Distribution line tree barrier defect automatic detection method and device based on laser point cloud

Abstract

The invention relates to a laser point cloud-based distribution line defect automatic detection method and device. According to the method, through processing and calculation of reading, classification, filtering and the like of the laser radar point cloud, identification of the electric power part of the 110KV line channel is completed, the problem that the classification of the line point cloud at a low voltage level is easily interfered by virtual points and outliers is solved, and high-precision tree obstacle distance detection is realized. According to the distribution line tree barrier defect automatic detection method and device based on the laser point cloud provided by the invention, the three-dimensional point cloud data is calculated and preprocessed based on the computer processing equipment, and a filtering algorithm and a detection process are correspondingly improved, so that the distribution line tree barrier defect automatic detection method and device can be applied to a line channel with a relatively low voltage level; and a parallel processing mechanism can be utilized to improve the calculation efficiency, and basic support is provided for digital transformation of a power grid operation and maintenance specialty.

Description

Technical Field

[0001] The present invention relates to the field of laser radar technology, and in particular to a method and device for automatically detecting tree obstacle defects in distribution lines based on laser point clouds. Background Art

[0002] As a three-dimensional sensor, lidar can obtain accurate three-dimensional spatial information of the target. It is often used in conjunction with drones as a carrier platform as a technical means of scanning the geographic environment. Intelligent inspections based on this technology have been widely used in high-voltage transmission lines, becoming an important technical support means for the digital transformation of transmission network operation and maintenance.

[0003] 110kV lines, as a crucial component of both transmission and distribution, carry the backbone power transmission of distribution networks and are widely used in urban and industrial power supply. Their maintenance is crucial to public safety and the stability of the power system. Using LiDAR (LiDAR) to identify the power lines, towers, and terrain features along 110kV lines and construct a digital sandbox for automated defect detection is a key component of the digital transformation of power grid operations and maintenance. However, due to their low voltage levels and relatively short tower heights, LiDAR point cloud processing technology faces difficulties in segmentation due to the line's susceptibility to interference from terrain, hindering widespread adoption. Existing digital elevation model (DEM) construction methods primarily target point cloud maps with regular distribution and smooth terrain. However, line point clouds are typically distributed in long strips along the corridor, with uneven terrain. Ground point filtering is often affected by occlusion caused by high objects, resulting in ground holes and sparse point clouds, leading to significant errors in relative elevation calculations. When segmenting components such as power lines and towers, virtual points generated by reflective surfaces and noise outliers caused by ground distribution become factors that must be considered during scanning. 2D projection and Hough transform, commonly used for higher voltage transmission lines, will be difficult to meet inspection needs at lower line heights and higher accuracy requirements.

[0004] For transmission and distribution lines with lower voltage levels, the development of an automated detection method and device for transmission and distribution line defects based on airborne laser point clouds will provide important support for improving line inspection efficiency and the digital transformation of transmission and distribution operation and maintenance. Summary of the Invention

[0005] This invention provides a method and device for automated detection of transmission and distribution line defects based on laser point clouds for 110KV power transmission and distribution lines. This method can automatically process and classify point cloud data, detect tree hazards based on the classified line point clouds, and visualize the potential defects. The automated detection method includes the following steps:

[0006] Step S1: Read the original .las file collected by the drone and complete data conversion and line slicing;

[0007] Step S2: Complete raster segmentation based on the plane coordinate system and construct a digital elevation model (DEM) for each grid;

[0008] Step S3: Based on the constructed digital elevation model DEM, the power line point cloud is screened based on the point cloud spatial distribution density and the neighborhood covariance matrix eigenvalue, and the power line extraction and fitting are completed through random sampling iteration;

[0009] Step S4: Based on the constructed digital elevation model (DEM), combined with the spatial density and spatial scale of the point cloud, complete the classification and denoising of the towers;

[0010] Step S5: Perform distance detection of tree obstacle hazards based on the classified point cloud.

[0011] The device includes: a display, a processor, a memory, a data bus, and a computer program. The computer program is stored in the memory and is configured to be executed by the processor to implement an automated detection method, and the DEM construction and power line extraction calculation method can be executed in parallel by multiple corresponding processors.

[0012] The specific description of steps S1-S5 is as follows:

[0013] For step S1: read the original .1as file collected by the drone, complete data conversion and line slicing, the specific steps are as follows

[0014] Step S1.1: Based on the system parameters and projection method of the geographic coordinate system used by the LiDAR acquisition system, the starting longitude and latitude of the line to be detected are converted into the corresponding plane coordinate starting point, and the corresponding line channel point cloud is cropped accordingly;

[0015] For step S2: complete the raster segmentation based on the plane coordinate system and construct a digital elevation model DEM for each grid, the specific steps are as follows:

[0016] Step S2.1: Divide the grid according to the XOY plane coordinates and calculate the extreme value x of the X-axis coordinate min , x max , Y-axis coordinate extreme value y min ,y max ; with (x min ,y min ) as the starting point, first divide it along the X axis according to the distance interval Δ, and divide it into Then make a strip along the Y axis at the same intervals. The grids are marked as

[0017] Step S2.2: For grid n i Perform Cloth Simulation filtering, set the grid resolution to according to the predetermined parameters, and iteratively calculate the ground points of each grid;

[0018] Step S2.3: According to the corresponding grid number n i , all point clouds are projected onto the XOY plane through two-dimensional projection;

[0019] Step S2.4: Let P g represents the ground point set, P h Represents a set of non-ground points, and uses the KNN nearest neighbor algorithm to search for the corresponding grid n i Internal P h The k nearest neighboring ground points of any point in;

[0020] Step S2.5: Count the relative height differences z between the k nearest ground points and the corresponding non-ground points in the Z-axis direction j , j∈{1, 2, ..., k} and the Euclidean distance difference d on the XOY plane j , j∈{1, 2, ..., k}, divided into the following two cases for discussion: 1) If the height difference distribution is less than or equal to 0.1m, then choose d j The z corresponding to the minimum point of the sort j As the relative height, assign it to the corresponding non-ground point; 2) If the height difference distribution is greater than 0.1m, it means that there are holes or sparse phenomena in the ground point cloud here, and choose d j z corresponding to the smallest three points in the sorting j Calculate using the following formula: The calculated result z is used as the relative height. By assigning relative height to each point, the DEM is constructed.

[0021] For step S3: Based on the constructed digital elevation model DEM, the power line point cloud is screened based on the spatial distribution density of the point cloud and the eigenvalues ​​of the neighborhood covariance matrix, and the power line extraction and fitting are completed through random sampling iteration. The specific steps are as follows:

[0022] Step S3.1: Based on the DEM constructed in step S1, filter the point cloud set P above the preset height above the ground. hl ;

[0023] Step S3.2: Construct a multi-leaf node KDTree for the point cloud above the specific height selected in step S2.1;

[0024] Step S3.3: Set the search radius r pca , linear threshold va pca , distribution threshold vc pca, create the basic processing function worker according to the above parameters: in KDTree with r pca Search for the point p as the radius i Neighborhood of point p i ∈P hl , calculate the corresponding center point p i The covariance matrix and matrix eigenvalues ​​of ; sorted from small to large as λ1, λ2, λ3, and the corresponding eigenvectors are ξ1, ξ2, ξ3; compare Is it greater than va pca , whether |ξ1| is less than vc pca ,If both criteria are met, it is judged to be a power line point;

[0025] Step S3.4: Press P to execute the basic processing function worker. hl The points in the order are distributed to the corresponding processor threads in sequence, and the returned results are arranged in the same order to obtain the coarse classification points of the power lines;

[0026] Step S3.5: Cluster the obtained coarse classification points of power lines according to density using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm;

[0027] Step S3.6: The point cloud clusters obtained by clustering are distinguished according to the number and distribution length, and the power line point clouds are screened out;

[0028] Step S3.7: The power line point cloud obtained by clustering is iteratively fitted using the Random Sample Consensus (RANSAC) algorithm. The fitting target formula is:

[0029] Where xyz is the three-dimensional coordinate of the point cloud, and the rest are parameters to be fitted.

[0030] For step S4: Based on the constructed digital elevation model DEM, combined with the spatial density and spatial scale of the point cloud, complete the classification and denoising of the towers. The specific steps are as follows:

[0031] Step S4.1: Using the extracted coordinates of both ends of the power line as the quick positioning coordinates of the tower insulator contact point;

[0032] Step S4.2: Based on the initial position of the tower insulator obtained by rapid positioning, the radius r is calculated linearly according to PCA. pca Make corrections to compensate for the error caused by the search radius;

[0033] Step S4.3: Calculate the corrected tower insulator coordinates, and based on this, calculate the tower axis coordinates from the tower radius. Step S4.4: Cut a cylinder with the tower axis coordinates as the axis to obtain a coarsely classified tower point cloud.

[0034] Step S4.5: Based on the coarsely classified tower point cloud, use density statistics method to remove noise generated during the acquisition process.

[0035] Regarding step S5: performing distance detection of tree obstacle hazards based on the classified point cloud, the specific steps are as follows:

[0036] Step S5.1: Calculate the distance between the classified power line point cloud set and the ground object point cloud set, and preliminarily screen out the potential danger point set based on the tree barrier potential danger judgment distance;

[0037] Step S5.2: Density clustering is performed on the initially screened hidden danger point set to eliminate outlier noise points with fewer neighborhood points and capture the main branches and leaves of the trees.

[0038] The beneficial effects brought by the present invention are:

[0039] In view of the channel characteristics of the 110KV transmission and distribution lines with lower voltage levels, the relevant steps in the design point cloud processing and segmentation process are improved. The correction of the impact of ground voids and sparse point clouds when building digital high-rise models, the iterative processing of virtual points and outliers when segmenting line components and calculating distances are considered. This makes the segmented line point cloud and the distance detection of tree obstacles more consistent with the actual line conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0042] Attachment Figure 1 A simplified flowchart of an automated detection method for power transmission and distribution line defects based on airborne laser point clouds

[0043] Attachment Figure 2 Detailed flow chart of an automated detection method for power transmission and distribution line defects based on airborne laser point cloud

[0044] Attachment Figure 3 Side view of a single-stage circuit slice of an embodiment of the present invention

[0045] Attachment Figure 4 A diagram showing the setting of PCI-E device memory pool entries in a system configuration file according to an embodiment of the present invention

[0046] Attachment Figure 5 Result diagram of identifying FC daughter card based on PCI-E in an embodiment of the present invention

[0047] Attachment Figure 6 Tower classification and denoising results of an embodiment of the present invention

[0048] Attachment Figure 7 Detection results of tree obstacle hazards according to an embodiment of the present invention

[0049] Attachment Figure 8 Schematic diagram of an automated detection device for power transmission and distribution line defects based on airborne laser point cloud DETAILED DESCRIPTION

[0050] In order to make the technical solution and the purpose of the present invention more clear and understandable, the implementation steps of the present invention will be described in detail below.

[0051] The implementation example of the present invention will be combined with a 110KV actual line channel in the northern Hebei region to perform point cloud segmentation and tree barrier hidden danger defect detection. This example is a specific example of the present invention to fully demonstrate the implementation ideas of the present invention, and its implementation examples are not limited to this example. A method and device for automatic detection of transmission and distribution line defects based on airborne laser point cloud

[0052] The overall process of laser point cloud proposed by the present invention is as follows: Figure 2 As shown, the specific implementation steps combined with the above example are as follows: Step S1: read the original .las file collected by the drone, complete data conversion and line slicing;

[0053] Step S1.1: Based on the system parameters and projection method of the geographic coordinate system used by the lidar acquisition system, the coordinate system used in this example is the WGS-84 coordinate system and the UTM mapping method. The starting longitude and latitude of the line to be detected is determined by the tower ledger information, and the corresponding coordinate system parameters and mapping method are used to convert it into the corresponding plane coordinate starting point, and the corresponding line channel point cloud is clipped accordingly, as shown in the attached figure. Figure 3 shown.

[0054] Step S2: Complete raster segmentation based on the plane coordinate system and construct a digital elevation model (DEM) for each grid.

[0055] Step S2.1: Divide the grid according to the XOY plane coordinates and calculate the extreme value x of the X-axis coordinate min , x max , Y-axis coordinate extreme value y min ,y max; with (x min ,y min ) as the starting point, first divide the X axis according to the distance interval Δ = 20m, and divide it into Then make a strip along the Y axis at the same intervals. The grids are marked as

[0056] Step S2.2: For grid n i Perform Cloth Simulation filtering, set the grid resolution to 0.1Δ, and iteratively calculate the ground points of each grid;

[0057] Step S2.3: According to the corresponding grid number n i , all point clouds are projected onto the XOY plane through two-dimensional projection;

[0058] Step S2.4: Let P g represents the ground point set, P h Represents a set of non-ground points, and uses the KNN nearest neighbor algorithm to search for the corresponding grid n i Internal P h The k nearest neighboring ground points of any point in , where k is set to 8;

[0059] Step S2.5: Count the relative height differences z between the k nearest ground points and the corresponding non-ground points in the Z-axis direction j , j∈{1, 2, ..., k} and the Euclidean distance difference d on the XOY plane j , j∈{1, 2, ..., k}, divided into the following two cases for discussion: 1) If the height difference distribution is less than or equal to 0.1m, then choose d j The z corresponding to the minimum point of the sort j As the relative height, assign it to the corresponding non-ground point; 2) If the height difference distribution is greater than 0.1m, it means that there are holes or sparse phenomena in the ground point cloud here, and choose d j z corresponding to the smallest three points in the sorting j Calculate using the following formula: The calculated result z is used as the relative height. By assigning relative height to each point, the DEM is constructed.

[0060] Step S3: Based on the constructed digital elevation model DEM, the power lines are extracted and fitted through iteration.

[0061] Step S3.1: Based on the DEM constructed in step S1, filter the point cloud set P above a specific height above the ground. hl ;

[0062] Step S3.2: Construct a multi-leaf node KDTree for the point cloud above the specific height selected in step S2.1. In this example, the leaf node is set to 4 to reduce the KDTre e Depth;

[0063] Step S3.3: Set the search radius r pca , linear threshold va pca , distribution threshold vc pca , create the basic processing function worker according to the above parameters: in KDTree with r pca Search for the point p as the radius i Neighborhood of point p i ∈P hl , calculate the corresponding center point p i The covariance matrix and matrix eigenvalues ​​of ; sorted from small to large as λ1, λ2, λ3, and the corresponding eigenvectors are ξ1, ξ2, ξ3; compare Is it greater than va pca , whether |ξ1| is less than vc pca , if it meets the requirements, it is judged to be a power line point;

[0064] Step S3.4: Press P to execute the basic processing function worker. hl The points in the order are distributed to the corresponding processor threads in turn, and the returned results are arranged in the same order to obtain the coarse classification points of the power lines. In this example, an 8-core processor is used. To ensure normal task scheduling and parallel speed, the number of threads is set to 6.

[0065] Step S3.5: Cluster the obtained power line rough classification points according to density using the DBSCAN algorithm

[0066] Step S3.6: The point cloud clusters obtained by clustering are distinguished according to the number and distribution length of power lines, and the point cloud clusters with too few numbers or too short lengths are eliminated to screen out the power line point clouds. The power line point cloud extraction results are shown in the attached figure. Figure 4 shown.

[0067] Step S3.7: Perform RANSAC iterative fitting on the power line point cloud obtained by clustering. The fitting formula is:

[0068]

[0069] In this example, the number of iterations of the power line RANSAC fitting is set to 20, the cluster search radius is set to 1m, and the number of domain points is set to 10. The fitted power model is shown in the attached figure. Figure 5 shown.

[0070] Step S4: Based on the constructed digital elevation model DEM, complete the classification and denoising of the towers;

[0071] Step S4.1: Using the extracted coordinates of both ends of the power line as the quick positioning coordinates of the tower insulator contact point;

[0072] Step S4.2: Based on the initial position of the tower insulator obtained by rapid positioning, the radius r is calculated linearly according to PCA. pca Correction is performed to compensate for the error caused by the search radius. In this example, r pca =1.5m;

[0073] Step S4.3: Calculate the corrected tower insulator coordinates, and based on this, calculate the tower axis coordinates from the tower radius. Step S4.4: Cut a cylinder with the tower axis coordinates as the axis to obtain a coarsely classified tower point cloud.

[0074] Step S4.5: Based on the coarse classification of the tower point cloud, use the density statistics method to remove the noise generated during the acquisition process. The result is shown in the attached figure. Figure 6 As shown;

[0075] Step S5: Detect the distance of tree obstacles based on the classified point cloud

[0076] Step S5.1: Calculate the distance between the classified power line point cloud set and the ground object point cloud set, and preliminarily screen out the potential danger point set based on the tree barrier potential danger judgment distance;

[0077] Step S5.2: Perform density clustering on the initially screened hidden danger point set, remove outlier noise points with fewer neighborhood points, and capture the main branches and leaves of the trees. The results are shown in the attached figure. Figure 7 shown.

[0078] As described above, the embodiment of the present invention completes the identification of power components of the 110KV line channel through the reading, classification, filtering and other processing and calculation of the airborne lidar point cloud, solves the problem of being easily interfered by virtual points and outliers in the point cloud classification of lower voltage level lines, and realizes high-precision tree obstacle distance detection.

[0079] The method and device for automated detection of transmission and distribution line defects based on airborne laser point clouds proposed in the present invention perform computational preprocessing of three-dimensional point cloud data using computer processing equipment. By improving the filtering algorithm and detection process accordingly, the present invention can be applied to line channels with lower voltage levels, and can utilize parallel processing mechanisms to improve computing efficiency, providing basic support for the digital transformation of power grid operation and maintenance professionals.

[0080] Without departing from the basic technical ideas and spirit of the present invention, any obvious modifications, equivalent replacements, or further optimizations made to the above details should be included in the scope of the claims of the present invention.

Claims

1. A method for automatically detecting tree obstacle defects in distribution lines based on laser point cloud, characterized in that: The system can automatically read the .las file obtained by laser radar scanning, complete the construction of rasterized digital elevation model (DEM) through data conversion and preprocessing, and assign height information to the point cloud; combine the spatial distribution density of the point cloud with the eigenvalues ​​of the neighborhood covariance matrix to complete the classification and extraction of power lines, and remove noise and error samples through iteration, and complete the digital model fitting of power lines based on random sampling; combine the spatial distribution density and spatial scale of the point cloud to complete the rapid positioning and denoising of the tower; finally, realize the tree obstacle hidden danger detection based on the classified line point cloud, and display the hidden danger defect points through visualization. The said automated detection method includes the following steps: Step S1: Read the original .las file collected by the drone and complete data conversion and line slicing; Step S2: Complete raster segmentation based on the plane coordinate system and construct a digital elevation model (DEM) for each grid; Step S3: Based on the constructed digital elevation model DEM, the power line point cloud is screened based on the point cloud spatial distribution density and the neighborhood covariance matrix eigenvalue, and the power line extraction and fitting are completed through random sampling iteration; Step S4: Based on the constructed digital elevation model (DEM), combined with the spatial density and spatial scale of the point cloud, complete the classification and denoising of the towers; Step S5: Perform distance detection of tree obstacle hazards based on the classified line point cloud.

2. The method according to claim 1, characterized in that In step S1, the specific steps are: Step S1.1: Based on the system parameters and projection method of the geographic coordinate system used by the LiDAR acquisition system, the starting longitude and latitude of the line to be detected are converted into the corresponding plane coordinate starting point, and the corresponding line channel point cloud is cropped accordingly.

3. The method according to claim 1, characterized in that In step S2, the specific steps are: Step S2.1: Divide the grid according to the XOY plane coordinates and calculate the extreme value x of the X-axis coordinate min ,x max ,Y-axis coordinate extreme value y min ,y max ; with (x min ,y min) As the starting point, first divide the X axis according to the distance interval Δ, and divide it into Then make a strip along the Y axis at the same intervals. The grid is marked as Step S2.2: For grid n i Perform Cloth Simulation filtering, set the grid resolution according to the predetermined parameters, and iteratively calculate the ground points of each grid; Step S2.3: According to the corresponding grid number n i , all point clouds are projected onto the XOY plane through two-dimensional projection; Step S2.4: Let P g represents the ground point set, P h Represents a set of non-ground points, and uses the KNN nearest neighbor algorithm to search for the corresponding grid n i Internal P h The k nearest neighboring ground points of any point in ; Step S2.5: Count the relative height differences z between the k nearest ground points and the corresponding non-ground points in the Z-axis direction j ,j∈{1,2,…,k} and the Euclidean distance difference d on the XOY plane j ,j∈{1,2,…,k}, divided into the following two cases: 1) If the height difference distribution is less than or equal to 0.1m, then choose d j The z corresponding to the minimum point of the sort j As the relative height, assign it to the corresponding non-ground point; 2) If the height difference distribution is greater than 0.1m, it means that there are holes or sparse phenomena in the ground point cloud here, and choose d j z corresponding to the smallest three points in the sorting j Calculate using the following formula: The calculated result z is used as the relative height; by assigning relative height to each point, DEM construction is completed.

4. The method according to claim 1, wherein In step S3, the specific steps are: Step S3.1: Based on the DEM constructed in step S1, filter the point cloud set P above the preset height above the ground. hl ; Step S3.2: Construct a multi-leaf node KDTree for the point cloud above the specific height selected in step S2.1; Step S3.3: Set the search radius r pca , linear threshold va pca , distribution threshold vc pca , create the basic processing function worker according to the above parameters: in KDTree, use r pca Search for the point p as the radius i Neighborhood of point p i ∈P hl , calculate the corresponding center point p i The covariance matrix and matrix eigenvalues ​​of ; sorted from small to large as λ1, λ2, λ3, the corresponding eigenvectors are ξ1, ξ2, ξ3; compare Is it greater than va pca , whether |ξ1| is less than vc pca ,If both criteria are met, it is judged to be a power line point; Step S3.4: Press P to execute the basic processing function worker. hl The points in the order are distributed to the corresponding processor threads in turn, and the returned results are arranged in the same order to obtain the coarse classification points of the power lines; Step S3.5: cluster the obtained power line coarse classification points according to density using the spatial density clustering DBSCAN algorithm; Step S3.6: The point cloud clusters obtained by clustering are distinguished according to the number and distribution length, and the power line point clouds are screened out; Step S3.7: The power line point cloud obtained by clustering is subjected to random sampling algorithm RANSAC iterative fitting, and the fitting target formula is: Where xyz is the three-dimensional coordinate of the point cloud, and the rest are parameters to be fitted.

5. The method according to claim 1, characterized in that In step S4, the specific steps are: Step S4.1: Using the extracted coordinates of both ends of the power line as the quick positioning coordinates of the tower insulator contact point; Step S4.2: Based on the initial position of the tower insulator obtained by rapid positioning, the radius r is calculated linearly according to PCA. pca Make corrections to compensate for the error caused by the search radius; Step S4.3: Calculate the corrected coordinates of the tower insulator, and based on this, calculate the coordinates of the tower axis using the tower radius; Step S4.4: Using the tower axis coordinate as the axis, perform cylindrical cutting to obtain a coarsely classified tower point cloud; Step S4.5: Based on the coarsely classified tower point cloud, use density statistics method to remove noise generated during the acquisition process.

6. The method according to claim 1, characterized in that In step S5, the specific steps are: Step S5.1: Calculate the distance between the classified power line point cloud set and the ground object point cloud set, and preliminarily screen out the potential danger point set based on the tree barrier potential danger judgment distance; Step S5.2: Density clustering is performed on the initially screened hidden danger point set to eliminate outlier noise points with fewer neighborhood points and capture the main branches and leaves of the trees.

7. An automated detection device for tree obstacle defects in distribution lines based on laser point cloud, characterized in that: include: monitor; processor; Memory; Data bus; as well as computer programs; Wherein, the computer program is stored in the memory and is configured to be executed by the processor to implement the automatic detection method for tree barrier defects in distribution lines based on laser point cloud as claimed in any one of claims 1 to 6, and the DEM construction and power line extraction calculation method can be executed in parallel by multiple corresponding processors.

Images