A method for automatically classifying laser point clouds of key elements of high-voltage overhead transmission lines

Abstract

The application discloses a kind of high-voltage overhead transmission line key element laser point cloud automatic classification method, it is related to overhead transmission line technical field, including S1, collection high-voltage overhead transmission line point cloud data, and reference overhead line account information is carried out to the point cloud data of collection Preprocessing;S2, for the point cloud data after preprocessing, using irregular triangle net TIN progressive encryption filtering algorithm separates ground point cloud data and non-ground point cloud data;S3, for ground point cloud data, using weighted least square method carries out point cloud smoothing processing;S4, for non-ground point cloud data, using different algorithm carries out refinement classification and extraction.The application uses irregular triangle net TIN progressive encryption filtering algorithm, improves the efficiency and accuracy of overhead transmission line point cloud automatic classification and extraction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of overhead transmission lines, and more particularly to a laser point cloud automatic classification method for key elements of high-voltage overhead transmission lines. BACKGROUND

[0002] The point cloud data of the transmission line obtained by the airborne laser radar is very large. After the front-end data acquisition, the point cloud data needs to be pre-processed and post-processed, and the post-processing includes steps such as thinning, filtering classification, model reconstruction, etc. of the point cloud. The automatic classification of the point cloud is a key step in data post-processing.

[0003] The point cloud data has the characteristics of discrete type, multi-echo, reflection intensity information, high density, data massiveness, data missing, etc. The point cloud data processing and analysis efficiency is low, and it takes at least 48 hours to generate a channel danger channel hidden danger report, resulting in low efficiency of transmission line inspection operation. The automatic classification of the point cloud has the following problems in actual processing and analysis: (1) The accuracy of the automatic classification of the point cloud is low, and the post-processing relies on manual processing. At the same time, due to the fact that the automatic degree of some algorithms cannot meet the requirements of the post-data processing, a large amount of manual intervention is required for traditional transmission line modeling, and the efficiency is low and the accuracy is poor. (2) The automatic classification algorithm has poor universality. Most classification algorithms are based on the similarity of local point clouds, such as comparing local slope difference and elevation difference, object shape, etc. At present, there is no algorithm for automatic subdivision of ground object categories. (3) The transmission line account information is not fully utilized. China's power grid construction has developed a complete transmission line equipment account specification. The transmission line account information can be obtained by calling the PMS2.0 data interface or accessing the database table, which contains coordinate information of the equipment account data. The data accuracy and efficiency of the transmission line automatic classification tower and ground wire are greatly improved. (4) The time performance and space performance of the automatic classification algorithm are poor. Due to the discreteness of the point cloud data and the instability of the algorithm, the execution process is time-consuming, the space occupation is large, and the time and memory space consumed are large.

[0004] Therefore, how to improve the efficiency and accuracy of the automatic classification of the point cloud of the overhead transmission line is a problem to be solved by those skilled in the art. SUMMARY

[0005] Therefore, the present application provides a laser point cloud automatic classification method for key elements of high-voltage overhead transmission lines.

[0006] In order to achieve the above purpose, the present application adopts the following technical solutions:

[0007] A laser point cloud automatic classification method for key elements of high-voltage overhead transmission lines, comprising the following steps:

[0008] To achieve the above object, the present application adopts the following technical solutions:

[0009] A high-voltage overhead transmission line key element laser point cloud automatic classification method, comprising the following steps:

[0010] S1, collecting high-voltage overhead transmission line point cloud data, and referencing overhead line account information to preprocess the collected point cloud data;

[0011] S2, for the preprocessed point cloud data, using an irregular triangle network TIN progressive encryption filtering algorithm to separate ground point cloud data and non-ground point cloud data;

[0012] S3, for the ground point cloud data, using a weighted least squares method for point cloud smoothing;

[0013] S4, for non-ground point cloud data, using different algorithms for fine classification and extraction.

[0014] Preferably, step S1 references overhead line account information to preprocess the point cloud data, specifically including:

[0015] Using a denoising algorithm based on local spatial distribution statistics of point cloud to denoise the collected high-voltage overhead transmission line point cloud data;

[0016] According to the tower coordinate position in the overhead line account information, the denoised high-voltage overhead transmission line point cloud data is segmented to generate multi-standard las format point cloud data.

[0017] Preferably, step S2 includes the following steps:

[0018] S21, obtaining the preprocessed point cloud data;

[0019] S22, grid dividing the preprocessed point cloud data, taking the lowest point in each grid as the initial point to generate an initial irregular triangle network TIN;

[0020] S23, calculating the perpendicular distance of the point cloud data point to the nearest triangular face and the angle between the data point and the three vertices of the triangular face, if the maximum value of the perpendicular distance and the angle is less than a set threshold, the data point is added to the ground point cloud data set, and the generated initial irregular triangle network TIN is encrypted;

[0021] S24, repeating steps S22-S23, iteratively processing all point cloud data until all point cloud data is traversed.

[0022] Preferably, in step S4, different algorithms are used for fine classification of non-ground point cloud data, specifically including:

[0023] a. Extracting the tower point cloud data by using the Kd-tree clustering method;

[0024] b. Extracting the power line point cloud data by using the center line fitting method combined with Hough transform and least squares, and separating the ground wire point cloud data and the conductor point cloud data;

[0025] c. Extracting the building point cloud data and low vegetation point cloud data by using the random sample consensus algorithm and the least squares fitting plane region growing algorithm.

[0026] Preferably, the method a of extracting the tower point cloud data by using the Kd-tree algorithm and the bounding box specifically comprises:

[0027] According to the tower coordinate position information, the tower point cloud data is obtained, and the tower point cloud data is divided into tower top point cloud data and tower body point cloud data according to the tower and pole geometric characteristics;

[0028] For the tower top point cloud data: an initial OBB bounding box of the tower top point cloud data is constructed, and a Kd-tree is recursively constructed with the initial OBB bounding box as the root node, the tower top point cloud data in the initial OBB bounding box is horizontally sliced, and the point cloud data in each horizontal slice is circularly fitted to extract the point cloud data in the horizontal slice, and the Kd-tree is used to search and merge all the point cloud data in the horizontal slice to extract the tower and pole point cloud data;

[0029] For the tower body point cloud data: an initial OBB bounding box of the tower body point cloud data is constructed, and an initial OBB fitting is performed on the tower body point cloud data in the initial OBB bounding box, according to the fitting result of the initial OBB, the tower edges of the tower body point cloud data are identified, and an accurate OBB fitting is performed on the identified tower edges.

[0030] Preferably, the method b of extracting the power line point cloud data by using the center line fitting method combined with Hough transform and least squares, and separating the ground wire point cloud data and the conductor point cloud data specifically comprises:

[0031] b1. Obtain an elevation value image of non-ground point cloud data, perform Canny operator detection on the elevation value image, and extract an edge image;

[0032] b2. Detect line segments in the edge image by using Hough transform, and calculate the slope and intercept of each line segment in the edge image;

[0033] b3. Cluster the line segments in the edge image according to the calculated slope and intercept, and obtain m clusters of line segments;

[0034] b4. Fit the coordinate point data in the m clusters of line segments by using the least squares method, thereby fitting m line segments;

[0035] b5, clustering the line segment slopes, eliminating the line segment class with an element less than a certain threshold, and obtaining the fitted power line;

[0036] b6, using the elevation feature detection to fit the power line, and obtaining the ground wire point cloud data and the conductor point cloud data.

[0037] Preferably, the random sample consensus algorithm and the least square fitting plane region growing algorithm are used in the method c to extract the building point cloud data and the low vegetation point cloud data, and specifically comprising:

[0038] c1, classifying the non-ground point cloud data according to the random sample consensus algorithm to obtain an inner point data set and an outer point data set;

[0039] c2, clustering the point cloud data in the inner point data set and the outer point data set, respectively;

[0040] c3, using the least square fitting plane region growing algorithm to screen the clustered point cloud data to obtain the building point cloud data and the low vegetation point cloud data.

[0041] Preferably, the random sample consensus algorithm and the least square fitting plane region growing algorithm are used in the method c to extract the building point cloud data and the low vegetation point cloud data, and specifically comprising:

[0042] Using the random sample consensus algorithm to fit and cluster the non-ground point cloud data to extract the preliminary point cloud data meeting the target;

[0043] Using the least square fitting plane region growing algorithm again according to the preliminary point cloud data to extract the final target point cloud data;

[0044] Using the three-dimensional coordinate values of the final target point cloud data to calculate the 3*3 covariance matrix of the target data point;

[0045] Determining three eigenvalues λx, λy, λz of the target data point in X, Y and Z directions from the covariance matrix;

[0046] If the eigenvalue λz of the target data point in Z direction is greater than the eigenvalues λx, λy of the target data point in X and Y directions, the target data point is a point in the building point cloud data; otherwise, if the eigenvalue λz of the target data point in Z direction is less than or equal to the eigenvalues λx, λy of the target data point in X and Y directions, the target data point is a point in the low vegetation point cloud data.

[0047] According to the above technical solution, compared with the prior art, the present application provides a kind of high-voltage overhead transmission line key element laser point cloud automatic classification method, with the following beneficial effects:

[0048] 1. Optimize the initial triangulation: While keeping the original mesh size unchanged, move the mesh window, extract the lowest point in the mesh, and add it to the initial triangulation. This method effectively solves the information loss caused by an excessively large mesh and greatly reduces the number of point cloud iterations, effectively improving the computation speed and meeting the basic requirements of automation.

[0049] 2. Based on the centerline fitting method combining Hough transform and least squares, electric power lines are extracted; by making full use of the shape characteristics of the tower itself, the minimum bounding box and fitting circle of the point cloud are calculated in real time, providing an effective solution for tower extraction and point cloud classification.

[0050] 3. An improved algorithm combining the random sampling consensus algorithm and the region growth algorithm is used to accurately extract buildings. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0052] Figure 1 This is a schematic diagram of the overall process flow of the method provided by the present invention;

[0053] Figure 2 A schematic diagram illustrating the detailed process of the method provided by this invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] like Figure 1 , Figure 2 This invention discloses an automatic classification method for laser point clouds of key elements of high-voltage overhead transmission lines, comprising the following steps:

[0056] S1. Collect point cloud data of high-voltage overhead transmission lines and preprocess the collected point cloud data by referencing overhead line ledger information.

[0057] Specifically, this includes: using a denoising algorithm based on the local spatial distribution statistics of point clouds to denoise the collected point cloud data of high-voltage overhead transmission lines;

[0058] The point cloud data of the high-voltage overhead transmission line after the noise removal processing is segmented according to the tower coordinate position in the overhead line account information, and the point cloud data in multiple standard las formats is generated.

[0059] The noise removal algorithm based on the local spatial distribution statistics of the point cloud can calculate the number of point clouds in a space range of 1 cubic meter, and if the point cloud distribution in a certain local space is very small (the number of point clouds is less than 10), the point cloud is considered as a noise point.

[0060] S2, for the point cloud data after preprocessing, an irregular triangle network TIN progressive encryption filtering algorithm is used to separate ground point cloud data and non-ground point cloud data;

[0061] In the embodiment of the application, the ground point cloud data mainly refers to the ground surface point cloud data representing the ground surface point cloud, and other data other than the ground point cloud data are referred to as non-ground point cloud data.

[0062] Specifically, it comprises:

[0063] S21, obtaining the point cloud data after preprocessing;

[0064] S22, the point cloud data after preprocessing is divided into a grid, and the lowest point in each grid is taken as an initial point to generate an initial irregular triangle network TIN;

[0065] S23, the perpendicular distance of the point cloud data point to the nearest triangular face and the included angle between the data point and the three vertices of the triangular face are calculated, if the maximum value of the perpendicular distance and the included angle is less than a set threshold value, the data point is added to the ground point cloud data set, and the generated initial irregular triangle network TIN is encrypted;

[0066] S24, repeating steps S22-S23, and iteratively processing all point cloud data until all point cloud data is traversed.

[0067] S3, for the ground point cloud data, a weighted least square method is used for point cloud smoothing processing;

[0068] The weighted least square method is used for point cloud smoothing processing on the ground points obtained in the above step, the point cloud thickness can be reduced to within 5cm, and the features of the smoothed ground objects are well preserved, meeting the requirements of high-precision data classification and analysis processing.

[0069] S4, for the non-ground point cloud data, different algorithms are used for fine classification and extraction, specifically comprising:

[0070] a, using a Kd-tree clustering method to extract tower point cloud data;

[0071] b. The center line fitting method combining Hough transform and least squares is used to extract power line point cloud data, and ground wire point cloud data and conductor point cloud data are separated;

[0072] c. The random sample consensus algorithm and least squares fitting plane region growing algorithm are used to extract building point cloud data and low vegetation point cloud data.

[0073] The Kd-tree clustering method in method a for extracting tower point cloud data specifically includes:

[0074] The tower point cloud data is obtained according to the tower coordinate position information, and the tower point cloud data is divided into tower top point cloud data and tower body point cloud data according to the tower and pole geometric characteristics;

[0075] For the tower top point cloud data: an initial OBB bounding box of the tower top point cloud data is constructed, and a Kd-tree is recursively constructed with the initial OBB bounding box as the root node. The tower top point cloud data in the initial OBB bounding box is horizontally sliced, and the point cloud data in each horizontal slice is fitted with a circle to extract the point cloud data in the horizontal slice. The Kd-tree is used to search and merge all the point cloud data in the horizontal slices to extract the tower and pole point cloud data.

[0076] For the tower body point cloud data: an initial OBB bounding box of the tower body point cloud data is constructed, and an initial OBB fitting is performed on the tower body point cloud data in the initial OBB bounding box. According to the results of the initial OBB fitting, the tower edges of the tower body are identified, and the tower body point cloud data of the identified tower edges is accurately fitted with an OBB.

[0077] The biggest feature of OBB is its arbitrary direction, which makes it possible to enclose the object as closely as possible according to the shape characteristics of the enclosed object. The OBB bounding box is more closely approximated to the object than the AABB bounding box and the enclosing sphere, which can significantly reduce the number of bounding boxes and avoid intersection detection between a large number of bounding boxes.

[0078] In method b, the center line fitting method combining Hough transform and least squares is used to extract power line point cloud data, and ground wire point cloud data and conductor point cloud data are separated, which specifically includes:

[0079] b1. Obtain an elevation value image of non-ground point cloud data, perform Canny operator detection on the elevation value image, and extract an edge image;

[0080] b2. Use Hough transform to detect line segments in the edge image, and calculate the slope and intercept of each line segment in the edge image;

[0081] b3. Cluster the line segments in the edge image according to the calculated slope and intercept to obtain m clusters of line segments;

[0082] Specifically, the line segments detected by the Hough transform can be considered as collinear if the slopes and intercepts of the line segments are close to each other, clustering is performed according to the collinearity criterion, the ranges of all slopes and intercepts are counted, the slope threshold Ta and the intercept threshold Tb are set according to the range, and all line segments are scanned in turn, if the slope difference and the intercept difference of two line segments are less than Ta and Tb respectively, it is considered that they belong to the same cluster, and thus m cluster line segment groups are obtained.

[0083] b4, respectively, the coordinate point data in the m cluster line segment groups are fitted by the least square method, and thus m line segments are fitted;

[0084] b5, the line segment slopes are clustered, the line segment class with an element less than a certain threshold is removed, and the fitted power line is obtained;

[0085] b6, the fitted power line is detected using the elevation feature, and the ground wire point cloud data and the conductor point cloud data are obtained.

[0086] In the method c, the random sample consensus algorithm and the least square fitting plane region growing algorithm are used to extract the building point cloud data and the low vegetation point cloud data, and the method specifically comprises:

[0087] The random sample consensus algorithm is used to fit and cluster the non-ground point cloud data, and the preliminary point cloud data meeting the target are extracted;

[0088] The least square fitting plane region growing algorithm is used again according to the preliminary point cloud data to extract the final target point cloud data;

[0089] The 3*3 covariance matrix of the target data point is calculated using the three-dimensional coordinate values of the final target point cloud data;

[0090] The three eigenvalues λx, λy and λz of the target data point in the X, Y and Z directions are determined from the covariance matrix;

[0091] If the eigenvalue λz of the target data point in the Z direction is greater than the eigenvalues λx and λy of the target data point in the X and Y directions, the target data point is a point in the building point cloud data; otherwise, if the eigenvalue λz of the target data point in the Z direction is less than or equal to the eigenvalues λx and λy of the target data point in the X and Y directions, the target data point is a point in the low vegetation point cloud data.

[0092] All the calculation models in the application can be realized based on an open distributed computing architecture, and efficient and automatic fine classification of laser point cloud of key elements of overhead transmission lines is obtained. The specific process of distributed monitoring is as follows:

[0093] (1) The distributed algorithm monitoring tool is deployed on the master node machine, the specific information of the sub-nodes and the classification is obtained by reading the configuration file, and the information is displayed in a table;

[0094] (2) Remote start the classification processing algorithm of the corresponding node machine by checking the relevant nodes, and show the execution progress of each sub-node (each node machine should be deployed with point cloud classification function);

[0095] (3) The distributed monitoring tool is to split a "huge" classification task into multiple sub-tasks, and start multiple classification processing processes, each process automatically takes point cloud classification sub-tasks and executes.

[0096] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part.

[0097] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for automatic classification of laser point clouds of key elements of high voltage overhead power transmission lines, characterized in that, The method comprises the following steps: S1, collecting high-voltage overhead transmission line point cloud data, and referencing overhead line account information to preprocess the collected point cloud data, wherein the overhead line account information comprises tower coordinate position information; the preprocessing of the point cloud data by referencing the overhead line account information specifically comprises: adopting a denoising algorithm based on local spatial distribution statistics of point cloud to perform denoising processing on the collected high-voltage overhead transmission line point cloud data; segmenting the denoised high-voltage overhead transmission line point cloud data according to the tower coordinate position in the overhead line account information, and generating multi-arch standard las format point cloud data; S2, for the preprocessed point cloud data, adopting an irregular triangle network (TIN) progressive encryption filtering algorithm to separate ground point cloud data and non-ground point cloud data; S3, for the ground point cloud data, adopting a weighted least squares method to perform point cloud smoothing processing; S4, for the non-ground point cloud data, utilizing different algorithms to perform fine classification and extraction; for the non-ground point cloud data, utilizing different algorithms to perform fine classification, specifically comprising: a, extracting tower point cloud data by utilizing a Kd-tree algorithm and an OBB bounding box; b, extracting power line point cloud data by combining a Hough transform and a center line fitting method, and separating ground line point cloud data and conductor point cloud data; c, extracting building point cloud data and low vegetation point cloud data by adopting a random sample consensus algorithm and a least squares fitting plane region growing algorithm.

2. The method for automatic classification of laser point clouds of key elements of high voltage overhead power lines according to claim 1, characterized in that, Step S2 comprises the following steps: S21, obtaining the preprocessed point cloud data; S22, performing grid division on the preprocessed point cloud data, taking the lowest point in each grid as an initial point to generate an initial irregular triangle network (TIN); S23, calculating the perpendicular distance of a point cloud data point to the nearest triangular face and the included angle between the data point and the three vertices of the triangular face, if the maximum value of the perpendicular distance and the included angle is less than a set threshold, the data point is added to the ground point cloud data set, and the generated initial irregular triangle network (TIN) is encrypted; S24, repeating steps S22-S23, and iteratively processing all point cloud data until all point cloud data is traversed.

3. The method for automatic classification of laser point clouds of key elements of high voltage overhead power lines according to claim 1, characterized in that, In method a, the extraction of tower point cloud data by utilizing a Kd-tree algorithm and a bounding box specifically comprises: obtaining tower point cloud data according to tower coordinate position information, and dividing the tower point cloud data into tower top point cloud data and tower body point cloud data according to tower geometric characteristics; for the tower top point cloud data: constructing an initial OBB bounding box of the tower top point cloud data, and recursively constructing a Kd-tree with the initial OBB bounding box as a root node, performing horizontal slicing on the tower top point cloud data in the initial OBB bounding box, and performing circle fitting on the point cloud data in each horizontal slice to extract the point cloud data in the horizontal slice, and utilizing the Kd-tree to search and merge all point cloud data in the horizontal slices to extract tower point cloud data; For the tower body point cloud data: construct the initial OBB bounding box of the tower body point cloud data, perform initial OBB fitting on the tower body point cloud data in the initial OBB bounding box, and identify the tower ridge of the tower body according to the result of the initial OBB fitting. The tower body point cloud data of the identified tower ridge is subjected to accurate OBB fitting.

4. The method for automatic classification of laser point clouds of key elements of high voltage overhead power lines according to claim 1, characterized in that, In method b, the center line fitting method combining Hough transform and least square is used to extract power line point cloud data, and the ground wire point cloud data and the conductor point cloud data are separated, which specifically includes: b1, obtaining the elevation value image of the non-ground point cloud data, performing Canny operator detection on the elevation value image, and extracting the edge image; b2, using Hough transform to detect the line segments in the edge image, and calculating the slope and intercept of each line segment in the edge image; b3, clustering the line segments in the edge image according to the calculated slope and intercept, and obtaining m clusters of line segment groups; b4, fitting the coordinate point data in the m clusters of line segment groups with the least square method, thereby fitting m line segments; b5, clustering the line segment slopes, removing the line segment class with an element less than a certain threshold, and obtaining the fitted power line; b6, using the elevation feature to detect the fitted power line, and obtaining the ground wire point cloud data and the conductor point cloud data.

5. The method for automatic classification of laser point clouds of key elements of high voltage overhead power lines according to claim 1, characterized in that, In method c, the random sample consensus algorithm and the least square fitting plane region growing algorithm are used to extract building point cloud data and low vegetation point cloud data, which specifically includes: Using the random sample consensus algorithm to fit and cluster the non-ground point cloud data, and extracting the preliminary point cloud data meeting the target; Using the least square fitting plane region growing algorithm again according to the preliminary point cloud data to extract the final target point cloud data; Using the three-dimensional coordinate values of the final target point cloud data to calculate the 3*3 covariance matrix of the target data points; Determining three eigenvalues λx, λy, λz of different target data points in X, Y and Z directions from the covariance matrix; If the eigenvalue λz of the target data point in the Z direction is greater than the eigenvalues λx, λy of the target data point in the X and Y directions, the target data point is a point in the building point cloud data; otherwise, if the eigenvalue λz of the target data point in the Z direction is less than or equal to the eigenvalues λx, λy of the target data point in the X and Y directions, the target data point is a point in the low vegetation point cloud data.

Images