A template matching-based insulator identification method for overhead transmission lines

Abstract

The present invention relates to an overhead transmission line insulator identification method based on template matching, which obtains the transmission line point cloud data to be identified; pre-processes the point cloud data; identifies the target of the insulator on the pre-processed point cloud data; infers the unidentified insulators based on the correctly identified insulators; accurately and efficiently identifies the insulators, which is of great significance to the intelligent monitoring of the transmission line. Not only is the efficiency greatly improved, but the three-dimensional information of the insulator can also be accurately obtained, which provides important data support for the drone route planning and insulator fault detection of power inspection.

Description

Technical Field

[0001] The invention relates to an overhead transmission line insulator identification method based on template matching, belonging to the technical field of transmission lines. Background Art

[0002] Insulators are key components used in large quantities in transmission lines, but they are exposed to the outside for a long time and are prone to failure. Therefore, it is necessary to conduct effective inspections on insulators in transmission lines in a timely manner, obtain the spatial position of insulators, and take corresponding measures.

[0003] Traditional manual inspections mainly use telescopes and infrared thermal imagers to conduct close-range inspections and tests of line equipment. This traditional power inspection method is inefficient and highly risky.

[0004] Therefore, drones are gradually being used in conjunction with airborne equipment (such as cameras, infrared thermal imagers, laser radars, synthetic aperture radars, etc.) to conduct line inspections. Compared with aerial images, airborne laser radar (LIDAR) can obtain point cloud data such as three-dimensional coordinates, intensity, and multiple echoes, and has the characteristics of high measurement accuracy, rich information, and strong anti-active interference capabilities. Therefore, the positioning of insulators through airborne laser radar is the mainstream direction today.

[0005] At the same time, since the insulators in the transmission line model established by laser point cloud were mostly determined by manual marking of the photo points, which was inefficient and had low accuracy, how to achieve efficient and intelligent positioning of insulators on the transmission line point cloud has become an urgent need in today's power inspection field.

[0006] For example: Patent CN115170631A proposes a method for identifying and locating the grabbing points of insulator detection tools based on solid-state laser radar point cloud; Patent CN114897877A proposes a method for building and training an insulator recognition model to perform insulator recognition.

[0007] However, these methods all require manual segmentation of towers and power lines, which results in high computational complexity and low recognition efficiency. Summary of the invention

[0008] The purpose of the present invention is to provide an overhead transmission line insulator identification method based on template matching, so as to solve the problem that the point cloud data collected by the drone-mounted laser radar proposed in the above background technology has high density but uneven distribution, and the insulator position obtained when the insulator surface texture information is incomplete has low accuracy or large calculation redundancy.

[0009] The technical solution of the present invention is as follows:

[0010] An overhead transmission line insulator identification method based on template matching comprises the following steps:

[0011] S1. Obtaining point cloud data of the power transmission line to be identified;

[0012] S2. Preprocess the point cloud data, including:

[0013] S2.1. Analyze the strength values ​​of different parts of the tower and use the strength value filtering to remove the pole point cloud;

[0014] S2.2, use principal component analysis to calculate the local point cloud eigenvalues, and delete the redundant flat area point clouds based on the local entropy function and spatial distribution characteristics constructed by the eigenvalues;

[0015] S2.3, avoid point cloud holes by grid patching;

[0016] S3. Perform insulator target recognition on the pre-processed point cloud data, including:

[0017] FPFH is selected as the feature descriptor, and the SAC-IA algorithm is improved by adding distance constraints between sampling points and adaptively adjusting parameters to complete the pose estimation of the insulator.

[0018] S4. Calculate the unidentified insulators based on the correctly identified insulators.

[0019] Preferably, in S1, point cloud data including ground points, tower points, and overhead power line points are acquired by an airborne laser scanning device.

[0020] Preferably, intensity value filtering first needs to use bilateral filtering to smooth the point cloud intensity value, and then traverse the point cloud intensity information of the entire tower, retaining points with intensity values ​​within the threshold, otherwise remove them.

[0021] Preferably, the strength values ​​of the segmented insulator area and the non-insulator area are counted, the average value and standard deviation of the strength values ​​are calculated to analyze the strength values ​​of different parts of the tower, the pole point cloud is removed by intensity value filtering, and the influence of noise is reduced by bilateral filtering;

[0022] Preferably, the weight value in bilateral filtering is composed of a spatial domain weight and an intensity domain weight, and the formula is as follows:

[0023]

[0024] Where: d and σ d are the standard deviation of the spatial domain weight function and the standard deviation of the intensity domain weight function, d d and d i is the distance value and intensity difference between the calculated point and its neighboring points;

[0025] Bilateral filtering corrects its own intensity value through the distance and intensity value of the neighboring points. The formula is as follows:

[0026]

[0027] Where: I p Indicates the intensity value of the calculation point, I out Indicates the corrected intensity value.

[0028] Preferably, let the coordinates of the neighboring points of point P be P i =[x i ,y i , z i ], the calculation formula of the local covariance matrix M of point P is as follows:

[0029]

[0030] Where: n represents the number of neighboring points of point P; is the centroid of the neighborhood point set of point P; the matrix M has three non-negative eigenvalues ​​λ1≥λ2≥λ3 and has the following characteristics:

[0031] 1) When λ1>>λ2>>λ3, the local neighborhood is determined to be linear;

[0032] 2) When λ1≈λ2>>λ3, the local neighborhood is determined to be a plane;

[0033] 3) When λ1≈λ2≈λ3, the local neighborhood is determined to be a surface;

[0034] Spatial distribution characteristics, the formula is as follows:

[0035]

[0036] The local entropy function is as follows:

[0037] E L =-S 1D ln(S 1D )-S 2D ln(S 2D )-S 3D ln(S 3D )

[0038] The local entropy value E L Points whose eigenvalue relationship is smaller than the set threshold and conforms to the plane features are deleted.

[0039] Preferably, the point cloud data processed by S2.2 is divided into a retained point cloud set and a point cloud set to be deleted, the point cloud set to be deleted is rasterized, a minimum cuboid is created to enclose the point cloud set to be deleted, the cuboid is evenly divided into multiple grids, and the number of point clouds in each grid is calculated. If the number of point clouds in the grid is greater than a set threshold, the centroid of all points in the grid is found, and the centroid point is added to the retained point cloud set.

[0040] Preferably, the specific process of the improved SAC-IA algorithm is as follows:

[0041] S3.1. Calculate the direction information of each point cloud: N P ,N Q , and then calculate the FPFH of each point cloud: F P ,F Q , where P is the template point cloud and Q is the target point cloud;

[0042] S3.2. Randomly sample M points in template P: P1, P2, P3, ..., P M , the distance between sampling points must meet the set minimum threshold D min ;

[0043] Calculate the random points P1, P2, P3, ..., P in the template M The Euclidean distance between all points

[0044] For the target point cloud, the candidate point for the i-th corresponding point is The candidate point for the jth corresponding point is The selection of points satisfies the following formula:

[0045]

[0046] Where μ is the distance error coefficient, Must be greater than D min ;

[0047] S3.3. Find M points in the target point cloud Q that have similar FPFH to the M points in S3.2: Q1, Q2, Q3, ..., Q M ;

[0048] S3.4, the singular value decomposition method is used to calculate the rigid body transformation matrix T of the M group of points, and the matrix is ​​applied to the template point cloud: P ′ =P×T;

[0049] S3.5. Calculation of P ′ The registration error between Q and Q stores the corresponding rigid body transformation matrix T and the feature point Euclidean error matrix ε dist ;

[0050] S3.6. Iterate S3.2-S3.5 to select the rigid body transformation matrix T corresponding to the minimum error, and stop when the maximum number of iterations is reached.

[0051] Preferably, the feature parameters are adaptively adjusted according to the point cloud resolution without manual adjustment of the parameters. is the average value of the minimum distance between points, and the formula is as follows:

[0052] d p =min(dis(p,q)),q=1,2,…,n,p≠q

[0053]

[0054] Where dis(p,q) is the distance between the query point p and any other point q, and the closest distance between point q and any other point is d. p , and finally the average value of the minimum distance of the entire point cloud is

[0055] Preferably, the insulator inference method is: the insulator template obtains a new spatial position through the calculated rigid body transformation matrix T, thereby obtaining the maximum three-dimensional coordinate point and the minimum three-dimensional coordinate point of the multiple transformed templates, and calculating the vectors of the two points (x_(max)-x_(min), y_(max)-y_min, z_(max)-z_(min));

[0056] By judging whether the Z-axis component in the vector is within the defined interval, the erroneous insulators can be preliminarily eliminated; then, the minimum three-dimensional coordinate points are sorted according to the size of the Z-axis, and the minimum three-dimensional coordinate points are projected to the XY plane. At least two groups of insulator points with close positions in the XY plane are obtained by clustering, and the Euclidean distance between the cluster centers of each group is calculated. If the Euclidean distance d_(x) of the horizontal distribution of insulators in the tower is satisfied, it is finally determined as an insulator point, and the erroneous points are eliminated at the same time.

[0057] The present invention has the following beneficial effects:

[0058] The present invention can realize efficient and intelligent positioning of insulators from complex transmission line point clouds; there is no need to manually segment towers and power lines, the calculation complexity is simple and the recognition efficiency is high;

[0059] In view of the fact that the point cloud data collected by airborne laser radar generally lacks surface texture information of insulators and has uneven density distribution, accurate and efficient identification of insulators is of great significance for intelligent monitoring of transmission lines. Not only is the efficiency greatly improved, but the three-dimensional information of insulators can also be accurately obtained, providing important data support for drone route planning and insulator fault detection for power inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a flow chart of the detection method of the present invention;

[0061] Figure 2 It is a flow chart of the pretreatment method of the present invention;

[0062] Figure 3 A fixed local coordinate system uvw is defined for the present invention;

[0063] Figure 4 The insulator estimation method of the present invention;

[0064] Figure 5 This is the final segmentation effect of the present invention. DETAILED DESCRIPTION

[0065] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0066] Example:

[0067] S1. Use airborne laser scanning equipment to obtain point cloud data of transmission lines, including ground points, tower points, and overhead power line points.

[0068] S2. Preprocess the point cloud data of the transmission line to obtain point cloud data. The intensity value filtering first needs to use bilateral filtering to smooth the point cloud intensity value, and then traverse the point cloud intensity information of the entire tower, retaining points with intensity values ​​within the threshold, otherwise remove them. The intensity value filtering improves the efficiency and accuracy of insulator detection by eliminating most of the pole body points.

[0069] like Figure 2 As shown in Figure 1, the method for preprocessing point cloud data is:

[0070] The intensity values ​​of the segmented insulator area and non-insulator area are counted, and the average and standard deviation of the intensity values ​​are calculated to analyze the intensity values ​​of different parts of the tower. Most of the pole point clouds are removed by intensity value filtering; due to noise interference during the laser radar acquisition process, the intensity values ​​of the point clouds in the same dielectric material will also be different, and the influence of noise can be reduced by bilateral filtering. The weight value in bilateral filtering is composed of the spatial domain weight and the intensity domain weight, as shown in the following formula:

[0071]

[0072] Where: d and σ d are the standard deviation of the spatial domain weight function and the standard deviation of the intensity domain weight function, d d and d i It is the distance value and intensity difference between the calculated point and its neighboring points. Bilateral filtering corrects its own intensity value through the distance and intensity value of the neighboring points, as shown in the following formula:

[0073]

[0074] Where: I p Indicates the intensity value of the calculation point, I out Indicates the corrected intensity value.

[0075] The principal component analysis method is used to calculate the local point cloud eigenvalues, and the redundant flat area point clouds are deleted according to the local entropy function and spatial distribution characteristics constructed by the eigenvalues; the coordinates of the neighborhood points of point P are set as P i =[x i ,y i , z i ], the local covariance matrix M of point P is calculated as follows:

[0076]

[0077] Where: n represents the number of neighboring points of point P; is the centroid of the neighborhood point set of point P; since M is a symmetric semi-positive definite matrix, there must be three non-negative eigenvalues. Let λ1≥λ2≥λ3 be the three eigenvalues ​​of matrix M, and have the following characteristics:

[0078] 1) When λ1>>λ2>>λ3, the local neighborhood is determined to be linear.

[0079] 2) When λ1≈λ2>>λ3, the local neighborhood is determined to be a plane.

[0080] 3) When λ1≈λ2≈λ3, the local neighborhood is determined to be a surface.

[0081] The spatial distribution characteristics of the local point cloud can be intuitively represented by the dimensional features, as shown in the following formula:

[0082]

[0083] The local entropy function of the calculation point is as follows:

[0084] E L =-S 1D ln(S 1D )-S 2D ln(S 2D )-S 3D ln(S 3D )

[0085] Local entropy function E L The larger the value, the more obvious the change in concave and convex the point is. L Points whose eigenvalues ​​are smaller than the threshold and whose relationship conforms to the plane features are deleted, thus achieving the purpose of deleting non-feature points in point cloud simplification.

[0086] The point cloud data processed by the above method is divided into a retained point cloud set and a point cloud set to be deleted. Sometimes point cloud holes will appear in the retained point cloud set, and point cloud holes can be avoided by grid patching. First, the point set to be deleted is rasterized. Create a minimum cuboid to enclose the point cloud data, and divide the cuboid into multiple grids according to the specified step size. Calculate the number of point clouds in each grid, and determine the threshold based on the point cloud density in the grid. If the number of point clouds in the grid is greater than the threshold, it means that point cloud holes will be generated after deleting all points in the grid. To ensure the integrity of the data, find the centroid of all points in the grid and add the centroid point to the retained point cloud set.

[0087] S3. Perform insulator target recognition on the pre-processed point cloud data. The method for identifying insulators on the point cloud data is as follows:

[0088] Selecting fast point feature histogram FPFH as the feature descriptor can better obtain the geometric feature information of the point cloud.

[0089] To calculate two points p s and p t And their corresponding normal vectors n s and n t The relative angle deviation between them requires the definition of a fixed local coordinate system uvw, such as Figure 3 As shown, where:

[0090]

[0091] In the uvw coordinate system, the normal vector n between any two points s and n t The characteristic deviation between them can be expressed as an angle:

[0092] α=v·n t

[0093]

[0094] θ=arctan(w·n t ,u·n t )

[0095] The triple (α, φ, θ) can represent the relative relationship between all two points in the k-neighborhood of the query point p. Calculate the (α, φ, θ) between the query point and other points in the neighborhood, and then use the neighboring SPFH values ​​to calculate the final FPFH, as follows:

[0096]

[0097] Where: i Represents the query point p to the neighboring point pi distance.

[0098] The SAC-IA algorithm is improved by adding distance constraints between sampling point pairs and adaptively adjusting parameters to complete the insulator posture estimation. The specific process of the improved SAC-IA algorithm is as follows:

[0099] 1) Calculate the direction information of each point cloud: N P ,N Q , and then calculate the FPFH of each point cloud: F P ,F Q , where P is the template point cloud and Q is the target point cloud.

[0100] 2) Randomly sample M points in the template P: P1, P2, P3, ..., P M , the distance between sampling points must meet the set minimum threshold D min . Calculate the random points P1, P2, P3, ..., P in the template M The Euclidean distance between all points For the target point cloud, the candidate point for the i-th corresponding point is The candidate point for the jth corresponding point is The selection of points satisfies the following formula:

[0101]

[0102] Where μ is the distance error coefficient, Must be greater than D min .

[0103] 3) Find M points in the target point cloud Q that have similar FPFH to the M points in (2): Q1, Q2, Q3, ..., Q M

[0104] 4) The singular value decomposition method is used to calculate the rigid body transformation matrix T of the M group of points, and the matrix is ​​applied to the template point cloud: P ′ =P×T.

[0105] 5) Calculate P ′ The registration error between Q and Q stores the corresponding transformation matrix T and the feature point Euclidean error matrix ε dist .

[0106] 6) Iterate steps 2-5 to select the transformation matrix corresponding to the minimum error, and stop when the maximum number of iterations is reached.

[0107] Adaptively adjust feature parameters according to point cloud resolution without manual adjustment of parameters, which can reduce the impact of point cloud sparsity and the resolution of point cloud. is the average value of the minimum distance between points, and the formula is as follows:

[0108] d p =min(dis(p,q)),q=1,2,…,n,p≠q

[0109]

[0110] Where dis(p,q) is the distance between the query point p and any other point q, and the closest distance between point q and any other point is d. p , and finally the average value of the minimum distance of the entire point cloud is The smaller it is, the denser the point cloud distribution is and the higher the resolution is.

[0111] S4. Based on the identified correct insulators, calculate the unidentified insulators. Figure 4 As shown, the insulator inference method is as follows: the insulator template obtains a new spatial position through the calculated transformation matrix, thereby obtaining the maximum three-dimensional coordinate point and the minimum three-dimensional coordinate point of multiple transformed templates, and calculating the vector of the two points (x_(max)-x_(min), y_(max)-y_min, z_(max)-z_(min)). Since the insulators in the tower are all in a vertical state and almost parallel to the Z axis, the wrong insulators can be preliminarily eliminated by judging whether the Z-axis component in the vector is within the defined interval. Subsequently, the minimum three-dimensional coordinate points are sorted according to the size of the Z axis, and the minimum three-dimensional coordinate points are projected onto the XY plane. Since the insulators in the tower are symmetrical in three-dimensional space with the pole body as the center, at least two groups of insulator points with close positions in the XY plane can be obtained by clustering, and the Euclidean distance between the cluster centers of each group is calculated. If the Euclidean distance d_(x) of the horizontal distribution of the insulators in the tower is satisfied, it is finally determined as an insulator point, and the wrong point is eliminated at the same time.

[0112] like Figure 5 As shown, the left part is the undivided state, the middle part is the divided tower, and the right part is the divided insulator.

[0113] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for identifying insulators of overhead transmission lines based on template matching, characterized in that: The following steps are involved: S1. Obtaining point cloud data of the power transmission line to be identified; S2. Preprocess the point cloud data, including: S2.

1. Analyze the strength values ​​of different parts of the tower and use the strength value filtering to remove the pole point cloud; S2.2, use principal component analysis to calculate the local point cloud eigenvalues, and delete the redundant flat area point clouds based on the local entropy function and spatial distribution characteristics constructed by the eigenvalues; S2.3, avoid point cloud holes by grid patching; S3. Perform insulator target recognition on the pre-processed point cloud data, including: FPFH is selected as the feature descriptor, and the SAC-IA algorithm is improved by adding distance constraints between sampling points and adaptively adjusting parameters to complete the pose estimation of the insulator. S4, inferring the unidentified insulators based on the correctly identified insulators; The specific process of the improved SAC-IA algorithm is as follows: S3.

1. Calculate the direction information of each point cloud: , and then calculate the FPFH of each point cloud: , where P is the template point cloud and Q is the target point cloud; S3.

2. Randomly sample M points in template P: , the distance between sampling points must meet the set minimum threshold ; Calculate random points in the template The Euclidean distance between all points ; For the target point cloud, the candidate point for the i-th corresponding point is , the candidate point for the jth corresponding point is , the selection of points satisfies the following formula: In the formula is the distance error coefficient, Greater than ; S3.

3. Find M points in the target point cloud Q that have similar FPFH to the M points in S3.2: ; S3.4, the singular value decomposition method is used to calculate the rigid body transformation matrix T of the M group of points, and the matrix is ​​applied to the template point cloud: ; S3.

5. Calculation The registration error between Q and Q stores the corresponding rigid body transformation matrix T and the feature point Euclidean error matrix ; S3.

6. Iterate S3.2-S3.5 to select the rigid body transformation matrix T corresponding to the minimum error, and stop when the maximum number of iterations is reached.

2. The method for identifying insulators of overhead transmission lines based on template matching according to claim 1, characterized in that: In S1, airborne laser scanning equipment is used to obtain point cloud data, including ground points, tower points, and overhead power line points.

3. The method for identifying insulators of overhead transmission lines based on template matching according to claim 1, characterized in that: Intensity value filtering first needs to use bilateral filtering to smooth the point cloud intensity value, and then traverse the point cloud intensity information of the entire tower, retaining points with intensity values ​​within the threshold, otherwise remove them.

4. The method for identifying insulators of overhead power transmission lines based on template matching according to claim 3, characterized in that: The strength values ​​of the segmented insulator area and non-insulator area are statistically analyzed, and the average and standard deviation of the strength values ​​are calculated to analyze the strength values ​​of different parts of the tower. The pole point cloud is eliminated by intensity value filtering, and the noise effect is reduced by bilateral filtering.

5. The method for identifying insulators of overhead power transmission lines based on template matching according to claim 4, characterized in that: The weight value in bilateral filtering consists of spatial domain weight and intensity domain weight, and the formula is as follows: Where: and are the standard deviation of the spatial domain weight function and the standard deviation of the intensity domain weight function, and is the distance value and intensity difference between the calculated point and its neighboring points; Bilateral filtering corrects its own intensity value through the distance and intensity value of the neighboring points. The formula is as follows: Where: represents the intensity value of the calculation point, Indicates the corrected intensity value.

6. The method for identifying insulators of overhead power transmission lines based on template matching according to claim 1, characterized in that: set up The coordinates of the neighboring points of a point are , The local covariance matrix of the point The calculation formula is as follows: Where: express The number of neighboring points of a point; for The centroid of the neighborhood point set of a point; matrix There are three non-negative eigenvalues , and has the following characteristics: 1) When , the local neighborhood is determined to be linear; 2) When , the local neighborhood is determined to be a plane; 3) When , the local neighborhood is determined to be a surface; Spatial distribution characteristics, the formula is as follows: The local entropy function is as follows: The local entropy Points whose eigenvalue relationship is smaller than the set threshold and conforms to the plane features are deleted.

7. The method for identifying insulators of overhead power transmission lines based on template matching according to claim 6, characterized in that: The point cloud data processed by S2.2 is divided into a retained point cloud set and a point cloud set to be deleted. The point cloud set to be deleted is rasterized, and a minimum cuboid is created to enclose the point cloud set to be deleted. The cuboid is evenly divided into multiple grids, and the number of point clouds in each grid is calculated. If the number of point clouds in the grid is greater than the set threshold, the centroid of all points in the grid is calculated, and the centroid point is added to the retained point cloud set.

8. The method for identifying insulators of overhead power transmission lines based on template matching according to claim 1, characterized in that: Adaptively adjust feature parameters according to point cloud resolution without manual adjustment of parameters. is the average value of the minimum distance between points, and the formula is as follows: In the formula is the distance between the query point p and any other point q, and the closest distance between point q and other points is calculated as , and finally the average value of the minimum distance of the entire point cloud is .

9. The method for identifying insulators of overhead power transmission lines based on template matching according to claim 1, characterized in that: The insulator inference method is as follows: the insulator template obtains a new spatial position through the calculated rigid body transformation matrix T, thereby obtaining the maximum three-dimensional coordinate point and the minimum three-dimensional coordinate point of multiple transformed templates, and calculating the vector of the two points (x_(max)-x_(min), y_(max)-y_min, z_(max)-z_(min)); By judging whether the Z-axis component in the vector is within the defined interval, the erroneous insulators can be preliminarily eliminated; then, the minimum three-dimensional coordinate points are sorted according to the size of the Z-axis, and the minimum three-dimensional coordinate points are projected onto the XY plane. At least two groups of insulator points with close positions in the XY plane are obtained by clustering, and the Euclidean distance between the cluster centers of each group is calculated. If the Euclidean distance d_(x) of the horizontal distribution of insulators in the tower is satisfied, it is finally determined as an insulator point, and the erroneous points are eliminated at the same time.

Images