Ultra-high voltage maintenance operation path assessment method, device, equipment and storage medium

Abstract

The present invention relates to the field of ultra-high voltage maintenance technology, and discloses an ultra-high voltage maintenance operation path assessment method, device, equipment and storage medium. The method removes outliers, performs thinning and registration processing on a data set to obtain a data set; segments the data set based on a straight line detection method and a regional growth segmentation algorithm to obtain a power line point cloud, a tower point cloud and an insulator point cloud; inputs the insulator point cloud into a bag-of-words model to obtain an insulator classification result; simulates the target distance between the operator and the target conductor and tower according to the insulator classification result, the power line and the tower point cloud, and compares with a preset safety threshold to determine whether the operator is safe. The present invention quickly locates in massive data through outlier removal, thinning and registration. Classification is performed by straight line detection to solve the discrete problem of point cloud distribution, and the regional growth segmentation algorithm improves the edge processing capability and makes segmentation more accurate, and then the distance calculation is made more accurate through cylinder simulation to improve the accuracy of judgment.

Description

Technical Field

[0001] The present invention relates to the field of ultra-high voltage maintenance technology, and in particular to an ultra-high voltage maintenance operation path assessment method, device, equipment and storage medium. Background Art

[0002] Overhead ultra-high voltage transmission lines are an important way of power transmission in the power industry. The safe and stable operation of transmission lines directly affects the high reliability and stable power supply of the power system. Therefore, regular and effective inspection and maintenance of transmission lines has become an important task for the power industry. At present, the inspection of overhead ultra-high voltage transmission lines mainly relies on manual and mechanical operations. Transmission line inspection must ensure that the various distances between transmission conductors and line corridors meet the requirements of safety distance regulations, especially the safety of maintenance personnel in manual operations.

[0003] The traditional safety distance assessment method requires the use of image recognition, analysis, and calculation techniques to obtain assessment results, which involves complex calculations such as ground filtering, DEM construction, object classification, and wire fitting. There is a possibility of data deviation and it is not accurate enough, which will affect the accuracy of the judgment results. Summary of the invention

[0004] The main purpose of the present invention is to provide a method, device, equipment and storage medium for evaluating the ultra-high voltage maintenance operation path, aiming to solve the technical problems such as safety hazards existing in the maintenance process of existing overhead ultra-high voltage transmission lines, and the low accuracy of existing safety assessment methods and the possibility of misjudgment.

[0005] To achieve the above object, the present invention provides a method for evaluating an ultra-high voltage maintenance operation path, comprising:

[0006] Acquire a point cloud data set of an ultra-high voltage transmission channel, and remove outliers from the point cloud data set to obtain an initial point cloud data set;

[0007] Preprocessing the initial point cloud data set based on a voxel segmentation and thinning algorithm and a normal distribution point cloud registration algorithm to obtain a target point cloud data set;

[0008] Performing point cloud segmentation on the data of the target point cloud data set based on a power line segmentation method of straight line detection to obtain a power line point cloud;

[0009] Performing point cloud segmentation on the target point cloud data set based on a region growing segmentation algorithm to obtain a tower point cloud and an insulator point cloud;

[0010] Inputting the insulator point cloud into a bag-of-words model to output an insulator classification result;

[0011] Simulate the target distance between the operator and the target conductor and the tower according to the insulator classification result, the power line point cloud and the tower point cloud;

[0012] Whether the operator is safe is determined based on the comparison result between the target distance and the preset safety threshold.

[0013] In some embodiments, the step of acquiring a point cloud dataset of an ultra-high voltage power transmission channel and removing outliers from the point cloud dataset to obtain an initial point cloud dataset includes:

[0014] Collect point cloud data sets of ultra-high voltage transmission channels based on LiDAR technology;

[0015] Using a straight-through filter to remove outliers from the point cloud data set in three dimensional directions to obtain filtered point cloud data;

[0016] A statistical filter is used to remove sparse outlier points from the filtered point cloud data to obtain an initial point cloud data set.

[0017] In some embodiments, the step of removing outliers from the point cloud data set in three dimensional directions using a straight-through filter to obtain filtered point cloud data includes:

[0018] Extracting a point cloud of the point cloud data set;

[0019] A straight-through filter is used to obtain the maximum value and the minimum value of the point cloud in three dimensional directions respectively;

[0020] Construct a three-dimensional value range parallel to the coordinate axis according to the maximum value point and the minimum value point;

[0021] Points in the point cloud data set that are not within the value range are removed to obtain filtered point cloud data.

[0022] In some embodiments, the preprocessing of the initial point cloud dataset based on the voxel segmentation thinning algorithm and the normal distribution point cloud registration algorithm to obtain the target point cloud dataset includes:

[0023] Constructing an octree structure based on the data of the initial point cloud data set according to a voxel segmentation and thinning algorithm;

[0024] Resampling the data foot points in the octree structure based on a K-neighborhood algorithm to obtain a subset of the initial point cloud data set;

[0025] Point cloud registration is performed on a subset of the initial point cloud data set according to a normal distribution point cloud registration algorithm to obtain a target point cloud data set.

[0026] In some embodiments, performing point cloud registration on a subset of the initial point cloud dataset according to a normal distribution point cloud registration algorithm to obtain a target point cloud dataset includes:

[0027] Dividing the point cloud of the subset into three-dimensional cells according to a normal distribution point cloud registration algorithm;

[0028] Establishing a continuously differentiable probability distribution function for points within the three-dimensional cell;

[0029] Point cloud registration is performed by obtaining the maximum value of the probability distribution function to obtain a target point cloud data set.

[0030] In some embodiments, simulating the target distance between the operator and the target conductor and the tower according to the insulator classification result, the power line point cloud, and the tower point cloud includes:

[0031] Determining the shape of the insulator according to the insulator classification result, and identifying the insulator endpoints according to the shape;

[0032] Determine the starting point and the end point of the motion path according to the end points of the insulator;

[0033] Selecting a number of path points according to the morphology of the insulator;

[0034] Connecting the starting point, the end point, and the path point based on the shortest path algorithm to form a target operation path;

[0035] The target distance between the operator and the target conductor and tower is simulated according to the target operation path, the power line point cloud, the tower point cloud and the cylinder.

[0036] In some embodiments, inputting the insulator point cloud into a bag-of-words model to output an insulator classification result includes:

[0037] Using a deep learning algorithm to extract features from the insulator point cloud to obtain feature points;

[0038] Building a bag-of-words dictionary based on the feature points based on the K-means algorithm, and obtaining feature vectors corresponding to the feature points;

[0039] The similarity between the feature vectors is calculated, and the categories are divided according to the similarity to output the insulator classification result.

[0040] In addition, to achieve the above-mentioned purpose, the present invention also proposes an ultra-high voltage maintenance operation path evaluation device, comprising:

[0041] A point cloud removal module is used to obtain a point cloud data set of an ultra-high voltage transmission channel, and remove outliers from the point cloud data set to obtain an initial point cloud data set;

[0042] A point cloud processing module, used for preprocessing the initial point cloud data set based on a voxel segmentation and thinning algorithm and a normal distribution point cloud registration algorithm to obtain a target point cloud data set;

[0043] A point cloud classification module, used for performing point cloud segmentation on the data of the target point cloud data set based on a power line segmentation method of straight line detection to obtain a power line point cloud;

[0044] A point cloud segmentation module is used to perform point cloud segmentation on the data of the target point cloud data set based on a region growing segmentation algorithm to obtain a tower point cloud and an insulator point cloud;

[0045] An insulator classification module, used for inputting the insulator point cloud into a bag-of-words model to output an insulator classification result;

[0046] A distance calculation module, used for simulating the target distance between the operator and the target conductor and the tower according to the insulator classification result, the power line point cloud and the tower point cloud;

[0047] The safety judgment module is used to judge whether the operator is safe based on the comparison result of the target distance and the preset safety threshold.

[0048] In addition, to achieve the above-mentioned objectives, the present invention also proposes an ultra-high voltage maintenance operation path assessment device, the ultra-high voltage maintenance operation path assessment device comprising: a memory, a processor, and an ultra-high voltage maintenance operation path assessment program stored on the memory and executable on the processor, the ultra-high voltage maintenance operation path assessment program being configured to implement the ultra-high voltage maintenance operation path assessment method as described above.

[0049] In addition, to achieve the above-mentioned purpose, the present invention also proposes a storage medium, which stores an ultra-high voltage maintenance operation path evaluation program, and the ultra-high voltage maintenance operation path evaluation program is used to enable the processor to implement the ultra-high voltage maintenance operation path evaluation method described above when executing.

[0050] The present invention obtains a point cloud data set of an ultra-high voltage transmission channel, removes outliers from the point cloud data set to obtain an initial point cloud data set; pre-processes the initial point cloud data set based on a voxel segmentation thinning algorithm and a normal distribution point cloud registration algorithm to obtain a target point cloud data set; performs point cloud segmentation on the data of the target point cloud data set based on a power line segmentation method of straight line detection to obtain a power line point cloud; performs point cloud segmentation on the data of the target point cloud data set based on a regional growing segmentation algorithm to obtain a tower point cloud and an insulator point cloud; inputs the insulator point cloud into a bag of words model to output an insulator classification result; simulates the target distance between the operator and the target conductor and tower according to the insulator classification result, the power line point cloud and the tower point cloud; and determines whether the operator is safe according to the comparison result of the target distance and a preset safety threshold. In the present invention, through outlier removal, thinning and registration, one or some points can be quickly located in a massive amount of point cloud data, which is convenient for subsequent processing. Point cloud classification based on power line segmentation using straight line detection can solve the discreteness problem of point cloud distribution to a large extent. It can improve the processing capability at the edge based on the region growing segmentation algorithm and obtain more accurate classification results. Then, operation simulation can be performed using a cylinder to make distance calculation more accurate, thereby further improving the accuracy of distance judgment. This improves the accuracy of safety judgment and solves the safety hazards in the maintenance process of existing overhead ultra-high voltage transmission lines, as well as the technical problems of low accuracy and misjudgment of existing safety assessment methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of the structure of an ultra-high voltage maintenance operation path assessment device in a hardware operating environment involved in an embodiment of the present invention;

[0052] Figure 2 It is a flow chart of an embodiment of an ultra-high voltage maintenance operation path evaluation method of the present invention;

[0053] Figure 3 A schematic diagram showing a three-dimensional neighborhood elevation diagram of an ultra-high voltage maintenance operation path assessment method according to an embodiment of the present invention;

[0054] Figure 4 A schematic diagram of different types of insulators in an ultra-high voltage maintenance operation path assessment method according to an embodiment of the present invention;

[0055] Figure 5 A schematic diagram of an optimal separation hyperplane in an ultra-high voltage maintenance operation path evaluation method according to an embodiment of the present invention;

[0056] Figure 6 It is a structural block diagram of an embodiment of an ultra-high voltage maintenance operation path evaluation device of the present invention.

[0057] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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.

[0059] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0060] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0061] Reference Figure 1 , Figure 1 It is a schematic diagram of the structure of an ultra-high voltage maintenance operation path assessment device in the hardware operating environment involved in the embodiment of the present invention.

[0062] like Figure 1As shown, the ultra-high voltage maintenance operation path assessment device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (Wireless-Fidelity, Wi-Fi) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM memory) or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0063] Those skilled in the art will understand that Figure 1 The structure shown in does not constitute a limitation on the ultra-high voltage maintenance work path assessment equipment, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0064] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and an ultra-high voltage maintenance operation path evaluation program.

[0065] exist Figure 1 In the ultra-high voltage maintenance operation path assessment device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the ultra-high voltage maintenance operation path assessment device of the present invention can be set in the ultra-high voltage maintenance operation path assessment device, and the ultra-high voltage maintenance operation path assessment device calls the ultra-high voltage maintenance operation path assessment program stored in the memory 1005 through the processor 1001, and executes the ultra-high voltage maintenance operation path assessment method provided by the embodiment of the present invention.

[0066] The present invention provides an ultra-high voltage maintenance operation path assessment method, device, equipment and storage medium.

[0067] The embodiment of the present invention provides a method for evaluating an ultra-high voltage maintenance operation path, referring to Figure 2 , Figure 2The present invention is a flowchart of an embodiment of an ultra-high voltage maintenance operation path assessment method.

[0068] like Figure 2 As shown, the ultra-high voltage maintenance operation path assessment method includes:

[0069] Step S100: acquiring a point cloud data set of an ultra-high voltage transmission channel, and removing outliers from the point cloud data set to obtain an initial point cloud data set;

[0070] Step S200: preprocessing the initial point cloud data set based on a voxel segmentation and thinning algorithm and a normal distribution point cloud registration algorithm to obtain a target point cloud data set;

[0071] Step S300: performing point cloud segmentation on the data of the target point cloud data set based on a power line segmentation method of straight line detection to obtain a power line point cloud;

[0072] Step S400: performing point cloud segmentation on the data of the target point cloud data set based on a region growing segmentation algorithm to obtain a tower point cloud and an insulator point cloud;

[0073] Step S500: inputting the insulator point cloud into a bag-of-words model to output an insulator classification result;

[0074] Step S600: simulating the target distance between the operator and the target conductor and the tower according to the insulator classification result, the power line point cloud and the tower point cloud;

[0075] Step S700: judging whether the operator is safe based on the comparison result between the target distance and the preset safety threshold.

[0076] It should be noted that the executing entity in this embodiment may be an ultra-high voltage maintenance operation path assessment device, which may be a computer device with data processing functions, or other devices that can achieve the same or similar functions. This embodiment does not limit this. In this embodiment, a computer device is used as an example for explanation.

[0077] It should be noted that this embodiment is described by taking the method of ultra-high voltage maintenance based on the path method as an example. Specifically, a multi-line segment method is used to plan the live working path and calculate the working safety.

[0078] It can be understood that the core steps of the ultra-high voltage maintenance operation path assessment method proposed in this embodiment include: Step 1: Outlier removal; Step 2: Point cloud thinning: Resample the original point cloud data to improve the point cloud processing efficiency without losing key features; Step 3: Point cloud alignment: Align multiple segments of point cloud data according to feature points to ensure that there is no data offset; Step 4: Point cloud segmentation (tower body, power line, insulator); Step 5: Insulator classification Step 6: Draw the path according to the insulator position and shape; Step 7: Operation simulation.

[0079] In one embodiment, a point cloud dataset of an ultra-high voltage transmission channel is obtained, and outliers are removed from the point cloud dataset to obtain an initial point cloud dataset, including: collecting the point cloud dataset of the ultra-high voltage transmission channel based on LiDAR technology; using a straight-through filter to remove outliers from the point cloud dataset in three dimensions to obtain filtered point cloud data; and using a statistical filter to remove sparse outliers from the filtered point cloud data to obtain an initial point cloud dataset.

[0080] Specifically, a straight-through filter is used to remove outliers from the point cloud data set in three dimensional directions to obtain filtered point cloud data, including: extracting the point cloud of the point cloud data set; using a straight-through filter to respectively obtain the maximum value point and the minimum value point of the point cloud in three dimensional directions; constructing a three-dimensional value range parallel to the coordinate axis according to the maximum value point and the minimum value point; and removing points in the point cloud data set that are not within the value range to obtain filtered point cloud data.

[0081] It is understandable that in the process of collecting and acquiring point cloud data through LiDAR technology, due to the problem of measurement errors, some sparse outliers will be generated. At the same time, the formation of these outliers will also be affected by many factors. Therefore, the distribution of outliers is irregular, without a specific position and state. From the perspective of local density, outliers are very likely to exist in isolation, and there is also the possibility of forming high-density point clusters. On the one hand, these points will have a great interference with the algorithm in the subsequent point cloud classification process, affecting the accuracy and speed of classification, and will cause erroneous calculations, resulting in the failure of subsequent data processing; on the other hand, due to the process of measuring transmission lines and towers, there will always be a large number of outliers that are far away from the measured objects, which is of little significance to the process and implementation of point cloud classification, so they need to be eliminated.

[0082] Step 1: Use a straight-through filter to remove points that are not in the given value range in the three dimensions of x, y, and z. The idea of ​​a straight-through filter is to use each dimension and the value range under that dimension to determine whether each point is within the value range of that dimension, and if not, remove it. The specific implementation steps of the straight-through filter are as follows:

[0083] (1) Obtain a point cloud dataset L = {l, i = 1, 2, ..., n};

[0084] (2) Obtain the maximum and minimum points in the three dimensions x, y, and z, respectively, and denote them as x max 、x min ,y max ,y min 、z max 、z min ;

[0085] (3) Based on the extreme points, a three-dimensional range S parallel to the coordinate axis can be constructed i {[x min , x max ]×[y min ,y max ]×[z min , z max ]}, remove the points that are not in the range and obtain the filtered point cloud data.

[0086] The second step is to use a statistical filter to remove sparse outliers. Through analysis, it is found that the spatial distribution of point cloud data is discrete. In the main point cloud, the distribution of points is relatively concentrated, the density is high, and the distance between them and the neighboring points is small; while the outliers are relatively sparse, far away from the main point cloud or more dispersedly distributed around it, and the distance between them and the neighboring points is large. This embodiment adopts a statistical filtering algorithm based on this idea, and uses statistical analysis of the neighborhood of each point to remove outliers.

[0087] For example, the statistical filter algorithm can be interpreted as calculating the average distance between all points in the point cloud and their nearest t neighboring points, assuming that all average values ​​conform to the Gaussian distribution, calculating their standard deviation according to the Gaussian formula, and setting a distance threshold. When the average distance calculated for a point in the point cloud is greater than this threshold, the point is determined to be an outlier and removed. The specific implementation steps are as follows:

[0088] (1) After the straight-through filter processing, the point cloud data set R = {r, i = 1, 2, 3, ..., n} is obtained;

[0089] (2) For any point, calculate the average distance between it and the k points closest to it, denoted as A i ;

[0090] (3) Since all average distances A i Obeying Gaussian distribution, A i The mean value μ and standard deviation σ can be calculated by the following formula:

[0091]

[0092]

[0093] (4) Set the distance threshold to d, that is, the points in the point cloud that can be considered as outliers are all points whose distance from their neighbors is outside the range of (μ-t·σ, μ+1·σ). Among them, the proportional coefficient r is a constant, and its size is determined by the number of points in the field;

[0094] (5) Traverse all point clouds and remove all A i >d, and observe the denoising effect.

[0095] In one embodiment, the initial point cloud dataset is preprocessed based on a voxel segmentation and thinning algorithm and a normal distribution point cloud registration algorithm to obtain a target point cloud dataset, including: constructing an octree structure based on the data of the initial point cloud dataset according to the voxel segmentation and thinning algorithm; resampling the data feet in the octree structure based on a K-neighborhood algorithm to obtain a subset of the initial point cloud dataset; and performing point cloud registration on the subset of the initial point cloud dataset according to the normal distribution point cloud registration algorithm to obtain a target point cloud dataset.

[0096] Specifically, the voxel thinning algorithm can also be referred to as spatial thinning. The principle of the algorithm is to quickly construct an octree structure for massive point cloud data, and combine the K-neighborhood method to determine discrete points and remove them, so as to achieve the purpose of point cloud data thinning. It is a simple and effective point cloud thinning algorithm. The voxel thinning algorithm can sample high-density and high-redundancy point clouds according to a certain interval distance, so that the data foot points after thinning are evenly distributed with equal interval distances, effectively remove redundant point clouds, and at the same time retain the original feature points of the point cloud as much as possible. The octree structure used in voxel thinning is to divide the three-dimensional space in a recursive manner. Each node of the octree represents a volume element of a cube, and each node has 8 child nodes. The initial cube bounding box of the point cloud is continuously divided into 8 equal cube units, and this splitting process is recursively repeated for each cube. The stopping condition is that there are no more points (such as 3 points) in the cube unit or when a maximum preset condition is reached. By constructing an octree structure for point cloud data, data management and processing are facilitated. This structure can quickly determine points in specific cube units and cube triplets. For example, one or some points can be quickly located in massive point cloud data, making it easier to process them later.

[0097] After the octree is constructed, the K-neighborhood is established, and then the neighborhood of each point is statistically analyzed. Outliers are processed by calculating the distance from the point to the neighborhood in the point cloud. For each point in the space, the average distance and standard deviation of the points in its neighborhood are obtained. Assuming that the obtained values ​​are distributed in a Gaussian shape, a threshold for outliers is set, and points with an average distance outside the threshold are defined as outliers and removed from the point cloud space.

[0098] In one embodiment, point cloud registration is performed on a subset of the initial point cloud data set according to a normal distribution point cloud registration algorithm to obtain a target point cloud data set, including: dividing the point cloud of the subset into three-dimensional cells according to the normal distribution point cloud registration algorithm; establishing a continuously differentiable probability distribution function for the points in the three-dimensional cells; and performing point cloud registration by finding the maximum value of the probability distribution function to obtain the target point cloud data set.

[0099] Specifically, the Normal Distributions Transform (NDT) algorithm actually transforms the process of finding the optimal transformation matrix of two point clouds into the process of finding the maximum value of the probability distribution function. The NDT algorithm divides the point cloud evenly into three-dimensional cells of the same size, and achieves the normal distribution description of the entire point cloud by establishing a continuously differentiable probability distribution function for the points in each cell. n} is divided into cells of fixed size. For a point x in a cell, its probability distribution function in the cell is:

[0100]

[0101] Where c is a constant, is the centroid of the cell, and C is the covariance matrix, which is obtained by the following formula:

[0102]

[0103]

[0104] Suppose the cell contains m points x i .

[0105] The evaluation function can be established by calculating the probability distribution function in the formula for all points in the point cloud X:

[0106]

[0107] It represents the point x i The probability of being in the reference point cloud after being transformed by θ (parameters related to rotation and translation), T(θ, x i ) is x i The position of the point after transformation:

[0108]

[0109] Among them, θ=[t|r|φ], t=[t x , t y , t z ], s = sinφ, c = cosφ.

[0110] The NDT registration process is to continuously solve θ to make s(θ) reach the optimal value. The Hessian matrix method is usually used to solve the minimum value of s(θ). Let f = s(θ) and use the equation in each iterative calculation:

[0111] HΔθ=-g

[0112] Among them, g is the gradient transpose of function f, and its elements are H is the Hessian matrix of f, The formula is updated continuously iteratively to reach the required convergence condition and complete the registration.

[0113] In this embodiment, the NDT algorithm is used for alignment, which has the following advantages: the requirements for initial alignment are not high, and even if the error of the initial value is large, a good alignment effect can be achieved; there is no need to perform feature matching between point clouds, which avoids problems that occur in feature matching, such as the influence of point cloud noise, object movement, and point cloud overlap on feature matching.

[0114] Specifically, point cloud segmentation: This embodiment includes a power line segmentation method based on straight line detection and an RPCA segmentation algorithm based on region growing.

[0115] Exemplarily, power line segmentation based on straight line detection: the point cloud data after preprocessing is mainly divided into four parts: poles, insulators, power lines and vegetation. Different segmentation methods need to be implemented for different point cloud distribution characteristics. Since the environment is flat and the distance between the power lines and the ground objects and vegetation is far, the elevation threshold method used in this embodiment can preliminarily divide the power lines from the ground objects and vegetation. Among them, the power lines appear as a linear distribution composed of many point clouds. Therefore, the Hough transform method is used in this embodiment to extract and segment the power lines, which can solve the discreteness problem of point cloud distribution to a large extent. Specifically, the specific steps based on the Hough transform are as follows:

[0116] (1) Detect point cloud data based on the elevation threshold and use the Canny operator (edge ​​detection operator) to obtain the boundary of the point cloud.

[0117] (2) Perform Hough transform on the boundary point cloud to detect straight lines.

[0118] (3) Count the slope and intercept of the detected linear point cloud and set the slope threshold d A and the intercept threshold d B .

[0119] (4) Re-scan all linear point clouds, and d A and d BCompare the sizes. If the comparison relationship between two linear point clouds is consistent, they are grouped together.

[0120] (5) Perform fitting processing on each group of generated point clouds to obtain the power line fitting results.

[0121] Exemplarily, the point cloud segmentation implemented by the RPCA segmentation algorithm based on region growing includes: eigenvalue and eigenvector calculation and normal vector calculation.

[0122] Specifically, eigenvalue and eigenvector calculation: The variables of the three-dimensional point cloud are the point sets of three coordinate systems x, y, and z. It is necessary to analyze the degree of data dispersion and obtain three principal components. Since the distribution of points on the spatial plane does not completely follow the Gaussian distribution, the distribution of points is the most discrete based on the direction perpendicular to the plane, so the variance is the largest; based on the direction parallel to the plane, the distribution of points is the most concentrated, so the variance is the smallest. In the traditional PCA algorithm, the principal component is expressed as a linear combination of a small amount of data. In the process of variance calculation, the corresponding principal component is found by calculating the difference between the eigenvector of the obtained scatter matrix and the mean of the variance. However, since this method is highly sensitive to outliers of edge and local features, this embodiment adopts the PPR idea (projection pursuit regression model) instead of the calculation of the covariance matrix to obtain its eigenvalues ​​and eigenvectors. Among them, the principle of the projection pursuit regression model: a method for processing and analyzing high dimensions, which solves the shortcomings of traditional interval methods (PCA, etc.). By maximizing the selected projection index, finding the projection direction that best reflects the data characteristics, and projecting the high-dimensional data into a low-dimensional space for analysis.

[0123] Specifically, assuming that there are n points in the point cloud data, these points are first processed by feature centering to obtain the processed feature point set P:

[0124]

[0125] From this we can get the mean vector p of the point set:

[0126]

[0127] The point set of the matrix is ​​decomposed into point sets of three dimensions: x, y, and z, that is, p1 = (p xi , p yi , p zi )(i=1,2,3,...,n), the above formula can be replaced by:

[0128]

[0129] The covariance matrix X defined by it is as follows, which defines the local geometric characteristics of the data point set P:

[0130]

[0131] Substituting formula p into the above formula X, the covariance can be expressed as cov(p ri , p si ), where r∈(x,y,z), s∈(x,y,z), and the following changed covariance matrix X can be obtained:

[0132]

[0133] Next, PPR is used to take the variance σ of the data scatter as the projection index. Let v be the projection direction vector and the eigenvalue be λ. The calculation formula of the variance σ is:

[0134]

[0135] When the variance σ value is the largest, the obtained vector v1 is the eigenvector, and the projection spread of the data in the direction of this vector reaches the maximum. The vector v1 is the eigenvector of the largest eigenvalue λ1 in the covariance matrix X.

[0136] Continue to perform projection pursuit on the data. This time, select the data space perpendicular to the vector v1, and still find the vector v2 that makes the variance σ value reach the maximum, and solve the eigenvalue λ2. Repeat the above steps to obtain the eigenvector v3 and eigenvalue λ3. Finally, the eigenvector matrix V can be obtained:

[0137]

[0138] Among them, the eigenvectors v1, v2, and v3 are arranged in descending order according to the corresponding eigenvalues ​​λ1, λ2, and λ3, which are defined as λ1≥λ2≥λ3.

[0139] Specifically, normal vector calculation: In the process of fitting the plane using the RPCA (Robust Principal Component Analysis) method, there may be a few noise points whose distance from the desired plane point cloud is relatively small relative to the range of the plane itself, and the impact on the fitting effect can be ignored, but it is more difficult to process at the edge. In addition, since the proportion of insulator point cloud data is extremely small, these points are distributed in the same direction within a certain range and clustered into a cluster. Therefore, this embodiment uses a three-dimensional sphere neighborhood partition method to calculate the plane normal vector. Construct a three-dimensional spherical neighborhood T and obtain all neighborhood points in the neighborhood. Use the eigenvector v3 as the initial normal vector to calculate a fitting plane M. The point in the neighborhood T is the neighborhood point, the center of the neighborhood sphere is O, and the radius is r. The steps for normal vector calculation are as follows:

[0140] (1) Calculate the distance d1 from all neighborhood points to M, where the error is θ, and set the weight of the point whose distance d1 is less than or equal to 2θ to 1, and the weight of the point whose distance is greater than 2θ to 0. Then, recalculate the previously obtained fitting plane M;

[0141] (2) Repeat the normal vector calculation step (1), record the plane parameters during the fitting process, and stop fitting until the change is less than the threshold or the size remains unchanged. The fitted plane obtained at this time is M', and the normal vector of M' is the required normal vector.

[0142] according to Figure 3 The three-dimensional neighborhood elevation diagram shown shows that the center o of the sphere is the point to be found, the radius is r, and the neighborhood points of the point to be found are the points in the same plane as the point to be found. The normal vector of the point to be found is determined by calculating the neighborhood points.

[0143] In the traditional region growing segmentation method, since the inaccuracy of the normal line of the range point near the region boundary will affect the accuracy of the segmentation, this embodiment adopts the RPCA method to calculate and obtain a more accurate eigenvalue λ3, and the calculation of the low deflection angle σ adopts the normal vector obtained after three-dimensional sphere fitting. The low deflection angle σ between a point and its adjacent points:

[0144]

[0145] in, and Represent the two unit normals of the i-th point and one of its neighboring points respectively.

[0146] On the other hand, an important part of the region growing method is the selection of seed points. The selection in traditional methods is relatively random. Since the boundaries of point cloud data are relatively fuzzy, errors are prone to occur. Therefore, in this embodiment, the points in the boundary neighborhood are used as seed points, and region growth search is performed on their neighborhood.

[0147] The calculation steps of the improved region growing method in this embodiment are as follows:

[0148] (1) Let the plane point set be S and the boundary point set be U. Then select the region V formed by the intersection of the plane point set S and the boundary point set V (i.e., V = S ∪ U), use the points in this region as seed points, and create an empty current region list D, a point set list Q of seed points (i∈Q), and a list L. The seed point set Q is used to record the seed points after screening, and the list L records the final segmentation results.

[0149] (2) Select the points with the lowest λ3 value in region V. If they are not in the current domain list D and the seed point set list Q, add them and remove the point from region V.

[0150] (3) In the seed point set Q, search for each point q1 to find its nearest neighbor point q k , and determine a threshold μ based on the above fitting process:

[0151] a. Calculate q i and its neighboring point q k Low deflection angle σ;

[0152] b. Find all points in region V where σ≤μ;

[0153] c. Add these points to the current domain list D and the seed point set list Q, and remove the point from the region V.

[0154] (4) Repeat the calculation steps (2)-(3) of the improved region growing method until the point set in region V becomes an empty set;

[0155] (5) Sort the domains in list L to obtain the segmentation results.

[0156] After the above processing, the plane parameter d can be obtained, and the point clouds on the same plane will be divided into the same cluster.

[0157] Specifically, insulator classification: bag of words model. The bag of words (BoW) method is a commonly used method for processing and retrieving data. It is mainly used for natural language processing. Its origin comes from text classification. The BoW method represents the image as a combination of words of different frequencies by creating a large number of words. The bag of words model can be used for image classification. The main idea of ​​the BoW model can be reflected in two points: the disorder and independence of words. Construct a BoW set, which contains many words. The disorder means that these words are stored in any position in the BoW. No matter how their relative positions change, it will not affect the results of feature extraction and classification; and independence means that each word exists independently, and there will be no dependence and interference between each other, and there will be no confusion of features during classification. In practical applications, when images need to be classified, the words in a bag of words model correspond to the feature descriptors in the image.

[0158] In one embodiment, the insulator point cloud is input into a bag-of-words model to output an insulator classification result, including: using a deep learning algorithm to perform feature extraction on the insulator point cloud to obtain feature points; constructing a bag-of-words dictionary based on the feature points based on a K-means algorithm, and obtaining feature vectors corresponding to the feature points; calculating the similarity between the feature vectors, dividing the categories according to the similarity, and outputting the insulator classification result.

[0159] Specifically, in the point cloud data obtained based on LiDAR technology, the actual description of a point cloud data image is expressed as a collection of multiple feature points. Therefore, the essence of using the BoW model to classify point cloud data is similarity calculation. Taking an image taken in the laboratory as an example, it can be summarized as "five students are studying with three computers, with two drones and a binocular camera next to them." Assuming that this image is modeled, the feature points extracted from it include "classmates", "computers", "drones", and "binocular cameras". These feature points are combined as words, and finally a bag of words related to the laboratory will be constructed. After constructing the bag of words C, the image contains the above-mentioned feature words that already exist in the bag of words C, and they are recorded as α1, α2, α3, and α4 respectively. Then in this image, according to the words it contains in the bag of words C, the image can be expressed as a formula:

[0160] C=5α1+3α2+2α3+1α4

[0161] Since the order of all words in the bag-of-words dictionary is fixed, we can use the vector [5, 3, 2, 1] T To represent the information of the image.

[0162] The above example illustrates the basic principle of the bag of words used for image representation. In the classification of power inspection information based on the bag of words, the main steps of the bag of words model application can be divided into: first, feature extraction of the collected relevant power tower point cloud data (insulator point cloud data); second, constructing a bag of words with the extracted feature points to form a set, and obtaining the feature vector corresponding to the feature point; finally, given the data to be classified, it is divided into its corresponding category by calculating the similarity between the feature vectors.

[0163] Specifically, insulator feature extraction: In the transmission line, there are many disc-shaped insulators hanging at the connection points of the power lines on the power towers, which can increase the creepage distance for power transmission. Such insulators are usually made of glass or ceramics and are called insulators. The power tower insulator point cloud information collected in this chapter mainly includes three types of insulators: suspension insulators, transverse insulators and tension insulators (for example Figure 4As shown). Suspension insulators are installed in a suspended manner on the power towers of transmission lines, that is, vertically to the ground. This method allows the suspension insulators to bear the full weight of the transmission line. Horizontal insulators are mostly used on terminal power towers. Compared with suspension insulators, their diameter is larger and can bear the tension of long-distance transmission lines. Tension insulators are installed horizontally on power towers and can bear the full tension of transmission lines. Compared with suspension insulators, the difference between horizontal insulators and suspension insulators is that there is a certain angle between the transmission lines on both sides of the tower where the tension insulator is located, and there is an arc between the suspension points on both sides of the tension insulator.

[0164] It should be noted that due to the uneven distribution of point clouds and the unstructured distribution, there are fewer feature points and the information is incomplete. Therefore, this embodiment uses a deep learning algorithm to extract deep features from the three-dimensional insulator point cloud. Based on the traditional CNN idea, an X-convolution (X-Conv) operation based on point cloud is proposed. The main idea of ​​the X-convolution operation is to converge the features of other points in the neighborhood of the selected point to the point through a recursive convolution operation, so that the selected point contains more information.

[0165] Specifically, the bag-of-words dictionary is constructed based on the k-Means algorithm: The K-Means algorithm is widely used and is simple and effective to implement. In the initial step of clustering, the K-Means algorithm needs to determine the cluster center, take all the insulator point cloud data as samples, use the three-dimensional deep learning network to extract the deep features of each point cloud data, and cluster all the sample deep features through the K-means algorithm to obtain the corresponding cluster centers. At the same time, they are regarded as words, and these words are combined together to form a bag-of-words dictionary. The size of the bag-of-words dictionary corresponds to the number of words contained in the bag-of-words dictionary. Therefore, this embodiment uses the K-means algorithm clustering to construct a bag-of-words dictionary.

[0166] The K-Means algorithm is used to cluster the deep feature points of the insulator point cloud data. The specific implementation steps are as follows:

[0167] (1) Establish a set of insulator point cloud information N{N1, N2, N i ,…,N n}(i=1, 2, 3, ..., n);

[0168] (2) Assuming that the clustering is into k categories, k cluster centers can be generated, among which the jth cluster center is represented by s j (j=1, 2, 3, ..., k);

[0169] (3) For each point N in the set N j , calculate it and all s j The distance d j, choose the one with the smallest distance j , point N j and j Classified into one category.

[0170] (4) Set a d j The convergence threshold t is set. When the distance calculated for each point is less than t, the algorithm converges and clustering ends. Otherwise, repeat the K-Means algorithm clustering steps (2)-(3).

[0171] After calculation by the K-means algorithm, the feature descriptors of the insulator point cloud data are clustered to form a bag-of-words dictionary containing k words.

[0172] Specifically, the SVM classification method is implemented as follows: The support vector machine (SVM) algorithm can be divided into two cases: linearly separable and linearly inseparable. Simply put, the sample points that can be classified with a straight line are linearly distinguishable, and the optimal classification hyperplane can be directly found in the sample space to be classified. The other case is linear inseparability. For this case, two steps are required: the first step is to use a nonlinear mapping to transform the vector points in the original data set into a higher-dimensional space; the second step is to try to find the optimal classification hyperplane in this high-dimensional space. The basic idea of ​​the SVM algorithm is to solve the separating hyperplane that can correctly divide the training data set and maximize the geometric interval or margin, such as Figure 5 shown.

[0173] Specifically, the SVM classifier uses a separating hyperplane to classify different samples. The linear equation of the separating hyperplane is:

[0174] ω T x+b=0

[0175] Among them, x is the input vector or training instance, ω is the weight vector or the slope of the separation surface, and b is the bias term.

[0176] Assume that a training data set T = {(x1, y1)(x2, y2), ..., (x N ,y N )}, where x1∈R n , i = 1, 2, ..., N, x1 is the i-th feature value, y i is the class label. When it is equal to +1, it is a positive example; when it is -1, it is a negative example. Assuming that the training data set is linearly separable, the classification can be completed when ω and b are determined. The classification result is as follows:

[0177]

[0178] The support vector is the sample that satisfies the equality. The sum of the distances from two heterogeneous support vectors to the separating hyperplane is shown in the formula:

[0179]

[0180] In order to maximize the distance, the existing training set is used to train the parameters ω and b to maximize the value of γ. Maximizing 1 / ||ω|| is equivalent to ||ω|| 2 , so finding the maximum value of γ is equivalent to finding the formula:

[0181]

[0182] Using the Lagrange multiplier method, the above formula can be written as follows:

[0183]

[0184] Among them, ||ω|| 2 / 2 is the objective function, is the constraint condition. The SVM objective function requires a minimum value, and from the structure of the constraint condition, we can see that when the constraint condition is to be small or even extremely small, y i (ω T x+b) will become very large, that is, the distance between the current sample vector and the partitioning hyperplane is large, and the classification effect is better. Taking the partial derivative of ω and b for the above formula and setting the partial derivative to 0, we can get the formula:

[0185]

[0186]

[0187] Substitute the above two formulas into the formula, eliminate ω and b in L(ω, b, α), and solve for α i After that, we can find ω and b to get the model formula:

[0188]

[0189] For the case of linear inseparability, the kernel function is introduced into the support vector machine, and the radial basis kernel function (RBF) is introduced as follows:

[0190] K(x i ,y i ) = exp(-γ||x i -x j ||,γ>0

[0191] Specifically, the specific steps for classification based on insulator point cloud data are:

[0192] Step 1: Input the insulator point cloud data, and use the PointCNN deep learning network to aggregate and extract the features of each point cloud data set of different types of insulators to obtain multiple aggregate features. It includes three parameter settings: the initial input N1=2048, which is obtained by computing all point clouds by the random downsampling method; the initial convolution kernel size is K1=4, where the value of K is the k value obtained by the k-nearest neighbor algorithm for the point; the initial channel size is C1=3, which indicates the feature dimension of the initial point.

[0193] Step 2: Use the k-means algorithm to build a visual word dictionary. Use the k-means algorithm to cluster all the point cloud features obtained in step 1 based on the classification of the insulator point cloud data to generate a visual dictionary. Given that the number of words in the bag-of-words model dictionary has a great impact on the accuracy of the model (the fewer words, the worse the bag-of-words model effect). K-means clusters 1000 categories, that is, the word bag size is set to 1000;

[0194] Step 3: Use the vocabulary in the dictionary to represent the features of the insulator point cloud data to be classified. Calculate the distance between each aggregate feature of the insulator point cloud data and each word in the visual dictionary. The word with the closest distance in the dictionary represents an aggregate feature. Count the frequency of each word in the dictionary in the current point cloud data and represent it as a 1000-component statistical histogram (a 1000-dimensional numerical vector when calculated);

[0195] Step 4: After completing the construction of the bag-of-words model, construct the training set and the set to be classified. Use the SVM classifier for training, divide the insulator point cloud data into three categories, train the classification model of each class separately, and randomly extract equal amounts of positive and negative samples for training. When using the BRF kernel function, the parameter Y=0.4 is set, which is obtained by repeated experimental modification. After the training is completed, use the trained SVM classifier to classify the point cloud data of the set to be classified, and obtain the classification results of the insulator.

[0196] In one embodiment, the target distance between the operator and the target conductor and the pole tower is simulated according to the insulator classification result, the power line point cloud and the pole tower point cloud, including: determining the shape of the insulator according to the insulator classification result, and identifying the insulator endpoints according to the shape; determining the starting point and the end point of the motion path according to the insulator endpoints; selecting a number of path points according to the shape of the insulator; connecting the starting point, the end point and the path points based on the shortest path algorithm to form a target working path; simulating the target distance between the operator and the target conductor and the pole tower according to the target working path, the power line point cloud, the pole tower point cloud and the cylinder.

[0197] Specifically, the target operation path is drawn as follows: ① Confirm the path endpoints: According to the shape of the insulator, identify the insulator endpoints (for example, two), and determine the starting point and end point of the motion path according to the two endpoints. ② Determine the path points: Appropriate path points can be manually selected on the insulator. These points should be located in an area where the operator can walk safely. ③ Connect the path points: Use the shortest path algorithm to connect the path points to form a complete path.

[0198] Specifically, operation simulation: after determining the moving path (i.e., the target operation path), use a cylinder to simulate the human body and calculate the distance to the conductor and the tower. The specific method is as follows:

[0199] To find the shortest distance from a point on the cylinder to a set of discrete points, you can use the following steps:

[0200] 1) Cylinder parameters: Get the parameters of the cylinder, including the coordinates of the center point of the bottom surface (C x , C y , C z ), radius R and height H.

[0201] 2) Point cloud data: Get point cloud data, which is a set of coordinates of a group of points. These points represent the positions of discrete points, and it is necessary to find the point with the shortest distance from these points to the cylinder.

[0202] 3) Calculate the shortest distance: For each point, perform the following steps:

[0203] a. Coordinate transformation: transform the coordinates of the point from the global coordinate system to the cylindrical coordinate system. This can be achieved by subtracting the coordinates of the center point of the cylinder from the coordinates of the point. The coordinates of the point in the cylindrical coordinate system (P x ',P y ',P z ').

[0204] b. Horizontal distance calculation: Calculate the coordinates of the horizontal projection point on the bottom surface of the cylinder (P x ',P y ', 0), where the Z coordinate is 0. The horizontal distance is equal to the two-dimensional Euclidean distance from the point to the projection point, that is,

[0205]

[0206] c. Vertical distance calculation: Calculate the vertical distance from the point to the horizontal projection point on the bottom of the cylinder. This can be achieved by calculating the Euclidean distance between the two points, that is,

[0207] H'=abs(P z -C z )

[0208] d. Calculation of the shortest distance point: The shortest distance point is the projection point of the point on the surface of the cylinder. It can be calculated by adding the horizontal distance R and the vertical distance to the coordinates of the center point of the bottom surface of the cylinder. The coordinates of the shortest distance point are (C x , C y ,0)+(R,0,H′).

[0209] 4) Find the point with the minimum distance: In step 3), the shortest distance from each point to the cylinder is calculated. Finally, the point with the minimum distance is found among these calculation results. This point is the point on the cylinder closest to the point in the point cloud, and is also the point on the cylinder closest to the point cloud.

[0210] It should be noted that the above calculation of the distance between the simulated cylinder and the two types of point clouds can effectively find the point on the cylinder with the minimum distance from a group of points, no matter how these points are distributed. After obtaining the distance between the cylinder and the wire and the tower, the distance between the cylinder and the wire (the first shortest distance) and the distance between the cylinder and the tower (the second shortest distance) are combined, that is, the first shortest distance and the second shortest distance are summed to obtain the target distance. The threshold can be used to determine whether the operator is safe, that is, the comparison result between the target distance and the preset safety threshold can be used to determine whether the operator is safe.

[0211] In this embodiment, the point cloud data set of the ultra-high voltage transmission channel is obtained, and outliers are removed from the point cloud data set to obtain an initial point cloud data set; the initial point cloud data set is preprocessed based on the voxel segmentation thinning algorithm and the normal distribution point cloud registration algorithm to obtain a target point cloud data set; the data of the target point cloud data set is segmented based on the power line segmentation method of straight line detection to obtain a power line point cloud; the data of the target point cloud data set is segmented based on the regional growing segmentation algorithm to obtain a tower point cloud and an insulator point cloud; the insulator point cloud is input into a bag of words model to output an insulator classification result; the target distance between the operator and the target conductor and the tower is simulated according to the insulator classification result, the power line point cloud and the tower point cloud; and the operator is judged whether it is safe according to the comparison result of the target distance and the preset safety threshold. In this embodiment, through outlier removal, thinning and registration, one or some points can be quickly located in the massive point cloud data, which is convenient for subsequent processing. Point cloud classification based on power line segmentation using straight line detection can solve the discreteness problem of point cloud distribution to a large extent. It can improve the processing capability at the edge based on the region growing segmentation algorithm and obtain more accurate classification results. Then, operation simulation can be performed using a cylinder to make distance calculation more accurate, thereby further improving the accuracy of distance judgment. This improves the accuracy of safety judgment and solves the safety hazards in the maintenance process of existing overhead ultra-high voltage transmission lines, as well as the technical problems of low accuracy and misjudgment of existing safety assessment methods.

[0212] In addition, an embodiment of the present invention also proposes a storage medium, on which is stored an ultra-high voltage maintenance operation path evaluation program, which, when executed by a processor, implements the steps of the ultra-high voltage maintenance operation path evaluation method described above.

[0213] Reference Figure 6 , Figure 6 It is a structural block diagram of an embodiment of an ultra-high voltage maintenance operation path evaluation device of the present invention.

[0214] like Figure 6 As shown, the ultra-high voltage maintenance operation path evaluation device comprises:

[0215] The point cloud removal module 10 is used to obtain a point cloud data set of the ultra-high voltage transmission channel, and remove outliers from the point cloud data set to obtain an initial point cloud data set;

[0216] A point cloud processing module 20, configured to pre-process the initial point cloud data set based on a voxel segmentation and thinning algorithm and a normal distribution point cloud registration algorithm to obtain a target point cloud data set;

[0217] A point cloud classification module 30, configured to perform point cloud segmentation on the data of the target point cloud data set based on a power line segmentation method of straight line detection to obtain a power line point cloud;

[0218] A point cloud segmentation module 40 is used to perform point cloud segmentation on the data of the target point cloud data set based on a region growing segmentation algorithm to obtain a tower point cloud and an insulator point cloud;

[0219] An insulator classification module 50, used for inputting the insulator point cloud into a bag-of-words model to output an insulator classification result;

[0220] A distance calculation module 60, configured to simulate the target distance between the operator and the target conductor and the tower according to the insulator classification result, the power line point cloud and the tower point cloud;

[0221] The safety judgment module 70 is used to judge whether the operator is safe according to the comparison result between the target distance and the preset safety threshold.

[0222] In this embodiment, through outlier removal, thinning and registration, one or some points can be quickly located in the massive point cloud data, which is convenient for subsequent processing. Point cloud classification based on the power line segmentation method of straight line detection can largely solve the discrete problem of point cloud distribution, improve the processing ability at the edge based on the regional growth segmentation algorithm, and obtain more accurate classification results. Then, through the operation simulation of the cylinder, the distance calculation is more accurate, thereby further improving the distance judgment accuracy, thereby improving the accuracy of safety judgment, and solving the safety hazards in the existing overhead ultra-high voltage transmission line maintenance process, and the existing safety assessment method is not accurate and has misjudgment and other technical problems.

[0223] In addition, technical details that are not described in detail in the embodiment of the ultra-high voltage maintenance operation path assessment device can be referred to the ultra-high voltage maintenance operation path assessment method described above provided in any embodiment of the present invention, and will not be repeated here.

[0224] It should be understood that the above is only an example and does not constitute any limitation on the technical solution of the present invention. In specific applications, technicians in this field can make settings as needed, and the present invention does not limit this.

[0225] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present invention. In practical applications, technicians in this field can select part or all of them according to actual needs to achieve the purpose of the present embodiment, and no limitation is made here.

[0226] In addition, it should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0227] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0228] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory (ROM) / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0229] The above are only preferred 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 evaluating an ultra-high voltage maintenance operation path, characterized in that: The ultra-high voltage maintenance operation path assessment method comprises: Acquire a point cloud data set of an ultra-high voltage transmission channel, and remove outliers from the point cloud data set to obtain an initial point cloud data set; Preprocessing the initial point cloud data set based on a voxel segmentation and thinning algorithm and a normal distribution point cloud registration algorithm to obtain a target point cloud data set; Performing point cloud segmentation on the data of the target point cloud data set based on a power line segmentation method of straight line detection to obtain a power line point cloud; Performing point cloud segmentation on the target point cloud data set based on a region growing segmentation algorithm to obtain a tower point cloud and an insulator point cloud; Inputting the insulator point cloud into a bag-of-words model to output an insulator classification result; Simulate the target distance between the operator and the target conductor and the tower according to the insulator classification result, the power line point cloud and the tower point cloud; Determining whether the operator is safe based on a comparison result between the target distance and a preset safety threshold; Among them, simulating the target distance between the operator and the target conductor and the pole tower according to the insulator classification result, the power line point cloud and the pole tower point cloud includes: determining the shape of the insulator according to the insulator classification result, and identifying the insulator endpoints according to the shape; determining the starting point and the end point of the motion path according to the insulator endpoints; selecting a number of path points according to the shape of the insulator; connecting the starting point, the end point and the path points based on the shortest path algorithm to form a target working path; simulating the target distance between the operator and the target conductor and the pole tower according to the target working path, the power line point cloud, the pole tower point cloud and the cylinder.

2. The ultra-high voltage maintenance operation path assessment method according to claim 1, characterized in that: The step of obtaining a point cloud data set of the ultra-high voltage transmission channel and removing outliers from the point cloud data set to obtain an initial point cloud data set includes: Collect point cloud data sets of ultra-high voltage transmission channels based on LiDAR technology; Using a straight-through filter to remove outliers from the point cloud data set in three dimensional directions to obtain filtered point cloud data; A statistical filter is used to remove sparse outlier points from the filtered point cloud data to obtain an initial point cloud data set.

3. The ultra-high voltage maintenance operation path assessment method according to claim 2, characterized in that: The step of removing outliers from the point cloud data set in three dimensional directions by using a straight-through filter to obtain filtered point cloud data includes: Extracting a point cloud of the point cloud data set; A straight-through filter is used to obtain the maximum value and the minimum value of the point cloud in three dimensional directions respectively; Construct a three-dimensional value range parallel to the coordinate axis according to the maximum value point and the minimum value point; Points in the point cloud data set that are not within the value range are removed to obtain filtered point cloud data.

4. The ultra-high voltage maintenance operation path assessment method according to claim 1, characterized in that: The preprocessing of the initial point cloud data set based on the voxel segmentation thinning algorithm and the normal distribution point cloud registration algorithm to obtain the target point cloud data set includes: Constructing an octree structure based on the data of the initial point cloud data set according to a voxel segmentation and thinning algorithm; Resampling the data foot points in the octree structure based on a K-neighborhood algorithm to obtain a subset of the initial point cloud data set; Point cloud registration is performed on a subset of the initial point cloud data set according to a normal distribution point cloud registration algorithm to obtain a target point cloud data set.

5. The ultra-high voltage maintenance operation path assessment method according to claim 4, characterized in that: The step of performing point cloud registration on a subset of the initial point cloud data set according to a normal distribution point cloud registration algorithm to obtain a target point cloud data set includes: Dividing the point cloud of the subset into three-dimensional cells according to a normal distribution point cloud registration algorithm; Establishing a continuously differentiable probability distribution function for points within the three-dimensional cell; Point cloud registration is performed by obtaining the maximum value of the probability distribution function to obtain a target point cloud data set.

6. The ultra-high voltage maintenance operation path assessment method according to claim 1, characterized in that: The step of inputting the insulator point cloud into a bag-of-words model to output an insulator classification result includes: Using a deep learning algorithm to extract features from the insulator point cloud to obtain feature points; Building a bag-of-words dictionary based on the feature points based on the K-means algorithm, and obtaining feature vectors corresponding to the feature points; The similarity between the feature vectors is calculated, and the categories are divided according to the similarity to output the insulator classification result.

7. An ultra-high voltage maintenance operation path assessment device, characterized in that: The ultra-high voltage maintenance operation path assessment device comprises: A point cloud removal module is used to obtain a point cloud data set of an ultra-high voltage transmission channel, and remove outliers from the point cloud data set to obtain an initial point cloud data set; A point cloud processing module, used for preprocessing the initial point cloud data set based on a voxel segmentation and thinning algorithm and a normal distribution point cloud registration algorithm to obtain a target point cloud data set; A point cloud classification module, used for performing point cloud segmentation on the data of the target point cloud data set based on a power line segmentation method of straight line detection to obtain a power line point cloud; A point cloud segmentation module is used to perform point cloud segmentation on the data of the target point cloud data set based on a region growing segmentation algorithm to obtain a tower point cloud and an insulator point cloud; An insulator classification module, used for inputting the insulator point cloud into a bag-of-words model to output an insulator classification result; A distance calculation module, used for simulating the target distance between the operator and the target conductor and the tower according to the insulator classification result, the power line point cloud and the tower point cloud; A safety judgment module, used to judge whether the operator is safe according to the comparison result between the target distance and the preset safety threshold; Among them, simulating the target distance between the operator and the target conductor and the pole tower according to the insulator classification result, the power line point cloud and the pole tower point cloud includes: determining the shape of the insulator according to the insulator classification result, and identifying the insulator endpoints according to the shape; determining the starting point and the end point of the motion path according to the insulator endpoints; selecting a number of path points according to the shape of the insulator; connecting the starting point, the end point and the path points based on the shortest path algorithm to form a target working path; simulating the target distance between the operator and the target conductor and the pole tower according to the target working path, the power line point cloud, the pole tower point cloud and the cylinder.

8. An ultra-high voltage maintenance operation path assessment device, characterized in that: The ultra-high voltage maintenance operation path assessment device includes: a memory, a processor, and an ultra-high voltage maintenance operation path assessment program stored in the memory and executable on the processor, wherein the ultra-high voltage maintenance operation path assessment program is configured to implement the ultra-high voltage maintenance operation path assessment method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium stores an ultra-high voltage maintenance operation path evaluation program, and the ultra-high voltage maintenance operation path evaluation program is used to enable the processor to implement the ultra-high voltage maintenance operation path evaluation method according to any one of claims 1 to 6 when executed.

Images