Coordinate data global alignment method and device for inspection scene

Abstract

The embodiment of the invention provides an inspection scene-oriented coordinate data global alignment method and device, and the method comprises the steps: constructing a feature description vector set through filtering noise reduction and local curvature key point screening, screening a seed corresponding relation step by step in combination with a spatial distance constraint and a normal vector consistency constraint, and constructing a geometric compatibility matrix; and extracting a global consistency weight sequence through eigenvalue decomposition to construct a weighted covariance matrix, executing singular value decomposition to solve a rotation matrix and a translation vector to complete corridor coordinate alignment, and finally outputting ultralimit vegetation positioning information to be called by cutting operation. The defects of the traditional technology in the aspects of coordinate data preprocessing and feature extraction, mismatching robust screening and geometric compatibility modeling, global consistency weight solution, alignment result output and the like are effectively overcome; and a technical guarantee is provided for intelligent global alignment of coordinate data and precise positioning of overrun vegetation in a power transmission corridor unmanned aerial vehicle inspection scene.

Description

Technical Field

[0001] This application relates to the field of electronic digital data processing, specifically to a method and apparatus for global alignment of coordinate data in inspection scenarios. Background Technology

[0002] Existing global alignment methods for coordinate data in inspection scenarios have significant shortcomings. Traditional systems perform poorly in the preprocessing and feature extraction of data collected by UAVs. They typically lack the ability to perform systematic filtering, noise reduction, and downsampling on both the source and reference coordinate datasets. They also fail to effectively screen key data points from standardized corridor coordinate datasets and calculate feature descriptors to obtain feature description vector sets based on local curvature threshold conditions. This results in insufficient noise suppression and key feature point extraction accuracy of coordinate data in the complex terrain of power transmission corridors, making it difficult to provide a high-quality feature representation foundation for subsequent cross-dataset matching.

[0003] Furthermore, existing technologies suffer from bottlenecks in initial correspondence selection and geometric compatibility modeling. Most systems lack the ability to sequentially perform spatial distance constraint selection and normal vector consistency constraint selection on the initial correspondence set to obtain a seed correspondence set, calculate the compatibility score of each matching pair based on the rigid body transformation distance preservation property, and construct a geometric compatibility matrix. This results in mismatched pairs causing significant interference in subsequent pose estimation, affecting the robustness and accuracy of global alignment, especially in actual inspection scenarios with dense repetitive structures and unstable point cloud overlap rates in power transmission corridors.

[0004] Existing systems have technical shortcomings in solving for globally consistent weights and outputting coordinate alignment based on geometric compatibility matrices. They lack a complete closed-loop alignment execution mechanism that involves performing eigenvalue decomposition on the geometric compatibility matrix to extract principal eigenvectors, obtaining a globally consistent weight sequence, constructing a weighted covariance matrix, performing singular value decomposition to solve for rotation and translation vectors, and then completing corridor coordinate alignment and outputting over-limit vegetation location information for use in pruning operations. This affects the inspection system's ability to accurately locate over-limit vegetation targets and coordinate subsequent intelligent pruning operations. Solving these problems is crucial for improving the accuracy and intelligence level of global coordinate data alignment in power transmission corridor UAV inspection scenarios. Summary of the Invention

[0005] To address the problems in existing technologies, this application provides a method and apparatus for global alignment of coordinate data in inspection scenarios. It can effectively solve the shortcomings of traditional technologies in coordinate data preprocessing and feature extraction, robust screening of mismatches and geometric compatibility modeling, global consistency weight solution and alignment result output, etc., and provide technical support for intelligent global alignment of coordinate data and accurate positioning of over-limit vegetation in the UAV inspection scenario of power transmission corridors.

[0006] To solve at least one of the above problems, this application provides the following technical solution: Firstly, this application provides a method for global alignment of coordinate data in inspection scenarios, including: The source coordinate dataset and reference coordinate dataset collected by the UAV on the power transmission corridor are obtained. The source coordinate dataset and the reference coordinate dataset are filtered, denoised and downsampled to obtain a standardized corridor coordinate dataset. Key data points are selected from the standardized corridor coordinate dataset based on the local curvature threshold condition and feature descriptors are calculated to obtain a feature description vector set. The feature description vector set is subjected to cross-dataset similarity matching to obtain an initial correspondence set. Spatial distance constraint filtering and normal vector consistency constraint filtering are performed on the initial correspondence set in sequence to obtain a seed correspondence set. Based on the rigid body transformation distance preservation property, the compatibility score of each matching pair in the seed correspondence set is calculated to obtain a geometric compatibility matrix. The geometric compatibility matrix is ​​subjected to eigenvalue decomposition and the principal eigenvectors are extracted to obtain a globally consistent weight sequence. The globally consistent weight sequence is used as coefficients to construct a weighted covariance matrix and singular value decomposition is performed to obtain a rotation matrix and a translation vector. The rotation matrix and the translation vector are applied to the source coordinate dataset to complete the corridor coordinate alignment and output the over-limit vegetation location information for use in the pruning operation.

[0007] Furthermore, it also includes: calculating the average spatial distance between each data point in the source coordinate dataset and the reference coordinate dataset and its nearest neighbor data point in the neighborhood to obtain the neighborhood distance value; comparing the neighborhood distance value with the global average distance and determining noise points based on the standard deviation threshold condition; removing the data points determined to be noise points from the source coordinate dataset and the reference coordinate dataset to obtain the denoised coordinate dataset. The three-dimensional space covered by the denoised coordinate dataset is divided into uniform voxel grid cells. The centroid coordinates of all data points in each voxel grid cell are calculated to obtain the grid centroid point. The grid centroid points are then aggregated to form a standardized corridor coordinate dataset.

[0008] Furthermore, it also includes: extracting neighborhood point sets for each data point in the standardized corridor coordinate dataset and constructing a neighborhood covariance matrix; performing eigenvalue decomposition on the neighborhood covariance matrix to obtain the minimum eigenvalue and the maximum eigenvalue; calculating the ratio of the minimum eigenvalue to the maximum eigenvalue to obtain the local curvature value; comparing the local curvature value with a preset local curvature threshold condition and selecting data points whose curvature values ​​meet the threshold condition as key data points; A local reference coordinate system is established with each key data point as the center, and the neighborhood space is divided into multiple sub-regions. The radial distance distribution, the angle distribution between the normal vector and the relative height distribution of the data points in each sub-region are statistically analyzed to obtain a sub-region statistical histogram. The statistical histograms of each sub-region are spliced ​​together in a fixed order to form a feature description vector. The feature description vectors corresponding to all the key data points are aggregated to obtain a feature description vector set.

[0009] Furthermore, it also includes: searching for the feature description vector with the smallest Euclidean distance in the feature description vector set of the reference coordinate dataset for each key data point in the source coordinate dataset to obtain the nearest neighbor match and the second nearest neighbor match; calculating the ratio of the nearest neighbor match distance to the second nearest neighbor match distance to obtain the distance ratio value; comparing the distance ratio value with the preset ratio value test threshold condition and retaining the matching pairs that meet the threshold condition to form an initial correspondence set; For each matching pair in the initial correspondence set, the spatial Euclidean distance between the source data point and the target data point is calculated to obtain the matching spatial distance. The matching spatial distance is compared with the preset maximum displacement threshold condition, and matching pairs that exceed the threshold condition are removed to obtain the distance-filtered set. For each matching pair in the distance-filtered set, the angle between the normal vector of the source data point and the normal vector of the target data point is calculated to obtain the normal vector angle value. The normal vector angle value is compared with the preset angle threshold condition, and matching pairs that meet the threshold condition are retained to form the seed correspondence set.

[0010] Furthermore, it also includes: calculating the spatial distance between corresponding two points in the source coordinate dataset for any two matching pairs in the seed correspondence set to obtain the source end distance, calculating the spatial distance between corresponding two points in the reference coordinate dataset to obtain the target end distance, and taking the absolute value of the difference between the source end distance and the target end distance to obtain the distance difference. The compatibility score is calculated by inputting the distance difference into a kernel function that monotonically decreases with the distance difference. The compatibility scores between all pairs of matching pairs in the seed correspondence set are written into a symmetric matrix according to the row and column indices to obtain a geometric compatibility matrix. The diagonal elements of the geometric compatibility matrix are set to the maximum value to indicate that each matching pair is completely compatible with itself.

[0011] Furthermore, it also includes: performing eigenvalue decomposition on the geometric compatibility matrix to obtain an eigenvalue set and a corresponding eigenvector set, extracting the largest eigenvalue from the eigenvalue set and reading the eigenvector corresponding to the largest eigenvalue as the principal eigenvector, and normalizing each component in the principal eigenvector to the interval between zero and one to obtain a globally consistent weight sequence; Each weight value in the global consistency weight sequence is used as the coefficient of the corresponding matching pair. Based on the coefficient, a weighted cross covariance matrix is ​​constructed for the source data point coordinates and target data point coordinates in the seed correspondence set. Singular value decomposition is performed on the weighted cross covariance matrix to obtain a rotation matrix. A translation vector is calculated based on the rotation matrix and the weighted centroid coordinates.

[0012] Furthermore, it also includes: performing matrix multiplication on the coordinates of each data point in the source coordinate dataset and the rotation matrix to obtain the rotated coordinates, performing vector addition on the rotated coordinates and the translation vector to obtain the transformed coordinates, and aggregating all the transformed coordinates to form an aligned corridor coordinate dataset; Identify vegetation data points and traverse data points from the aligned corridor coordinate dataset, calculate the spatial distance between each vegetation data point and the nearest traverse data point to obtain the tree-line distance value, compare the tree-line distance value with a preset safe distance threshold condition, and filter vegetation data points that are less than the threshold condition as over-limit vegetation points, extract the spatial coordinates of each over-limit vegetation point to form over-limit vegetation positioning information, and output it for use in UAV trimming operation path planning.

[0013] Secondly, this application provides a global coordinate data alignment device for inspection scenarios, comprising: The data filtering module is used to acquire the source coordinate dataset and reference coordinate dataset collected by the UAV on the power transmission corridor, perform filtering, noise reduction and downsampling on the source coordinate dataset and the reference coordinate dataset to obtain a standardized corridor coordinate dataset, and filter key data points from the standardized corridor coordinate dataset based on the local curvature threshold condition and calculate feature descriptors to obtain a feature description vector set. The data alignment module is used to perform cross-dataset similarity matching on the feature description vector set to obtain an initial correspondence set, and to perform spatial distance constraint filtering and normal vector consistency constraint filtering on the initial correspondence set to obtain a seed correspondence set. Based on the rigid body transformation distance preservation property, the module calculates the compatibility score of each matching pair in the seed correspondence set to obtain a geometric compatibility matrix. The pruning module is used to perform eigenvalue decomposition on the geometric compatibility matrix and extract the principal eigenvectors to obtain a globally consistent weight sequence. The globally consistent weight sequence is used as coefficients to construct a weighted covariance matrix and perform singular value decomposition to obtain a rotation matrix and a translation vector. The rotation matrix and the translation vector are applied to the source coordinate dataset to complete the corridor coordinate alignment and output the over-limit vegetation positioning information for the pruning operation to call.

[0014] As can be seen from the above technical solution, this application provides a method and device for global alignment of coordinate data in inspection scenarios. It constructs a feature description vector set by filtering and denoising and selecting key points of local curvature. It then combines spatial distance constraints and normal vector consistency constraints to progressively select seed correspondences and construct a geometric compatibility matrix. Finally, it extracts a global consistency weight sequence through eigenvalue decomposition to construct a weighted covariance matrix and performs singular value decomposition to solve the rotation matrix and translation vector to complete the corridor coordinate alignment. The final output is the location information of over-limit vegetation for use in pruning operations. This effectively solves the shortcomings of traditional technologies in coordinate data preprocessing and feature extraction, robust screening of mismatches and geometric compatibility modeling, and global consistency weight solution and alignment result output. It provides technical support for intelligent global alignment of coordinate data and accurate positioning of over-limit vegetation in power transmission corridor UAV inspection scenarios. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the global alignment method for coordinate data in an inspection scenario as described in this application embodiment. Figure 2 This is a structural diagram of the global coordinate data alignment device for inspection scenarios in this application embodiment. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0019] In view of the problems existing in the prior art, this application provides a method and device for global alignment of coordinate data in inspection scenarios. It constructs a feature description vector set by filtering and denoising and selecting key points of local curvature. It then combines spatial distance constraints and normal vector consistency constraints to progressively select seed correspondences and construct a geometric compatibility matrix. Finally, it extracts a global consistency weight sequence through eigenvalue decomposition to construct a weighted covariance matrix and performs singular value decomposition to solve the rotation matrix and translation vector to complete the corridor coordinate alignment. The final output is the location information of over-limit vegetation for use in pruning operations. This effectively solves the shortcomings of traditional technologies in coordinate data preprocessing and feature extraction, robust screening of mismatches and geometric compatibility modeling, and global consistency weight solution and alignment result output. It provides technical support for intelligent global alignment of coordinate data and accurate positioning of over-limit vegetation in power transmission corridor UAV inspection scenarios.

[0020] To effectively address the shortcomings of traditional technologies in coordinate data preprocessing and feature extraction, robust screening of mismatches and geometric compatibility modeling, and global consistency weight calculation and alignment result output, and to provide technical support for intelligent global alignment of coordinate data and precise positioning of over-limit vegetation in power transmission corridor UAV inspection scenarios, this application provides an embodiment of a global alignment method for coordinate data in inspection scenarios. See [link to embodiment]. Figure 1 The global alignment method for coordinate data in the inspection scenario specifically includes the following: Step S101: Obtain the source coordinate dataset and reference coordinate dataset collected by the UAV on the power transmission corridor. Perform filtering, noise reduction and downsampling on the source coordinate dataset and the reference coordinate dataset to obtain a standardized corridor coordinate dataset. Based on the local curvature threshold condition, select key data points from the standardized corridor coordinate dataset and calculate feature descriptors to obtain a feature description vector set. This embodiment acquires source coordinate datasets and reference coordinate datasets of the power transmission corridor from an airborne laser sensor on a UAV. The source coordinate dataset is a set of three-dimensional spatial coordinate points collected during the current flight, while the reference coordinate dataset is a set of coordinate points collected during historical baseline flights and manually verified. The two datasets record the spatial distribution of tower structures, conductor sag, and vegetation canopy within the corridor area at their respective acquisition times, forming the raw input for subsequent alignment calculations.

[0021] After acquiring the source coordinate dataset and the reference coordinate dataset, this embodiment performs statistical filtering noise reduction processing on them. Specifically, for each data point in the dataset, the average spatial distance between it and its nearest neighbor data points is calculated, and this distance value is recorded as the neighborhood distance value. This embodiment calculates the global mean and standard deviation of the neighborhood distance values ​​of all data points, and identifies data points whose neighborhood distance values ​​exceed several times the global mean standard deviation as noise points and removes them. This processing targets isolated outliers caused by sensor multipath reflection and electromagnetic interference, removing them from the dataset to obtain the noise-reduced coordinate dataset.

[0022] Based on the denoised coordinate dataset, this embodiment performs voxel rasterization downsampling to compress the data size. This embodiment divides the three-dimensional space covered by the denoised coordinate dataset into cubic raster cells with uniform side lengths. The mean coordinate of all data points falling within the same raster cell is calculated as the centroid coordinate point of that raster. The centroids of all raster cells are aggregated to form a standardized corridor coordinate dataset. This standardized corridor coordinate dataset compresses the number of points to a fixed proportion of the original size while preserving the geometric structure of the corridor space, reducing the computational overhead of subsequent matrix operations.

[0023] Accordingly, this embodiment performs key data point screening on the standardized corridor coordinate dataset. For each data point in the dataset, its neighborhood point set is extracted and a neighborhood covariance matrix is ​​constructed. Eigenvalue decomposition is performed on the neighborhood covariance matrix to obtain three eigenvalues. This embodiment calculates the ratio of the minimum eigenvalue to the maximum eigenvalue as the local curvature value of the data point. The local curvature value reflects the degree of curvature of the local surface where the data point is located. The local curvature value is compared with a preset local curvature threshold condition, and data points whose curvature values ​​meet the threshold condition are selected as key data points. These key data points typically correspond to geometrically significant locations such as tower nodes, conductor sag apexes, and vegetation branch forks.

[0024] After the key data points are selected, this embodiment calculates feature descriptors for each key data point to generate a distinguishable digital representation. A local reference coordinate system is established with the key data point as the center, and its neighborhood space is divided into multiple sub-regions according to three dimensions: spherical shell radius, elevation angle, and azimuth angle. The radial distance distribution, normal vector angle distribution, and relative height distribution of data points in each sub-region are statistically analyzed, and the statistical results are encoded into a sub-region statistical histogram. In this embodiment, the statistical histograms of each sub-region are concatenated in a fixed order to form the feature description vector of the key data point. The feature description vectors corresponding to all key data points are aggregated to obtain a feature description vector set, which serves as the input data source for cross-dataset similarity matching in the subsequent step S201.

[0025] Step S102: Perform cross-dataset similarity matching on the feature description vector set to obtain an initial correspondence set. Perform spatial distance constraint filtering and normal vector consistency constraint filtering on the initial correspondence set in sequence to obtain a seed correspondence set. Calculate the compatibility score of each matching pair in the seed correspondence set based on the rigid body transformation distance preservation property to obtain a geometric compatibility matrix. In this embodiment, the feature description vectors of key data points corresponding to the source coordinate dataset and the reference coordinate dataset are read from the feature description vector set output in step S101 above. Cross-dataset similarity matching is performed to establish an initial correspondence. For the feature description vector of each key data point in the source coordinate dataset, the feature description vector with the smallest Euclidean distance in the feature description vector set of the reference coordinate dataset is searched as the nearest neighbor match, and the feature description vector with the second smallest Euclidean distance is recorded as the second nearest neighbor match. In this embodiment, the ratio of the nearest neighbor match distance to the second nearest neighbor match distance is calculated to obtain the distance ratio. The distance ratio is compared with a preset ratio test threshold condition, and the matching pairs with a distance ratio less than the threshold condition are retained. This ratio test strategy filters out fuzzy matching cases where there are multiple similar candidates in the feature space, and the matching pairs that pass the test are aggregated to form an initial correspondence set.

[0026] Based on the initial set of correspondences, this embodiment performs spatial distance constraint filtering to eliminate physically impossible erroneous matches. For each matching pair in the initial set of correspondences, the spatial Euclidean distance between the source data point coordinates and the target data point coordinates is calculated to obtain the matching spatial distance. This embodiment presets a maximum displacement threshold condition based on the UAV flight path planning and position error range. The matching spatial distance is compared with the maximum displacement threshold condition. Matching pairs that exceed the threshold condition exceed a reasonable range in spatial displacement and are judged as erroneous matches and removed from the set. The remaining matching pairs form the distance-filtered set.

[0027] After the distance-filtered set is generated, this embodiment continues to perform normal vector consistency constraint filtering. Rigid body transformation does not change the angle relationship between surface normal vectors. Based on this, the source data point normal vector and the target data point normal vector are read for each matching pair in the distance-filtered set, and the angle between the two normal vectors is calculated to obtain the normal vector angle value. This embodiment compares the normal vector angle value with a preset angle threshold condition. Matching pairs whose normal vector angle values ​​exceed the threshold condition are inconsistent in local geometric orientation and are discarded. This constraint has a targeted filtering effect on cross-line mismatches between parallel conductor segments in the transmission corridor. The filtered matching pairs are aggregated to form a seed correspondence set.

[0028] Accordingly, this embodiment constructs a geometric compatibility matrix for the seed correspondence set based on the rigid body transformation distance preservation property. The distance preservation property states that if two correspondences are both correct matches, the spatial distance between corresponding points in the source coordinate dataset should be equal to the spatial distance between corresponding points in the reference coordinate dataset. In this embodiment, for any two matching pairs in the seed correspondence set, the spatial distance between corresponding points in the source coordinate dataset is calculated to obtain the source-end distance, and the spatial distance between corresponding points in the reference coordinate dataset is calculated to obtain the target-end distance. The absolute value of the difference between the source-end distance and the target-end distance is then used to obtain the distance difference.

[0029] After the distance difference is calculated, this embodiment inputs it into a kernel function to quantify the compatibility between two matching pairs. The kernel function is a function that monotonically decreases with the distance difference; it outputs the maximum compatibility score when the distance difference is zero, and the output score rapidly decays and approaches zero as the distance difference increases. In this embodiment, compatibility scores are calculated pairwise for all matching pairs in the seed correspondence set. Each score is written into a symmetric matrix structure according to the row and column indices of the matching pair. The diagonal elements of the symmetric matrix are set to their maximum values ​​to represent that each matching pair is completely compatible with itself, thereby constructing a geometric compatibility matrix. This geometric compatibility matrix encodes the topological constraints between all candidate matching pairs and serves as the input data source for eigenvalue decomposition and weight extraction in the subsequent step S103.

[0030] Step S103: Perform eigenvalue decomposition on the geometric compatibility matrix and extract the principal eigenvectors to obtain a globally consistent weight sequence. Use the globally consistent weight sequence as coefficients to construct a weighted covariance matrix and perform singular value decomposition to obtain a rotation matrix and a translation vector. Apply the rotation matrix and the translation vector to the source coordinate dataset to complete the corridor coordinate alignment and output the over-limit vegetation positioning information for use in the pruning operation.

[0031] This embodiment performs eigenvalue decomposition on the geometric compatibility matrix output in step S102 to extract global consistency information. The eigenvalue decomposition process decomposes the geometric compatibility matrix into a set of eigenvalues ​​and a corresponding set of eigenvectors. In this embodiment, the eigenvalue with the largest value is located from the eigenvalue set, and the eigenvector corresponding to this largest eigenvalue is read as the principal eigenvector. The magnitude of each component in the principal eigenvector reflects the degree to which the corresponding matching pair belongs to the globally consistent correct set. A larger component value indicates that the matching pair generally has high compatibility with other correct matches, while a smaller component value indicates that the matching pair has geometric contradictions with most matches.

[0032] Based on the principal feature vector, this embodiment performs normalization processing to generate a globally consistent weight sequence. This embodiment reads the original values ​​of each component in the principal feature vector, maps each component to the interval between zero and one to obtain normalized weight values, and arranges all normalized weight values ​​in their original index order to form a globally consistent weight sequence. Each weight value in the globally consistent weight sequence corresponds one-to-one with each matching pair in the seed correspondence set. Matching pairs with weight values ​​close to one are determined as high-confidence correct matches, while matching pairs with weight values ​​close to zero are determined as hidden incorrect matches.

[0033] Accordingly, this embodiment uses the globally consistent weight sequence as coefficients to construct a weighted cross-covariance matrix to solve for the rotation matrix. For each matching pair in the seed correspondence set, the source data point coordinates and target data point coordinates are weighted with their corresponding weight values. After calculating the weighted centroid coordinates, each coordinate point is decentroided. In this embodiment, the weighted cross-covariance matrix is ​​constructed using the weight values ​​as coefficients to pair the decentroided source data point coordinates and target data point coordinates. This matrix encodes the weighted spatial correlation between the source coordinate dataset and the reference coordinate dataset.

[0034] After the weighted cross-covariance matrix is ​​constructed, this embodiment performs singular value decomposition (SVD) to extract the rotation matrix and translation vector. SVD decomposes the weighted cross-covariance matrix into three components: a left singular matrix, a singular value diagonal matrix, and a right singular matrix. In this embodiment, the right singular matrix is ​​multiplied by the transpose of the left singular matrix to obtain the rotation matrix. Based on the rotation matrix and the aforementioned weighted centroid coordinates, this embodiment calculates the difference between the weighted centroid of the target dataset and the weighted centroid of the rotated source dataset to obtain the translation vector. The rotation matrix and the translation vector together constitute the rigid body transformation parameters.

[0035] Based on the rotation matrix and the translation vector, this embodiment performs a rigid body transformation on the source coordinate dataset to complete the corridor coordinate alignment. The coordinates of each data point in the source coordinate dataset are multiplied by the rotation matrix to obtain the rotated coordinates. The rotated coordinates are then added to the translation vector to obtain the transformed coordinates. All transformed coordinates are converged to form the aligned corridor coordinate dataset. The aligned corridor coordinate dataset and the reference coordinate dataset are in the same spatial coordinate system, allowing for unified spatial analysis.

[0036] After the aligned corridor coordinate dataset is generated, this embodiment performs over-limit vegetation identification and location information extraction. This embodiment distinguishes vegetation data points from guide wire data points in the aligned corridor coordinate dataset based on geometric features and spatial distribution patterns. For each vegetation data point, the spatial distance between it and the nearest guide wire data point is calculated to obtain the tree-line distance value. The tree-line distance value is compared with a preset safety distance threshold, and vegetation data points with tree-line distance values ​​less than the threshold are selected as over-limit vegetation points. This embodiment extracts the three-dimensional spatial coordinates of each over-limit vegetation point to form over-limit vegetation location information. This over-limit vegetation location information is output to the upper-level operation and maintenance management system for use in UAV trimming operation path planning.

[0037] As described above, the global alignment method for coordinate data in inspection scenarios provided in this application can construct a feature description vector set through filtering and noise reduction and screening of local curvature key points. It combines spatial distance constraints and normal vector consistency constraints to screen seed correspondences step by step and construct a geometric compatibility matrix. It extracts a global consistency weight sequence through eigenvalue decomposition to construct a weighted covariance matrix and performs singular value decomposition to solve the rotation matrix and translation vector to complete the corridor coordinate alignment. Finally, it outputs the location information of over-limit vegetation for use in pruning operations. This method effectively solves the shortcomings of traditional technologies in coordinate data preprocessing and feature extraction, robust screening of mismatches and geometric compatibility modeling, and global consistency weight solution and alignment result output. It provides technical support for intelligent global alignment of coordinate data and accurate positioning of over-limit vegetation in the UAV inspection scenario of power transmission corridors.

[0038] In one embodiment of the global coordinate data alignment method for inspection scenarios in this application, it may further include the following: Step S201: Calculate the average spatial distance between each data point in the source coordinate dataset and the reference coordinate dataset and its nearest neighbor data point to obtain the neighborhood distance value. Compare the neighborhood distance value with the global average distance and determine noise points based on the standard deviation threshold condition. Remove the data points determined to be noise points from the source coordinate dataset and the reference coordinate dataset to obtain the denoised coordinate dataset. Step S202: Divide the three-dimensional space covered by the noise reduction coordinate dataset into uniform voxel grid cells, calculate the centroid coordinates of all data points in each voxel grid cell to obtain the grid centroid point, and aggregate the grid centroid points to form a standardized corridor coordinate dataset.

[0039] This embodiment reads the 3D spatial coordinates of each data point from the source coordinate dataset and reference coordinate dataset acquired by the UAV's onboard laser sensor, and performs neighborhood distance statistics to identify noise points. For any data point in the source coordinate dataset, this embodiment searches for its nearest neighboring data points in 3D space, calculates the Euclidean distance between the data point and each neighboring data point, and obtains the arithmetic mean. This arithmetic mean is recorded as the neighborhood distance value of the data point. The neighborhood distance value characterizes the sparsity of the data point in the local space. The neighborhood distance values ​​of normal data points are distributed within a relatively concentrated numerical range, while noise points exhibit significantly larger neighborhood distance values ​​due to their distance from the main point cloud.

[0040] After calculating the neighborhood distance values ​​of all data points in the source coordinate dataset, this embodiment performs global statistical analysis to establish a noise judgment benchmark. All neighborhood distance values ​​are aggregated into a sample set, and the arithmetic mean of this sample set is calculated to obtain the global average distance. The standard deviation of this sample set is then calculated to obtain the distance dispersion metric. In this embodiment, the global average distance and the distance dispersion metric are input into the noise judgment criteria. The noise judgment criteria have an upper bound based on a standard deviation threshold; data points whose neighborhood distance values ​​exceed a certain multiple of the global average distance by a certain number of standard deviations are judged as noise points.

[0041] Accordingly, this embodiment performs the same neighborhood distance calculation and noise determination process on the reference coordinate dataset. The neighborhood distance value of each data point in the reference coordinate dataset is also calculated and compared with its independent global average distance. Data points that meet the standard deviation threshold are determined to be noise points. In this embodiment, data points determined to be noise points in both the source and reference coordinate datasets are removed from their respective datasets. The removal operation eliminates isolated outliers caused by factors such as sensor multipath reflection, electromagnetic interference, and bird flyovers. The remaining data points are aggregated to form a denoised coordinate dataset. The denoised coordinate dataset includes two parts: a denoised subset of the source coordinates and a denoised subset of the reference coordinates.

[0042] After the denoised coordinate dataset is generated, this embodiment performs voxel rasterization downsampling to compress the data size. This embodiment calculates the extreme range of coordinates in three-dimensional space for the denoised coordinate dataset, and divides this spatial range into regularly arranged uniform voxel raster units, using a preset raster side length as the unit. Each voxel raster unit is a cubic spatial region with equal side lengths, and adjacent raster units have no overlap or gaps, collectively covering the complete spatial bounding box of the denoised coordinate dataset.

[0043] Based on the voxel grid cell division results, this embodiment performs centroid calculation on the data points within each grid cell. For any non-empty voxel grid cell, the coordinates of all data points falling within the spatial range of that cell are read, and the arithmetic mean of the coordinate components is calculated for each of the three coordinate axes. The mean values ​​of the three coordinate components are combined to obtain the grid centroid of that grid cell. The grid centroid is located at the geometric center of the data points within the grid cell, and a single coordinate point represents the spatial location information of this local area.

[0044] In this embodiment, the centroids of all non-empty voxel grid cells are aggregated to form a standardized corridor coordinate dataset. The number of data points in the standardized corridor coordinate dataset is determined by both the grid side length and the original data spatial density, compressing the point cloud size to a controllable range while preserving the macroscopic geometry of the transmission corridor. This standardized corridor coordinate dataset serves as the input data source for the subsequent step S203, performing local curvature calculation and key data point selection. Its compressed data size reduces the computational overhead of constructing the neighborhood covariance matrix and eigenvalue decomposition.

[0045] In one embodiment of the global coordinate data alignment method for inspection scenarios in this application, it may further include the following: Step S301: Extract neighborhood point sets for each data point in the standardized corridor coordinate dataset and construct a neighborhood covariance matrix. Perform eigenvalue decomposition on the neighborhood covariance matrix to obtain the minimum and maximum eigenvalues. Calculate the ratio of the minimum eigenvalue to the maximum eigenvalue to obtain the local curvature value. Compare the local curvature value with a preset local curvature threshold condition and select data points whose curvature values ​​meet the threshold condition as key data points. Step S302: Establish a local reference coordinate system with each key data point as the center and divide the neighborhood space into multiple sub-regions. Statistically calculate the radial distance distribution, the angle distribution between the normal vectors, and the relative height distribution of the data points in each sub-region to obtain a sub-region statistical histogram. Concatenate the statistical histograms of each sub-region in a fixed order to form a feature description vector. Aggregate the feature description vectors corresponding to all the key data points to obtain a feature description vector set.

[0046] This embodiment reads the three-dimensional spatial coordinates of each data point from the standardized corridor coordinate dataset output in step S202, and performs neighborhood point set extraction to construct the basis for local geometric analysis. For any data point in the standardized corridor coordinate dataset, this embodiment sets a spherical search range centered on that data point, and aggregates all other data points falling within this spherical range to form the neighborhood point set of that data point. The size of the neighborhood point set is determined by the search radius and the local point cloud density, and contains a sufficient number of neighborhood points to support the stable estimation of the subsequent covariance matrix.

[0047] Based on the neighborhood point set, this embodiment constructs a neighborhood covariance matrix to encode local geometric distribution features. The arithmetic mean of the coordinates of each data point in the neighborhood point set is calculated to obtain the neighborhood centroid coordinates. Subtracting the neighborhood centroid coordinates from the coordinates of each neighborhood point yields decentroided coordinates. This embodiment constructs a 3x3 covariance matrix for all decentroided coordinates. Each element of the neighborhood covariance matrix is ​​filled with the covariance values ​​of each component of the decentroided coordinates. This matrix encodes the distribution pattern and principal direction information of the neighborhood point set in three-dimensional space.

[0048] After the neighborhood covariance matrix is ​​constructed, this embodiment performs eigenvalue decomposition to extract geometric curvature information. The eigenvalue decomposition process decomposes the neighborhood covariance matrix into three eigenvalues ​​and three corresponding eigenvectors. This embodiment identifies the smallest and largest eigenvalues ​​from the three eigenvalues. The smallest eigenvalue corresponds to the dispersion of the neighborhood point set along the normal vector direction, and the largest eigenvalue corresponds to the extension of the neighborhood point set along the principal tangential direction. The relative magnitude of these two values ​​reflects the curvature characteristics of the local surface.

[0049] Accordingly, this embodiment calculates the ratio of the minimum eigenvalue to the maximum eigenvalue to obtain the local curvature value. When the neighborhood point set is distributed in an approximately planar region, the minimum eigenvalue approaches zero while the maximum eigenvalue remains relatively large, and the ratio of the two approaches zero, representing a low-curvature flat region. When the neighborhood point set is distributed in a sharp turning region, the minimum eigenvalue increases relatively, and the increased ratio of the two represents a high-curvature geometrically significant region. This embodiment compares the local curvature value of each data point with a preset local curvature threshold condition, and selects data points whose local curvature values ​​exceed the threshold condition as key data points. These key data points correspond to geometrically significant locations such as the intersection of tower crossarms, the extreme point of conductor sag, and the bifurcation of vegetation branches.

[0050] After the key data points are selected, this embodiment establishes a local reference coordinate system for each key data point to support rotation-invariant feature descriptions. Using the key data point as the origin, the eigenvector corresponding to the largest eigenvalue of its neighborhood covariance matrix is ​​set as the principal axis direction of the local coordinate system, and the eigenvector corresponding to the smallest eigenvalue is set as the normal vector direction. The direction of the third coordinate axis is determined according to the right-hand rule. This local reference coordinate system ensures that the calculation of the feature description vector is unaffected by rotation of the global coordinate system.

[0051] Based on the aforementioned local reference coordinate system, this embodiment divides the neighborhood space of key data points into multiple sub-regions to construct a spatial histogram. This embodiment divides the neighborhood space into several concentric spherical shells along the radial distance dimension, each shell layer into several latitude zones along the elevation dimension, and each latitude zone into several longitude sectors along the azimuth dimension. These three dimensions are combined to form multiple sub-regions. For each sub-region, the radial distance distribution, the angle distribution between the data point normal vector and the key data point normal vector, and the height difference distribution relative to the key data point are statistically analyzed. The statistical results are encoded into a sub-region statistical histogram.

[0052] In this embodiment, the statistical histograms of each sub-region are concatenated in a fixed order of radial layer index, latitude zone index, and longitude sector index to form the feature description vector of the key data point. The feature description vector is a numerical sequence of fixed dimensions, with each dimension representing the geometric and statistical characteristics of the corresponding sub-region. This embodiment repeats the complete process of establishing a local reference coordinate system, dividing sub-regions, calculating statistical histograms, and concatenating vectors for all key data points, aggregating the feature description vectors corresponding to all key data points to obtain a feature description vector set. This feature description vector set serves as the input data source for cross-dataset similarity matching in the subsequent step S401.

[0053] In one embodiment of the global coordinate data alignment method for inspection scenarios in this application, it may further include the following: Step S401: Search for the feature description vector with the smallest Euclidean distance in the feature description vector set of the reference coordinate dataset for the feature description vector of each key data point in the source coordinate dataset to obtain the nearest neighbor match and the second nearest neighbor match. Calculate the ratio of the nearest neighbor match distance to the second nearest neighbor match distance to obtain the distance ratio value. Compare the distance ratio value with the preset ratio test threshold condition and retain the matching pairs that meet the threshold condition to form an initial correspondence set. Step S402: Calculate the spatial Euclidean distance between the source data point and the target data point for each matching pair in the initial correspondence set to obtain the matching spatial distance. Compare the matching spatial distance with the preset maximum displacement threshold condition and remove matching pairs that exceed the threshold condition to obtain a distance-filtered set. Calculate the angle between the source data point normal vector and the target data point normal vector for each matching pair in the distance-filtered set to obtain the normal vector angle value. Compare the normal vector angle value with the preset angle threshold condition and retain the matching pairs that meet the threshold condition to form a seed correspondence set.

[0054] In this embodiment, the feature description vectors of key data points corresponding to the source coordinate dataset and the reference coordinate dataset are read from the feature description vector set output in step S302, and cross-dataset similarity matching is performed to establish candidate correspondences. For any key data point in the source coordinate dataset, this embodiment extracts its feature description vector and calculates the Euclidean distance for each feature description vector in the reference coordinate dataset. The Euclidean distance quantifies the degree of difference between the two feature description vectors in the high-dimensional feature space; the smaller the distance value, the more similar the local geometric features of the two key data points.

[0055] Based on the Euclidean distance calculation results, this embodiment identifies nearest neighbor matching and second nearest neighbor matching to construct a discriminative criterion. The key data point corresponding to the feature description vector with the smallest Euclidean distance to the source data point's feature description vector in the reference coordinate dataset is denoted as the nearest neighbor matching point, and the key data point corresponding to the feature description vector with the second smallest Euclidean distance is denoted as the second nearest neighbor matching point. The nearest neighbor matching distance and the second nearest neighbor matching distance are their corresponding Euclidean distance values, and this embodiment calculates their ratio to obtain the distance ratio. The distance ratio reflects the discriminative power of nearest neighbor matching relative to second nearest neighbor matching; a smaller ratio indicates that nearest neighbor matching has a significant advantage in the feature space.

[0056] Accordingly, this embodiment compares the distance ratio with a preset ratio test threshold to filter fuzzy matches. When the distance ratio is less than the ratio test threshold, the distance difference between the nearest neighbor and the second nearest neighbor in the feature space is sufficiently significant, and this embodiment determines that the match has reliable discriminative power and retains it. When the distance ratio exceeds the ratio test threshold, there are multiple similar candidate points in the feature space, and the reliability of the matching result is insufficient; this embodiment discards the match. All matching pairs that pass the ratio test are aggregated to form an initial correspondence set, and each matching pair in the initial correspondence set records the corresponding mapping between the source data point index and the target data point index.

[0057] After the initial correspondence set is generated, this embodiment performs spatial distance constraint filtering to eliminate physically impossible erroneous matches. For each matching pair in the initial correspondence set, this embodiment reads the three-dimensional spatial coordinates of the source data point in the source coordinate dataset and the three-dimensional spatial coordinates of the target data point in the reference coordinate dataset, and calculates the spatial Euclidean distance between them to obtain the matching spatial distance. The matching spatial distance represents the apparent displacement of the same physical location in two acquisitions, and this displacement should be within the range of UAV positioning error and corridor deformation.

[0058] This embodiment compares the matched spatial distance with a preset maximum displacement threshold to perform filtering. The maximum displacement threshold is set based on a combination of the UAV carrier's positioning accuracy and the structural stability of the power transmission corridor. Matching pairs whose spatial distance exceeds this threshold exceed a reasonable physical range in terms of spatial displacement. In this embodiment, matching pairs that exceed the threshold are determined to be cross-regional erroneous matches and are removed from the set. The remaining matching pairs are then aggregated to form a distance-filtered set.

[0059] Based on the distance-filtered set, this embodiment continues to perform normal vector consistency constraint filtering to filter out erroneous matches with inconsistent local geometric orientations. Rigid body transformations maintain the angular relationship between surface normal vectors, and the normal vectors of the source and target data points in a correctly matched pair should be approximately parallel. In this embodiment, for each matching pair in the distance-filtered set, the normal vectors of the source and target data points are read, and the spatial angle between the two normal vectors is calculated to obtain the normal vector angle value.

[0060] This embodiment compares the included angle value of the normal vector with a preset included angle threshold condition to complete the final screening. Matching pairs whose included angle value exceeds the included angle threshold condition have significant differences in the orientation of the local curved surface, which does not meet the geometric constraint characteristics of rigid body transformation. This constraint has a targeted filtering effect on cross-line mismatches between parallel conductors in the power transmission corridor scenario. The normal vector of the conductor surface points radially outward, and mismatches between adjacent conductors are effectively identified due to the difference in the orientation of the normal vector. This embodiment retains and aggregates the matching pairs that meet the included angle threshold condition to form a seed correspondence set. The seed correspondence set serves as the input data source for constructing the geometric compatibility matrix in the subsequent step S501.

[0061] In one embodiment of the global coordinate data alignment method for inspection scenarios in this application, it may further include the following: Step S501: For any two matching pairs in the seed correspondence set, calculate the spatial distance between the corresponding two points in the source coordinate dataset to obtain the source end distance, calculate the spatial distance between the corresponding two points in the reference coordinate dataset to obtain the target end distance, and take the absolute value of the difference between the source end distance and the target end distance to obtain the distance difference. Step S502: Input the distance difference into a kernel function that monotonically decreases with the distance difference to calculate the compatibility score. Write the compatibility scores of all matching pairs in the seed correspondence set into a symmetric matrix according to the row and column indices to obtain the geometric compatibility matrix. Set the diagonal elements of the geometric compatibility matrix to the maximum value to indicate that each matching pair is completely compatible with itself.

[0062] This embodiment reads all matching pair information from the seed correspondence set output in step S402 and performs pairwise distance calculation to quantify the geometric compatibility between matching pairs. For any first and second matching pairs selected from the seed correspondence set, this embodiment extracts the three-dimensional spatial coordinates of the source data points in the first matching pair and the three-dimensional spatial coordinates of the source data points in the second matching pair in the source coordinate dataset, respectively, and calculates the spatial Euclidean distance between the two source data points to obtain the source-end distance. The source-end distance represents the spatial interval between the corresponding physical locations of the two matching pairs in the source coordinate dataset.

[0063] Based on the selection of the same matching pairs, this embodiment performs corresponding distance calculations in the reference coordinate dataset. The three-dimensional spatial coordinates of the target data points in the first matching pair and the second matching pair in the reference coordinate dataset are extracted, and the spatial Euclidean distance between these two target data points is calculated to obtain the target end distance. The target end distance represents the spatial interval between the corresponding physical locations of the two matching pairs in the reference coordinate dataset. If both matching pairs are correct matches, the source end distance and the target end distance should be equal.

[0064] Accordingly, this embodiment calculates the distance difference based on the distance preservation property of rigid body transformation to verify the geometric consistency between matching pairs. Rigid body transformation only includes two types of operations: rotation and translation. It does not change the spatial distance between any two points. Therefore, the difference between the source end distance and the target end distance of a correctly matched pair should be close to zero. In this embodiment, the absolute value of the difference between the source end distance and the target end distance is taken to obtain the distance difference value. The distance difference value quantifies the degree to which the two matching pairs deviate from the rigid body transformation constraints. The smaller the distance difference value, the more likely the two matching pairs belong to the same correctly matched set.

[0065] After the distance difference is calculated, this embodiment inputs it into a kernel function to generate a normalized compatibility score. The kernel function is a mapping function that monotonically decreases with the distance difference; it outputs the maximum compatibility score when the distance difference is zero, and the output score rapidly decays and approaches zero as the distance difference increases. This embodiment uses a Gaussian kernel function, dividing the square of the distance difference by a bandwidth parameter and taking the negative exponent. The bandwidth parameter is set according to the expected measurement error range in the power transmission corridor scenario. This kernel function maps the distance difference to a compatibility score within the zero-to-one interval, facilitating the numerical stability of subsequent matrix operations.

[0066] Based on the aforementioned compatibility score calculation method, this embodiment performs pairwise pairing calculations on all matching pairs in the seed correspondence set to construct a geometric compatibility matrix. Assuming the seed correspondence set contains several matching pairs, this embodiment establishes a square matrix structure where the number of rows and columns are equal to the number of matching pairs. For any two matching pairs corresponding to any row and column indices, their distance difference is calculated and input into a kernel function to obtain a compatibility score, which is then written into the corresponding position in the matrix. Since the distance difference calculation process is insensitive to the order of the two matching pairs, the geometric compatibility matrix exhibits a symmetrical structure.

[0067] In this embodiment, the diagonal elements of the geometric compatibility matrix are set to the maximum value to characterize the self-compatibility property. The diagonal positions correspond to the compatibility calculation of the same matching pair with itself; the distance difference is always zero, and the kernel function outputs the maximum score. This embodiment explicitly assigns a value of one to the diagonal elements to ensure the normalization of the matrix values. The geometric compatibility matrix encodes the pairwise geometric constraints between all matching pairs in the seed correspondence set. The geometric compatibility matrix serves as the input data source for eigenvalue decomposition and principal eigenvector extraction in the subsequent step S601.

[0068] In one embodiment of the global coordinate data alignment method for inspection scenarios in this application, it may further include the following: Step S601: Perform eigenvalue decomposition on the geometric compatibility matrix to obtain an eigenvalue set and a corresponding eigenvector set. Extract the largest eigenvalue from the eigenvalue set and read the eigenvector corresponding to the largest eigenvalue as the principal eigenvector. Normalize each component in the principal eigenvector to the interval between zero and one to obtain a globally consistent weight sequence. Step S602: Use each weight value in the global consistency weight sequence as the coefficient of the corresponding matching pair. Based on the coefficient, construct a weighted cross covariance matrix for the source data point coordinates and target data point coordinates in the seed correspondence set. Perform singular value decomposition on the weighted cross covariance matrix to obtain a rotation matrix. Calculate the translation vector based on the rotation matrix and the weighted centroid coordinates.

[0069] This embodiment performs eigenvalue decomposition on the geometric compatibility matrix output in step S502 to extract global geometric consistency information. Eigenvalue decomposition decomposes the geometric compatibility matrix into a set of eigenvalues ​​and a corresponding set of eigenvectors. Each eigenvalue represents the scaling strength of the matrix along the direction of the corresponding eigenvector, and each eigenvector is a unit vector corresponding one-to-one with an eigenvalue. This embodiment compares the values ​​of each eigenvalue in the eigenvalue set and identifies the eigenvalue with the largest value as the principal eigenvalue. The principal eigenvalue corresponds to the most prominent structural pattern in the geometric compatibility matrix.

[0070] Based on the principal feature value, this embodiment reads its corresponding feature vector as the principal feature vector to encode the global consistency information of the matching pairs. The dimension of the principal feature vector is equal to the number of matching pairs in the seed correspondence set, and each component corresponds one-to-one with each matching pair in index order. Positions with larger component values ​​in the principal feature vector correspond to high-confidence matching pairs that are generally compatible with other correct matches, while positions with smaller component values ​​correspond to suspicious matching pairs that have geometric contradictions with most matches. This distribution characteristic stems from the clustering effect of the geometric compatibility matrix on the subset of correct matches.

[0071] Accordingly, this embodiment performs normalization processing on the principal feature vector to generate a globally consistent weight sequence. The original values ​​of each component in the principal feature vector are read, and each component value is divided by the maximum value of the component, or an interval mapping method is used to map all component values ​​to the interval between zero and one to obtain normalized weight values. In this embodiment, all normalized weight values ​​are arranged in their original index order to form a globally consistent weight sequence. Each weight value in the globally consistent weight sequence quantifies the confidence level of the corresponding matching pair belonging to the globally consistent correct set.

[0072] After the globally consistent weight sequence is generated, this embodiment uses it as a set of seed correspondences for coefficient pairs to construct a weighted cross-covariance matrix. First, the weighted centroid coordinates are calculated. The coordinates of the source data points for each matching pair in the source coordinate dataset are weighted and summed using the corresponding weight values ​​as coefficients. The summation result is divided by the total weights to obtain the source-side weighted centroid coordinates. The same weighted summation and normalization operations are performed on the target data point coordinates for each matching pair in the reference coordinate dataset to obtain the target-side weighted centroid coordinates. These weighted centroid coordinates ensure that high-confidence matching pairs dominate the centroid calculation.

[0073] Based on the weighted centroid coordinates, this embodiment performs decentroiding on the coordinates of each data point and constructs a weighted cross-covariance matrix. The decentized source coordinates are obtained by subtracting the source-end weighted centroid coordinates from the coordinates of each source data point, and the decentized target coordinates are obtained by subtracting the target-end weighted centroid coordinates from the coordinates of each target data point. This embodiment performs a weighted summation on the outer product of the decentized source coordinates and the decentized target coordinates using weight values ​​as coefficients. The summation result is organized into a 3x3 matrix to obtain the weighted cross-covariance matrix, which encodes the weighted spatial correlation between the source coordinate dataset and the reference coordinate dataset.

[0074] This embodiment performs singular value decomposition (SVD) on the weighted cross-covariance matrix to extract the rotation matrix and translation vector. SVD decomposes the weighted cross-covariance matrix into a product of three components: a left singular matrix, a singular value diagonal matrix, and the transpose of the right singular matrix. In this embodiment, the right singular matrix is ​​multiplied by the transpose of the left singular matrix to obtain the rotation matrix. This rotation matrix is ​​a 3x3 orthogonal matrix with a determinant of positive one, representing the rotation transformation required to align the source coordinate dataset to the reference coordinate dataset. Based on the rotation matrix, this embodiment calculates the difference between the weighted centroid coordinates of the target end and the weighted centroid coordinates of the source end after rotation to obtain the translation vector. The rotation matrix and the translation vector together serve as the rigid body transformation parameters for the coordinate transformation performed in the subsequent step S701.

[0075] In one embodiment of the global coordinate data alignment method for inspection scenarios in this application, it may further include the following: Step S701: Perform matrix multiplication on the coordinates of each data point in the source coordinate dataset and the rotation matrix to obtain the rotated coordinates; perform vector addition on the rotated coordinates and the translation vector to obtain the transformed coordinates; and aggregate all the transformed coordinates to form the aligned corridor coordinate dataset. Step S702: Identify vegetation data points and traverse data points from the aligned corridor coordinate dataset, calculate the spatial distance between each vegetation data point and the nearest traverse data point to obtain the tree-line distance value, compare the tree-line distance value with a preset safe distance threshold condition, and filter vegetation data points that are less than the threshold condition as over-limit vegetation points, extract the spatial coordinates of each over-limit vegetation point to form over-limit vegetation positioning information, and output it for use in UAV trimming operation path planning.

[0076] In this embodiment, rigid body transformation parameters are read from the rotation matrix and translation vector output in step S602, and coordinate transformation is performed on the source coordinate dataset to complete the corridor coordinate alignment. For any data point in the source coordinate dataset, this embodiment extracts its three-dimensional spatial coordinates and organizes them into a three-dimensional column vector. The rotation matrix and the column vector are then multiplied to obtain the rotated coordinates. The matrix multiplication operation rotates the source data point coordinates around the origin to an orientation consistent with the reference coordinate dataset, and the rotation angle and rotation axis are implicitly encoded by the rotation matrix.

[0077] Based on the rotated coordinates, this embodiment continues to perform a translation transformation to complete the rigid body transformation process. A vector addition operation is performed on the rotated coordinates and the translation vector, adding the three components of the rotated coordinates to the corresponding components of the translation vector to obtain the transformed coordinates. This vector addition operation moves the rotated source data point coordinates along the translation vector direction to a spatial position aligned with the reference coordinate dataset; the translation distance and direction are explicitly represented by the translation vector.

[0078] This embodiment repeatedly performs matrix multiplication and vector addition operations on all data points in the source coordinate dataset, converging all transformed coordinates to form an aligned corridor coordinate dataset. The coordinates of each data point in the aligned corridor coordinate dataset are in the same spatial coordinate system as the reference coordinate dataset, eliminating coordinate system differences caused by positioning drift and attitude deviations between UAV flights. This alignment result allows for overlay analysis and change detection of power transmission corridor point cloud data collected across flights within a unified coordinate framework.

[0079] After the aligned corridor coordinate dataset is generated, this embodiment performs land cover classification to identify vegetation data points and traverse data points. This embodiment distinguishes different land cover types based on the geometric distribution characteristics and spatial location patterns of the data points. Traverse data points exhibit an approximately parabolic sag distribution along the erection direction and are located within a fixed height range, while vegetation data points show an irregular cluster distribution with a large range of height variations. Local geometric features are extracted from each data point in the aligned corridor coordinate dataset, and combined with elevation distribution information, data points conforming to traverse geometric features are marked as traverse data points, and data points conforming to vegetation geometric features are marked as vegetation data points.

[0080] Based on the classification results of the vegetation data points and conductor data points, this embodiment calculates the tree-line distance to assess the degree of safety threat posed by vegetation to the conductors. For each vegetation data point, this embodiment searches for the conductor data point with the smallest spatial distance among all conductor data points, and calculates the spatial Euclidean distance between the vegetation data point and the nearest conductor data point to obtain the tree-line distance value. The tree-line distance value characterizes the closest spatial interval between the vegetation canopy and the transmission conductor, and this interval value is directly related to the discharge risk and the line safety margin.

[0081] This embodiment compares the treeline distance values ​​of each vegetation data point with a preset safety distance threshold to filter out vegetation points exceeding the limit. The safety distance threshold is set based on the safety clearance requirements corresponding to the voltage level. Vegetation data points with treeline distance values ​​less than this threshold encroach on the conductor's safety clearance area and are thus identified as vegetation points exceeding the limit. This embodiment extracts the three-dimensional spatial coordinates of each vegetation point exceeding the limit to form vegetation location information. This location information includes the longitude coordinates, latitude coordinates, altitude, and corresponding treeline distance values ​​of the vegetation points. The vegetation location information is output to the upper-level operation and maintenance management system for use by the UAV trimming operation path planning module to generate autonomous flight routes and trimming operation instructions.

[0082] To effectively address the shortcomings of traditional technologies in coordinate data preprocessing and feature extraction, robust screening of mismatches and geometric compatibility modeling, and global consistency weight calculation and alignment result output, and to provide technical support for intelligent global alignment of coordinate data and precise positioning of over-limit vegetation in power transmission corridor UAV inspection scenarios, this application provides an embodiment of a coordinate data global alignment device for inspection scenarios, which implements all or part of the aforementioned coordinate data global alignment method for inspection scenarios. See [link to embodiment]. Figure 2 The global coordinate data alignment device for inspection scenarios specifically includes the following components: Data filtering module 10 is used to acquire source coordinate dataset and reference coordinate dataset collected by UAV on the power transmission corridor, perform filtering, noise reduction and downsampling on the source coordinate dataset and the reference coordinate dataset to obtain a standardized corridor coordinate dataset, and filter key data points from the standardized corridor coordinate dataset based on the local curvature threshold condition and calculate feature descriptors to obtain a feature description vector set. Data alignment module 20 is used to perform cross-dataset similarity matching on the feature description vector set to obtain an initial correspondence set, and to perform spatial distance constraint filtering and normal vector consistency constraint filtering on the initial correspondence set to obtain a seed correspondence set. Based on the rigid body transformation distance preservation property, the compatibility score of each matching pair in the seed correspondence set is calculated to obtain a geometric compatibility matrix. The pruning module 30 is used to perform eigenvalue decomposition on the geometric compatibility matrix and extract the principal eigenvector to obtain a globally consistent weight sequence. The globally consistent weight sequence is used as coefficients to construct a weighted covariance matrix and perform singular value decomposition to obtain a rotation matrix and a translation vector. The rotation matrix and the translation vector are applied to the source coordinate dataset to complete the corridor coordinate alignment and output the over-limit vegetation positioning information for the pruning operation to call.

[0083] As described above, the coordinate data global alignment device for inspection scenarios provided in this application can construct a feature description vector set through filtering and noise reduction and screening of local curvature key points. It combines spatial distance constraints and normal vector consistency constraints to screen seed correspondences step by step and construct a geometric compatibility matrix. It extracts a global consistency weight sequence through eigenvalue decomposition to construct a weighted covariance matrix and performs singular value decomposition to solve the rotation matrix and translation vector to complete the corridor coordinate alignment. Finally, it outputs the location information of over-limit vegetation for use in pruning operations. It effectively solves the shortcomings of traditional technologies in coordinate data preprocessing and feature extraction, robust screening of mismatches and geometric compatibility modeling, and global consistency weight solution and alignment result output. It provides technical support for intelligent global alignment of coordinate data and accurate positioning of over-limit vegetation in the UAV inspection scenario of power transmission corridors.

[0084] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the global alignment method for coordinate data in an inspection scenario.

[0085] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned global coordinate data alignment method for inspection scenarios.

[0086] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the aforementioned global coordinate data alignment method for inspection scenarios.

[0087] In this embodiment of the invention, a feature description vector set is constructed by filtering and denoising and selecting key points of local curvature. The seed correspondence is selected step by step by combining spatial distance constraints and normal vector consistency constraints, and a geometric compatibility matrix is ​​constructed. The global consistency weight sequence is extracted by eigenvalue decomposition to construct a weighted covariance matrix, and singular value decomposition is performed to solve the rotation matrix and translation vector to complete the corridor coordinate alignment. Finally, the location information of over-limit vegetation is output for use in the pruning operation. This effectively solves the shortcomings of traditional technologies in coordinate data preprocessing and feature extraction, robust screening of mismatches and geometric compatibility modeling, and global consistency weight solution and alignment result output. It provides technical support for intelligent global alignment of coordinate data and accurate positioning of over-limit vegetation in the scenario of UAV inspection of power transmission corridors.

[0088] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0089] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0090] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0091] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0092] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for global alignment of coordinate data in inspection scenarios, characterized in that, The method includes: The source coordinate dataset and reference coordinate dataset collected by the UAV on the power transmission corridor are obtained. The source coordinate dataset and the reference coordinate dataset are filtered, denoised and downsampled to obtain a standardized corridor coordinate dataset. Key data points are selected from the standardized corridor coordinate dataset based on the local curvature threshold condition and feature descriptors are calculated to obtain a feature description vector set. The feature description vector set is subjected to cross-dataset similarity matching to obtain an initial correspondence set. Spatial distance constraint filtering and normal vector consistency constraint filtering are performed on the initial correspondence set in sequence to obtain a seed correspondence set. Based on the rigid body transformation distance preservation property, the compatibility score of each matching pair in the seed correspondence set is calculated to obtain a geometric compatibility matrix. The geometric compatibility matrix is ​​subjected to eigenvalue decomposition and the principal eigenvectors are extracted to obtain a globally consistent weight sequence. The globally consistent weight sequence is used as coefficients to construct a weighted covariance matrix and singular value decomposition is performed to obtain a rotation matrix and a translation vector. The rotation matrix and the translation vector are applied to the source coordinate dataset to complete the corridor coordinate alignment and output the over-limit vegetation location information for use in the pruning operation.

2. The global coordinate data alignment method for inspection scenarios according to claim 1, characterized in that, The process of acquiring source coordinate datasets and reference coordinate datasets collected by the UAV from the power transmission corridor, and performing filtering, noise reduction, and downsampling on the source coordinate datasets and the reference coordinate datasets to obtain a standardized corridor coordinate dataset includes: For each data point in the source coordinate dataset and the reference coordinate dataset, calculate the average spatial distance between it and its nearest neighbor data point in the neighborhood to obtain the neighborhood distance value. Compare the neighborhood distance value with the global average distance and determine noise points based on the standard deviation threshold condition. Remove the data points determined to be noise points from the source coordinate dataset and the reference coordinate dataset to obtain the denoised coordinate dataset. The three-dimensional space covered by the denoised coordinate dataset is divided into uniform voxel grid cells. The centroid coordinates of all data points in each voxel grid cell are calculated to obtain the grid centroid point. The grid centroid points are then aggregated to form a standardized corridor coordinate dataset.

3. The global coordinate data alignment method for inspection scenarios according to claim 1, characterized in that, The process of selecting key data points from the standardized corridor coordinate dataset based on the local curvature threshold condition and calculating feature descriptors to obtain a feature description vector set includes: For each data point in the standardized corridor coordinate dataset, a neighborhood point set is extracted and a neighborhood covariance matrix is ​​constructed. Eigenvalue decomposition is performed on the neighborhood covariance matrix to obtain the minimum eigenvalue and the maximum eigenvalue. The ratio of the minimum eigenvalue to the maximum eigenvalue is calculated to obtain the local curvature value. The local curvature value is compared with a preset local curvature threshold condition, and data points whose curvature values ​​meet the threshold condition are selected as key data points. A local reference coordinate system is established with each key data point as the center, and the neighborhood space is divided into multiple sub-regions. The radial distance distribution, the angle distribution between the normal vector and the relative height distribution of the data points in each sub-region are statistically analyzed to obtain a sub-region statistical histogram. The statistical histograms of each sub-region are spliced ​​together in a fixed order to form a feature description vector. The feature description vectors corresponding to all the key data points are aggregated to obtain a feature description vector set.

4. The global coordinate data alignment method for inspection scenarios according to claim 1, characterized in that, The step of performing cross-dataset similarity matching on the feature description vector set to obtain an initial correspondence set, and then sequentially performing spatial distance constraint filtering and normal vector consistency constraint filtering on the initial correspondence set to obtain a seed correspondence set, includes: For each key data point in the source coordinate dataset, the feature description vector with the smallest Euclidean distance is searched in the feature description vector set of the reference coordinate dataset to obtain the nearest neighbor match and the second nearest neighbor match. The distance ratio is calculated by comparing the distance ratio with the preset ratio test threshold condition and retaining the matching pairs that meet the threshold condition to form an initial correspondence set. For each matching pair in the initial correspondence set, the spatial Euclidean distance between the source data point and the target data point is calculated to obtain the matching spatial distance. The matching spatial distance is compared with the preset maximum displacement threshold condition, and matching pairs that exceed the threshold condition are removed to obtain the distance-filtered set. For each matching pair in the distance-filtered set, the angle between the normal vector of the source data point and the normal vector of the target data point is calculated to obtain the normal vector angle value. The normal vector angle value is compared with the preset angle threshold condition, and matching pairs that meet the threshold condition are retained to form the seed correspondence set.

5. The global coordinate data alignment method for inspection scenarios according to claim 1, characterized in that, The calculation of compatibility scores for each matching pair in the seed correspondence set based on the rigid body transformation distance preservation property to obtain the geometric compatibility matrix includes: For any two matching pairs in the seed correspondence set, calculate the spatial distance between their corresponding two points in the source coordinate dataset to obtain the source end distance, calculate the spatial distance between their corresponding two points in the reference coordinate dataset to obtain the target end distance, and take the absolute value of the difference between the source end distance and the target end distance to obtain the distance difference. The compatibility score is calculated by inputting the distance difference into a kernel function that monotonically decreases with the distance difference. The compatibility scores between all pairs of matching pairs in the seed correspondence set are written into a symmetric matrix according to the row and column indices to obtain a geometric compatibility matrix. The diagonal elements of the geometric compatibility matrix are set to the maximum value to indicate that each matching pair is completely compatible with itself.

6. The global coordinate data alignment method for inspection scenarios according to claim 1, characterized in that, The step of performing eigenvalue decomposition on the geometric compatibility matrix and extracting the principal eigenvectors to obtain a globally consistent weight sequence, using the globally consistent weight sequence as coefficients to construct a weighted covariance matrix and performing singular value decomposition to obtain a rotation matrix and a translation vector, includes: Eigenvalue decomposition is performed on the geometric compatibility matrix to obtain a set of eigenvalues ​​and a set of corresponding eigenvectors. The largest eigenvalue is extracted from the set of eigenvalues ​​and the eigenvector corresponding to the largest eigenvalue is read as the principal eigenvector. Each component in the principal eigenvector is normalized to the interval between zero and one to obtain a globally consistent weight sequence. Each weight value in the global consistency weight sequence is used as the coefficient of the corresponding matching pair. Based on the coefficient, a weighted cross covariance matrix is ​​constructed for the source data point coordinates and target data point coordinates in the seed correspondence set. Singular value decomposition is performed on the weighted cross covariance matrix to obtain a rotation matrix. A translation vector is calculated based on the rotation matrix and the weighted centroid coordinates.

7. The global coordinate data alignment method for inspection scenarios according to claim 1, characterized in that, The step of applying the rotation matrix and the translation vector to the source coordinate dataset to align the corridor coordinates and output the location information of over-limit vegetation for use in the pruning operation includes: The coordinates of each data point in the source coordinate dataset are multiplied by the rotation matrix to obtain the rotated coordinates. The rotated coordinates are then added by the translation vector to obtain the transformed coordinates. All the transformed coordinates are then aggregated to form the aligned corridor coordinate dataset. Identify vegetation data points and traverse data points from the aligned corridor coordinate dataset, calculate the spatial distance between each vegetation data point and the nearest traverse data point to obtain the tree-line distance value, compare the tree-line distance value with a preset safe distance threshold condition, and filter vegetation data points that are less than the threshold condition as over-limit vegetation points, extract the spatial coordinates of each over-limit vegetation point to form over-limit vegetation positioning information, and output it for use in UAV trimming operation path planning.

8. A global coordinate data alignment device for inspection scenarios, characterized in that, The device includes: The data filtering module is used to acquire the source coordinate dataset and reference coordinate dataset collected by the UAV on the power transmission corridor, perform filtering, noise reduction and downsampling on the source coordinate dataset and the reference coordinate dataset to obtain a standardized corridor coordinate dataset, and filter key data points from the standardized corridor coordinate dataset based on the local curvature threshold condition and calculate feature descriptors to obtain a feature description vector set. The data alignment module is used to perform cross-dataset similarity matching on the feature description vector set to obtain an initial correspondence set, and to perform spatial distance constraint filtering and normal vector consistency constraint filtering on the initial correspondence set to obtain a seed correspondence set. Based on the rigid body transformation distance preservation property, the module calculates the compatibility score of each matching pair in the seed correspondence set to obtain a geometric compatibility matrix. The pruning module is used to perform eigenvalue decomposition on the geometric compatibility matrix and extract the principal eigenvectors to obtain a globally consistent weight sequence. The globally consistent weight sequence is used as coefficients to construct a weighted covariance matrix and perform singular value decomposition to obtain a rotation matrix and a translation vector. The rotation matrix and the translation vector are applied to the source coordinate dataset to complete the corridor coordinate alignment and output the over-limit vegetation positioning information for the pruning operation to call.

Images