Point cloud power line recognition method and system based on reflectivity and spatial morphology

Abstract

The present invention discloses a method and system for identifying power lines from point clouds based on reflectivity and spatial morphology. The method first obtains and segments the original point cloud data of a lidar to obtain a point cloud segmentation area. Secondly, based on the density and elevation information of the point cloud segmentation area, the first power line point cloud is screened, and based on the spatial characteristics of the point to adjacent points, the second power line point cloud is screened. Then, for each point in the second power line point cloud, based on a preset reflectivity threshold, the third power line point cloud is screened. Finally, morphological processing is performed on each point in the third power line point cloud to screen out the target point cloud. The system includes a data collection module, a region segmentation module, a ground object separation module, a feature processing module, a reflectivity processing module, and a morphological processing module for power grid environment point clouds. The present invention can automatically and real-time identify power lines, effectively reducing the workload of line inspection employees and improving the maintenance efficiency of power lines.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a point cloud power line identification method and system based on reflectivity and spatial morphology. Background Art

[0002] As an important part of the power grid system, power grid lines carry the lifeline of electric energy transportation required for national production and life, and are an indispensable link to ensure the power system. With the rapid development of the national economy, the scale of the power grid has expanded rapidly and the structure has become increasingly complex, which has brought new challenges to the periodic inspection of power lines to ensure the safety of the power grid. The traditional method is mainly based on manual ground inspection, which is time-consuming and labor-intensive, has a long inspection cycle, and is heavily dependent on terrain conditions. It cannot meet the needs of the modernization of the power industry. In recent years, the development of drones and airborne laser scanning technology has brought hope for unmanned real-time inspection of power lines. Airborne laser radar has the characteristics of high-precision ranging, high safety, high anti-interference and adaptability, which helps to achieve accurate inspection and maintenance of power lines and helps to improve the reliability and safety of the power system.

[0003] Power grid lines are divided into transmission lines and distribution lines. Transmission lines use towers, the power lines are thicker and the surroundings are open, while distribution lines mostly use electric poles, the power lines are thinner and the surrounding environment is more complex. At present, power line inspection mainly uses two methods to obtain laser point cloud data: manually operated drone flight or drone flight according to a predetermined route. The existing power line extraction method is mainly applied to transmission lines. After the inspection, the obtained laser point cloud data is processed offline in non-real time. This method requires the cruise route to be known in advance, and at the same time, it relies heavily on the performance of the computing device, the algorithm complexity is high, and the running time is long, which cannot meet the needs of real-time detection on the end-side device (drone). Therefore, on the end-side device (drone), based on laser point cloud data, how to accurately and quickly extract power line point clouds, how to accurately fit and reconstruct power lines, and how to formulate routes in real time to achieve the goal of real-time automatic inspection of power lines has important research significance.

[0004] Based on this, it is necessary to provide a method and system for real-time extraction of power lines in distribution lines to address the technical problems that the above-mentioned existing technologies do not fully consider the complexity of distribution network lines and the low performance of terminal equipment, resulting in the inability to conduct real-time inspections, low extraction accuracy, large amount of calculations, and complex parameter threshold settings, resulting in low efficiency. Summary of the invention

[0005] In view of the above problems, the present invention proposes a classification and recognition method and system based on spatial morphological feature extraction, which is applied to the field of power line recognition.

[0006] The present invention uses spatial grid technology to process power line point clouds. Through power line characteristic information such as point cloud density, elevation information, spatial dimension characteristics, morphological characteristics, and reflectivity, the extraction of power line point clouds is not restricted by factors such as equipment, corridor terrain, environment, and point cloud density, improving the accuracy and efficiency of extraction and identifying power lines in real time.

[0007] The first aspect of the present invention discloses a power line classification and recognition method based on the extraction of spatial morphological characteristics. This method includes the following steps:

[0008] S1. Obtain the original point cloud data of the lidar, and segment the original point cloud data to obtain several point cloud segmentation areas.

[0009] S2. Based on the overall density information and elevation information of each point cloud segmentation area, screen out the first power line point cloud.

[0010] S3. For each point in the first power line point cloud, based on the spatial characteristics of the point to all adjacent points within the preset adjacent area threshold, screen out the second power line point cloud regarded as a power line point.

[0011] S4. For each point in the second power line point cloud, based on the preset reflectivity threshold, screen out the third power line point cloud regarded as a power line point.

[0012] S5. For each point in the third power line point cloud, perform morphological processing, screen out the target point cloud, and obtain the recognized power line.

[0013] In the power line extraction method described above, the original point cloud data includes a point set and the coordinates and reflectivity information corresponding to each point in the point set. The point set includes ground points, vegetation points, utility pole points, noise points, and power line points.

[0014] Further, the specific steps of segmenting the power line lidar point cloud to obtain several point cloud segmentation areas include:

[0015] S1.1. Perform equidistant segmentation of the original point cloud data in the X-axis direction to obtain the first three-dimensional spatial grid corresponding to the original point cloud data.

[0016] S1.2. Perform equidistant segmentation of each first three-dimensional spatial grid in the Y-axis direction to obtain the second three-dimensional spatial grid corresponding to the original point cloud data.

[0017] S1.3. Project each second three-dimensional spatial network onto the XY plane to obtain several point cloud segmentation areas.

[0018] Further, the specific steps of screening out the first power line point cloud based on the overall density information and elevation information of each point cloud segmentation area include:

[0019] S2.1. Set a distance threshold. If there is no point in the adjacent area of the point cloud segmentation area whose distance from the highest elevation of this point cloud segmentation area is less than the distance threshold, then this segmentation area is regarded as an area without power lines.

[0020] S2.2. Set a density threshold. If there is a point in the adjacent area of the point cloud segmentation area whose distance from the highest elevation of this area is less than the distance threshold, and the point cloud density in this area is less than the density threshold, then all points in this area are regarded as the first component of the first power line point cloud.

[0021] S2.3. Set an elevation difference. If the point cloud density in this area is greater than the density threshold, and the difference between the highest elevation and the lowest elevation in this area exceeds the elevation difference, then this area is regarded as a power line, and the optimal height threshold is calculated according to the elevation distribution in this area.

[0022] S2.4. If the point cloud in this area whose elevation is higher than the optimal height threshold is regarded as the second component of the first power line point cloud.

[0023] S2.5. Combine the first component of the first power line point cloud and the second component of the first power line point cloud to obtain the first power line point cloud.

[0024] Further, the specific process of screening out the second power line point cloud corresponding to the points regarded as power line points based on the spatial characteristics of all adjacent points within the preset adjacent area threshold includes:

[0025] S3.1. Construct a k-dimensional tree kd-tree with the first power line point cloud, and obtain all points within the adjacent range of each point from the kd-tree to get its adjacent area.

[0026] S3.2. Obtain and calculate the covariance matrix within the adjacent area of each point and the three eigenvalues and three eigenvectors corresponding to this matrix.

[0027] S3.3. Divide this area into linear, planar, and mesh-like according to the eigenvalue situation, and set a feature threshold h. Among them, if the difference between the three eigenvalues is less than h, it is regarded as mesh-like; if the difference between two eigenvalues is less than h and is significantly greater than the third eigenvalue, it is regarded as planar; if one eigenvalue is much greater than the other two eigenvalues, it is regarded as linear; if one eigenvalue is much greater than the other two eigenvalues, and the eigenvector corresponding to this eigenvalue is not parallel to the z-axis, it is regarded as the second power line point cloud.

[0028] Further, the specific steps of screening out the target point cloud regarded as power line points based on morphological processing:

[0029] S5.1. Obtain the point cloud segmentation area where the third power line point cloud is located, with the segmentation area as the smallest unit.

[0030] S5.2. First, perform closing operation, that is, dilate first and then erode, to fill the cracks.

[0031] S5.3. Then, perform opening operation, that is, erode first and then dilate, to remove isolated small dots, burrs, and small bridges.

[0032] S5.4. Finally, obtain the point cloud segmentation area where power lines exist, and all the third power line point clouds within this area are the target point clouds.

[0033] The specific steps of the dilation operation are as follows: Traverse each point cloud segmentation area. If there are third power line point clouds in this point cloud segmentation area, then its adjacent segmentation areas are also regarded as having third power line point clouds.

[0034] The specific steps of the erosion operation are as follows: Traverse each point cloud segmentation area. If all its adjacent segmentation areas have third power line point clouds, then this segmentation area is regarded as having third power line point clouds; otherwise, this segmentation area is regarded as not having third power line point clouds.

[0035] The second aspect of the present invention discloses a power line recognition and tracking system based on unmanned aerial vehicle lidar detection data. The system includes a data collection module for grid environment point clouds, a region segmentation module, a ground object separation module, a feature processing module, a reflectivity processing module, and a morphological processing module.

[0036] The said data collection module is used to collect grid environment point cloud data by the lidar carried by the unmanned aerial vehicle, and send the collected grid environment point cloud data to the segmentation module.

[0037] The said region segmentation module is used to segment the original point cloud data collected by the data collection module into several point cloud segmentation areas, and count the density, average elevation, and highest elevation information in each area.

[0038] The said ground object separation module processes the point cloud segmentation areas through the density and elevation information of the point cloud segmentation areas, separates the ground, buildings, and some low vegetation points, and retains the points in the air, that is, the first power line point clouds.

[0039] The said feature processing module, based on the spatial features of each point in the first power line point cloud to all adjacent points within the preset adjacent area threshold, removes the vegetation points and tower pole points in the air, and retains most of the power line points, that is, the second power line point clouds.

[0040] The said reflectivity processing module, based on each point in the second power line point cloud, according to the preset reflectivity threshold, eliminates the points with larger reflectivity, and screens out the third power line point clouds.

[0041] The described morphological processing module is used to remove the areas of isolated small points, burrs, and small bridges in the third power line point cloud, obtaining a target point cloud with less noise.

[0042] Advantages of the present invention: The present invention utilizes the power grid point cloud data collected in real time by an unmanned aerial vehicle carrying an airborne lidar. Taking the power line as the target and vegetation, the ground, tower poles, etc. as ground object points, and using the characteristics of the power line in space and reflectivity, it can effectively segment the original point cloud and identify the power line points. The invention can automatically and real-time identify the power line, and based on the real-time identified power line, it can automatically plan the flight path of the unmanned aerial vehicle, effectively reducing the workload of line inspection workers and improving the inspection efficiency of the power line. Brief Description of the Drawings

[0043] Figure 1 is the original point cloud image;

[0044] Figure 2 is the first power line point cloud image;

[0045] Figure 3 is the second power line point cloud image;

[0046] Figure 4 is the third power line point cloud image;

[0047] Figure 5 is the target point cloud image;

[0048] Figure 6 is the schematic diagram of the power line recognition system. Detailed Implementation Manner

[0049] This example discloses a classification and recognition method based on spatial morphological feature extraction. The method includes: As an optional implementation manner, as Figure 1 shown, in this example, the original data is segmented to obtain several point cloud segmentation regions, which specifically include:

[0050] Based on the line distance between power lines in the region, determine the segmentation distance d.

[0051] Perform equidistant segmentation of the power line laser point cloud in the X-axis direction to obtain the first three-dimensional space grid corresponding to the power line laser point cloud.

[0052] Perform equidistant segmentation of each first three-dimensional space grid in the Y-axis direction to obtain the second three-dimensional space grid corresponding to the power line laser point cloud.

[0053] Project each second three-dimensional space network onto the XY plane to obtain several point cloud segmentation regions.

[0054] As an optional implementation manner, as Figure 2As shown, in this example, based on the overall density information and elevation information of each point cloud segmentation area, the first power line point cloud is selected, which specifically includes:

[0055] For any point cloud segmentation region R 00 , if there is no neighboring region satisfying |R ij -T 00 | ≤ ε, then this point cloud segmentation region is regarded as a region without power lines, where -1 ≤ i, j ≤ 1, otherwise:

[0056] Obtain the number of all points in this region as the point cloud density ρ. If ρ is less than the set density threshold ρ set , then this region is regarded as the first component of the first power line point cloud, otherwise:

[0057] Calculate the highest elevation h max , the lowest elevation h min , and the average elevation h ave . If h max -h min ≥ h set , then use the average elevation h ave as the best height threshold.

[0058] For each point in this region, if its height satisfies h ≥ h ave , then this point is regarded as the second component of the first power line point cloud. Combine the first component of the first power line point cloud and the second component of the first power line point cloud to obtain the first power line point cloud.

[0059] As an alternative implementation, as Figure 3 shown, in this example, the spatial features of all neighboring points within the preset neighboring region threshold are preset, and the second power line point cloud corresponding to the power line points regarded as being selected specifically includes:

[0060] Use the kd-tree algorithm with the second power line point as the search point to obtain the information of all neighboring points within the preset radius threshold of this search point.

[0061] Obtain and calculate the covariance matrix C = (c ij ) 3×3 in the adjacent region of each point, where 1 ≤ i, j ≤ 3, respectively representing the X direction, Y direction, and Z direction, and c ij is the calculated covariance value. Calculate the corresponding eigenvalues λ1, λ2, λ3 of this matrix and the corresponding eigenvectors {a i1 , a i2 , a i3}.

[0062] Calculate If there exists an eigenvalue λ of this search point i1≈1, then it is considered that this point forms a line, and continue the following judgment. If the corresponding eigenvector {a i1 , a i2 , a i3} × {0, 0, 1} ≈ {0, 0, 0}, then it is considered that this point may be the point of the telegraph pole and is perpendicular to the ground. Otherwise, it is regarded as the point cloud point of the second power line.

[0063] As an optional implementation manner, as Figure 4 shown, in this example, based on the preset reflectivity threshold, the specific steps for screening out the point cloud regarded as the third power line include:

[0064] Statistically analyze the reflectivity of the original point cloud data, and use the optimal threshold method to determine the reflectivity threshold. Assume that the reflectivity histogram can be assumed to be two distributions p0(r), p1(r). For the power line ω0 and the non-power line ω1, there is h(r) = P0p0(r) + P1p1(r), where P0 and P1 are the prior probabilities of ω0 and ω1. The optimal reflectivity threshold selects T as the value of r to minimize the probability of misclassified points:

[0065]

[0066] Traverse all points of the second point cloud, and the points less than the threshold T are regarded as the points of the third power line point cloud.

[0067] As an optional implementation manner, as Figure 5 shown, in this example, based on the morphological processing, the specific steps for screening out the target point cloud regarded as the power line are as follows:

[0068] Obtain the point cloud segmentation area where the third power line point cloud is located, set the values of these segmentation areas to 1, and set the values of the remaining segmentation areas to 0.

[0069] First, perform a closing operation, that is, dilate first and then erode, to make up for cracks.

[0070] Then perform an opening operation, that is, erode first and then dilate, to remove isolated small points, burrs, and small bridges.

[0071] Finally, obtain the point cloud segmentation area where the power line exists, and obtain all the third power line point clouds in this area, which is the fourth power line point cloud, that is, the target point cloud.

[0072] Among them, dilation expands the target area of the segmentation area to make it closer to the surrounding segmentation areas. The dilation operation can be expressed by the following formula:

[0073]

[0074] Among them, (x, y) represents the coordinates of the segmentation area, (i, j) represents the relative coordinates of the dilation kernel, and S represents the element shape of the dilation kernel. f(x, y) represents the value of the input segmentation area, and Dilated(x, y) represents the value after the dilation operation.

[0075] Among them, erosion shrinks the target area of the segmentation area, making it closer to the core part of the target area. The erosion operation can be expressed by the following formula:

[0076]

[0077] Generally, the frequency for the drone to receive instructions is 5 - 50 hz, while the frequency of the radar scan data is 10 hz. Therefore, 100 ms is selected as a processing cycle. In this example, the running time and running effect of each stage are shown in Table 1:

[0078] Table 1

[0079] Step 1 Step 2 Step 3 Step 4 Step 5 Running Time (ms) 13.57 2.40 60.59 0.02 2.24

[0080] In the above table, 24,000 original point clouds are received every 100 ms. The total running time of the above five steps is 78.82 ms, which can be processed within one processing cycle for real-time power line extraction.

[0081] Among them, the total number of test samples is 24,000. After step five, the target point cloud is obtained. The evaluation indexes of this algorithm are shown in Table 2:

[0082] Table 2

[0083] Accuracy Precision Recall F1 Score 99.2％ 90.28 79.2％ 84.5％

[0084] The accuracy rate of this algorithm reaches 99.2%, which can better extract the power line point cloud, providing conditions for accurately tracking the power line. At the same time, the running time of this algorithm meets the real-time requirements and can process the original data in real time.

[0085] As Figure 6 shown, the second aspect of the present invention discloses a power line recognition and tracking system based on drone lidar detection data. The system includes a data collection module for power grid environment point clouds, a region segmentation module, a ground object separation module, a feature processing module, a reflectivity processing module, and a morphological processing module.

[0086] The said data collection module is used to collect the power grid environment point cloud data by the lidar carried on the drone and send the collected power grid environment point cloud data to the segmentation module.

[0087] The region segmentation module is used to segment the original point cloud data collected by the data collection module into a number of point cloud segmentation areas, and to count the density, average elevation and maximum elevation information in each area.

[0088] The ground feature separation module processes the point cloud segmentation area through the point cloud segmentation area density and elevation information, separates the ground, buildings, and some low vegetation points, and retains the points in the sky, namely the first power line point cloud.

[0089] The feature processing module removes the vegetation points and tower points in the air based on the spatial features of all adjacent points within a preset neighboring area threshold from each point in the first power line point cloud, and retains most of the power line points, namely the second power line point cloud.

[0090] The reflectivity processing module removes points with a larger reflectivity based on each point in the second power line point cloud according to a preset reflectivity threshold, and screens out a third power line point cloud.

[0091] The morphological processing module is used to remove isolated small dots, burrs and small bridge areas in the third power line point cloud to obtain a target point cloud with less noise.

Claims

1. A method for identifying power lines in point clouds based on reflectivity and spatial morphology, characterized in that It includes the following steps: S1. Obtain the original point cloud data of the lidar, segment the original point cloud data, and obtain several point cloud segmentation regions; S2. Based on the overall density information and elevation information of each point cloud segmentation region, filter out the first power line point cloud; The specific implementation is as follows: S2.

1. Set a distance threshold. If there is no point with the highest elevation in the adjacent area of the point cloud segmentation region whose distance from the highest elevation of this point cloud segmentation region is less than the distance threshold, then this segmentation region is regarded as an area without power lines; S2.

2. Set a density threshold. If there is a point with the highest elevation in the adjacent area of the point cloud segmentation region whose distance from the highest elevation of this region is less than the distance threshold, and the point cloud density in this region is less than the density threshold, then all points in this region are regarded as the first component of the first power line point cloud; S2.

3. Set an elevation difference value. If the point cloud density in this region is greater than the density threshold, and the difference between the highest elevation and the lowest elevation in this region exceeds the elevation difference value, then calculate the height threshold according to the elevation distribution in this region; S2.

4. The point cloud with an elevation higher than the height threshold in this region is regarded as the second component of the first power line point cloud; S2.

5. Combine the first component of the first power line point cloud and the second component of the first power line point cloud to obtain the first power line point cloud; S3. For each point in the first power line point cloud, based on the spatial characteristics of this point to all adjacent points within the preset adjacent area threshold, filter out the second power line point cloud; The specific implementation is as follows: S3.

1. Construct a k-dimensional tree kd-tree with the first power line point cloud, and obtain all points within the adjacent range of each point from the kd-tree to obtain its adjacent area; S3.

2. Obtain and calculate the covariance matrix within the adjacent area of each point and the three corresponding eigenvalues and three eigenvectors of this matrix; S3.

3. Divide this region into linear, planar, and mesh-shaped according to the eigenvalue situation, set a feature threshold h. Among them, if the difference between the three eigenvalues is less than h, it is regarded as mesh-shaped; if the difference between two eigenvalues is less than h and is significantly greater than the third eigenvalue, it is regarded as planar; if one eigenvalue is much greater than the other two eigenvalues, it is regarded as linear; If one eigenvalue is much greater than the other two eigenvalues, and the eigenvector corresponding to this eigenvalue is not parallel to the z-axis, it is regarded as the second power line point cloud; S4. For each point in the second power line point cloud, based on the preset reflectivity threshold, filter out the third power line point cloud; The specific implementation is as follows: Statistical reflectivity of the original point cloud data, and use the optimal threshold method to determine the reflectivity threshold; The reflectivity histogram is assumed to be two distributions p0(r), p1(r). For the power line ω0 and the non-power line ω1, there is h(r) = P0p0(r) + P1p1(r), where P0, P1 are the prior probabilities of ω0, ω1, and the optimal reflectivity threshold selects T as the value of r to minimize the probability of misclassified points: Traverse all points of the second point cloud, and points less than the threshold T are regarded as points of the third power line point cloud; S5. For each point in the third power line point cloud, perform morphological processing, filter out the target point cloud, and obtain the recognized power line; The specific implementation is as follows: S5.

1. Obtain the point cloud segmentation area where the third power line point cloud is located, with the segmentation area division as the smallest unit; S5.

2. Perform a closing operation, that is, first perform a dilation operation and then an erosion operation to repair cracks; S5.

3. Perform an opening operation, that is, first perform an erosion operation and then a dilation operation to remove isolated small points, burrs, and small bridges; S5.

4. Obtain the point cloud segmentation area with power lines, and obtain all the third power line point clouds within this area, which are the target point clouds.

2. The method for identifying power lines in point clouds based on reflectivity and spatial morphology according to claim 1, wherein The original point cloud data described in step S1 includes a point set and the coordinate and reflectivity information corresponding to each point in the point set; The point set includes ground points, vegetation points, utility pole points, noise points, and power line points.

3. The method for identifying power lines in point clouds based on reflectivity and spatial morphology according to claim 2, characterized in that The specific process of segmenting the original point cloud data described in step S1 to obtain several point cloud segmentation areas is as follows: S1.

1. Perform equidistant segmentation of the original point cloud data in the X-axis direction to obtain the first three-dimensional space grid corresponding to the original point cloud data; S1.

2. Perform equidistant segmentation of each first three-dimensional space grid in the Y-axis direction to obtain the second three-dimensional space grid corresponding to the original point cloud data; S1.

3. Project each second three-dimensional space network onto the XY plane to obtain several point cloud segmentation areas.

4. The method for identifying power lines in point clouds based on reflectivity and spatial morphology according to claim 3, wherein, The dilation operation is specifically: traverse each point cloud segmentation area. If there is a third power line point cloud in this point cloud segmentation area, its adjacent segmentation area is also regarded as having a third power line point cloud; The erosion operation is specifically: traverse each point cloud segmentation area. If there are third power line point clouds in all its adjacent segmentation areas, this segmentation area is regarded as having a third power line point cloud; otherwise, this segmentation area is regarded as not having a third power line point cloud.

5. A point cloud power line recognition system based on reflectivity and spatial morphology, which is used to implement the point cloud power line recognition method according to any one of claims 1 to 4, and is characterized in that It includes a data collection module, a region segmentation module, a ground object separation module, a feature processing module, a reflectivity processing module, and a morphological processing module for power grid environment point clouds; The data collection module is used to collect power grid environment point cloud data by the lidar carried by the unmanned aerial vehicle; The region segmentation module is used to segment the original point cloud data collected by the data collection module into several point cloud segmentation areas, and count the density, average elevation, and highest elevation information in each area; The ground object separation module processes the point cloud segmentation area through the density and elevation information of the point cloud segmentation area, separates the ground, buildings, and some low-lying vegetation points, and retains the first power line point cloud; The feature processing module, based on the spatial features of each point in the first power line point cloud to all adjacent points within the preset adjacent area threshold, removes the vegetation points and tower pole points in the air, and retains the second power line point cloud; The reflectivity processing module, based on each point in the second power line point cloud, according to the preset reflectivity threshold, eliminates the points with larger reflectivity and screens out the third power line point cloud; The morphological processing module is used to remove the isolated small points, burrs, and small bridges in the third power line point cloud to obtain a target point cloud with less noise.

Images