud, the centroid and the covariance matrix are... A symmetric matrix whose elements represent the covariance between different coordinate dimensions; Through singular value decomposition Let's calculate, first calculate Rotation matrix Translation vector ,in It is orthogonal matrix, It is orthogonal matrix, It is diagonal matrix, The elements on the diagonal are called singular values, and and The column vectors are called the left singular vector and the right singular vector, respectively; specifically: to calculate the right singular vector: first calculate the matrix It is a The square array, for The orthogonal matrix formed by the eigenvectors obtained through eigenvalue decomposition is... , The column vectors are the right singular vectors; to calculate the left singular vectors: calculate the matrix. It is a The square array, for The orthogonal matrix formed by the eigenvectors obtained through eigenvalue decomposition is... , The column vectors are the left singular vectors.
[0013] In one implementation of this application, step S3, updating the pose of the residual feature cloud, specifically involves applying the calculated transformation matrix to the residual feature cloud for each feature point in the residual feature cloud. Through matrix multiplication The new coordinates are obtained, thus yielding the new residual cloud attitude.
[0014] In one implementation of this application, step S4 specifically includes: Output the final transformation matrix. Top left corner The submatrix is a rotation matrix, which represents the rotation of the incomplete point cloud relative to the complete point cloud in three-dimensional space. Its elements satisfy the properties of a rotation matrix, determining the rotation of the point cloud in three-dimensional space. Combinations of rotation angles along the axis are used to rotate the source point cloud to a direction that is better aligned with the target point cloud; The top right corner The submatrix is a translation vector, whose elements correspond to... The translation amount in the axial direction is used to move the rotated source point cloud to the corresponding position of the target point cloud, so as to realize the translation and alignment of the point cloud in space.
[0015] On the other hand, this application also provides a substation equipment point cloud replacement device based on feature point matching. The device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to complete the aforementioned substation equipment point cloud replacement method based on feature point matching.
[0016] This application provides a method and device for replacing point clouds of substation equipment based on feature point matching, which has the following beneficial effects: 1. This method constructs a second-order derivative correlation matrix based on the local neighborhood derivative matrix, and combines eigenvalue weighting and response function calculation to enhance the feature point recognition capability of key geometric structures such as corners and edges. This can effectively improve the problem of poor feature point extraction results caused by the lack of obvious geometric features in point clouds.
[0017] 2. This method introduces a 3D nonmaximum suppression mechanism, which retains significant feature points by comparing local neighborhood response values, avoiding redundancy and improving distribution uniformity. This improves the computational burden of large-scale substation site cloud models.
[0018] 3. This method uses Singular Value Decomposition (SVD) of Dynamic Covariance Matrix to accurately calculate the rotation matrix and translation vector, thereby improving the point cloud alignment accuracy. It also designs a dual convergence judgment condition error threshold and a maximum number of iterations to prevent algorithm divergence and reduce computation time, thus optimizing the problem of excessive iterations and premature termination in traditional ICP. Attached Figure Description
[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart of a substation equipment point cloud replacement method based on feature point matching is provided for an embodiment of this application; Figure 2 A schematic diagram of the defective cloud model provided in the embodiments of this application; Figure 3 This application provides a point cloud image after feature point extraction of a residual point cloud. Figure 4 This is a schematic diagram of a substation equipment point cloud replacement device based on feature point matching, provided as an embodiment of this application. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0021] This application provides a method and device for replacing point clouds of substation equipment based on feature point matching. The technical solution proposed in this application will be described in detail below with reference to the accompanying drawings.
[0022] Figure 1 This document presents a flowchart illustrating a point cloud replacement method for substation equipment based on feature point matching, as provided in an embodiment of this application. Figure 1 As shown, the method mainly includes the following steps: Step 1: Feature point extraction from point cloud data.
[0023] 1. Input the incomplete point cloud P and the complete point cloud Q, for the incomplete 3D point cloud dataset. , where each point ; 2. Constructing a local domain, specifically including: Determine the neighborhood radius. Based on factors such as the density of the point cloud and the size of the objects, select a suitable neighborhood radius. For point clouds of large buildings, the radius needs to be large, such as... The unit is meters; Construct a neighborhood set of points. For each point... Iterate through all points in the point cloud and calculate the Euclidean distance between any two points: , distance Collect the points less than or equal to form a point. local neighborhood point set ; Calculate the derivative correlation matrix Specifically, it includes: Calculate the first derivative using the numerical difference method. For The first derivative of the direction, at point 1 local neighborhood Internal calculation: set up yes The point in the middle, ,in It is the number of neighboring points; The approximate value of the first derivative in the direction is: ,in It is a tiny interval that can be estimated based on the sampling interval of the point cloud. For example, if the sampling interval of the point cloud in each coordinate axis direction is... ,but Similarly, calculate direction and The first derivatives in the directions are denoted as follows: and ; Constructing the second derivative correlation matrix Second derivative correlation matrix The format is as follows: ; 4. Eigenvalue analysis, specifically including: Calculate the eigenvalues. Solve for the eigenvalues of matrix H. For each point... Correlation matrix Calculate its eigenvalues ; Eigenvalue sorting. To facilitate subsequent calculations and comparisons, the eigenvalues are sorted in descending order, i.e. ; 5. Response function calculation and threshold determination, specifically including: Eigenvalue weighting. Eigenvalues represent the spatial distribution of a point cloud. The eigenvector corresponding to the largest eigenvalue is the dominant direction of the point cloud distribution, i.e., the direction with the largest variance. Smaller eigenvalues correspond to directions with smaller variances, reflecting the geometric structure of the point cloud. By weighting the eigenvalues, the geometric features of the point cloud become more apparent, facilitating better identification of feature points. The weighting function is: ; Calculate the response function value. Choose an empirical constant. Pick Calculate the response function For each point The response function value is: ; Set a threshold and evaluate feature points. Traditional thresholding can lead to dense and unevenly distributed feature points. To address this, 3D Non-Maximum Suppression (3D-NMS) is introduced to retain the strongest feature points through local neighborhood comparison. exist local neighborhood If other feature points exist inside. satisfy This indicates that within this local neighborhood, These are more prominent feature points, while These are relatively redundant feature points, therefore... Suppression, that is, not treating it as the final feature point; If in local neighborhood Inside, If the response value is greater than the response value of all other feature points in the neighborhood, then The point with the largest response within the local neighborhood is retained as the final feature point.
[0024] Step 2: Iterative process of the ICP algorithm for point cloud data.
[0025] 1. Nearest Neighbor Search. Constructing an efficient data structure for the complete point cloud Q. Trees are used for fast nearest neighbor searches. In the point cloud alignment process, it is necessary to find each feature point in the residual point cloud. Nearest neighbor in a complete point cloud To calculate the correspondence and transformation matrix, the specific method is as follows: starting from the root node, based on the dimension and segmentation value represented by the current node, compare the query point with the current node to determine whether to search the left or right subtree, until a leaf node is found. Then, determine the search path within the leaf node and its adjacent nodes. The nearest neighbor point; a schematic diagram of the residual cloud model is shown below. Figure 2 As shown.
[0026] For feature points in the incomplete point cloud, their nearest neighbors are found in the complete point cloud. The nearest neighbor relationship can be determined using Euclidean distance, and corresponding point pairs are recorded. The point cloud image after feature point extraction from the incomplete point cloud is shown below. Figure 3 As shown.
[0027] 2. Calculate the error between corresponding point pairs. Find the corresponding point pairs and calculate the distance error between them. Let the corresponding point pair be... Then the error Represents Euclidean distance; Calculate either the sum or the average of the errors for all corresponding point pairs, and use this as an error index for the current attitude to evaluate the quality of the attitude.
[0028] 3. Calculate the transformation matrix. This includes: Let the point matrix of the residual defect cloud be... ,in The number of points corresponds to the point matrix of the complete point cloud. ; Calculate the covariance matrix of the point cloud ,in and They are respectively and For a 3D point cloud, the centroid and the covariance matrix are... A symmetric matrix whose elements represent the covariance between different coordinate dimensions; Through singular value decomposition Let's calculate, first calculate Rotation matrix Translation vector ,in It is orthogonal matrix, It is orthogonal matrix, It is diagonal matrix, The elements on the diagonal are called singular values, and and The column vectors are called the left singular vector and the right singular vector, respectively; Calculate the right singular vector: First calculate the matrix It is a The square array, for The orthogonal matrix formed by the eigenvectors obtained through eigenvalue decomposition is... , The column vectors are the right singular vectors; Calculate the left singular vector: Calculate the matrix It is a The square array, for The orthogonal matrix formed by the eigenvectors obtained through eigenvalue decomposition is... , The column vectors are the left singular vectors.
[0029] 4. Update the pose of the residual feature cloud. Apply the calculated transformation matrix to the residual feature cloud for each feature point in the residual feature cloud. Through matrix multiplication The new coordinates are obtained, thus yielding the new residual cloud attitude.
[0030] Step 3: ICP convergence judgment.
[0031] 1. Set convergence conditions. Specifically, this includes: Set error threshold When the average error of the corresponding point pair is less than the threshold, the algorithm is considered to have converged. Set the maximum number of iterations This prevents the algorithm from getting stuck in an infinite loop.
[0032] 2. Determine convergence. This specifically includes: After each iteration, calculate the current average error and compare it with the threshold. At the same time, check whether the number of iterations exceeds the maximum value. If the convergence condition is met, stop the iteration; otherwise, return to step two to continue the next iteration.
[0033] Step 4: Output the fully aligned point cloud data results.
[0034] Output the final transformation matrix. Top left corner The submatrix is a rotation matrix, which represents the rotation of the incomplete point cloud relative to the complete point cloud in three-dimensional space. Its elements satisfy the properties of a rotation matrix, determining the rotation of the point cloud in three-dimensional space. Combinations of rotation angles along the axis are used to rotate the source point cloud to a direction that is better aligned with the target point cloud; The top right corner The submatrix is a translation vector, whose elements correspond to... The translation amount along the axis is used to move the rotated source point cloud to the corresponding position of the target point cloud, realizing the translational alignment of the point cloud in space. This matrix accurately describes the attitude alignment relationship between the residual point cloud and the complete point cloud, and can be used for subsequent coordinate transformation, attitude adjustment and other operations.
[0035] Step 5: Through the above steps, the pose of the Kuli point cloud is matched with that of the target point cloud, thereby directly replacing the target point cloud with the Kuli point cloud model.
[0036] The above describes a substation equipment point cloud replacement system based on feature point matching, as provided in this application. Based on the same inventive concept, this application also provides a substation equipment point cloud replacement device based on feature point matching. Figure 4 A schematic diagram of a substation equipment point cloud replacement device based on feature point matching is provided for an embodiment of this application, as shown below. Figure 4 As shown, the device mainly includes: at least one processor 401; and a memory 402 communicatively connected to the at least one processor; wherein the memory 402 stores instructions that can be executed by the at least one processor 401, and the instructions are executed by the at least one processor 401 to enable the at least one processor 401 to complete the aforementioned substation equipment point cloud replacement method based on feature point matching.
[0037] Below is an application example of this application in a specific scenario. A 220kV outdoor substation is conducting 3D digital modeling of its equipment. Point cloud data collection and complete model construction are required for the core equipment within the substation—Main Transformer No. 1 (model SFSZ11-240000 / 220). When using a LiDAR scanner (scanning accuracy ±2mm) for multi-angle scanning on-site, two significant deficiencies were observed in the collected point cloud data due to obstruction by the top bushings of the main transformer and interference from the side fences: firstly, the point cloud was missing in the area at the root of the high-voltage bushing (approximately 0.8m × 1.2m); secondly, the point cloud was sparse and distorted at the connection between the bottom of the tank and the foundation (approximately 1.5m × 0.6m). The existing substation equipment point cloud database stores complete point cloud models of main transformers of the same model and installation method. The point cloud replacement method described in this application is needed to fill in the incomplete point cloud data collected on-site with the complete point cloud from the database, generating a high-precision complete 3D model to provide data support for subsequent equipment inspection, fault simulation, and maintenance path planning. The specific implementation steps are as follows: (a) Data input and preprocessing.
[0038] Input data: Defect cloud P collected on-site: containing 186,000 3D points, with coordinates ranging from... The missing area is marked as P defect (23,500 points are missing / distorted); Complete point cloud database Q: Standard point cloud of the same model main transformer, containing 212,000 3D points, with coordinate range of... The point cloud density is consistent with the data collected on site (average point spacing 0.005m).
[0039] Preprocessing: Denoising is performed on P and Q respectively (isolated points that are more than 3 times the standard deviation from the neighborhood mean are removed), and valid point cloud data is retained.
[0040] (ii) Extraction and matching of key feature points.
[0041] Local neighborhood construction: Set the neighborhood radius r=0.3m (combining the main transformer structure size and point cloud density, taking into account both feature extraction accuracy and computational efficiency); taking a certain feature candidate point Pi=(85.6,37.8,6.2) (a point near the high-voltage side bushing) in the residual defect cloud P as an example, traverse all points in P, calculate the Euclidean distance d(Pi,Pj), and collect 32 points with a distance ≤0.3m to form a local neighborhood point set N(Pi).
[0042] Derivative matrix and eigenvalue calculation: Sampling interval h = 0.002m (matching the sampling accuracy of lidar scanning), the first derivative of each point in N(Pi) in the x, y, and z directions is calculated using the numerical difference method. For example, the approximate first derivative in the x direction is... (s is the sorted index of the points in the neighborhood); construct the second derivative correlation matrix H (3×3 symmetric matrix), and solve for the eigenvalues of H to obtain the eigenvalues. Sort in descending order as .
[0043] Response function and 3D-NMS screening: An empirical constant k=0.04 was set, and the response function R=λ1λ2λ3-k(λ1+λ2+λ3)²=2.86×0.72×0.18-0.04×(2.86+0.72+0.18)²≈0.37-0.04×14.24≈0.37-0.57≈-0.20 (the negative value here is because the point is in a smooth region and will be screened later). After calculating the R value for all candidate points in P, 3D-NMS was performed: with each candidate point as the center and a suppression radius of 0.5m, the point with the largest R value in its neighborhood was retained. Finally, 1280 key feature points (covering areas with significant geometric features such as sleeve flanges, oil tank welds, and lifting ramps) were selected from P, and 1320 corresponding feature points were selected from Q.
[0044] kd-tree matching and error calculation: Construct a kd-tree for the complete point cloud Q, and find the corresponding point Qi for each feature point Pi in P through nearest neighbor search. For example, feature point P of P 102 =(86.3,38.1,7.5) (edge of the sleeve flange), matched to feature point Q of Q. 102 =(12.7,7.3,7.5); Calculate the error between corresponding point pairs. (Euclidean distance), the average error of all point pairs under the initial attitude is 0.23m, and the total error is 1280×0.23≈294.4m.
[0045] (III) Solving for the transformation matrix.
[0046] Convergence condition setting: Set the error threshold (Meets the accuracy requirement of ±0.06m for 3D modeling of substations), with a maximum number of iterations N=50 (to avoid excessive iteration leading to computational redundancy).
[0047] Covariance matrix and SVD decomposition: Residual feature cloud feature point matrix P=[p1,p2,...,p 1280 The complete point cloud feature point matrix Q = [q1, q2, ..., q] 1280 ]; Calculate the centroid:, =(1 / (1280))∑1² 80 qi=(12.5,7.2,4.5); covariance matrix H=(1 / 1280)∑1² 80 (pi- (qi-) ) The calculation yields:
[0048] Performing SVD decomposition on H yields U (left singular vector matrix), S (singular value matrix), and V. (Right singular vector matrix), calculate the rotation matrix R=VU Translation vector :
[0049] t=( 73.6, 30.8, 0.0) The final transformation matrix T is: .
[0050] (iv) Iterative verification and convergence.
[0051] Pose update: Substitute all feature points in the residual feature cloud P into the formula This yields the updated feature point coordinates, for example... =R×(86.3,38.1,7.5)+(-73.6,-30.8,0.0)=(12.7,7.3,7.5), and Q 102 Fully aligned.
[0052] Iterative verification: After the first iteration, the average error of the corresponding point pairs decreased to 0.11m (still greater than 0.11m). The iteration continued; by the 28th iteration, the average error had decreased to 0.043m (less than...). =0.05m), and the number of iterations (28) < the maximum number of iterations (50), the convergence condition is met, and the iteration stops.
[0053] (v) Point cloud replacement and result output.
[0054] Replacement execution: Apply the converged transformation matrix T to the complete point cloud Q, so that the coordinate system of Q is aligned with the coordinate system of the residual point cloud P collected on site (i.e., Qtransformed=R×Q+t). Replacement of missing regions: Retain the valid point cloud in P (non-Pdefect region), and replace P_defect with the point cloud of the corresponding missing region in Qtransformed (root of the casing, bottom of the tank) to generate the complete point cloud Pcomplete.
[0055] III. Verification of Application Effects
[0056] Accuracy verification: The replaced complete point cloud Pcomplete was checked in the field. Ten key dimensions were selected (such as bushing center distance, oil tank height, and base length). The measurement results showed that the error between the measurement results and the equipment design drawings was ≤0.05m, which met the accuracy requirements of substation digital modeling. Practicality verification: Pcomplete can be directly imported into the 3D operation and maintenance platform, supporting equipment surface defect detection (such as bushing crack identification), virtual path planning for maintenance personnel (such as avoiding obstructed areas), and equipment collision simulation (such as spatial interference verification when hoisting new components), significantly improving the digitalization level of substation operation and maintenance.
[0057] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.
[0058] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0059] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.
Claims
1. A method for replacing point clouds of substation equipment based on feature point matching, characterized in that, The method includes the following steps: Step S1: Input the residual point cloud and the complete point cloud of the substation equipment, and use the local neighborhood derivative matrix, eigenvalue weighting and 3D-NMS to filter key feature points, construct the complete point cloud kd tree, match the nearest neighbor points and calculate the error; Step S2: Set the error threshold and the maximum number of iterations, calculate the covariance matrix based on the corresponding point pairs, and solve the rotation matrix and translation vector through SVD to form the transformation matrix; Step S3: Apply the transformation matrix to update the residual cloud attitude, iterate and check the error and number of iterations. If it does not converge, repeat the solution of the matrix until the convergence condition is met. Step S4: Output the alignment result and transformation matrix, and replace the incomplete point cloud with the complete point cloud from the database.
2. The substation equipment point cloud replacement method based on feature point matching according to claim 1, characterized in that, Step S1 specifically includes: Input a fragmented point cloud P and a complete point cloud Q, and you have a fragmented 3D point cloud dataset. Each point is represented as: ; Construct a local neighborhood and determine its radius; For each point Iterate through all points in the point cloud and calculate the Euclidean distance between any two points: , distance Collect the points less than or equal to form a point. local neighborhood point set ; Calculate the derivative correlation matrix H and analyze the eigenvalues; Calculate the response function and determine the threshold.
3. The substation equipment point cloud replacement method based on feature point matching according to claim 2, characterized in that, Calculate the derivative correlation matrix H, specifically including: The first derivative is calculated using the numerical difference method, specifically: for The first derivative of the direction, at point 1 local neighborhood Internal calculation: set up yes The point in the middle, ,in It is the number of neighboring points; The approximate value of the first derivative in the direction is: ,in It is the interval, estimated based on the sampling interval of the point cloud; direction and The first derivatives in the directions are denoted as follows: and ; Constructing the second derivative correlation matrix The format is as follows: 。 4. The substation equipment point cloud replacement method based on feature point matching according to claim 2, characterized in that, Analyze eigenvalues, specifically including: Solve for the eigenvalues of matrix H, for each point Correlation matrix Calculate its eigenvalues ; Sort the eigenvalues in descending order, i.e. .
5. The substation equipment point cloud replacement method based on feature point matching according to claim 2, characterized in that, Calculating the response function and determining the threshold specifically includes: The eigenvalues are weighted using the following weighting function: Choose an empirical constant k and calculate the response function. For each point The response function value is: ； Set a threshold and identify feature points.
6. The substation equipment point cloud replacement method based on feature point matching according to claim 1, characterized in that, In step S1, constructing a complete point cloud kd-tree specifically involves: Determine each feature point in the residual cloud. Nearest neighbor in a complete point cloud Record the corresponding point pairs; Calculate the distance error between corresponding point pairs, let the corresponding point pairs be... Then the error Represents Euclidean distance; The sum of errors for all corresponding point pairs is calculated and used as an error index under the current attitude to evaluate the quality of the attitude.
7. The substation equipment point cloud replacement method based on feature point matching according to claim 1, characterized in that, The process of forming the transformation matrix in step S2 specifically includes: Let the point matrix of the residual defect cloud be... ,in The number of points corresponds to the point matrix of the complete point cloud. ; Calculate the covariance matrix of the point cloud ,in and They are respectively and For a 3D point cloud, the centroid and the covariance matrix are... A symmetric matrix whose elements represent the covariance between different coordinate dimensions; Through singular value decomposition Let's calculate, first calculate Rotation matrix Translation vector ,in It is orthogonal matrix, It is orthogonal matrix, It is diagonal matrix, The elements on the diagonal are called singular values, and and The column vectors are called the left singular vector and the right singular vector, respectively; specifically: to calculate the right singular vector: first calculate the matrix It is a The square array, for The orthogonal matrix formed by the eigenvectors obtained through eigenvalue decomposition is... , The column vectors are the right singular vectors; to calculate the left singular vectors: calculate the matrix. It is a The square array, for The orthogonal matrix formed by the eigenvectors obtained through eigenvalue decomposition is... , The column vectors are the left singular vectors.
8. The substation equipment point cloud replacement method based on feature point matching according to claim 1, characterized in that, In step S3, updating the pose of the residual feature cloud specifically involves applying the calculated transformation matrix to the residual feature cloud for each feature point in the residual feature cloud. Through matrix multiplication The new coordinates are obtained, thus yielding the new residual cloud attitude.
9. The substation equipment point cloud replacement method based on feature point matching according to claim 1, characterized in that, Step S4 specifically includes: Output the final transformation matrix. Top left corner The submatrix is a rotation matrix, which represents the rotation of the incomplete point cloud relative to the complete point cloud in three-dimensional space. Its elements satisfy the properties of a rotation matrix, determining the rotation of the point cloud in three-dimensional space. Combinations of rotation angles along the axis are used to rotate the source point cloud to a direction that is better aligned with the target point cloud; The top right corner The submatrix is a translation vector, whose elements correspond to... The translation amount in the axial direction is used to move the rotated source point cloud to the corresponding position of the target point cloud, so as to realize the translation and alignment of the point cloud in space.
10. A substation equipment point cloud replacement device based on feature point matching, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform a substation equipment point cloud replacement method based on feature point matching as described in any one of claims 1-9.