A point cloud processing method for power cable identification and reconstruction

Abstract

The present invention relates to the field of cable identification technology, and discloses a point cloud processing method for power cable identification and reconstruction, comprising: reading point cloud data: reading point cloud data from a depth camera, a lidar, and three-dimensional coordinate point cloud data from a binocular camera; accumulating multiple frames of point cloud: accumulating multiple frames of point cloud to reduce the impact of missing and sparse point clouds on the final processing results; operating instructions: issuing operating commands for operating specific cables; cropping point cloud: cropping the scene to crop out the cable to be operated and a small area around it from the scene. The present invention performs excellently in the field of live working. Compared with point cloud processing methods in other fields or with a certain degree of versatility, this method has higher accuracy in cable identification, and after testing, the accuracy rate can be stably maintained at above 90%.

Description

Technical Field

[0001] The present invention relates to the technical field of cable identification, and in particular to a point cloud processing method for power cable identification and reconstruction. Background Art

[0002] With the emergence of high-precision sensors and the increasing maturity of computer vision technology, 3D scene modeling, recognition, and reconstruction technologies are being widely applied in various fields, such as autonomous driving, terrain mapping, and path planning. Coincidentally, this technology is also a rising star in the field of live-line work. Due to the high risk of live-line work, most live-line work is now carried out by robots. Robots rely primarily on sensors to identify the working cables. While traditional binocular vision sensors offer high accuracy, they are significantly affected by lighting. Therefore, sensors such as LiDAR are needed outdoors to overcome this shortcoming. Processing point cloud data is an essential step in this process.

[0003] Currently, most common point cloud processing processes are only targeted at the above-mentioned specific technical fields, while the field of live working has some problems that are not involved in other fields. For example, in strong light and backlight conditions outdoors, the point cloud data of power cables is sparse, severely missing, and has severe noise; the black insulation of power cables has low reflectivity, resulting in a weak radar echo signal; the exposed metal part of the cable exhibits high-brightness optical characteristics under sunlight, causing the binocular camera based on visible light technology to fail and rely on radar. However, in strong light and backlight environments, the laser signal emitted by the radar is severely interfered by the strong light, and the collected point cloud is either missing or has severe noise, which makes cable identification difficult. Therefore, point cloud processing methods in other fields or point cloud processing methods with a certain degree of generality have poor applicability in the field of live working. How to solve the above problems is exactly what the point cloud processing method for power cable identification and reconstruction aims to solve. Summary of the Invention

[0004] The purpose of the present invention is to provide a point cloud processing method for power cable identification and reconstruction, so as to solve the problems raised in the background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a point cloud processing method for power cable identification and reconstruction, comprising:

[0006] Read point cloud data: read point cloud data from depth cameras, lidars, and 3D coordinate point cloud data from binocular cameras;

[0007] Accumulate multi-frame point clouds: accumulate point clouds over multiple frames to reduce the impact of missing and sparse point clouds on the final processing results;

[0008] Operation instructions: Issue operation instructions for specific cables;

[0009] Cropping point cloud: cropping the scene, cutting out the cable to be operated and the small area around it;

[0010] Point cloud downsampling: Gridding the numerous (dense) point cloud data voxels to perform preliminary filtering on the point cloud, reduce the data volume, and improve the algorithm processing speed;

[0011] Point cloud clustering and segmentation: separate cables from surrounding scattered noise;

[0012] Point cloud filtering: remove isolated points and perform further filtering on the point cloud;

[0013] Point cloud color filtering: Combine the cable point cloud texture and color features to perform the final filtering on the point cloud;

[0014] Point cloud fitting: Perform straight line or curve fitting on the filtered points and obtain the reconstructed point coordinates of the cable point cloud.

[0015] As a preferred embodiment of the present invention, the specific steps of point cloud downsampling are:

[0016] Step 1.1: Search the coordinates of all point clouds, find the maximum values in the X, Y, and Z directions, and use them to build a large cubic voxel grid that encompasses all point clouds;

[0017] Step 1.2: If the side length of the large cube grid is greater than the preset side length, divide it into several small voxel grids along the X, Y, and Z directions until the side lengths of all voxel grids are less than or equal to the preset side lengths;

[0018] Step 1.3: When the side lengths of all voxel grids are less than or equal to the preset side lengths, calculate the centroid of each voxel grid and finally obtain the centroid coordinates of the voxel grid;

[0019] Step 1.4: Calculate the distance between all points in the voxel grid and the centroid. In subsequent processing, this point will replace all points in the voxel grid.

[0020] As a preferred embodiment of the present invention, the specific steps of point cloud clustering and segmentation are:

[0021] Step 2.1: Set the cropped point cloud data as the initial point set and use the data in the initial point set to construct the KD_TREE data structure;

[0022] Step 2.2: Set the subscripts of all points;

[0023] Step 2.3: Declare two vector containers, one for storing the subscripts of points and the other for storing the distance between two points;

[0024] Step 2.4: Randomly select a point from the initial point set as the fixed point, and calculate the distance between all other points and the fixed point;

[0025] Step 2.5: Calculate the average distance between all points and the fixed point in step 2.4, and use 1 / 4 of this value as the preset threshold;

[0026] Step 2.6: Query all points in the initial point set. If the distance between the point and the fixed point is less than the preset threshold, store the point in the target point set. Store the point index and the distance between the point and the fixed point in two vector containers respectively, and mark the point as searched.

[0027] Step 2.7: Determine whether all points in the target point set are marked as searched. If so, proceed to step 2.10; otherwise, proceed to step 2.8.

[0028] Step 2.8: Change the distance value of the points marked as searched in the corresponding vector container to 0, and classify the points in the target point set according to whether the distance is 0;

[0029] Step 2.9: Set the point corresponding to the maximum distance value in the vector container used to store the distance between two points as the new fixed point, return to step 2.5, and redeclare the container and subscript;

[0030] Step 2.10: Determine whether all points in the initial point set are marked as searched. If so, complete the cluster segmentation. If not, classify all points not marked as searched into a new point set, select a new fixed point in the point set, and return to step 2.5;

[0031] Step 2.11: After the point cloud segmentation process is completed, several point cloud cluster subsets can be obtained. The bounding box of each subset is calculated according to the preset algorithm, and then the bounding box is selected based on the geometric characteristics of the cable.

[0032] As a preferred embodiment of the present invention, the specific steps of the point cloud filtering are:

[0033] Step 3.1: Calculate the distance between all points in the point cloud;

[0034] Step 3.2: Calculate the mean and standard deviation of the distances between all points;

[0035] Step 3.3: Calculate the global distance threshold of the point cloud data;

[0036] Step 3.4: Calculate the average distance between a point and all other points in its neighborhood. If the average is greater than the global distance threshold of the point cloud data, the point is determined to be an isolated outlier and removed from the point cloud. If the average is less than the global distance threshold of the point cloud data, the point is retained and marked.

[0037] Step 3.5: If all remaining points in the point cloud are marked points, the point cloud filtering is completed.

[0038] As a preferred embodiment of the present invention, the specific steps of the point cloud color filtering are:

[0039] Using LiDAR and RGB cameras, the points in the world coordinate system of the LiDAR point cloud data and the points in the camera coordinate system of the RGB camera can be mapped one-to-one through a transformation matrix, thereby assigning the RGB value of each pixel of the RGB image in the pixel coordinate system to the corresponding point in the point cloud data in the world coordinate system;

[0040] The point cloud data and its corresponding RGB values are matched with the power cable image for texture and feature matching. If the final calculated matching error is within the error threshold, the point cloud data is retained; otherwise, the point cloud data is deleted to achieve color filtering of the point cloud.

[0041] As a preferred embodiment of the present invention, the specific steps of the point cloud fitting are:

[0042] If the geometric features of the cable within the identified range are close to a straight line, a straight line fitting algorithm is used for fitting. If the geometric features of the cable within the identified range are curved (non-straight), curve fitting is used for fitting.

[0043] Compared with the existing technology, the present invention provides a point cloud processing method for power cable identification and reconstruction, which has the following beneficial effects:

[0044] 1. It performs well in the field of live working. Compared with other fields or some general point cloud processing methods, this method has higher accuracy in cable identification, and after testing, the accuracy rate can be stably maintained at above 90%;

[0045] 2. The point cloud processing method for power cable identification is not only applicable to the processing of radar point cloud data, but also to the processing of multi-fusion sensor point cloud data;

[0046] 3. After multiple filtering processes such as voxel rasterization, isolated point filtering, and texture filtering, more than 95% of the noise in the point cloud has been cleaned up. This enables the point cloud processing method for power cable identification to cope with various adverse conditions, including but not limited to strong light, backlight, and complex spatial environments;

[0047] 4. The algorithm is highly efficient, takes less time to process point cloud data, and has a high degree of automation, which can effectively reduce manual participation. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0049] Figure 1 This is a schematic diagram of the overall flow of a point cloud processing method for power cable identification and reconstruction according to the present invention;

[0050] Figure 2 This is a schematic diagram of a point cloud downsampling sub-process of a point cloud processing method for power cable identification and reconstruction according to the present invention;

[0051] Figure 3 Schematic diagram of point cloud clustering and segmentation sub-processes of a point cloud processing method for power cable identification and reconstruction according to the present invention;

[0052] Figure 4 The present invention is a schematic diagram of a point cloud straight line fitting process of a point cloud processing method for power cable identification and reconstruction. DETAILED DESCRIPTION

[0053] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0054] See also Figure 1-2 The present invention provides a technical solution: a point cloud processing method for power cable identification and reconstruction, comprising:

[0055] Read point cloud data: read point cloud data from depth cameras, lidars, and 3D coordinate point cloud data from binocular cameras;

[0056] Accumulate multi-frame point clouds: accumulate point clouds over multiple frames to reduce the impact of missing and sparse point clouds on the final processing results;

[0057] Operation instructions: Issue operation instructions for specific cables;

[0058] Cropping point cloud: cropping the scene, cutting out the cable to be operated and the small area around it;

[0059] Point cloud downsampling: Gridding the numerous (dense) point cloud data voxels to perform preliminary filtering on the point cloud, reduce the data volume, and improve the algorithm processing speed;

[0060] Point cloud clustering and segmentation: separate cables from surrounding scattered noise;

[0061] Point cloud filtering: remove isolated points and perform further filtering on the point cloud;

[0062] Point cloud color filtering: Combine the cable point cloud texture and color features to perform the final filtering on the point cloud;

[0063] Point cloud fitting: Perform straight line or curve fitting on the filtered points and obtain the reconstructed point coordinates of the cable point cloud.

[0064] In this embodiment, the specific steps of point cloud downsampling are:

[0065] Step 1.1: Search the coordinates of all point clouds and find the maximum value X in the X, Y, and Z directions max ,Y max ,Z max and minimum value X min ,Y min ,Z min , and use this to build a large cubic voxel grid that encompasses all point clouds;

[0066] Step 1.2: If the side length of the large cube grid is greater than the preset side length (the preset side length is 1 cm), then divide it into several small voxel grids along the X, Y, and Z directions until the side lengths of all voxel grids are less than or equal to the preset side lengths;

[0067] Step 1.3: When the side lengths of all voxel grids are less than or equal to the preset side length, calculate the center of gravity of each voxel grid. The calculation formula for the center of gravity is:

[0068]

[0069] Where n is the number of all points in the voxel grid, (x i ,y i ,z i ) is the point cloud coordinate within the voxel grid, and the final coordinates of the center of gravity of the voxel grid are (x0, y0, z0);

[0070] Step 1.4: Calculate the distance between all points in the voxel grid and the centroid using the following formula:

[0071]

[0072] Substitute the coordinates of all points in the voxel grid into the formula and find the minimum distance. The point corresponding to the minimum value is the point closest to the center of gravity. In subsequent processing, this point will replace all points in the voxel grid.

[0073] In this embodiment, the specific steps of point cloud clustering and segmentation are:

[0074] Step 2.1: Set the cropped point cloud data as the initial point set U1, and use the data in U1 to construct the KD_TREE data structure;

[0075] Step 2.2: Set the subscripts of the points: Set the subscripts of all points to: P 11 , P 12 ,...P 1n ;

[0076] Step 2.3: Declare two vector containers, one for V subscript , used to store the subscript of the point, and the other is V distance , used to store the distance between two points;

[0077] Step 2.4: Randomly select a point P in U1 1j As a fixed point, and calculate the distance between all other points and this fixed point;

[0078] Step 2.5: Calculate the average distance between all points and the fixed point in step 2.4, and use 1 / 4 of this value as the preset threshold u;

[0079] Step 2.6: Query all points in U1. If the point is to the fixed point P 1j If the distance is less than the threshold u, the point is stored in the target point set U2, and the subscript of the point is stored in V subscript , store the distance from this point to the fixed point in V distance , and mark the point as searched;

[0080] Step 2.7: Determine whether all midpoints in U2 are marked as searched. If so, proceed to step 2.10; otherwise, proceed to step 2.8.

[0081] Step 2.8: Mark the points you have searched for in V distance The distance value in is changed to 0, and the midpoints of U2 are classified according to whether the distance is 0;

[0082] Step 2.9: V distance The point corresponding to the maximum distance value is set as the new fixed point, and the process returns to step 2.5. At the same time, the container and subscript are redeclared.

[0083] Step 2.10: Determine whether all points in U1 are marked as searched. If so, complete the cluster segmentation. If not, classify all points not marked as searched into a new point set, select a new fixed point in the point set, and return to step 2.5;

[0084] Step 2.11: After the point cloud segmentation process is completed, several point cloud cluster subsets are obtained. A bounding box is calculated for each subset based on the preset algorithm. The bounding boxes are then selected based on the geometric characteristics of the cable. For example, for straight cables, if the aspect ratio of the bounding box is greater than 10, it is retained; otherwise, it is deleted. This method filters out noisy point cloud data and obtains the point cloud data of the power cable. This step is also combined with the cable recognition results of the binocular camera to further ensure the accuracy of the recognition results.

[0085] In this embodiment, the specific steps of the point cloud filtering are:

[0086] Step 3.1: Calculate the distance between all points in the point cloud (the point cloud here has been downsampled and filtered), and record the calculated result as d ij , represents the point (x i ,y i ,z i ) and point (x j ,y j ,z j ) between the two locations;

[0087] Step 3.2: Calculate the mean and standard deviation of the distances between all points using the following formula:

[0088]

[0089]

[0090] Step 3.3: Calculate the global distance threshold of the point cloud data. The calculation formula is as follows:

[0091] d0=μ+λσ

[0092] Where λ is the preset standard deviation coefficient.

[0093] Step 3.4: Calculate the average distance between a point and all other points in its neighborhood, denoted as Compare and d0, if Then the point is determined to be an isolated outlier and removed from the point cloud; if Then keep the point and mark it;

[0094] Step 3.5: If all remaining points in the point cloud are marked points, the point cloud filtering is completed.

[0095] In this embodiment, the specific steps of the point cloud color filtering are:

[0096] The device equipped with the point cloud processing method for power cable identification and reconstruction integrates multiple sensors, including a lidar and an RGB camera. According to the EPnP principle, points in the world coordinate system of the radar point cloud data and points in the camera coordinate system of the RGB camera can be mapped one-to-one through a transformation matrix. This principle can be used to assign the RGB value of each pixel of the RGB image in the pixel coordinate system to the corresponding point in the point cloud data in the world coordinate system.

[0097] The point cloud data and its corresponding RGB values are matched with the power cable image for texture and feature matching. If the final calculated matching error is within the error threshold, the point cloud data is retained; otherwise, the point cloud data is deleted to achieve color filtering of the point cloud.

[0098] In this embodiment, the specific steps of the point cloud fitting are:

[0099] If the geometric features of the cable within the identified range are close to a straight line, a straight line fitting algorithm is used for fitting. The straight line fitting sub-process of the point cloud uses a straight line fitting algorithm based on RANSAC. The algorithm steps are described in detail as follows:

[0100] Step 5.1: Randomly select two points and mark them (marked points can be reselected and marked);

[0101] Step 5.2: Using the coordinates of the two points mentioned in step 5.1, calculate the equation of the line in space determined by these two points. The equation of the line is:

[0102]

[0103] where the vector is the direction vector of the line, and the two points in (1) are recorded as P0(x0,y0,z0) and P1(x1,y1,z1) respectively; at the same time, declare a vector container v model , save the straight line equation in a container;

[0104] Step 5.3: Substitute the horizontal coordinates of the remaining points into the line equation described in step 5.2 to obtain their vertical and horizontal coordinates. For example, if the original point coordinates are P(x, y, z), the coordinates obtained after substituting them into the line equation are P′(x′, y′, z′). Then, based on the coordinates of the two points, calculate the error distance between points P and P′. The calculation formula is as follows:

[0105]

[0106] Step 5.4: Declare two vector containers v s and v d, and initialize the error threshold to 10mm. Compare the error distance obtained in step 5.3 with the error threshold. If it is less than or equal to the error threshold, it means that the point supports the model (linear equation). If it is greater than the error threshold, it means that the point does not support the model. After traversing all points except P0 and P1, record the number of points that support the model and store the number value in the container v s , and calculate the average value of the error distance of all points, and store the average value in the container v d ;

[0107] Step 5.5: Determine whether all points in the point cloud have been marked. If not, return to step 5.1 to step 5.4; if so, in the container v s Find the maximum value in the container v and find the corresponding straight line equation. If there are two or more maximum values, then d Find the corresponding average distance error and select the straight line equation with the smaller average distance error as the optimal output of the point cloud straight line fitting sub-process;

[0108] Step 5.6: Mark the points that support the optimal model described in step 5.5 as key points in the point cloud data.

[0109] Step 5.7: Set a search radius. In this process, the search radius is set to 10mm.

[0110] Step 5.8: Starting from the first point cloud on one side of the line model fitted in step 5.5, search the line model in the positive (or negative) direction of the line direction vector. Use the circular search area with the search radius set in step 5.7 as the radius (the center of the circular search area is on the line). If there is a point in the area, the search area continues to move forward. If there is no point, interpolation is performed within the search area according to the preset algorithm until the search area moves forward to the last point on the other side of the line. Interpolation ends.

[0111] Step 5.8: Reconstruct the cable model based on the output results of the above steps;

[0112] If the geometric features of the cable within the recognition range are curves (not straight lines), curve fitting is used. The detailed description of the point cloud curve fitting sub-process is as follows:

[0113] Step 6.1: Domain division: Set the initial point coordinates on one side of the curved point cloud to (0, y1, z1) and the final point coordinates on the other side to (1, y2, z2). Divide the domain into several small intervals. Correspondingly, the curved point cloud is also divided into several regions. The number of regions depends on the geometric characteristics of the original cable within the recognition range. The longer the cable and the greater the curvature of the curved part, the more regions should be divided.

[0114] Step 6.2: Control point selection: Select a point in each small interval as the control point. The control point is the centroid of the point cloud area in the interval. The calculation method for selecting the centroid here is the same as the calculation method for selecting the centroid in the point cloud downsampling step.

[0115] Step 6.3: Curve fitting: Let the nth control point in (2) be c n , there are m control points, node vector U=(μ0,μ1,...,μ m+k ), k is the order of the B-spline curve. The B-spline curve function can be used to fit the curved point cloud. The fitting equation is as follows:

[0116]

[0117] Among them B n () is the basis function. The basis function of a k-order B-spline curve can be expressed as:

[0118]

[0119] Among them, the basis function B n () also satisfies the following differential equation:

[0120]

[0121] Step 6.4: Interpolation: During the actual fitting process, after the area is divided, some areas may not have point cloud data. In this case, there are two approaches: if the geometric characteristics of the original power cables in the area are close to straight lines, use the linear interpolation algorithm in this interval; if the geometric characteristics of the power cables in the area are curved (non-straight lines), interpolate along the B-spline curve trajectory calculated in step 6.3. The interpolation algorithm used here is the same as the interpolation method used in step 5.8.

[0122] It should be noted here that during actual operation, whether the cable to be processed is a straight line or a curve will be given in the operation instructions. For example, in the line operation, the line can be treated as a straight line, while in the lead operation, the lead will generally be treated as a curve.

[0123] Step 6.5: Reconstruct the cable model based on the output results of the above steps.

[0124] The above shows and describes the basic principles, main features and advantages of the present invention. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, from all points of view, the embodiments should be regarded as illustrative and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes that fall within the meaning and range of equivalents of the claims are included in the present invention. Any reference signs in the claims should not be construed as limiting the claim to which they relate.

[0125] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A point cloud processing method for power cable identification and reconstruction, characterized by: include: Read point cloud data: read point cloud data from depth cameras, lidars, and 3D coordinate point cloud data from binocular cameras; Accumulate multi-frame point clouds: accumulate point clouds over multiple frames to reduce the impact of missing and sparse point clouds on the final processing results; Operation instructions: Issue operation instructions for specific cables; Cropping point cloud: cropping the scene, cutting out the cable to be operated and the small area around it; Point cloud downsampling: Gridding the numerous point cloud data voxels to perform preliminary filtering on the point cloud, reduce the data volume, and improve the algorithm processing speed; Point cloud clustering and segmentation: separate cables from surrounding scattered noise; Point cloud filtering: remove isolated points and perform further filtering on the point cloud; Point cloud color filtering: Combine the cable point cloud texture and color features to perform the final filtering on the point cloud; Point cloud fitting: perform straight line or curve fitting on the filtered points and obtain the reconstructed point coordinates of the cable point cloud; The specific steps of point cloud downsampling are: Step 1.1: Search the coordinates of all point clouds, find the maximum values in the X, Y, and Z directions, and use them to build a large cubic voxel grid that encompasses all point clouds; Step 1.2: If the side length of the large cube grid is greater than the preset side length, divide it into several small voxel grids along the X, Y, and Z directions until the side lengths of all voxel grids are less than or equal to the preset side lengths; Step 1.3: When the side lengths of all voxel grids are less than or equal to the preset side lengths, calculate the centroid of each voxel grid and finally obtain the centroid coordinates of the voxel grid; Step 1.4: Calculate the distance between all points in the voxel grid and the centroid. In subsequent processing, this point will replace all points in the voxel grid.

2. The point cloud processing method for power cable identification and reconstruction according to claim 1, characterized in that: The specific steps of point cloud clustering and segmentation are: Step 2.1: Set the cropped point cloud data as the initial point set and use the data in the initial point set to construct the KD_TREE data structure; Step 2.2: Set the subscripts of all points; Step 2.3: Declare two vector containers, one for storing the subscripts of points and the other for storing the distance between two points; Step 2.4: Randomly select a point from the initial point set as the fixed point, and calculate the distance between all other points and the fixed point; Step 2.5: Calculate the average distance between all points and the fixed point in step 2.4, and use 1 / 4 of this value as the preset threshold; Step 2.6: Query all points in the initial point set. If the distance between the point and the fixed point is less than the preset threshold, store the point in the target point set. Store the point index and the distance between the point and the fixed point in two vector containers respectively, and mark the point as searched. Step 2.7: Determine whether all points in the target point set are marked as searched. If so, proceed to step 2.10; otherwise, proceed to step 2.

8. Step 2.8: Change the distance value of the points marked as searched in the corresponding vector container to 0, and classify the points in the target point set according to whether the distance is 0; Step 2.9: Set the point corresponding to the maximum distance value in the vector container used to store the distance between two points as the new fixed point, return to step 2.5, and redeclare the container and subscript; Step 2.10: Determine whether all points in the initial point set are marked as searched. If so, complete the cluster segmentation. If not, classify all points not marked as searched into a new point set, select a new fixed point in the point set, and return to step 2.5; Step 2.11: After the point cloud segmentation process is completed, several point cloud cluster subsets can be obtained. The bounding box of each subset is calculated according to the preset algorithm, and then the bounding box is selected based on the geometric characteristics of the cable.

3. The point cloud processing method for power cable identification and reconstruction according to claim 1, characterized in that: The specific steps of point cloud filtering are: Step 3.1: Calculate the distance between all points in the point cloud; Step 3.2: Calculate the mean and standard deviation of the distances between all points; Step 3.3: Calculate the global distance threshold of the point cloud data; Step 3.4: Calculate the average distance between a point and all other points in its neighborhood. If the average is greater than the global distance threshold of the point cloud data, the point is determined to be an isolated outlier and removed from the point cloud. If the average is less than the global distance threshold of the point cloud data, the point is retained and marked. Step 3.5: If all remaining points in the point cloud are marked points, the point cloud filtering is completed.

4. The point cloud processing method for power cable identification and reconstruction according to claim 1, characterized in that: The specific steps of point cloud color filtering are: Using LiDAR and RGB cameras, the points in the world coordinate system of the LiDAR point cloud data and the points in the camera coordinate system of the RGB camera can be mapped one-to-one through a transformation matrix, thereby assigning the RGB value of each pixel of the RGB image in the pixel coordinate system to the corresponding point in the point cloud data in the world coordinate system; The point cloud data and its corresponding RGB values are matched with the power cable image for texture and feature matching. If the final calculated matching error is within the error threshold, the point cloud data is retained; otherwise, the point cloud data is deleted to achieve color filtering of the point cloud.

5. The point cloud processing method for power cable identification and reconstruction according to claim 1, characterized in that: The specific steps of point cloud fitting are: If the geometric features of the cable within the identified range are close to a straight line, a straight line fitting algorithm is used for fitting; if the geometric features of the cable within the identified range are a curve, a curve fitting algorithm is used for fitting.

Images