A method for automatic classification of laser point clouds for power transmission lines

Abstract

The present invention relates to a method for automatic classification of laser point clouds of power transmission lines, comprising the following steps: rotating the laser point cloud between gears so that the power lines of two power transmission towers turn to the same direction; splitting the rotated laser point cloud into voxels; identifying the initial category voxels once according to the height difference; using the height difference to perform a second identification of the power line category voxels; using a clustering method to perform a third identification of the power line category voxels after the second identification; identifying the laser points of the tower category; identifying the laser points of the ground category. The classification method improves the automation of the laser point cloud processing algorithm, effectively improves the efficiency of point cloud analysis and processing, ensures timely discovery of defects and hidden dangers in the safe operation of the transmission line, and provides important technical support for the safe operation of the transmission line.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power inspection, and in particular to automatic classification of laser point clouds for power transmission lines. Background Art

[0002] With the rapid development of the national economy, the demand for high voltage, high power and long-distance power transmission has increased. The geographical environment that the line corridors pass through is becoming more and more complex, and its operation and maintenance are becoming increasingly difficult. The transmission line covers a wide area, and the terrain along the line is complex and diverse. When the line passes through a mountainous area covered with vegetation, it is impossible to measure the densely vegetated area, or the vegetation within the field of view is directly cut down before measurement, resulting in a lot of unnecessary vegetation cutting and damage to the surrounding environment. Airborne LiDAR uses drones, helicopters and other platforms. The high-frequency laser pulses it emits have a certain degree of penetration and can penetrate the dense vegetation canopy to obtain terrain information under the forest. It is very suitable for three-dimensional information acquisition in narrow and long areas, vegetation-covered areas, and complex terrain areas. This provides strong technical support for digital power grid construction and line safety inspections. Its applications can cover the detection of dangerous objects in line corridors, the fine measurement of the distance between power lines, the three-dimensional visualization management of transmission lines, the capacity increase analysis of transmission lines, and the assessment and management of tree felling. The laser point cloud data acquired by airborne radar is massive, dense, information-rich, and large in volume. However, the current laser point cloud processing algorithm has a low degree of automation and cannot automatically classify laser point cloud categories, thus failing to meet the needs of power operation and maintenance. Summary of the invention

[0003] The purpose of the present invention is to overcome the defects of the prior art, provide a method for automatic classification of laser point clouds of transmission lines, improve the degree of automation of laser point cloud processing algorithms, effectively improve the efficiency of point cloud analysis and processing, ensure timely discovery of defects and hidden dangers in the safe operation of transmission lines, and provide important technical support for the safe operation of transmission lines.

[0004] The technical solution of the present invention is: a method for automatic classification of laser point clouds of power transmission lines, comprising the following steps:

[0005] Step S100: rotating the laser point cloud between gears so that the power lines of the two transmission towers are turned to the same direction;

[0006] Step S200: splitting the rotated laser point cloud into voxels;

[0007] Step S300: identifying the initial category voxels based on the height difference;

[0008] Step S400: performing secondary identification on voxels of the power line category using height differences;

[0009] Step S500: using a clustering method to perform a third recognition on the power line category voxels after the second recognition;

[0010] Step S600: Identify the laser point of the tower type;

[0011] Step S700: Identify laser points of ground category.

[0012] Furthermore, in steps S300 , S400 and S500 , recognition is performed in units of voxels, and in steps S600 and S700 , recognition is performed in units of laser points.

[0013] Furthermore, the step S100 further includes the following steps:

[0014] Step S110: obtaining the center position coordinates of two transmission towers between the sections;

[0015] Step S120: Calculate the slope of the line connecting the two transmission towers according to the position coordinates;

[0016] Step S130: Calculate the rotation angle corresponding to the slope;

[0017] Step S140: rotating the laser point cloud between gears according to the rotation angle;

[0018] Step S150: Calculate the minimum coordinates of the laser point cloud in each direction after rotation;

[0019] Step S160: Setting the category of the laser point.

[0020] Furthermore, the step S200 further includes the following steps:

[0021] Step S210: placing the rotated laser point cloud coordinates into corresponding voxels;

[0022] Step S220: sorting all voxels according to altitude;

[0023] Step S230: creating an association relationship between voxel labels and voxel height differences.

[0024] Furthermore, the step S300 further includes the following steps:

[0025] Step S310: add up all the height differences in the association relationship and divide by the number of voxels to obtain an average height difference Zavg;

[0026] Step S320: Divide all voxels into two parts according to the average height difference Zavg, namely, voxels with a height difference greater than or equal to the average height difference and voxels with a height difference less than the average height difference;

[0027] Step S330: respectively calculating the average height differences ZavgTop and ZavgBottom for the two parts of voxels in step S320;

[0028] Step S340: Optimizing the average height difference Zavg;

[0029] Step S350: Identify the voxel category according to the average height difference Zavg.

[0030] Furthermore, the step S400 further includes the following steps:

[0031] Step S410: Calculate the average height difference ZavgCube of the power line category voxels;

[0032] Step S420: Divide the power line category voxels into two parts according to the average height difference ZavgCube, namely, voxels with a height difference greater than or equal to the average height difference and voxels with a height difference less than the average height difference;

[0033] Step S430: respectively calculate the average height differences ZavgCubeTop and ZavgCubeBottom for the two parts of voxels in step S420;

[0034] Step S440: Optimizing the average height difference ZavgCube;

[0035] Step S450: For each power line type voxel, the voxel type with a height difference greater than or equal to ZavgCube remains unchanged, and the voxel type with a height difference less than ZavgCube becomes a vegetation type.

[0036] Furthermore, the step S500 further includes: performing DBScan clustering on all power line type voxels, in the clustering results, the voxel category with the largest number remains unchanged, and the voxels corresponding to other clustering results are set as vegetation categories.

[0037] Furthermore, the step S600 further includes the following steps:

[0038] Step S610: Determine the width of the line between stalls;

[0039] Step S620: Identify the laser point of the pole tower type according to the line width between the sections.

[0040] Furthermore, the step S700 further includes the following steps:

[0041] Step S710: Calculate the height difference of voxels of all vegetation categories, and the minimum value is Heightmin;

[0042] Step S720: traverse all voxels, take the lowest point of each voxel as the reference, find the laser point whose height difference with the lowest point is less than or equal to Heightmin, set the category of the laser point to the ground category, otherwise keep the original category.

[0043] The present invention has the following beneficial effects: the present invention is a fast classification algorithm proposed according to the characteristics of laser point cloud data between transmission line sections. Compared with other algorithms, the algorithm has strong pertinence, good applicability and fast classification speed; the present invention proposes to use height difference to measure the continuity of laser point cloud data in spatial distribution, which is more accurate than directly using height, density and other data for classification; the present invention proposes an overall idea and algorithm for automatically classifying point cloud data between transmission line sections into poles, power lines, vegetation and ground categories, and the classification results are accurate, which improves the automation degree of the laser point cloud automatic classification algorithm and effectively improves the efficiency of laser point cloud analysis and processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a simplified flowchart of the method for automatic classification of laser point clouds of power transmission lines according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0047] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0048] A method for automatically classifying laser point clouds of power transmission lines is proposed. Before processing the laser point clouds, the laser point cloud data is first cut into files, and every two transmission towers and the channel between them are regarded as a file. The laser point cloud itself is discrete and has no attribute categories. According to the actual surface conditions of the transmission line, the categories of laser points are classified into initial category, power line category, vegetation category, tower category, and ground category.

[0049] The laser point cloud coordinates are three-dimensional coordinates, defining the east-west direction as the X-axis direction, the north-south direction as the Y-axis direction, and the altitude as the Z-axis direction.

[0050] The specific steps include:

[0051] Step S100: Rotate the laser point cloud between gears to make the power lines of two transmission towers turn to the same direction.

[0052] The direction can be the X-axis direction or the Y-axis direction.

[0053] Step S110: Obtain the center position coordinates of two transmission towers between the sections.

[0054] According to the position Xstart, Ystart of the starting tower and the position Xend, Yend of the ending tower, calculate the center point coordinates Xmiddle, Ymiddle of the two towers:

[0055] X_middle = (Xstart + Xend) / 2;

[0056] Y_middle=(Ystart+Yend) / 2;

[0057] Step S120: Calculate the slope of the line connecting the two transmission towers according to the position coordinates.

[0058] According to the position Xstart, Ystart of the starting tower and the position Xend, Yend of the ending tower, calculate the slope K formed by the two towers:

[0059] K=(Yend-Ystart) / (Xend-Xstart);

[0060] Step S130: Calculate the rotation angle corresponding to the slope.

[0061] According to the slope K calculated in the previous step, the angle α corresponding to the slope K is calculated according to the inverse tangent formula:

[0062] α=tan -1 K;

[0063] Step S140: Rotate the laser point cloud between the gears according to the rotation angle.

[0064] Rotate the laser point cloud in the current range according to the angle α, with the rotation center at Xmiddle and Ymiddle. If the rotation is set to be parallel to the X axis, the slope α≥0 means clockwise rotation, and the slope α<0 means counterclockwise rotation.

[0065] Step S150: Calculate the minimum coordinates of the rotated laser point cloud in each direction.

[0066] Calculate the coordinate values ​​X and Y of each point in the rotated laser point cloud data. The calculation formula is as follows:

[0067] X=(X-X_middle)×cosα-(Y-Y_middle)×sinα+X_middle;

[0068] Y=(X-X_middle)×sinα-(Y-Y_middle)×sinα+Y_middle;

[0069] According to the above formula, determine the minimum coordinate values ​​Xmin, Ymin, and Zmin in three directions of all laser points. Take Xmin as an example: first set Xmin to a very large value. If the rotated laser point cloud coordinate X≤Xmin, then refresh the Xmin value with X at this time, otherwise it will not be refreshed; then compare the X of the next laser point, if it is less than, refresh it, otherwise it will not be refreshed, and repeat this cycle until all laser point cloud data in the current range are processed. The same applies to Ymin and Zmin.

[0070] Step S160: Setting the category of the laser point.

[0071] At the beginning, the categories of all laser points are set to the initial category.

[0072] The laser point cloud between the gears is a long and narrow area. If it is directly divided into voxels, many empty areas will be generated, which greatly increases the amount of calculation and storage. Therefore, the laser point cloud between the gears is first rotated and then voxelized.

[0073] Step S200: split the rotated laser point cloud into voxels.

[0074] Step S210: placing the rotated laser point cloud coordinates into corresponding voxels.

[0075] Set the voxel width threshold CUBE, divide the grid in the XY direction by the width threshold CUBE, generate voxels, and put each laser point in the laser point cloud of each level into the voxel at the corresponding position according to the coordinates of the laser point cloud. The row and column of the voxel can be calculated according to the minimum value of the XY coordinates and the voxel width threshold. XIndex and YIndex are the voxel labels, and the voxel label indicates which row and column this is:

[0076] Xindex = (X-Xmin) / CUBE;

[0077] Yindex = (Y-Ymin) / CUBE;

[0078] Step S220: sort all voxels according to their altitudes.

[0079] The Z values ​​of all laser points in each voxel are sorted from high to low, and the maximum difference between two adjacent points is taken as the height difference of the voxel.

[0080] Step S230: creating an association relationship between voxel labels and voxel height differences.

[0081] Affected by the terrain and objects, the height of the lowest point of each voxel is not the same. If the lowest point of a voxel is used as the reference point, the lowest points of other voxels are translated to the height of the reference point, and the remaining points in the voxel are also offset according to the offset distance and offset direction, resulting in all voxels needing to traverse all the laser point clouds contained, which is too inefficient. Therefore, the height value of each voxel needs to be retained.

[0082] The subsequent steps of point cloud calculation are performed in voxels, creating an association between the height difference of the voxels and the point clouds in the voxels, and creating a "key-value pair" type association relationship Height-CubeIndex. The key is the height difference between the highest point height and the lowest point height of the current voxel; the value is the voxel subscript XIndex and YIndex of the current voxel, that is, the height difference of each voxel is retained through the "key-value pair".

[0083] Step S300: Identify the initial category voxels based on the height difference.

[0084] In order to avoid hidden dangers of high objects to transmission lines, high vegetation is not allowed under and near the power lines of the transmission lines. Therefore, there are fewer high objects and basically no high objects within a certain range of the lines. Therefore, near the power lines, the voxel height difference is large; in areas far away from the power lines, the voxel height difference is small.

[0085] Step S310: add up all the height differences in the association relationship Height_CubeIndex and divide the sum by the number of voxels to obtain an average height difference Zavg.

[0086] Step S320: Divide all voxels into two parts according to the average height difference Zavg, ie, voxels with a height difference greater than or equal to the average height difference and voxels with a height difference less than the average height difference.

[0087] Step S330: respectively calculating the average height differences ZavgTop and ZavgBottom for the above two parts of voxels.

[0088] Step S340: Optimize the average height difference Zavg.

[0089] Take the median ZavgMiddle of ZavgTop and ZavgBottom, and compare ZavgMiddle with Zavg. If the difference between the two is large, take the value of the current median ZavgMiddle as the value of Zavg, and repeat steps S320-S340; if the difference between Zavg and ZavgMiddle is very small, exit the loop, and obtain a stable average height difference Zavg after multiple rounds of optimization.

[0090] Step S350: Identify the category of the voxel according to the average height difference Zavg.

[0091] The voxels with height difference less than Zavg are set as vegetation category, otherwise they are set as power line category. The voxel category here means that all laser point clouds in each voxel are of the same category, all laser point clouds in the voxels of vegetation category are of vegetation category, and all laser point clouds in the voxels of power line category are of power line category.

[0092] Step S400: performing secondary identification on voxels of the power line category using the height difference.

[0093] Considering that the laser point cloud of the tower has strong vertical continuity, but the power lines often maintain a certain distance from the ground objects below (the ground and ground objects), the voxels containing vegetation and the voxels containing power lines can be separated according to the height difference.

[0094] Step S410: Calculate the average height difference ZavgCube of the power line category voxels.

[0095] Step S420: Divide the power line category voxels into two parts according to the average height difference ZavgCube, ie, voxels with a height difference greater than or equal to the average height difference and voxels with a height difference less than the average height difference.

[0096] Step S430: respectively calculating the average height differences ZavgCubeTop and ZavgCubeBottom for the above two parts of voxels.

[0097] Step S440: Optimize the average height difference ZavgCube.

[0098] Take the median value ZavgCubeMiddle of ZavgCubeTop and ZavgCubeBottom and compare it with ZavgCube. If the difference is large, take the value of the current median ZavgCubeMiddle as the value of ZavgCube and repeat steps S420 to S440; if the difference between ZavgCube and ZavgCubeMiddle is very small, exit the loop.

[0099] After the above processing, the pole tower will basically be intercepted in the middle of the tower body, and the power lines and ground objects will basically be intercepted between the power lines and the ground objects.

[0100] Step S450: For each power line type voxel, the voxel type with a height difference greater than or equal to ZavgCube remains unchanged, and the voxel type with a height difference less than ZavgCube becomes a vegetation type.

[0101] Step S500: using a clustering method to perform a third recognition on the power line category voxels after the second recognition.

[0102] After the above steps, the vegetation and power lines in the voxels have been roughly separated, but due to the interference of tall vegetation, screening is required. In the case of a bird's-eye view, considering that the lines account for the largest proportion of the laser point cloud between the stalls, and the tall objects in the stalls are also extracted at this time, it is necessary to exclude the interference of tall vegetation far away from the towers and power lines.

[0103] DBScan clustering is performed on all power line type voxels. In the clustering results, the voxel category with the largest number remains unchanged, and the voxels corresponding to the other clustering results are all interference and are set as vegetation categories.

[0104] Step S600: Identify the laser point of the tower type.

[0105] Step S610: Determine the width of the line between stalls.

[0106] Because the laser point cloud has been rotated to a horizontal state, select the column number XstartIndex of the voxel where the starting tower is located and the column number XendIndex of the voxel where the ending tower is located. Select all voxels containing power lines with column coordinates of XstartIndex+1, find the minimum and maximum Y values ​​among the power line points, and set them as YlineMin and YlineMax. Then select all voxels containing power lines with column coordinates of XstartIndex+2. If the Y value of a power line point in a voxel is greater than YlineMax, refresh YlineMax; if the Y value of a power line point in a voxel is less than YlineMin, refresh YlineMin. Until XstartIndex and XendIndex are equal. The calculation formula for line width lineWidth is as follows:

[0107] lineWidth=(YlineMin+YlineMax) / 2;

[0108] Step S620: Identify the laser point of the pole tower type according to the line width between the sections.

[0109] Since the tower coordinates are known, a circle is drawn with the starting tower and the ending tower coordinates as the center and lineWidth as the radius. All voxels are traversed. If a voxel intersects with any circle, the laser point cloud in the voxel is traversed. If the horizontal distance from the laser point to the tower coordinates is less than or equal to lineWidth, the category of the laser point is updated to the tower category, otherwise the original category is retained.

[0110] Step S700: Identify laser points of ground category.

[0111] Step S710: Calculate the height difference of voxels of all vegetation categories, with the minimum value being Heightmin.

[0112] Step S720: traverse all voxels, take the lowest point of each voxel as the reference, find the laser point whose height difference with the lowest point is less than or equal to Heightmin, set the category of the laser point to the ground category, otherwise keep the original category.

[0113] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A method for automatic classification of laser point clouds of power transmission lines, characterized in that: The following steps are involved: Step S100: rotating the laser point cloud between gears so that the power lines of the two transmission towers are turned to the same direction; Step S200: splitting the rotated laser point cloud into voxels; Step S300: sorting the Z values ​​of all laser points in each voxel from high to low, taking the maximum value of the difference between two adjacent points as the height difference of the voxel, and identifying the initial category voxels once according to the height difference; Step S400: performing secondary identification on voxels of the power line category using height differences; Step S500: using a clustering method to perform a third recognition on the power line category voxels after the second recognition; Step S600: Identify the laser point of the pole tower category: determine the line width between the gears and identify the laser point of the pole tower category according to the line width between the gears; Step S700: Identify the laser point of the ground category: calculate the height difference of all voxels of the vegetation category, with the minimum value being Heightmin; traverse all voxels, taking the lowest point of each voxel as the reference, find the laser point whose height difference with the lowest point is less than or equal to Heightmin, and set the category of the laser point to the ground category, otherwise keep the original category.

2. The method for automatic classification of laser point clouds of power transmission lines according to claim 1, characterized in that: In steps S300 , S400 and S500 , recognition is performed in units of voxels, and in steps S600 and S700 , recognition is performed in units of laser points.

3. The method for automatic classification of laser point clouds of power transmission lines according to claim 1, characterized in that: The step S100 further includes the following steps: Step S110: Obtain the center position coordinates of two transmission towers between the sections; Step S120: Calculate the slope of the line connecting the two transmission towers according to the position coordinates; Step S130: Calculate the rotation angle corresponding to the slope; Step S140: rotating the laser point cloud between gears according to the rotation angle; Step S150: Calculate the minimum coordinates of the rotated laser point cloud in each direction; Step S160: Set the category of the laser point.

4. The method for automatic classification of laser point clouds of power transmission lines according to claim 1, characterized in that: The step S200 further includes the following steps: Step S210: placing the rotated laser point cloud coordinates into corresponding voxels; Step S220: sorting all voxels according to altitude; Step S230: creating an association relationship between voxel labels and voxel height differences.

5. The method for automatic classification of laser point clouds of power transmission lines according to claim 1, characterized in that: The step S300 further includes the following steps: Step S310: add up all the height differences in the association relationship and divide by the number of voxels to obtain an average height difference Zavg; Step S320: Divide all voxels into two parts according to the average height difference Zavg, namely, voxels with a height difference greater than or equal to the average height difference and voxels with a height difference less than the average height difference; Step S330: respectively calculating the average height differences ZavgTop and ZavgBottom for the two parts of voxels in step S320; Step S340: Optimizing the average height difference Zavg; Step S350: Identifying the voxel category according to the average height difference Zavg.

6. The method for automatic classification of laser point clouds of power transmission lines according to claim 1, characterized in that: The step S400 further includes the following steps: Step S410: Calculate the average height difference ZavgCube of the power line category voxels; Step S420: Divide the power line category voxels into two parts according to the average height difference ZavgCube, namely, voxels with a height difference greater than or equal to the average height difference and voxels with a height difference less than the average height difference; Step S430: respectively calculate the average height differences ZavgCubeTop and ZavgCubeBottom for the two parts of voxels in step S420; Step S440: Optimizing the average height difference ZavgCube; Step S450: For each power line type voxel, the voxel type with a height difference greater than or equal to ZavgCube remains unchanged, and the voxel type with a height difference less than ZavgCube becomes a vegetation type.

7. The method for automatic classification of laser point clouds of power transmission lines according to claim 1, characterized in that: The step S500 further includes: performing DBScan clustering on all power line type voxels, in the clustering results, the voxel category with the largest number remains unchanged, and the voxels corresponding to other clustering results are set to vegetation categories.

Images