A laser point cloud power line repair method based on catenary algorithm

Abstract

The present invention relates to the field of computer technology, and in particular to a laser point cloud power line repair method based on a catenary algorithm. It includes a first server, which obtains the original point cloud data collected by a drone; enhances the original point cloud data; and also includes a second server, which cuts the enhanced data; the second server executes a square grid segmentation algorithm to obtain high-point point cloud data; the second server executes a DBSCAN clustering algorithm to obtain tower-like object set data, and the second server executes a DenseNet neural network algorithm to obtain electric tower point cloud data; the second server executes a yolov5 neural network algorithm to obtain electric wire point cloud data; the second server executes a Euclidean clustering algorithm to obtain single electric wire point cloud data; the second server executes a catenary equation algorithm to complete the missing data of the electric wire, and simulates the power line point cloud data. The present invention saves labor costs, solves the error problem caused by the missing power line point cloud, and ensures the timely detection of circuit hidden dangers and the safe operation of the power grid.

Description

Technical Field

[0001] The invention relates to the field of computer technology, and in particular to a laser point cloud power line repair method based on a catenary algorithm. Background Art

[0002] As the backbone of the power grid, the safety of overhead transmission lines plays a vital role in the reliability of power supply and the stable development of social economy.

[0003] With the continuous improvement of industrial production and the increasing enrichment of people's lives, power transmission safety has become increasingly important. It can be said that power transmission safety is the cornerstone of economic development. In the process of power transmission and transformation, the safety of transmission lines is of paramount importance. However, the environment in the distribution area of ​​transmission lines is complex and diverse. Illegal buildings, illegal construction and tree failures offline have caused great hidden dangers to the safe transmission of electric energy.

[0004] In the existing technology, circuit maintenance inspection mainly relies on manual inspection, LiDAR online monitoring technology and other methods; manual inspection has a large workload, high cost, low efficiency, and it is difficult to quickly and accurately discover potential problems. In addition, it is difficult for humans to reach lines in mountains, rivers, forests or virgin forests or uninhabited areas; LiDAR online detection often has problems such as partial missing of point cloud data in actual applications, resulting in large errors in the calculation of data such as span; therefore, it is necessary to develop a new method to solve the error problem caused by the missing point cloud of power lines. Summary of the invention

[0005] The purpose of the present invention is to address the deficiencies of the prior art and provide a laser point cloud power line repair method based on a catenary algorithm, which saves labor costs and solves the error problem caused by the missing power line point cloud.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The present invention relates to a laser point cloud power line repair method based on a catenary algorithm, specifically:

[0008] It includes a first server, wherein the first server is used for analyzing and storing data;

[0009] The first server obtains the original point cloud data collected by the drone via a wired or wireless method;

[0010] The first server enhances the original point cloud data to obtain enhanced data and stores the enhanced data in a first storage module;

[0011] Also included is a second server, which obtains the data in the first storage module and performs enhanced data segmentation;

[0012] The enhanced data cutting comprises the following steps:

[0013] (1) The second server obtains the original point cloud data in the first storage module, and executes a square grid segmentation algorithm to remove the ground and low near-ground objects to obtain high-point point cloud data, and stores it in the second storage module;

[0014] (2) The second server executes the DBSCAN clustering algorithm to obtain the tower-like object set data, and then the second server executes the DenseNet neural network algorithm to obtain the tower point cloud data, and stores it in the third storage module;

[0015] (3) The second server obtains the tower point cloud data in the third storage module, and divides the enhanced data into areas between two towers according to the location information of the tower point cloud data; the second server executes the yo l ov5 neural network algorithm to calculate the divided areas one by one to obtain the power line point cloud data;

[0016] (4) The second server executes the Euclidean clustering algorithm to obtain point cloud data of a single wire and stores the data in a fourth storage module;

[0017] (5) The second server retrieves the point cloud data of a single electric wire from the fourth storage module, and executes the catenary equation algorithm to complete the missing data of the electric wire, thereby simulating and obtaining the point cloud data of the power line.

[0018] The first server executes the CutMix algorithm to obtain image data generated by the point cloud, and the image data is the main view used for training.

[0019] The square grid segmentation algorithm divides the enhanced data into several square regions, obtains the lowest point in each region, and discards point cloud data that is not higher than 15m from the lowest point.

[0020] The above-mentioned second server executes the DBSCAN clustering algorithm to obtain the tower-like object set data, and then the second server obtains an axis-aligned bounding box based on the tower-like object set data; calculates the angle of the point cloud deviation from the x-axis based on the axis-aligned bounding box, rotates the point cloud so that the point cloud is parallel to the x-axis, and obtains the main view of the tower-like object set data at this time, and then the second server inputs the main view of the tower-like object set data into the DenseNet neural network to realize the recognition of the tower, and obtains the tower point cloud data.

[0021] Among them, the second server obtains the area between the two towers according to the tower point cloud data, and rotates the area so that the line between the two towers is parallel to the x-axis. The second server obtains the main view of the regional point cloud at this time and inputs the yo l ov5 neural network to select the area where the wires are located. The area where the wires are located is obtained according to the relationship between the selected area and the position of the towers.

[0022] To further illustrate, the catenary equation algorithm includes the following steps:

[0023] S1 obtains the coordinates of any three points on a single wire;

[0024] S2 Set the initial equation of the catenary Substitute the three points obtained into the equation to solve the values ​​of A, B, and C;

[0025] S3 obtains the tower boundary area based on the area between the towers, combines the values ​​of A, B, and C with the tower boundary area to generate data points for the power lines, and completes the missing data.

[0026] In the above catenary initial equation,

[0027] The beneficial effects of the present invention are as follows: the present invention utilizes deep learning and catenary algorithm to automatically repair the point cloud data of the electric tower, thereby providing more accurate data support for the later calculation of the span, safety angle, etc. of the point cloud data; it saves labor costs, solves the error problem caused by the missing point cloud of the power line, and realizes the fully automatic calculation of the span, the distance between the conductor and the ground wire, etc., thereby ensuring the timely detection of circuit hidden dangers and the safe operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0029] 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.

[0030] Example 1

[0031] A laser point cloud power line repair method based on catenary algorithm.

[0032] It includes a first server, wherein the first server is used for analyzing and storing data;

[0033] The first server obtains the original point cloud data collected by the drone via a wired or wireless method;

[0034] The first server enhances the original point cloud data to obtain enhanced data and stores the enhanced data in a first storage module;

[0035] Drones need to obtain a large amount of data for analysis and neural training. The huge amount of data is an important support for deep learning. However, the point cloud data of power transmission lines is limited, and direct training often faces problems such as overfitting and poor generalization ability. Therefore, it is very important to increase the amount of data through data enhancement.

[0036] The present invention adopts the data enhancement method of CutMix, which is to cut off a certain area of ​​the original picture and use part of other pictures to fill the cut area to generate a new picture. When labeling, the pictures that have been processed by CutMix are labeled with a separate category.

[0037] Also included is a second server, which obtains the data in the first storage module and performs enhanced data segmentation;

[0038] The enhanced data cutting comprises the following steps:

[0039] (1) The second server obtains the original point cloud data in the first storage module, and executes a square grid segmentation algorithm to remove the ground and low near-ground objects to obtain high-point point cloud data, and stores it in the second storage module;

[0040] (2) The second server executes the DBSCAN clustering algorithm to obtain the tower-like object set data, and then the second server executes the DenseNet neural network algorithm to obtain the tower point cloud data, and stores it in the third storage module;

[0041] (3) The second server obtains the tower point cloud data in the third storage module, and divides the enhanced data into areas between two towers according to the location information of the tower point cloud data; the second server executes the yo l ov5 neural network algorithm to calculate the divided areas one by one to obtain the power line point cloud data;

[0042] (4) The second server executes the Euclidean clustering algorithm to obtain point cloud data of a single wire and stores the data in a fourth storage module;

[0043] (5) The second server retrieves the point cloud data of a single electric wire from the fourth storage module, and executes the catenary equation algorithm to complete the missing data of the electric wire, thereby simulating and obtaining the point cloud data of the power line.

[0044] Among them, the first server executes the CutMix algorithm to obtain the image data generated by the point cloud, which is the main view used for training; the second server performs square grid cutting on the original point cloud data, and then the second server executes the DBSCAN clustering algorithm to obtain the main view of the image; the first server executes the CutMix algorithm to enhance the main view data.

[0045] Among them, the above-mentioned square grid segmentation algorithm divides the enhanced data into several square areas, obtains the lowest point in each area, and discards the point cloud data that is not higher than 15m from the lowest point; the grid cutting algorithm can approximate a complete model as a combination of multiple small grid models, and control the accuracy of the model by controlling the size and number of square areas; compared with the traditional CSF algorithm, which removes the ground and then uses the clustering algorithm to remove near-ground objects, this method has higher algorithm efficiency and performs better than the elevation filtering algorithm on point cloud data with mountainous landform characteristics.

[0046] The above-mentioned second server executes the DBSCAN clustering algorithm to obtain the tower-like object set data, and then the second server obtains an axis-aligned bounding box based on the tower-like object set data; calculates the angle of the point cloud deviation from the x-axis based on the axis-aligned bounding box, rotates the point cloud so that the point cloud is parallel to the x-axis, and obtains the main view of the tower-like object set data at this time, and then the second server inputs the main view of the tower-like object set data into the DenseNet neural network to realize the recognition of the tower, and obtains the tower point cloud data.

[0047] After removing the ground and low near-ground objects, the enhanced data only retains the top of the tower, the wires, and the top of tall near-ground objects. The point cloud data at this time is called "high-point point cloud"; at this time, the DBSCAN algorithm can be used to obtain the top of the tower and some tall near-ground objects, collectively referred to as "tower-like objects"; DBSCAN is a density-based clustering algorithm that can effectively remove objects similar to wires whose three-dimensional density is too different from that of towers. The main steps are: first determine the scanning radius and the minimum number of points contained in the cluster, and then start from any point selected as the visited point to find all points within the scanning radius of the point. If the number of points obtained is greater than the set minimum number of cluster points, the current point and all points within the scanning radius together form a cluster. According to this method, all points of the point cloud data are traversed in turn for recursive operations.

[0048] The second server obtains the area between the two towers based on the tower point cloud data, and rotates the area so that the line between the two towers is parallel to the x-axis. The second server obtains the main view of the regional point cloud at this time and inputs the yo lov5 neural network to select the area where the wires are located. The area where the wires are located is obtained based on the selected area and the positional relationship between the towers.

[0049] The Yo L OV5 neural network can use a single convolutional network for the entire image, dividing the image into grids to predict class probabilities and bounding boxes for each grid.

[0050] The separation of a single wire can be achieved through Euclidean clustering; Euclidean clustering is a clustering algorithm based on Euclidean distance. According to the set minimum number of cluster points, it can effectively extract a single wire from a wire group. However, due to the partial missing wire data in the point cloud, the obtained wires may be incomplete and a single wire may be divided into two in the cluster.

[0051] To further illustrate, the catenary equation algorithm includes the following steps:

[0052] S1 obtains the coordinates of any three points on a single wire;

[0053] S2 Set the initial equation of the catenary Substitute the three points obtained into the equation to solve the values ​​of A, B, and C;

[0054] S3 obtains the boundary area determined by the towers based on the area between the towers, combines the values ​​of A, B, and C with the boundary area determined by the towers to generate data points for the power lines, thereby completing the missing data.

[0055] In the above catenary initial equation,

[0056] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A laser point cloud power line repair method based on catenary algorithm, characterized by: It includes a first server, wherein the first server is used for analyzing and storing data; The first server obtains the original point cloud data collected by the drone via a wired or wireless method; The first server enhances the original point cloud data to obtain enhanced data and stores the enhanced data in a first storage module; Also included is a second server, which obtains the data in the first storage module and performs enhanced data segmentation; The enhanced data cutting comprises the following steps: (1) The second server obtains the original point cloud data in the first storage module, and executes a square grid segmentation algorithm to remove the ground and low near-ground objects to obtain high-point point cloud data, and stores it in the second storage module; (2) The second server executes the DBSCAN clustering algorithm to obtain the tower-like object set data, and then the second server executes the DenseNet neural network algorithm to obtain the tower point cloud data, and stores it in the third storage module; (3) The second server obtains the tower point cloud data in the third storage module, and divides the enhanced data into areas between two towers according to the location information of the tower point cloud data; The second server executes the yolov5 neural network algorithm to calculate the divided areas one by one to obtain the wire point cloud data; (4) The second server executes the Euclidean clustering algorithm to obtain point cloud data of a single wire and stores the data in a fourth storage module; (5) The second server retrieves the point cloud data of a single electric wire in the fourth storage module, and executes the catenary equation algorithm to fill in the missing data of the electric wire, thereby simulating the point cloud data of the electric wire; wherein, the second server obtains an axis-aligned bounding box according to the tower-like object set data; calculates the angle of deviation of the point cloud from the x-axis according to the axis-aligned bounding box, rotates the point cloud so that the point cloud is parallel to the x-axis, and obtains the main view of the tower-like object set data at this time, and then the second server inputs the main view of the tower-like object set data into the DenseNet neural network to realize the recognition of the electric tower, thereby obtaining the electric tower point cloud data; the second server obtains the area between the two electric towers according to the electric tower point cloud data, and rotates the area so that the line between the two electric towers is parallel to the x-axis; the second server obtains the main view of the regional point cloud at this time and inputs it into the yolov5 neural network to frame the area where the electric wire is located, and the area where the electric wire is located is obtained according to the relationship between the framed area and the position of the electric tower.

2. The laser point cloud power line repair method based on the catenary algorithm according to claim 1, characterized in that: The first server executes the CutMix algorithm to obtain image data generated by the point cloud, and the image data is a main view used for training.

3. The laser point cloud power line repair method based on the catenary algorithm according to claim 1, characterized in that: The square grid segmentation algorithm divides the enhanced data into several square regions, obtains the lowest point in each region, and discards point cloud data that is not higher than 15m from the lowest point.

4. The laser point cloud power line repair method based on the catenary algorithm according to claim 1, characterized in that: The catenary equation algorithm comprises the following steps: S1 obtains the coordinates of any three points on a single wire; S2 Set the initial equation of the catenary Substitute the three points obtained into the equation to solve the values ​​of A, B, and C. S3 obtains the tower boundary area based on the area between the towers, combines the values ​​of A, B, and C with the tower boundary area to generate data points for the power lines, and completes the missing data.

Images