A point cloud-based high-voltage transmission line three-dimensional model construction method

Abstract

The application provides a high-voltage transmission line three-dimensional model construction method based on point cloud, which can automatically identify power line point cloud data and three-dimensionally reconstruct power line features. The method is performed according to the following steps: S1. intercepting power corridor point cloud; S2. feature calculation; S3. power line point cloud preliminary selection; S4. point cloud spatial clustering; S5. point cloud segmented straight line detection; after spatial clustering, each clustering node data is obtained, and the data of each clustering node is reconstructed; S6. power multi-segment line generation; the generated segmented straight line features are sequentially connected, and three-dimensional power multi-segment line is generated. The application can quickly identify linear features in a power line scene, greatly exclude data participating in power line identification, lock target identification and reconstruction in a relatively small range, and therefore has high calculation efficiency.

Description

Technical Field

[0001] This invention relates to an automatic detection and three-dimensional model construction method for high-voltage transmission lines, which is particularly applicable to power line point cloud data obtained by unmanned aerial vehicle (UAV) inspection. Background Technology

[0002] Power transmission lines provide electricity for industry, agriculture, and residential use. Comprehensive inspections of the circuits are necessary during both the completion and routine operation and management phases. Traditional manual inspection is currently the main method of power line inspection in my country; however, it has several problems:

[0003] (1) High workload, low efficiency, and harsh conditions;

[0004] (2) It is impossible to carry out large-scale operations over large areas, resulting in high labor costs;

[0005] (3) Cannot operate around the clock;

[0006] (4) Spatial analysis cannot be performed.

[0007] With the development of the times, drone technology has made tremendous progress, enabling automated field inspection of power lines using consumer-grade drones. Drones can carry a variety of sensors, among which airborne lidar, with its active, all-weather, high efficiency, and high precision characteristics, is an important method for current power line inspection. From the massive amounts of airborne lidar point cloud data acquired, automatic detection of power line features and 3D reconstruction are essential steps to improve operational efficiency.

[0008] This invention studies a method for reconstructing the three-dimensional features of power lines based on three-dimensional point cloud information acquired by UAVs, providing technical support for the detection of power line features in high-voltage transmission lines. Summary of the Invention

[0009] The purpose of this invention is to provide a method for constructing a three-dimensional model of a high-voltage transmission line based on point cloud, which enables automatic identification of power line point cloud data and three-dimensional reconstruction of power line features.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a method for constructing a three-dimensional model of a high-voltage transmission line based on point clouds, which is carried out according to the following steps:

[0011] S1. Capture the point cloud of the power corridor. The power corridor line data is two-dimensional data. The calculation of the point cloud data capture is performed on the o-xy plane.

[0012] S2. Feature calculation;

[0013] 2.1 Three-dimensional grid division: After grid division, the point cloud data is distributed and stored in a dynamic array within each grid cell according to its spatial range, so as to facilitate neighborhood search. Grid cells without points are called empty grid cells.

[0014] 2.2 Second-level linear feature calculation: First, calculate the linear features in the grid cells, and then calculate the linear features of the point cloud data inside the linear grid cells;

[0015] S3. Initial selection of power line point cloud: Based on the linear feature values ​​calculated in S2, these are assigned to each point as scale feature values. Based on the feature distribution, a histogram is statistically analyzed to initially select the power line point cloud dataset.

[0016] S4. Point cloud spatial clustering,

[0017] S5. Point cloud segmented line detection: After spatial clustering, data of each cluster node is obtained, and the data of each cluster node is reconstructed.

[0018] S6. Power polyline generation: Connect the generated segmented straight line features in sequence to generate a three-dimensional power polyline. (1) Determine the power corridor line to which each straight line segment belongs based on the spatial range, and calculate the vector length of each endpoint relative to the starting point of the corridor. (2) Sort each straight line segment according to the vector length. (3) Link adjacent straight lines. (4) Generate power polyline.

[0019] Preferably, the S4 point cloud spatial clustering method uses a spatial three-dimensional grid to cluster the spatial point set, as follows:

[0020] (1) Identify all grid points as points to be classified;

[0021] (2) Select a cell to be classified in the grid structure. i,j,k Cluster nodes are created from this point. m Change the identifier of this point to a classified point;

[0022] (3) Search the 26 neighborhoods of the grid point to be classified for the grid point p. n And add cluster nodes m And change its category identifier;

[0023] (4) For cluster m All members in the cluster perform step (3) to process the cluster. m Expand until cluster m If no more members are added, then the cluster set is Cluster∪cluster. m ;

[0024] (5) Perform steps (2) to (4) on all current points to be classified until the number of grid cells to be classified is 0.

[0025] Furthermore, the S5 point cloud line detection involves segmenting the clustered node data for line detection:

[0026] (1) Point cloud segmentation

[0027] Point cloud segmentation is divided into two parts: point cloud file segmentation and segmentation within a file.

[0028] Point cloud classification: After clustering in S4, the point cloud data of each node is used to determine the classification of the power corridor to which each cluster node belongs.

[0029] Segmentation within a range: A power line in a range is generally quite long. Within each range, the point cloud data of the cluster nodes belonging to that range are divided by distance according to the corresponding length.

[0030] (2) Linear Model Detection

[0031] For the segmented point cloud data, the random sampling consistency method is used for line feature detection. The specific parameters and model in the detection are as follows:

[0032] Equation of the straight line:

[0033] Line model construction: Any two three-dimensional points p1 and p2 in space determine the direction dir of the line and the point p0 it passes through;

[0034] Scoring of a linear model: The distance from a sample point p0 to the linear model satisfies the following formula and can be considered as an interior point. The number of interior points of the model is the model score.

[0035]

[0036] Number of iterations: The number of times T is used to construct the linear model by randomly selecting sample points, which satisfies the following formula.

[0037]

[0038] The method designed in this invention can quickly identify linear features in power line scenes, greatly excluding data involved in power line identification, thus limiting target identification and reconstruction to a relatively small range, and therefore has high computational efficiency. This method also exhibits good accuracy and robustness, representing an important trend in future power line feature extraction and reconstruction. Attached Figure Description

[0039] Figure 1 This is a flowchart of the technology of the present invention.

[0040] Figure 2Schematic diagram of point cloud capture for power corridor

[0041] Figure 3 This is a visualization of the original point cloud data.

[0042] Figure 4 Comparison images were captured from the point cloud of the power corridor.

[0043] Figure 5 Construct a graph for the 3D grid (hide empty grid cells).

[0044] Figure 6 The RGB mapping effect is shown for scale features.

[0045] Figure 7 This is a linear characteristic distribution graph.

[0046] Figure 8 The initial selection of point cloud results for power lines.

[0047] Figure 9 This is the point cloud after spatial clustering.

[0048] Figure 10 To extract consistent three-dimensional straight line features for segmented random sampling.

[0049] Figure 11 The three-dimensional electric field line features reconstructed by this invention. Detailed Implementation

[0050] 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.

[0051] A method for constructing a 3D model of a high-voltage transmission line based on point clouds is carried out according to the following steps:

[0052] S1. Capture the point cloud of the power corridor. The power corridor line data is two-dimensional data. The calculation of the point cloud data capture is performed on the o-xy plane.

[0053] The so-called power corridor point cloud extraction refers to extracting point cloud data based on the power corridor lines and widths to reduce computational load and improve computational efficiency. The power corridor lines required for this task can be provided by the power line management department, or roughly drawn by the operator through human-computer interaction, as long as the deviation is not too large.

[0054] In the power corridor route, a rectangle is constructed for every two adjacent points. The point cloud within each rectangle is calculated according to its spatial relationship, thus extracting the corridor's point cloud. Figure 2 As shown, the polyline in the center is the polyline of the corridor route, and the two rectangular boxes on the left and right are the rectangles formed by the endpoints of the adjacent polylines.

[0055] Constructing rectangles directly using the endpoints of polylines can easily create data holes in the polyline transition areas (e.g.) Figure 2 (Regions a, b, c, d, e, f). To avoid this situation, the width of the rectangles at the endpoints is increased. Specifically, the corridor width is subtracted from and added to the lower left and upper right corners of the rectangles, thus increasing the size of each rectangle segment. Although this increases the corridor point cloud data corresponding to each line segment, the final corridor power point cloud data can be determined by constraining the distance from the point to the corridor polyline (the distance from the polyline in the center of the figure to the straight lines on both sides is the power corridor width L).

[0056] Assume the set of rectangles formed by the extended rectangles of the corridor polyline is R = {r i For any point p in the point cloud, the extended rectangle r to which it belongs is determined by the two-dimensional spatial coordinates of the lower left and upper right corners of the extended rectangle, where |0<i<n}. i If the starting coordinate of the line segment corresponding to the extended rectangle is p s The endpoint coordinates are p e Then the direction vector Then the perpendicular distance d from point pt to the line v It can be expressed as equation (4).

[0057] d v =s×sina (4)

[0058] Where s is the distance from point pt to point ps, s = |pp s |;a is p and p s The angle between the direction of the connecting line and the direction of the straight line segment. Additionally, d h =s×cosa cannot be greater than the length of the line segment in this range |p e -p s |

[0059] The power corridor route data is two-dimensional data, and the calculations for the extracted point cloud data of the corridor are all performed on the o-xy plane.

[0060] S2. Feature calculation;

[0061] 2.1 Three-dimensional grid division: After grid division, the point cloud data is distributed and stored in a dynamic array within each grid cell according to its spatial range, so as to facilitate neighborhood search. Grid cells without points are called empty grid cells.

[0062] Three-dimensional grid partitioning serves two purposes: accelerating neighborhood search calculations and detecting linear grid cells, thus accelerating power line feature detection. Specifically, it involves determining the spatial extent of the point cloud data to be calculated, that is, finding the maximum x value among the x, y, and z values. max ,y max ,z max Minimum value x min ,y min ,z min The grid size (cellsize) is determined by 10 to 20 times the horizontal distance between points. The grid row and column information is shown in the following formula (5).

[0063]

[0064] 2.2 Calculation of second-order linear characteristics

[0065] The so-called two-level linear feature calculation involves calculating the linear network features and linear point cloud features separately. The linear feature calculation uses the following method:

[0066] Neighborhood search: Searching for points of interest (x) i Calculate the mean of all neighboring points X within a radius R:

[0067]

[0068] Calculate the sample covariance:

[0069]

[0070] Eigenvalue decomposition: Eigenvalue decomposition is performed on the covariance matrix, yielding eigenvalues ​​lmt1 > lmt2 > lmt3. For surface objects, lmt1 and lmt2 are relatively close; for point objects, lmt1, lmt2, and lmt3 are all relatively close; for line objects, lmt1 is much larger than both lmt2 and lmt3, therefore, eigenvalues ​​can be selected...

[0071]

[0072] Feature selection is performed using linear features.

[0073] The second-level linear feature calculation first calculates the linear features in the grid cells, and then calculates the linear features of the point cloud data inside the linear grid cells.

[0074] 1) Grid feature calculation

[0075] Non-empty points in the statistical grid cells are used to create a new point cloud using the row, column, and height information of the non-empty grid cells, and its linear characteristics are calculated according to formula (8) to obtain the linear grid cells.

[0076] 2) Point cloud feature calculation

[0077] Calculate the linear eigenvalues ​​of the point cloud within the linear grid cell using formula (8).

[0078] By performing secondary feature calculations, point cloud data with linear features can be retrieved quickly.

[0079] S3. Initial selection of power line point cloud: Based on the linear feature values ​​calculated in S2, these are assigned to each point as scale feature values. Based on the feature distribution, a histogram is statistically analyzed to initially select the power line point cloud dataset.

[0080] S4. Point cloud spatial clustering,

[0081] S5. Point cloud segmented line detection: After spatial clustering, data of each cluster node is obtained, and the data of each cluster node is reconstructed.

[0082] S6. Power polyline generation: Connect the generated segmented straight line features in sequence to generate a three-dimensional power polyline. (1) Determine the power corridor line to which each straight line segment belongs based on the spatial range, and calculate the vector length of each endpoint relative to the starting point of the corridor. (2) Sort each straight line segment according to the vector length. (3) Link adjacent straight lines. (4) Generate power polyline.

[0083] The S4 point cloud spatial clustering method uses a spatial three-dimensional grid to cluster spatial point sets, as detailed below:

[0084] (1) Identify all grid points as points to be classified;

[0085] (2) Select a cell to be classified in the grid structure. i,j,k Cluster nodes are created from this point. m Change the identifier of this point to a classified point;

[0086] (3) Search the 26 neighborhoods of the cell to be classified. n And add cluster nodes m And change its category identifier;

[0087] (4) For cluster m All members in the cluster perform step (3) to process the cluster. m Expand until cluster m No more members are added. Therefore, the cluster set is Cluster∪cluster. m ;

[0088] (5) Perform steps (2) to (4) on all current points to be classified until the number of grid cells to be classified is 0.

[0089] The S5 point cloud line detection involves segmenting and detecting lines in the clustered node data.

[0090] (1) Point cloud segmentation

[0091] Point cloud segmentation consists of two parts: point cloud file segmentation and segmentation within a file.

[0092] Point cloud classification: After clustering in S4, the point cloud data of each node is used to determine the classification of the power corridor to which each cluster node belongs.

[0093] Segmentation within a range: A power line in a range is generally quite long. Within each range, the point cloud data of the cluster nodes belonging to each range are divided by distance according to the corresponding length.

[0094] (2) Linear Model Detection

[0095] For the segmented point cloud data, the random sampling consistency method is used for line feature detection. The specific parameters and model in the detection are as follows:

[0096] Equation of the straight line:

[0097] Line model construction: Any two three-dimensional points p1 and p2 in space determine the direction of the line dir and the point p0 it passes through.

[0098] Scoring of a linear model: The distance from a sample point p0 to the linear model satisfies the following formula and can be regarded as an interior point. The number of interior points of the model is the score of the model.

[0099]

[0100] Number of iterations: The number of times T is used to construct a linear model by randomly selecting sample points, which satisfies the following formula.

[0101]

[0102] Considering the presence of noise, two constraints are imposed on the detected linear model:

[0103] 1) Constraint on the number of consistent sets. If the number of consistent sets is less than a certain threshold, subsequent steps will not be performed.

[0104] 2) Directional constraints. Calculate the horizontal angle between the detected straight line and the corresponding power corridor segment. If the angle is greater than a certain value, the line will not be included in subsequent calculations.

[0105] The application method of the present invention is described below with reference to specific embodiments.

[0106] (1) Data Acquisition

[0107] A DJI M300RTK drone, equipped with its L1 laser payload, was used to collect data from power lines in a certain location. After preprocessing, multiple LAS point cloud files were obtained, with a total file size of approximately 1.63 GB and a total of 62,901,544 points. The point cloud data visualization effect is as follows. Figure 3 As shown in the image, the black and white areas represent the point cloud after being colored black and white according to elevation.

[0108] (2) Corridor point cloud data capture

[0109] Utilizing the provided power corridor multi-segment lines (such as Figure 3 The polylines at both ends (as shown) and power point cloud data are used to perform corridor extraction on the power point cloud according to the method of this invention, with a certain width set. The extraction effect is as follows: Figure 4 As shown: the middle polyline is the polyline of the power corridor, and the two side polylines are the result of the corridor polylines being offset left and right according to the set width. The left image is the original scanned point cloud data, and the right image is the extracted corridor point cloud data.

[0110] (3) Three-dimensional grid construction

[0111] To facilitate point cloud feature calculation, a 3D grid was constructed from the extracted power corridor point cloud data, such as... Figure 5 As shown.

[0112] (2) Eigenvalue Calculation and Statistics

[0113] According to the method of the present invention, the linear eigenvalues ​​of each point are calculated, and these are used as scale values ​​to map to the RGB color space, resulting in the rendering effect as follows. Figure 6 As shown in the diagram, the power lines are particularly prominent. Figure 6 The darker areas in the middle have a noticeable color difference from the ground features and the surrounding area.

[0114] (3) Initial selection of power line point cloud

[0115] To better identify the characteristics of power lines, we statistically analyzed the calculated linear characteristic values ​​and plotted a histogram of the linear characteristic distribution, as shown below. Figure 7 As shown in the histogram, the horizontal axis represents the interval distribution range of linear eigenvalues, and the vertical axis represents the number of points appearing in each interval.

[0116] The histogram shows a significant difference in scale characteristics between power line data and other point cloud information. Based on this histogram, a threshold can be selected to filter the point cloud data. A threshold range of 0–0.08 can be chosen to obtain initially selected power line point cloud data (e.g., ...). Figure 8 (As shown).

[0117] (4) Point cloud spatial clustering results

[0118] from Figure 8 As can be seen, the selected threshold range effectively extracts the power line point cloud data from the original dataset. However, a small number of discrete, random noise points still exist. To better extract the power line data, spatial clustering is performed on the initial power line point cloud data. The clustering effect is as follows: Figure 9 As shown in the figure (different shades in the figure represent different cluster sets).

[0119] (5) Results of power line segmentation inspection

[0120] Based on each cluster node, the point cloud is segmented according to distance, and then the random sampling consistency method is used for 3D straight line feature segmentation detection. The detection effect is as follows: Figure 10 As shown.

[0121] (6) Results of three-dimensional reconstruction of power lines

[0122] Finally, the generated segmented straight line features are sequentially connected to generate a three-dimensional electric field line polyline, such as... Figure 11 As shown.

[0123] This invention realizes a method for automatic identification and 3D reconstruction of power line features based on point clouds, which has the following advantages:

[0124] (1) Rapid Power Line Feature Identification Based on Linear Features. For rapid detection, the following aspects are employed: 1) Corridor point cloud extraction. Corridor point cloud data is rapidly extracted based on power corridor lines, significantly reducing the amount of data involved in detection; 2) Construction of a 3D grid index; 3) Linear feature calculation. Within a local neighborhood, linear feature values ​​of grid cells and point cloud data are calculated, and power line point cloud data is rapidly identified based on these feature values. Using linear features to identify power line point clouds eliminates the need for filtering and classification operations, greatly improving computational efficiency and making it well-suited for engineering applications.

[0125] (2) Automatic Reconstruction Method for Power Line Polysegmentation. This method is used because power lines are parabolic in space due to factors such as distance and gravity. Directly fitting the parabola has disadvantages such as large computational load, instability, and susceptibility to noise interference. This method extracts straight line segment features from the power line point cloud and then connects them sequentially into polysegments. It has better operability and robustness in reality and can be well applied in practical engineering.

[0126] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Although this specification describes embodiments, not every embodiment contains only one technical solution. This method of description is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for constructing a three-dimensional model of a high-voltage transmission line based on point clouds, characterized in that: The following steps were performed: S1. Capture the point cloud of the power corridor. The power corridor line data is two-dimensional data. The calculation of the point cloud data capture is performed on the o-xy plane. S2. Feature calculation; 2.1 Three-dimensional grid division: After grid division, the point cloud data is distributed and stored in a dynamic array within each grid cell according to its spatial range, so as to facilitate neighborhood search. Grid cells without points are called empty grid cells. 2.2 Second-level linear feature calculation: First, the linear features in the grid cells are calculated, and then the linear features of the point cloud data within the linear grid cells are calculated. The linear feature calculation adopts the following method: Neighborhood search: Searching for points of interest (x) i Calculate the mean of all neighboring points X within a radius R: （6） Calculate the sample covariance: （7） Eigenvalue decomposition: Eigenvalues ​​are obtained by performing eigenvalue decomposition on the covariance matrix, yielding eigenvalues ​​lmt1 > lmt2 > lmt3. For surface objects, lmt1 and lmt2 are relatively close; for point objects, lmt1, lmt2, and lmt3 are all relatively close; for line objects, lmt1 is much larger than both lmt2 and lmt3, therefore, the eigenvalues ​​can be selected... （8） As a linear feature, characteristic selection is performed; S3. Initial selection of power line point cloud: Based on the linear feature values ​​calculated in S2, assign them to each point as scale feature values, and statistically analyze the histogram based on the feature distribution to initially select the power line point cloud dataset. S4. Point cloud spatial clustering: The spatial point set is clustered using a 3D spatial grid method, as detailed below: (1) Identify all grid points as points to be classified; (2) Select a cell to be classified in the grid structure. i,j,k Cluster nodes are created from this point. Change the identifier of this point to a classified point; (3) Search the 26 neighborhoods of the grid point to be classified. and add cluster nodes. And change its category identifier; (4) All members execute step (3) to Expand until If no more members are added, then the cluster set... ; (5) Perform steps (2) to (4) on all current points to be classified until the number of grid cells to be classified is 0; S5. Point cloud segmented line detection: After spatial clustering, data for each cluster node is obtained, and the data for each cluster node is reconstructed; point cloud line detection involves segmenting and detecting lines in the data of the cluster nodes. (1) Point cloud segmentation: Point cloud segmentation consists of two parts: point cloud file segmentation and intra-file segmentation. Point cloud classification: After clustering in S4, the point cloud data of each node is used to determine the classification of the power corridor to which each cluster node belongs. Segmentation within a segment: A power line in a segment is generally quite long. Within each segment, the point cloud data of the cluster nodes belonging to each segment are divided by distance according to the corresponding length. (2) Linear model detection For the segmented point cloud data, the random sampling consistency method is used for line feature detection. The specific parameters and model in the detection are as follows: Equation of the straight line: (1) Linear model construction: any two 3D points in space and Determine the direction of the line and the points passed through ; Scoring of linear models: Sample points The distance to the straight-line model satisfies the following formula and can be considered as an interior point. The number of interior points of the model is the model's score. （2） Number of iterations: The number of times sample points are randomly selected to construct the linear model. Satisfy the following formula （3）； S6. Power polyline generation: Connect the characteristics of each segmented straight line in sequence to generate a three-dimensional power line polyline. (1) Determine the power corridor line to which each straight line segment belongs based on the spatial range, and calculate the vector length of each endpoint relative to the starting point of the corridor. (2) Sort each straight line segment according to the vector length. (3) Link adjacent straight lines. (4) Generate power polyline.

Images