Multi-source laser point cloud data adaptive registration method and system for power grid equipment

Abstract

The invention provides a power grid equipment-oriented multi-source laser point cloud data adaptive registration method and system, and relates to the technical field of power grid equipment, and the method comprises the steps: collecting source laser scanning data and target laser scanning data of power grid equipment, and carrying out the preprocessing; extracting key feature points of the power grid equipment and constructing a multi-dimensional local feature description vector; calculating a feature similarity matrix and establishing a dynamic matching threshold optimization model; and then constructing a registration optimization objective function of the geometric constraint of the power grid equipment, and carrying out iterative optimization by adopting a gradient descent method to obtain an optimal registration transformation matrix. According to the method, geometric constraint conditions and a self-adaptive matching threshold mechanism are introduced, so that the accuracy and robustness of power grid equipment laser point cloud data registration are improved, and the method can be effectively applied to three-dimensional data registration of complex power equipment.

Description

Technical Field

[0001] The present invention relates to power grid equipment technology, and in particular to a multi-source laser point cloud data adaptive registration method and system for power grid equipment. Background Art

[0002] Laser scanning technology has been widely used in the field of power inspection to obtain three-dimensional point cloud data of power grid equipment and provide support for the safe and stable operation of power systems. Point cloud registration is a key step in laser scanning data processing. Its purpose is to unify point cloud data from different perspectives or at different times into the same coordinate system for subsequent modeling, analysis and application. Traditional point cloud registration methods usually rely on manual selection of same-name points or registration based on global features, which is inefficient and the accuracy is greatly affected by human factors. For complex power grid equipment, due to its complex structure and rich details, traditional registration methods are difficult to effectively extract feature points, resulting in low registration accuracy and poor robustness.

[0003] Traditional registration methods are difficult to adapt to the structural characteristics of complex power grid equipment. Power grid equipment usually has complex geometric shapes and rich details. Traditional registration methods based on global features or manually selected same-name points are difficult to effectively extract feature points, resulting in low registration accuracy and poor robustness. Existing local feature descriptors are not comprehensive enough in describing the key feature points of power grid equipment. Some local feature descriptors only consider the geometric information of feature points, while ignoring their curvature information and spatial distribution information, resulting in inaccurate feature description and affecting the registration effect. The selection of matching thresholds usually relies on experience and lacks adaptability. Fixed matching thresholds are difficult to adapt to point cloud data of different scenarios and different data quality, resulting in unstable registration results. Summary of the invention

[0004] The embodiments of the present invention provide a method and system for adaptive registration of multi-source laser point cloud data for power grid equipment, which can solve the problems in the prior art.

[0005] According to a first aspect of the embodiments of the present invention, Provides a multi-source laser point cloud data adaptive registration method for power grid equipment, including: Collect source laser scanning data and target laser scanning data of power grid equipment, and pre-process the source laser scanning data and the target laser scanning data; use a statistical outlier filtering algorithm to remove noise points in the source laser scanning data and the target laser scanning data; use normal vector consistency analysis to extract key feature points of power grid equipment in the source laser scanning data and the target laser scanning data; construct a multidimensional local feature description vector based on the extracted key feature points of power grid equipment, wherein the multidimensional local feature description vector has curvature information, geometric shape information and spatial distribution information of the feature points; Based on the constructed multi-dimensional local feature description vector, a feature similarity matrix of key feature points of power grid equipment in the source laser scanning data and key feature points of power grid equipment in the target laser scanning data is calculated; a dynamic matching threshold optimization model is established according to the feature similarity matrix; the statistical distribution characteristics of the feature similarity matrix are calculated, and an initial matching threshold is determined in combination with the structural characteristics of the power grid equipment; an adaptive weight function is constructed based on the variance and kurtosis of the feature similarity matrix; the initial matching threshold is dynamically adjusted using the adaptive weight function to obtain an optimized feature point correspondence relationship; According to the optimized feature point correspondence, a random sampling consistency algorithm is used to eliminate erroneous matching point pairs; a registration optimization objective function of geometric constraints of power grid equipment is constructed; the structural characteristics of the power grid equipment are converted into geometric constraints, and the geometric constraints are introduced into the registration error calculation; the gradient descent method is used to iteratively optimize the registration optimization objective function to obtain an optimal registration transformation matrix; the optimal registration transformation matrix is ​​applied to the source laser scanning data to achieve accurate registration of the source laser scanning data and the target laser scanning data.

[0006] The key feature points of the power grid equipment in the source laser scanning data and the target laser scanning data are extracted by normal vector consistency analysis; a multidimensional local feature description vector is constructed according to the extracted key feature points of the power grid equipment, and the multidimensional local feature description vector has curvature information, geometric shape information and spatial distribution information of the feature points, including: Collecting point cloud data of power grid equipment, constructing a spherical neighborhood for each point in the point cloud data of power grid equipment, selecting a preset number of neighboring points in the spherical neighborhood, constructing a covariance matrix based on the neighboring points, performing eigenvalue decomposition on the covariance matrix to obtain three eigenvalues ​​and their corresponding eigenvectors, and using the eigenvector corresponding to the minimum eigenvalue as the normal vector of the point; Calculate the angle between the normal vector of the neighboring point in the spherical neighborhood and the normal vector of the center point, calculate the normal vector consistency metric value based on the angle, construct an adaptive threshold function according to the standard deviation of the consistency metric value in the spherical neighborhood, compare the normal vector consistency metric value with the calculation result of the adaptive threshold function, and determine the key feature points in the point cloud data of the power grid equipment; For the determined key feature points, the principal curvature is calculated based on the eigenvalues, and a shape index and a curvature index are constructed according to the principal curvature; a local coordinate system is established with the key feature point as the center, and the neighboring points in the spherical neighborhood are projected to the tangent plane, and the tangent plane is divided into uniform sectors, and the number of points in each sector is counted to form an azimuth histogram feature; the average distance, standard deviation, spatial density and sphericity coefficient of the neighboring points in the spherical neighborhood are calculated, and the shape index, the curvature index, the azimuth histogram feature, the average distance, the standard deviation, the spatial density and the sphericity coefficient are combined to construct a multi-dimensional local feature description vector, and the multi-dimensional feature description vector is used to characterize the feature information of the key feature point.

[0007] Based on the constructed multi-dimensional local feature description vector, a feature similarity matrix of key feature points of power grid equipment in the source laser scanning data and key feature points of power grid equipment in the target laser scanning data is calculated; and a dynamic matching threshold optimization model is established according to the feature similarity matrix, including: Acquire a multidimensional local feature description vector of a key feature point of a power grid device in the source laser scanning data and a multidimensional local feature description vector of a key feature point of a power grid device in the target laser scanning data, wherein the multidimensional local feature description vector includes a shape index, a curvature index, an azimuth histogram feature, an average distance, a standard deviation, a spatial density, and a sphericity coefficient; A first similarity is obtained by calculating the cosine distance based on the multidimensional local feature description vector of the key feature points of the power grid equipment in the source laser scanning data and the multidimensional local feature description vector of the key feature points of the power grid equipment in the target laser scanning data; a first distance matrix consisting of the key feature points of the power grid equipment in the source laser scanning data and their neighborhood points and a second distance matrix consisting of the key feature points of the power grid equipment in the target laser scanning data and their neighborhood points are obtained, and a second similarity of the local structure is calculated based on the first distance matrix and the second distance matrix; Determine a first weight coefficient and a second weight coefficient according to the local geometric significance measurement of the key feature points of the power grid equipment in the source laser scanning data and the key feature points of the power grid equipment in the target laser scanning data, add the product of the first similarity and the first weight coefficient, and the product of the second similarity and the second weight coefficient to obtain feature similarity; calculate the feature similarity of the key feature points of the power grid equipment in the source laser scanning data and the key feature points of the power grid equipment in the target laser scanning data in pairs to form a feature similarity matrix; Calculate the mean and standard deviation of all feature similarities in the feature similarity matrix, and construct an initial matching threshold based on the mean and the standard deviation; calculate the feature similarity variance between each key feature point of the power grid equipment in the source laser scanning data and the key feature points of the power grid equipment in all target laser scanning data, and calculate the feature similarity kurtosis between each key feature point of the power grid equipment in the target laser scanning data and the key feature points of the power grid equipment in all source laser scanning data; construct a first dynamic adjustment factor based on the feature similarity variance, and construct a second dynamic adjustment factor based on the feature similarity kurtosis; multiply the first dynamic adjustment factor by the second dynamic adjustment factor to obtain an adaptive weight function; multiply the initial matching threshold by the calculation result of the adaptive weight function to obtain a dynamic optimization matching threshold.

[0008] Calculating the statistical distribution characteristics of the feature similarity matrix, and determining the initial matching threshold in combination with the structural characteristics of the power grid equipment; constructing an adaptive weight function based on the variance and kurtosis of the feature similarity matrix; dynamically adjusting the initial matching threshold using the adaptive weight function, and obtaining the optimized feature point correspondence relationship includes: Calculate the mean, standard deviation, skewness and kurtosis of all elements in the feature similarity matrix, calculate the proportion of elements in the feature similarity matrix whose values ​​are greater than the mean, and use the difference between the proportion and a preset benchmark proportion as a distribution shift factor; correct the mean based on the distribution shift factor to obtain a corrected mean, and construct the weighted sum of the corrected mean and the standard deviation as an initial weighted matching threshold; Extracting row vectors and column vectors of the feature similarity matrix respectively, calculating the variance of each row vector to obtain a source data variance sequence, and calculating the variance of each column vector to obtain a target data variance sequence; calculating the source data cumulative distribution function based on the source data variance sequence, and calculating the target data cumulative distribution function based on the target data variance sequence; taking the difference between the source data cumulative distribution function and the target data cumulative distribution function as a structural consistency measure; Calculate the kurtosis of each row vector in the feature similarity matrix to obtain a source data kurtosis sequence, calculate the kurtosis of each column vector to obtain a target data kurtosis sequence; construct a source data distribution density function based on the source data kurtosis sequence, and construct a target data distribution density function based on the target data kurtosis sequence; use the overlapping area of ​​the source data distribution density function and the target data distribution density function as a distribution similarity measure; The structural consistency metric and the distribution similarity metric are weightedly combined to construct an adaptive weight function, wherein the weight coefficient of the adaptive weight function is dynamically adjusted based on the structural complexity of the power grid equipment; the initial weighted matching threshold is combined with the calculation result of the adaptive weight function to obtain a dynamically adjusted optimized matching threshold; the feature similarity matrix is ​​filtered according to the dynamically adjusted optimized matching threshold to obtain an optimized correspondence between feature points of power grid equipment.

[0009] According to the optimized feature point correspondence, a random sampling consistency algorithm is used to eliminate erroneous matching point pairs; a registration optimization objective function of geometric constraints of power grid equipment is constructed; the structural features of the power grid equipment are converted into geometric constraints, and the geometric constraints are introduced into the registration error calculation, including: Based on the corresponding relationship of the feature points, an error matching screening model is constructed, a support matrix is ​​determined by an iterative voting strategy, and a local consistency score is determined according to the support matrix; a cost function is obtained by weighted combination of the local consistency score and the Euclidean distance between the feature point pairs, and a random sampling consistency algorithm is used to screen the error matching point pairs based on the cost function; Extracting flatness constraints, perpendicularity constraints and parallelism constraints from the structural features of the power grid equipment, and establishing a set of constraint equations; using the calculation results of the point-to-plane distance, line-to-plane angle and plane-to-plane angle in the set of constraint equations as geometric constraint items, and constructing a registration objective function with a penalty factor; Based on the registration objective function, initial registration parameters are obtained, and the distance error from the feature point in the source point cloud to the corresponding feature point in the target point cloud is calculated; an adaptive threshold is determined according to the mean and standard deviation of the distance error, and feature point pairs whose distance error is greater than the adaptive threshold are eliminated; the feature point pairs after the outliers are eliminated are brought back into the registration objective function for optimization and solution to obtain optimized registration parameters; The geometric characteristic parameters of the power grid equipment are calculated according to the optimized alignment parameters, including flatness parameters, perpendicularity parameters and parallelism parameters; the geometric characteristic parameters are constructed as new constraint conditions, and the alignment objective function is updated; the alignment parameters are iteratively optimized based on the updated alignment objective function until the geometric characteristic parameters meet the preset accuracy requirements.

[0010] Iteratively optimizing the registration optimization objective function using a gradient descent method to obtain an optimal registration transformation matrix; applying the optimal registration transformation matrix to the source laser scanning data to achieve accurate registration of the source laser scanning data and the target laser scanning data includes: Constructing a registration optimization objective function of the source laser scanning data and the target laser scanning data, wherein the registration optimization objective function includes a point correspondence error term and a point cloud registration constraint term, and the point cloud registration constraint term includes a normal vector constraint, a curvature constraint, and a local shape constraint; Performing Lie algebra parameterization on the derivative of the registration optimization objective function with respect to the rotation matrix, constructing a gradient calculation formula for the translation vector, and using the derivative of the rotation matrix and the gradient of the translation vector as optimization parameters; An initial step value is set based on the optimization parameter, a current gradient direction of the registration optimization objective function is calculated in each iteration, and the rotation matrix and the translation vector are updated according to the current gradient direction; Adjusting the step size according to the updated registration optimization objective function value, increasing the step size when the registration optimization objective function value decreases, and decreasing the step size when the registration optimization objective function value increases; determining the iteration termination condition based on the gradient modulus value of the rotation matrix and the gradient modulus value of the translation vector; The rotation matrix obtained by iterative optimization is combined with the translation vector to form an optimal registration transformation matrix; the optimal registration transformation matrix is ​​applied to each data point in the source laser scanning data to obtain the transformed source laser scanning data; Calculate the point pair distance between the transformed source laser scanning data and the target laser scanning data in the overlapping area, extract the feature point pairs in the overlapping area, and verify the geometric feature consistency of the feature point pairs; when the point pair distance is less than a first preset threshold and the geometric feature consistency meets a second preset threshold, output the optimal registration transformation matrix to complete the precise registration of the source laser scanning data and the target laser scanning data.

[0011] According to a second aspect of the embodiments of the present invention, Provides a multi-source laser point cloud data adaptive registration system for power grid equipment, including: The first unit is used to collect source laser scanning data and target laser scanning data of power grid equipment, pre-process the source laser scanning data and the target laser scanning data; use a statistical outlier filtering algorithm to remove noise points in the source laser scanning data and the target laser scanning data; use normal vector consistency analysis to extract key feature points of power grid equipment in the source laser scanning data and the target laser scanning data; construct a multi-dimensional local feature description vector based on the extracted key feature points of power grid equipment, wherein the multi-dimensional local feature description vector has curvature information, geometric shape information and spatial distribution information of the feature points; The second unit is used to calculate the feature similarity matrix of the key feature points of the power grid equipment in the source laser scanning data and the key feature points of the power grid equipment in the target laser scanning data based on the constructed multi-dimensional local feature description vector; establish a dynamic matching threshold optimization model according to the feature similarity matrix; calculate the statistical distribution characteristics of the feature similarity matrix, and determine the initial matching threshold in combination with the structural characteristics of the power grid equipment; construct an adaptive weight function based on the variance and kurtosis of the feature similarity matrix; use the adaptive weight function to dynamically adjust the initial matching threshold to obtain an optimized feature point correspondence relationship; The third unit is used to eliminate erroneous matching point pairs by using a random sampling consistency algorithm according to the optimized feature point correspondence relationship; construct a registration optimization objective function of the geometric constraints of the power grid equipment; convert the structural characteristics of the power grid equipment into geometric constraints, and introduce the geometric constraints into the registration error calculation; use the gradient descent method to iteratively optimize the registration optimization objective function to obtain the optimal registration transformation matrix; apply the optimal registration transformation matrix to the source laser scanning data to achieve accurate registration of the source laser scanning data and the target laser scanning data.

[0012] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0013] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0014] The beneficial effects of this application are as follows: BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic diagram of a flow chart of a method for adaptive registration of multi-source laser point cloud data for power grid equipment according to an embodiment of the present invention; Figure 2 It is a structural schematic diagram of a multi-source laser point cloud data adaptive registration system for power grid equipment according to an embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0018] Figure 1 FIG. 1 is a flow chart of a method for adaptively registering multi-source laser point cloud data for power grid equipment according to an embodiment of the present invention. Figure 1 As shown, the method includes: S11. Collect source laser scanning data and target laser scanning data of power grid equipment, pre-process the source laser scanning data and the target laser scanning data; use statistical outlier filtering algorithm to remove noise points in the source laser scanning data and the target laser scanning data; use normal vector consistency analysis to extract key feature points of power grid equipment in the source laser scanning data and the target laser scanning data; construct a multi-dimensional local feature description vector based on the extracted key feature points of power grid equipment, wherein the multi-dimensional local feature description vector has curvature information, geometric shape information and spatial distribution information of feature points; S12. Based on the constructed multi-dimensional local feature description vector, calculate the feature similarity matrix of the key feature points of the power grid equipment in the source laser scanning data and the key feature points of the power grid equipment in the target laser scanning data; establish a dynamic matching threshold optimization model according to the feature similarity matrix; calculate the statistical distribution characteristics of the feature similarity matrix, and determine the initial matching threshold in combination with the structural characteristics of the power grid equipment; construct an adaptive weight function based on the variance and kurtosis of the feature similarity matrix; use the adaptive weight function to dynamically adjust the initial matching threshold to obtain the optimized feature point correspondence relationship; S13. According to the correspondence relationship of the optimized feature points, a random sampling consistency algorithm is used to eliminate erroneous matching point pairs; a registration optimization objective function of the geometric constraints of the power grid equipment is constructed; the structural characteristics of the power grid equipment are converted into geometric constraints, and the geometric constraints are introduced into the registration error calculation; the gradient descent method is used to iteratively optimize the registration optimization objective function to obtain the optimal registration transformation matrix; the optimal registration transformation matrix is ​​applied to the source laser scanning data to achieve accurate registration of the source laser scanning data and the target laser scanning data.

[0019] In an optional implementation, normal vector consistency analysis is used to extract key feature points of power grid equipment in the source laser scanning data and the target laser scanning data; a multidimensional local feature description vector is constructed based on the extracted key feature points of the power grid equipment, and the multidimensional local feature description vector has curvature information, geometric shape information and spatial distribution information of the feature points, including: Collecting point cloud data of power grid equipment, constructing a spherical neighborhood for each point in the point cloud data of power grid equipment, selecting a preset number of neighboring points in the spherical neighborhood, constructing a covariance matrix based on the neighboring points, performing eigenvalue decomposition on the covariance matrix to obtain three eigenvalues ​​and their corresponding eigenvectors, and using the eigenvector corresponding to the minimum eigenvalue as the normal vector of the point; Calculate the angle between the normal vector of the neighboring point in the spherical neighborhood and the normal vector of the center point, calculate the normal vector consistency metric value based on the angle, construct an adaptive threshold function according to the standard deviation of the consistency metric value in the spherical neighborhood, compare the normal vector consistency metric value with the calculation result of the adaptive threshold function, and determine the key feature points in the point cloud data of the power grid equipment; For the determined key feature points, the principal curvature is calculated based on the eigenvalues, and a shape index and a curvature index are constructed according to the principal curvature; a local coordinate system is established with the key feature point as the center, and the neighboring points in the spherical neighborhood are projected to the tangent plane, and the tangent plane is divided into uniform sectors, and the number of points in each sector is counted to form an azimuth histogram feature; the average distance, standard deviation, spatial density and sphericity coefficient of the neighboring points in the spherical neighborhood are calculated, and the shape index, the curvature index, the azimuth histogram feature, the average distance, the standard deviation, the spatial density and the sphericity coefficient are combined to construct a multi-dimensional local feature description vector, and the multi-dimensional feature description vector is used to characterize the feature information of the key feature point.

[0020] Acquire 3D point cloud data of power grid equipment, such as through laser scanner or drone photogrammetry. Assume that the acquired point cloud data contains millions of points, each of which contains 3D coordinate information (x, y, z).

[0021] For each point in the point cloud data, a spherical neighborhood is constructed with the point as the center. For example, the radius of the spherical neighborhood is set to 0.1 meters. Several neighboring points in the spherical neighborhood are selected, for example, the nearest 30 points are selected.

[0022] Using the selected neighboring points, a 3x3 covariance matrix is ​​constructed. The covariance matrix describes the distribution of these neighboring points in three-dimensional space.

[0023] Perform eigenvalue decomposition on the covariance matrix. Eigenvalue decomposition will obtain three eigenvalues ​​and corresponding three eigenvectors.

[0024] The eigenvector corresponding to the smallest eigenvalue among the three eigenvalues ​​is used as the normal vector of the point. The normal vector represents the normal direction of the local surface where the point is located.

[0025] Calculate the angle between the normal vectors of all neighboring points in the spherical neighborhood and the normal vector of the center point.

[0026] The normal vector consistency metric is calculated using these angles. For example, the average of the cosine values ​​of the angles can be used as the consistency metric. The higher the consistency metric value, the smoother the surface around the point.

[0027] Construct an adaptive threshold function based on the standard deviation of the consistency measure within the spherical neighborhood. For example, you can use the mean plus a multiple of the standard deviation as the threshold.

[0028] The calculated normal vector consistency metric value of each point is compared with the calculation result of the adaptive threshold function. If the consistency metric value is greater than the threshold, the point is determined to be a key feature point in the point cloud data of the power grid equipment.

[0029] For each point determined as a key feature point, the principal curvature of the point is calculated based on the three eigenvalues ​​obtained by eigenvalue decomposition. The principal curvature describes the curvature of the surface where the point is located. Assuming that the three eigenvalues ​​are λ1, λ2, and λ3, and λ1>=λ2>=λ3, the principal curvatures k1=λ1 and k2=λ2.

[0030] The shape index and curvature index are calculated based on the principal curvatures. The shape index and curvature index can more comprehensively describe the local geometric characteristics of the surface. For example, the shape index can be defined as (k1+k2) / (k1-k2), and the curvature index can be defined as (k1*k2) / (k1+k2).

[0031] A local coordinate system is established with the key feature point as the center. The x-axis and y-axis of the local coordinate system are located on the tangent plane perpendicular to the normal vector, and the z-axis is consistent with the direction of the normal vector.

[0032] Project the neighboring points in the spherical neighborhood onto the tangent plane.

[0033] The tangent plane is divided into several uniform sectors, for example, 8 sectors. The number of points in each sector is counted to form an azimuth histogram feature. The azimuth histogram feature describes the spatial distribution of points around the key feature point.

[0034] Calculates the average distance, standard deviation, spatial density, and sphericity of neighbor points in a spherical neighborhood. The average distance is the average distance from the neighbor points to the center point. The standard deviation is the standard deviation of the distance from the neighbor points to the center point. The spatial density is the number of neighbors divided by the volume of the spherical neighborhood. The sphericity describes how close the distribution shape of the neighbor points is to a sphere.

[0035] The shape index, curvature index, azimuth histogram features, average distance, standard deviation, spatial density and sphericity coefficient are combined to construct a multi-dimensional local feature description vector. This vector is used to characterize the feature information of key feature points. For example, the above features can be combined into a 15-dimensional vector (shape index and curvature index each occupy 1 dimension, azimuth histogram features 8 dimensions, average distance, standard deviation, spatial density, sphericity coefficient each occupy 1 dimension).

[0036] Assuming that the two principal curvatures of a key feature point are 0.2 and 0.1 respectively, the shape index is 0.5 and the curvature index is 0.017. The azimuth histogram feature is [10, 5, 2, 1, 3, 6, 8, 7], the average distance is 0.05 meters, the standard deviation is 0.01 meters, the spatial density is 5000, and the sphericity coefficient is 0.9. The multidimensional local feature description vector of the key feature point is: [0.5,0.017,10,5,2,1,3,6,8,7,0.05,0.01,5000,0.9].

[0037] The solution of this application can: Improve the accuracy and robustness of key feature point extraction for power grid equipment: This method can effectively distinguish key feature points from non-key feature points by analyzing the consistency of point cloud normal vectors, and can maintain high accuracy even in the presence of noise and missing data. Construct a more discerning multidimensional local feature description vector: This method combines a variety of geometric features and spatial distribution features to construct a more discerning multidimensional local feature description vector that can better characterize the characteristics of key feature points. Improve the efficiency of three-dimensional model registration and recognition of power grid equipment: The key feature points extracted and the multi-dimensional local feature description vector constructed using this method can effectively improve the efficiency of three-dimensional model registration and recognition of power grid equipment, thereby better supporting the automated inspection and maintenance of power grid equipment.

[0038] In an optional implementation, based on the constructed multidimensional local feature description vector, a feature similarity matrix of key feature points of power grid equipment in the source laser scanning data and key feature points of power grid equipment in the target laser scanning data is calculated; and establishing a dynamic matching threshold optimization model according to the feature similarity matrix includes: Acquire a multidimensional local feature description vector of a key feature point of a power grid device in the source laser scanning data and a multidimensional local feature description vector of a key feature point of a power grid device in the target laser scanning data, wherein the multidimensional local feature description vector includes a shape index, a curvature index, an azimuth histogram feature, an average distance, a standard deviation, a spatial density, and a sphericity coefficient; A first similarity is obtained by calculating the cosine distance based on the multidimensional local feature description vector of the key feature points of the power grid equipment in the source laser scanning data and the multidimensional local feature description vector of the key feature points of the power grid equipment in the target laser scanning data; a first distance matrix consisting of the key feature points of the power grid equipment in the source laser scanning data and their neighborhood points and a second distance matrix consisting of the key feature points of the power grid equipment in the target laser scanning data and their neighborhood points are obtained, and a second similarity of the local structure is calculated based on the first distance matrix and the second distance matrix; Determine a first weight coefficient and a second weight coefficient according to the local geometric significance measurement of the key feature points of the power grid equipment in the source laser scanning data and the key feature points of the power grid equipment in the target laser scanning data, add the product of the first similarity and the first weight coefficient, and the product of the second similarity and the second weight coefficient to obtain feature similarity; calculate the feature similarity of the key feature points of the power grid equipment in the source laser scanning data and the key feature points of the power grid equipment in the target laser scanning data in pairs to form a feature similarity matrix; Calculate the mean and standard deviation of all feature similarities in the feature similarity matrix, and construct an initial matching threshold based on the mean and the standard deviation; calculate the feature similarity variance between each key feature point of the power grid equipment in the source laser scanning data and the key feature points of the power grid equipment in all target laser scanning data, and calculate the feature similarity kurtosis between each key feature point of the power grid equipment in the target laser scanning data and the key feature points of the power grid equipment in all source laser scanning data; construct a first dynamic adjustment factor based on the feature similarity variance, and construct a second dynamic adjustment factor based on the feature similarity kurtosis; multiply the first dynamic adjustment factor by the second dynamic adjustment factor to obtain an adaptive weight function; multiply the initial matching threshold by the calculation result of the adaptive weight function to obtain a dynamic optimization matching threshold.

[0039] The laser scanning data power equipment key point matching method is used to improve the accuracy and efficiency of power equipment three-dimensional model matching.

[0040] First, extract the key feature points of power equipment from the source laser scanning data and the target laser scanning data. For example, we can extract the vertices, edge points and other points with significant geometric features of equipment such as transformers, circuit breakers and electric poles as key feature points.

[0041] Then, the extracted key feature points are described by multi-dimensional local features. Each key feature point uses a multi-dimensional vector to describe its local geometric features. This multi-dimensional vector contains multiple features, such as shape index, curvature index, azimuth histogram features, average distance, standard deviation, spatial density and sphericity coefficient. For example, the shape index of a point describes whether the shape of the neighborhood of the point is convex, flat or concave; the curvature index describes the degree of curvature of the surface of the neighborhood of the point; and the azimuth histogram features count the distribution of the neighborhood point cloud of the point in different directions. Assuming that a key feature point has a shape index of 0.8, a curvature index of 0.5, and an average distance of 0.2 meters, this indicates that the point is located in a relatively convex area with medium curvature and a dense neighborhood point cloud.

[0042] Next, the feature similarity matrix is ​​calculated. The similarity between key feature points is calculated by combining local feature similarity and local structural similarity. Local feature similarity is obtained by calculating the cosine distance of the multidimensional local feature description vectors of two key feature points. Local structural similarity is obtained by comparing the distance matrices of the neighborhood point clouds of two key feature points. For example, if the neighborhood point cloud distance matrices of two key feature points are very similar, their local structural similarity is high. Local geometric saliency is used to determine the weight coefficients of the two similarity indicators. For example, if a key feature point is located in a very significant geometric feature area, such as a sharp corner, the weight coefficient of its local feature similarity will be higher. The weighted local feature similarity and local structural similarity are added together to obtain the final feature similarity. The feature similarity between all source key feature points and all target key feature points is calculated to form a feature similarity matrix. For example, if the source data has 100 key feature points and the target data has 150 key feature points, the size of the feature similarity matrix is ​​100x150.

[0043] Subsequently, the matching threshold is dynamically optimized. First, an initial matching threshold is calculated based on the mean and standard deviation of all feature similarities in the feature similarity matrix. For example, the mean plus twice the standard deviation can be used as the initial threshold. Then, based on the variance and kurtosis of the feature similarities between each key feature point and all other key feature points, the first dynamic adjustment factor and the second dynamic adjustment factor are calculated. The variance reflects the degree of discreteness of the feature similarity, and the kurtosis reflects the sharpness of the feature similarity distribution. Multiplying these two dynamic adjustment factors together results in an adaptive weight function. Finally, multiplying the initial matching threshold by the adaptive weight function results in a dynamically optimized matching threshold.

[0044] Finally, key feature point matching is performed based on the dynamic optimization matching threshold. Each key feature point in the source data is matched with all points in the target data whose feature similarity is higher than the dynamic optimization matching threshold.

[0045] The solution of this application can: Improve matching accuracy: This method can more accurately identify and match key feature points of power equipment by combining multi-dimensional local feature description and dynamic matching threshold optimization strategy, thereby improving matching accuracy. Enhance robustness: This method takes into account local geometric saliency and uses an adaptive weight function, which can effectively deal with problems such as noise, occlusion and data loss, and enhance the robustness of matching. Improve efficiency: This method adopts a dynamic matching threshold optimization strategy, which can effectively reduce mismatches, thereby reducing the amount of calculation for subsequent processing and improving matching efficiency.

[0046] In an optional implementation, the statistical distribution characteristics of the feature similarity matrix are calculated, and an initial matching threshold is determined in combination with the structural characteristics of the power grid equipment; an adaptive weight function is constructed based on the variance and kurtosis of the feature similarity matrix; and the initial matching threshold is dynamically adjusted using the adaptive weight function to obtain an optimized feature point correspondence relationship, which includes: Calculate the mean, standard deviation, skewness and kurtosis of all elements in the feature similarity matrix, calculate the proportion of elements in the feature similarity matrix whose values ​​are greater than the mean, and use the difference between the proportion and a preset benchmark proportion as a distribution shift factor; correct the mean based on the distribution shift factor to obtain a corrected mean, and construct the weighted sum of the corrected mean and the standard deviation as an initial weighted matching threshold; Extracting row vectors and column vectors of the feature similarity matrix respectively, calculating the variance of each row vector to obtain a source data variance sequence, and calculating the variance of each column vector to obtain a target data variance sequence; calculating the source data cumulative distribution function based on the source data variance sequence, and calculating the target data cumulative distribution function based on the target data variance sequence; taking the difference between the source data cumulative distribution function and the target data cumulative distribution function as a structural consistency measure; Calculate the kurtosis of each row vector in the feature similarity matrix to obtain a source data kurtosis sequence, calculate the kurtosis of each column vector to obtain a target data kurtosis sequence; construct a source data distribution density function based on the source data kurtosis sequence, and construct a target data distribution density function based on the target data kurtosis sequence; use the overlapping area of ​​the source data distribution density function and the target data distribution density function as a distribution similarity measure; The structural consistency metric and the distribution similarity metric are weightedly combined to construct an adaptive weight function, wherein the weight coefficient of the adaptive weight function is dynamically adjusted based on the structural complexity of the power grid equipment; the initial weighted matching threshold is combined with the calculation result of the adaptive weight function to obtain a dynamically adjusted optimized matching threshold; the feature similarity matrix is ​​filtered according to the dynamically adjusted optimized matching threshold to obtain an optimized correspondence between feature points of power grid equipment.

[0047] In order to achieve accurate feature matching of power grid equipment, a feature point matching method based on adaptive weight function is proposed. This method dynamically adjusts the matching threshold by analyzing the statistical distribution characteristics of the feature similarity matrix and the structural characteristics of the power grid equipment, thereby obtaining an optimized feature point correspondence.

[0048] First, calculate the feature similarity matrix. Assume that there are two power grid devices to be matched, extract their feature points respectively and calculate the similarity between them to obtain a feature similarity matrix. For example, if device A has 5 feature points and device B has 6 feature points, then the similarity matrix is ​​a matrix with 5 rows and 6 columns. Each element in the matrix represents the similarity between a feature point of device A and a feature point of device B, and the similarity value is between 0 and 1.

[0049] Next, calculate the statistical distribution characteristics of the feature similarity matrix. Calculate the mean, standard deviation, skewness, and kurtosis of all elements in the matrix. For example, assume that the mean of all elements in the matrix is ​​0.5, the standard deviation is 0.2, the skewness is 0.1, and the kurtosis is 0.3. Then, calculate the percentage of elements in the matrix whose values ​​are greater than the mean. Assume that the percentage is 60%. Set a preset benchmark percentage, such as 50%. Use the difference between the actual percentage and the benchmark percentage (60%-50%=10%) as the distribution shift factor. Correct the mean based on the distribution shift factor. For example, add a portion of the 10% shift factor (such as 5%) to the mean value of 0.5 to obtain a corrected mean of 0.525. Construct the weighted sum of the corrected mean and standard deviation as the initial weighted matching threshold. For example, the weight of the corrected mean of 0.525 is 0.7, and the weight of the standard deviation of 0.2 is 0.3, then the initial weighted matching threshold is 0.525*0.7+0.2*0.3=0.4275.

[0050] Then, the row vectors and column vectors of the feature similarity matrix are extracted, and the variance of each row vector and column vector is calculated respectively to obtain the source data variance sequence and the target data variance sequence. The cumulative distribution function is calculated based on these two variance sequences respectively. The difference between the source data cumulative distribution function and the target data cumulative distribution function is used as the structural consistency measure.

[0051] At the same time, the kurtosis of each row vector and column vector in the feature similarity matrix is ​​calculated to obtain the source data kurtosis sequence and the target data kurtosis sequence. The distribution density function is constructed based on these two kurtosis sequences respectively. The overlapping area of ​​the source data distribution density function and the target data distribution density function is used as the distribution similarity measure.

[0052] Next, the structural consistency metric and the distribution similarity metric are weighted and combined to construct an adaptive weight function. The weight coefficient of the adaptive weight function is dynamically adjusted based on the structural complexity of the power grid equipment. For example, for equipment with simple structure, the weight of the structural consistency metric is greater; for equipment with complex structure, the weight of the distribution similarity metric is greater.

[0053] Finally, the initial weighted matching threshold is combined with the calculation result of the adaptive weight function to obtain the dynamically adjusted optimized matching threshold. The feature similarity matrix is ​​filtered according to the dynamically adjusted optimized matching threshold to obtain the optimized correspondence of the feature points of the power grid equipment. For example, all elements in the feature similarity matrix that are less than the dynamically adjusted optimized matching threshold are set to 0, and the remaining elements remain unchanged, and finally the optimized feature point correspondence is obtained.

[0054] The solution of this application can: Improve matching accuracy: This method dynamically adjusts the matching threshold through an adaptive weight function, which can more accurately identify the correspondence between feature points, thereby improving matching accuracy. Enhance robustness: This method takes into account the structural characteristics of power grid equipment and can adapt to equipment with different structural complexities, thereby enhancing the robustness of matching. Reduce the amount of calculation: This method simplifies the calculation process through statistical distribution characteristics, which can reduce the amount of calculation and improve matching efficiency.

[0055] In an optional implementation, according to the optimized feature point correspondence, a random sampling consistency algorithm is used to eliminate erroneous matching point pairs; a registration optimization objective function of geometric constraints of power grid equipment is constructed; the structural features of the power grid equipment are converted into geometric constraints, and the geometric constraints are introduced into the registration error calculation, including: Based on the corresponding relationship of the feature points, an error matching screening model is constructed, a support matrix is ​​determined by an iterative voting strategy, and a local consistency score is determined according to the support matrix; a cost function is obtained by weighted combination of the local consistency score and the Euclidean distance between the feature point pairs, and a random sampling consistency algorithm is used to screen the error matching point pairs based on the cost function; Extracting flatness constraints, perpendicularity constraints and parallelism constraints from the structural features of the power grid equipment, and establishing a set of constraint equations; using the calculation results of the point-to-plane distance, line-to-plane angle and plane-to-plane angle in the set of constraint equations as geometric constraint items, and constructing a registration objective function with a penalty factor; Based on the registration objective function, initial registration parameters are obtained, and the distance error from the feature point in the source point cloud to the corresponding feature point in the target point cloud is calculated; an adaptive threshold is determined according to the mean and standard deviation of the distance error, and feature point pairs whose distance error is greater than the adaptive threshold are eliminated; the feature point pairs after the outliers are eliminated are brought back into the registration objective function for optimization and solution to obtain optimized registration parameters; The geometric characteristic parameters of the power grid equipment are calculated according to the optimized alignment parameters, including flatness parameters, perpendicularity parameters and parallelism parameters; the geometric characteristic parameters are constructed as new constraint conditions, and the alignment objective function is updated; the alignment parameters are iteratively optimized based on the updated alignment objective function until the geometric characteristic parameters meet the preset accuracy requirements.

[0056] Point cloud registration method for 3D modeling of power equipment First, obtain the 3D point cloud data of the power equipment, including the source point cloud and the target point cloud. For example, use a 3D laser scanner to obtain the point cloud data of the transformer in the substation, where the point cloud obtained by one scan is used as the source point cloud and the point cloud obtained by another scan is used as the target point cloud. Assume that the source point cloud contains 10,000 points and the target point cloud contains 12,000 points.

[0057] Then, the source point cloud and the target point cloud are preprocessed, including denoising, filtering, etc. For example, a statistical filter is used to remove outliers in the point cloud, and a bilateral filter is used to smooth the point cloud surface.

[0058] Next, extract the feature points of the source point cloud and the target point cloud. For example, use the ISS algorithm to extract key points in the point cloud, assuming that 500 feature points are extracted from the source point cloud and 600 feature points are extracted from the target point cloud.

[0059] Afterwards, feature descriptors are calculated based on the extracted feature points. For example, the feature descriptors of each feature point are calculated using the FPFH algorithm.

[0060] According to the feature descriptors, the correspondence between the feature points of the source point cloud and the target point cloud is established. For example, the KD tree is used to search for the nearest neighbor feature points to obtain the initial feature point matching pairs. Assume that 400 pairs of initial matching points are obtained.

[0061] In order to eliminate incorrect matching point pairs, a random sampling consistency algorithm is used. First, a wrong matching screening model is constructed. A support matrix is ​​constructed based on the correspondence between feature points, and the support matrix is ​​determined using an iterative voting strategy. The local consistency score is determined based on the support matrix. The cost function is obtained by weighted combination of the local consistency score and the Euclidean distance between feature point pairs. Then, a random sampling consistency algorithm is used to screen incorrect matching point pairs based on the cost function. Assume that after screening, 300 matching point pairs remain.

[0062] Extract geometric constraints from the structural features of power equipment, such as flatness constraints, perpendicularity constraints, and parallelism constraints. Taking the transformer as an example, its top surface, side surface and other planes, as well as the perpendicular and parallel relationships between planes can be extracted. These constraints are converted into a set of constraint equations, such as the calculation results of the distance from a point to a plane, the line-plane angle, and the plane-plane angle, and are used as geometric constraints to construct a registration objective function with a penalty factor.

[0063] Use an optimization algorithm to obtain initial registration parameters based on the registration objective function. For example, use the Levenberg-Marquardt algorithm for optimization. Calculate the distance error from the feature point in the source point cloud to the corresponding feature point in the target point cloud. Determine the adaptive threshold based on the mean and standard deviation of the distance error. For example, set the threshold to the mean plus two times the standard deviation. Eliminate feature point pairs whose distance error is greater than the adaptive threshold. Assume that 10 pairs of feature point pairs are eliminated, leaving 290 pairs of matching point pairs. Substitute the feature point pairs after eliminating the outliers back into the registration objective function for optimization and solution to obtain the optimized registration parameters.

[0064] The geometric characteristic parameters of the power equipment, such as flatness parameters, perpendicularity parameters, and parallelism parameters, are calculated based on the optimized registration parameters. These geometric characteristic parameters are constructed as new constraints and the registration objective function is updated. The registration parameters are iteratively optimized based on the updated registration objective function until the geometric characteristic parameters meet the preset accuracy requirements. For example, the accuracy requirements for flatness, perpendicularity, and parallelism are all set to 0.5 cm.

[0065] The solution of this application can: Improve registration accuracy: This method converts the structural features of power equipment into geometric constraints and introduces them into the registration process, effectively reducing the registration error and improving the registration accuracy. Enhance robustness: This method uses a random sampling consistency algorithm to remove erroneous matching point pairs and uses an adaptive threshold to remove outliers, which enhances the robustness of the registration algorithm and enables it to cope with noise and missing data. Improve the degree of automation: This method iteratively optimizes the registration parameters until the geometric feature parameters meet the preset accuracy requirements, thereby automating the registration process and reducing manual intervention.

[0066] In an optional implementation, the registration optimization objective function is iteratively optimized using a gradient descent method to obtain an optimal registration transformation matrix; the optimal registration transformation matrix is ​​applied to the source laser scanning data to achieve accurate registration of the source laser scanning data and the target laser scanning data, including: Constructing a registration optimization objective function of the source laser scanning data and the target laser scanning data, wherein the registration optimization objective function includes a point correspondence error term and a point cloud registration constraint term, and the point cloud registration constraint term includes a normal vector constraint, a curvature constraint, and a local shape constraint; Performing Lie algebra parameterization on the derivative of the registration optimization objective function with respect to the rotation matrix, constructing a gradient calculation formula for the translation vector, and using the derivative of the rotation matrix and the gradient of the translation vector as optimization parameters; An initial step value is set based on the optimization parameter, a current gradient direction of the registration optimization objective function is calculated in each iteration, and the rotation matrix and the translation vector are updated according to the current gradient direction; Adjusting the step size according to the updated registration optimization objective function value, increasing the step size when the registration optimization objective function value decreases, and decreasing the step size when the registration optimization objective function value increases; determining the iteration termination condition based on the gradient modulus value of the rotation matrix and the gradient modulus value of the translation vector; The rotation matrix obtained by iterative optimization is combined with the translation vector to form an optimal registration transformation matrix; the optimal registration transformation matrix is ​​applied to each data point in the source laser scanning data to obtain the transformed source laser scanning data; Calculate the point pair distance between the transformed source laser scanning data and the target laser scanning data in the overlapping area, extract the feature point pairs in the overlapping area, and verify the geometric feature consistency of the feature point pairs; when the point pair distance is less than a first preset threshold and the geometric feature consistency meets a second preset threshold, output the optimal registration transformation matrix to complete the precise registration of the source laser scanning data and the target laser scanning data.

[0067] Laser scanning data registration method embodiment First, obtain the source laser scanning data and the target laser scanning data. For example, use a laser scanner to scan two overlapping scenes respectively to obtain two sets of point cloud data, each containing 10,000 three-dimensional coordinate points. The source point cloud data is recorded as P, and the target point cloud data is recorded as Q.

[0068] Next, the registration optimization objective function is constructed. This objective function aims to minimize the difference between the source point cloud and the target point cloud, while satisfying certain constraints to ensure the accuracy and stability of the registration. The objective function consists of two parts: the point correspondence error term and the point cloud registration constraint term. The point correspondence error term is used to measure the distance between the corresponding points in the source point cloud and the target point cloud. The point cloud registration constraint terms include normal vector constraints, curvature constraints, and local shape constraints, which are used to ensure that the geometric features of the registered point cloud remain consistent. For example, the normal vector constraint requires that the normal vector directions at the corresponding points are as close as possible, the curvature constraint requires that the curvature values ​​at the corresponding points are as similar as possible, and the local shape constraint requires that the local shapes around the corresponding points are as similar as possible.

[0069] Then, the objective function is optimized and solved to obtain the optimal registration transformation matrix. The gradient descent method is used for iterative optimization. First, the derivative of the rotation matrix is ​​parameterized by Lie algebra, and the gradient of the translation vector is calculated. The derivative of the rotation matrix and the gradient of the translation vector are used as optimization parameters. Set the initial step value, for example, 0.01. In each iteration, the current gradient direction of the objective function is calculated, and the rotation matrix and translation vector are updated according to the direction. Adjust the step size according to the updated objective function value. If the objective function value decreases, increase the step size, for example, multiply the step size by 1.2; if the objective function value increases, reduce the step size, for example, divide the step size by 2. The iteration termination condition is that the gradient modulus of the rotation matrix and the gradient modulus of the translation vector are both less than a preset threshold, for example, 0.0001.

[0070] After the iterative optimization is completed, the obtained rotation matrix and translation vector are combined to form the optimal registration transformation matrix. The matrix is ​​applied to each data point in the source laser scanning data to obtain the transformed source laser scanning data P'.

[0071] Finally, verify the registration result. Calculate the point pair distance between the transformed source laser scanning data P' and the target laser scanning data Q in the overlapping area. Extract the feature point pairs in the overlapping area, for example, select the points with larger curvature as feature points. Verify the geometric feature consistency of the feature point pairs, for example, compare the normal vector angle and curvature difference of the feature point pairs. If the point pair distance is less than a first preset threshold, such as 0.05, and the geometric feature consistency meets the second preset threshold, such as the normal vector angle is less than 5 degrees and the curvature difference is less than 0.1, then output the optimal registration transformation matrix to complete the precise registration of the source laser scanning data and the target laser scanning data.

[0072] The solution of this application can: Improve registration accuracy: By introducing point cloud registration constraints such as normal vector constraints, curvature constraints, and local shape constraints, the registration accuracy can be effectively improved to ensure that the registered point cloud maintains consistency in geometric features. Enhance registration stability: Adopting the gradient descent method for iterative optimization and dynamically adjusting the step size according to the objective function value can effectively enhance the stability of the registration and avoid falling into the local optimal solution. Improve registration efficiency: By parameterizing the derivative of the rotation matrix with Lie algebra, the calculation process can be simplified and the registration efficiency can be improved.

[0073] Figure 2 FIG. 1 is a schematic diagram of a structure of a multi-source laser point cloud data adaptive registration system for power grid equipment according to an embodiment of the present invention. Figure 2 As shown, the system comprises: The first unit is used to collect source laser scanning data and target laser scanning data of power grid equipment, pre-process the source laser scanning data and the target laser scanning data; use a statistical outlier filtering algorithm to remove noise points in the source laser scanning data and the target laser scanning data; use normal vector consistency analysis to extract key feature points of power grid equipment in the source laser scanning data and the target laser scanning data; construct a multi-dimensional local feature description vector based on the extracted key feature points of power grid equipment, wherein the multi-dimensional local feature description vector has curvature information, geometric shape information and spatial distribution information of the feature points; The second unit is used to calculate the feature similarity matrix of the key feature points of the power grid equipment in the source laser scanning data and the key feature points of the power grid equipment in the target laser scanning data based on the constructed multi-dimensional local feature description vector; establish a dynamic matching threshold optimization model according to the feature similarity matrix; calculate the statistical distribution characteristics of the feature similarity matrix, and determine the initial matching threshold in combination with the structural characteristics of the power grid equipment; construct an adaptive weight function based on the variance and kurtosis of the feature similarity matrix; use the adaptive weight function to dynamically adjust the initial matching threshold to obtain an optimized feature point correspondence relationship; The third unit is used to eliminate erroneous matching point pairs by using a random sampling consistency algorithm according to the optimized feature point correspondence relationship; construct a registration optimization objective function of the geometric constraints of the power grid equipment; convert the structural characteristics of the power grid equipment into geometric constraints, and introduce the geometric constraints into the registration error calculation; use the gradient descent method to iteratively optimize the registration optimization objective function to obtain the optimal registration transformation matrix; apply the optimal registration transformation matrix to the source laser scanning data to achieve accurate registration of the source laser scanning data and the target laser scanning data.

[0074] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0075] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0076] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An adaptive registration method for multi-source laser point cloud data for power grid equipment, characterized in that: include: Collect source laser scanning data and target laser scanning data of power grid equipment, and pre-process the source laser scanning data and the target laser scanning data; use a statistical outlier filtering algorithm to remove noise points in the source laser scanning data and the target laser scanning data; use normal vector consistency analysis to extract key feature points of power grid equipment in the source laser scanning data and the target laser scanning data; construct a multidimensional local feature description vector based on the extracted key feature points of power grid equipment, wherein the multidimensional local feature description vector has curvature information, geometric shape information and spatial distribution information of the feature points; Based on the constructed multi-dimensional local feature description vector, a feature similarity matrix of key feature points of power grid equipment in the source laser scanning data and key feature points of power grid equipment in the target laser scanning data is calculated; a dynamic matching threshold optimization model is established according to the feature similarity matrix; the statistical distribution characteristics of the feature similarity matrix are calculated, and an initial matching threshold is determined in combination with the structural characteristics of the power grid equipment; an adaptive weight function is constructed based on the variance and kurtosis of the feature similarity matrix; the initial matching threshold is dynamically adjusted using the adaptive weight function to obtain an optimized feature point correspondence relationship; According to the optimized feature point correspondence, a random sampling consistency algorithm is used to eliminate erroneous matching point pairs; a registration optimization objective function of geometric constraints of power grid equipment is constructed; the structural features of the power grid equipment are converted into geometric constraints, and the geometric constraints are introduced into the registration error calculation; The registration optimization objective function is iteratively optimized using a gradient descent method to obtain an optimal registration transformation matrix; the optimal registration transformation matrix is ​​applied to the source laser scanning data to achieve accurate registration of the source laser scanning data and the target laser scanning data.

2. The method according to claim 1, characterized in that The key feature points of the power grid equipment in the source laser scanning data and the target laser scanning data are extracted by normal vector consistency analysis; a multidimensional local feature description vector is constructed according to the extracted key feature points of the power grid equipment, and the multidimensional local feature description vector has curvature information, geometric shape information and spatial distribution information of the feature points, including: Collecting point cloud data of power grid equipment, constructing a spherical neighborhood for each point in the point cloud data of power grid equipment, selecting a preset number of neighboring points in the spherical neighborhood, constructing a covariance matrix based on the neighboring points, performing eigenvalue decomposition on the covariance matrix to obtain three eigenvalues ​​and their corresponding eigenvectors, and using the eigenvector corresponding to the minimum eigenvalue as the normal vector of the point; Calculate the angle between the normal vector of the neighboring point in the spherical neighborhood and the normal vector of the center point, calculate the normal vector consistency metric value based on the angle, construct an adaptive threshold function according to the standard deviation of the consistency metric value in the spherical neighborhood, compare the normal vector consistency metric value with the calculation result of the adaptive threshold function, and determine the key feature points in the point cloud data of the power grid equipment; For the determined key feature points, the principal curvature is calculated based on the eigenvalues, and a shape index and a curvature index are constructed according to the principal curvature; a local coordinate system is established with the key feature point as the center, and the neighboring points in the spherical neighborhood are projected to the tangent plane, and the tangent plane is divided into uniform sectors, and the number of points in each sector is counted to form an azimuth histogram feature; the average distance, standard deviation, spatial density and sphericity coefficient of the neighboring points in the spherical neighborhood are calculated, and the shape index, the curvature index, the azimuth histogram feature, the average distance, the standard deviation, the spatial density and the sphericity coefficient are combined to construct a multi-dimensional local feature description vector, and the multi-dimensional feature description vector is used to characterize the feature information of the key feature point.

3. The method according to claim 1, characterized in that Based on the constructed multi-dimensional local feature description vector, calculating a feature similarity matrix of key feature points of power grid equipment in the source laser scanning data and key feature points of power grid equipment in the target laser scanning data; Establishing a dynamic matching threshold optimization model according to the feature similarity matrix includes: Acquire a multidimensional local feature description vector of a key feature point of a power grid device in the source laser scanning data and a multidimensional local feature description vector of a key feature point of a power grid device in the target laser scanning data, wherein the multidimensional local feature description vector includes a shape index, a curvature index, an azimuth histogram feature, an average distance, a standard deviation, a spatial density, and a sphericity coefficient; A first similarity is obtained by calculating the cosine distance based on the multidimensional local feature description vector of the key feature points of the power grid equipment in the source laser scanning data and the multidimensional local feature description vector of the key feature points of the power grid equipment in the target laser scanning data; a first distance matrix consisting of the key feature points of the power grid equipment in the source laser scanning data and their neighborhood points and a second distance matrix consisting of the key feature points of the power grid equipment in the target laser scanning data and their neighborhood points are obtained, and a second similarity of the local structure is calculated based on the first distance matrix and the second distance matrix; Determine a first weight coefficient and a second weight coefficient according to the local geometric significance measurement of the key feature points of the power grid equipment in the source laser scanning data and the key feature points of the power grid equipment in the target laser scanning data, and add the product of the first similarity and the first weight coefficient to the product of the second similarity and the second weight coefficient to obtain feature similarity; Calculating the feature similarity of the key feature points of the power grid equipment in the source laser scanning data and the key feature points of the power grid equipment in the target laser scanning data in pairs to form a feature similarity matrix; Calculate the mean and standard deviation of all feature similarities in the feature similarity matrix, and construct an initial matching threshold based on the mean and the standard deviation; calculate the feature similarity variance between each key feature point of the power grid equipment in the source laser scanning data and the key feature points of the power grid equipment in all target laser scanning data, and calculate the feature similarity kurtosis between each key feature point of the power grid equipment in the target laser scanning data and the key feature points of the power grid equipment in all source laser scanning data; construct a first dynamic adjustment factor based on the feature similarity variance, and construct a second dynamic adjustment factor based on the feature similarity kurtosis; multiply the first dynamic adjustment factor by the second dynamic adjustment factor to obtain an adaptive weight function; multiply the initial matching threshold by the calculation result of the adaptive weight function to obtain a dynamic optimization matching threshold.

4. The method according to claim 1, characterized in that: Calculating the statistical distribution characteristics of the feature similarity matrix, and determining the initial matching threshold in combination with the structural characteristics of the power grid equipment; constructing an adaptive weight function based on the variance and kurtosis of the feature similarity matrix; dynamically adjusting the initial matching threshold using the adaptive weight function, and obtaining the optimized feature point correspondence relationship includes: Calculate the mean, standard deviation, skewness and kurtosis of all elements in the feature similarity matrix, calculate the proportion of elements in the feature similarity matrix whose values ​​are greater than the mean, and use the difference between the proportion and a preset benchmark proportion as a distribution shift factor; correct the mean based on the distribution shift factor to obtain a corrected mean, and construct the weighted sum of the corrected mean and the standard deviation as an initial matching threshold; Extracting row vectors and column vectors of the feature similarity matrix respectively, calculating the variance of each row vector to obtain a source data variance sequence, and calculating the variance of each column vector to obtain a target data variance sequence; calculating the source data cumulative distribution function based on the source data variance sequence, and calculating the target data cumulative distribution function based on the target data variance sequence; taking the difference between the source data cumulative distribution function and the target data cumulative distribution function as a structural consistency measure; Calculate the kurtosis of each row vector in the feature similarity matrix to obtain a source data kurtosis sequence, calculate the kurtosis of each column vector to obtain a target data kurtosis sequence; construct a source data distribution density function based on the source data kurtosis sequence, and construct a target data distribution density function based on the target data kurtosis sequence; use the overlapping area of ​​the source data distribution density function and the target data distribution density function as a distribution similarity measure; The structural consistency metric and the distribution similarity metric are weightedly combined to construct an adaptive weight function, wherein the weight coefficient of the adaptive weight function is dynamically adjusted based on the structural complexity of the power grid equipment; the initial matching threshold is combined with the calculation result of the adaptive weight function to obtain a dynamically optimized matching threshold; the feature similarity matrix is ​​filtered according to the dynamically optimized matching threshold to obtain an optimized correspondence between feature points of the power grid equipment.

5. The method according to claim 1, characterized in that According to the optimized feature point correspondence relationship, a random sampling consistency algorithm is used to eliminate erroneous matching point pairs; and a registration optimization objective function of geometric constraints of power grid equipment is constructed; Converting the structural features of the power grid equipment into geometric constraints and introducing the geometric constraints into the registration error calculation comprises: Based on the corresponding relationship of the feature points, an error matching screening model is constructed, a support matrix is ​​determined by an iterative voting strategy, and a local consistency score is determined according to the support matrix; a cost function is obtained by weighted combination of the local consistency score and the Euclidean distance between the feature point pairs, and a random sampling consistency algorithm is used to screen the error matching point pairs based on the cost function; Extracting flatness constraints, perpendicularity constraints and parallelism constraints from the structural features of the power grid equipment, and establishing a set of constraint equations; using the calculation results of the point-to-plane distance, line-to-plane angle and plane-to-plane angle in the set of constraint equations as geometric constraint items, and constructing a registration objective function with a penalty factor; Based on the registration objective function, initial registration parameters are obtained, and the distance error from the feature point in the source point cloud to the corresponding feature point in the target point cloud is calculated; an adaptive threshold is determined according to the mean and standard deviation of the distance error, and feature point pairs whose distance error is greater than the adaptive threshold are eliminated; the feature point pairs after the outliers are eliminated are brought back into the registration objective function for optimization and solution to obtain optimized registration parameters; The geometric characteristic parameters of the power grid equipment are calculated according to the optimized alignment parameters, including flatness parameters, perpendicularity parameters and parallelism parameters; the geometric characteristic parameters are constructed as new constraint conditions, and the alignment objective function is updated; the alignment parameters are iteratively optimized based on the updated alignment objective function until the geometric characteristic parameters meet the preset accuracy requirements.

6. The method according to claim 1, characterized in that Iteratively optimizing the registration optimization objective function using a gradient descent method to obtain an optimal registration transformation matrix; applying the optimal registration transformation matrix to the source laser scanning data to achieve accurate registration of the source laser scanning data and the target laser scanning data includes: Constructing a registration optimization objective function of the source laser scanning data and the target laser scanning data, wherein the registration optimization objective function includes a point correspondence error term and a point cloud registration constraint term, and the point cloud registration constraint term includes a normal vector constraint, a curvature constraint, and a local shape constraint; Performing Lie algebra parameterization on the derivative of the registration optimization objective function with respect to the rotation matrix, constructing a gradient calculation formula for the translation vector, and using the derivative of the rotation matrix and the gradient of the translation vector as optimization parameters; An initial step value is set based on the optimization parameter, a current gradient direction of the registration optimization objective function is calculated in each iteration, and the rotation matrix and the translation vector are updated according to the current gradient direction; Adjusting the step size according to the updated registration optimization objective function value, increasing the step size when the registration optimization objective function value decreases, and decreasing the step size when the registration optimization objective function value increases; determining the iteration termination condition based on the gradient modulus value of the rotation matrix and the gradient modulus value of the translation vector; The rotation matrix obtained by iterative optimization is combined with the translation vector to form an optimal registration transformation matrix; the optimal registration transformation matrix is ​​applied to each data point in the source laser scanning data to obtain the transformed source laser scanning data; Calculate the point pair distance between the transformed source laser scanning data and the target laser scanning data in the overlapping area, extract the feature point pairs in the overlapping area, and verify the geometric feature consistency of the feature point pairs; when the point pair distance is less than a first preset threshold and the geometric feature consistency meets a second preset threshold, output the optimal registration transformation matrix to complete the precise registration of the source laser scanning data and the target laser scanning data.

7. A multi-source laser point cloud data adaptive registration system for power grid equipment, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The first unit is used to collect source laser scanning data and target laser scanning data of power grid equipment, pre-process the source laser scanning data and the target laser scanning data; use a statistical outlier filtering algorithm to remove noise points in the source laser scanning data and the target laser scanning data; use normal vector consistency analysis to extract key feature points of power grid equipment in the source laser scanning data and the target laser scanning data; construct a multi-dimensional local feature description vector based on the extracted key feature points of power grid equipment, wherein the multi-dimensional local feature description vector has curvature information, geometric shape information and spatial distribution information of the feature points; The second unit is used to calculate the feature similarity matrix of the key feature points of the power grid equipment in the source laser scanning data and the key feature points of the power grid equipment in the target laser scanning data based on the constructed multi-dimensional local feature description vector; establish a dynamic matching threshold optimization model according to the feature similarity matrix; calculate the statistical distribution characteristics of the feature similarity matrix, and determine the initial matching threshold in combination with the structural characteristics of the power grid equipment; construct an adaptive weight function based on the variance and kurtosis of the feature similarity matrix; use the adaptive weight function to dynamically adjust the initial matching threshold to obtain an optimized feature point correspondence relationship; The third unit is used to remove erroneous matching point pairs using a random sampling consistency algorithm according to the optimized feature point correspondence relationship; construct a registration optimization objective function of the geometric constraints of the power grid equipment; convert the structural characteristics of the power grid equipment into geometric constraints, and introduce the geometric constraints into the registration error calculation; The registration optimization objective function is iteratively optimized using a gradient descent method to obtain an optimal registration transformation matrix; the optimal registration transformation matrix is ​​applied to the source laser scanning data to achieve accurate registration of the source laser scanning data and the target laser scanning data.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images