Real-time safety distance diagnosis method of laser point cloud for UAV power inspection

Abstract

The present invention proposes a real-time automatic safety distance diagnosis method for UAV power inspection LiDAR point cloud data. The LiDAR point cloud data collected by the UAV inspection system is used as the research object. According to its data characteristics, a three-step real-time safety diagnosis model application strategy is formulated. The multi-scale fast ground filtering, shape analysis and fast classification of objects, and fast calculation of safety distance defined in the three-step diagnosis model of the LiDAR point cloud data of the line corridor are concretized to complete the fast calculation of the safety distance measurement of the LiDAR point cloud data. This method can better solve the problem that the traditional safety distance measurement method is complicated in steps and difficult to calculate the safety distance in real time.

Description

Technical Field

[0001] The present invention relates to the application of LiDAR point cloud safety distance calculation in UAV power inspection, and proposes a new UAV power inspection LiDAR point cloud real-time safety distance diagnosis method. Background Art

[0002] Detection of safe distances between objects in transmission line corridors is an important part of power line inspection. Due to various objective reasons such as operating conditions, such potential safety hazards are difficult to detect or are detected in a short time. Tree barrier discharges and lines damaged by external forces can cause line tripping and power outages, and even worse, cause serious production safety accidents such as forest fires. [1] , directly endangering the safe and stable operation of the power system and the personal and property safety of residents within the radiation range of the line corridor. LiDAR is an important remote sensing method [2,3] , which can obtain the spatial geometric structure information of the detected object. Airborne laser radar (LiDAR) technology makes up for the limitations of low spatial positioning accuracy and low geometric structure measurement efficiency of traditional inspection methods, and can effectively solve the problem of accurate distance measurement between ground objects and conductors in the transmission line corridor that is difficult to solve with traditional line inspection technology. [4,5] The use of drones as a platform for the LiDAR system can effectively improve the flexibility and mobility of line inspections and save a lot of manpower, material resources and time costs required for manual / manned inspections. This is particularly important for overhead lines in difficult inspection areas such as mountainous areas.

[0003] According to the relevant definition of DL / T 741-2010 "Operation Regulations for Overhead Transmission Lines", the safe distance diagnosis of the transmission line corridor is to measure the distance from various objects in the line corridor, such as trees, vegetation, buildings, and crossings, to the power lines to determine whether the distance is within the safe range. Transmission line inspections must ensure that the distance between the transmission wires and various objects in the line corridor meets the requirements of the safe distance regulations.

[0004] Traditional safety distance calculation method [6] It involves complex calculations such as ground filtering, DEM construction, object classification, and wire fitting, making it difficult to calculate the safe distance in real time.

[0005] Related references are as follows:

[0006] [1] Liu Huaidong, Chen Wei, Gao Xiaohui, Lei Xiangdong. Analysis of a type of transmission line tree flash fault from the perspective of blackout [J]. Power System Technology, 2007(S1):67-69.

[0007] [2]Yang B,Chen C.Automatic registration of UAV-borne sequent images and LiDAR data[J].Isprs Journal of Photogrammetry & Remote Sensing,2015,101:262-274.

[0008] [3]Chen, Chi, Yang, et al. Dynamic occlusion detection and inpainting of in situ captured terrestrial laser scanning point clouds sequence [J]. ISPRSjournal of photogrammetry and remote sensing, 2016, 119 (Sep.): 90-107.

[0009] [4] Peng Xiangyang, Chen Chi, Xu Xiaogang, et al. Transmission channel safety distance diagnosis technology based on UAV laser scanning[J]. Power System Technology, 2014, 38(11):6.

[0010] [5]Franken P.Transmission Line monitoring through airborne modeling[OL].2003,http: / / www.fli-map.nl / downloads / downsload-articles / Transmission_Line_monitoring.

[0011] [6] Chen Chi, Peng Xiangyang, Song Shuang, Wang Ke, Qian Jinju, Yang Bisheng. Safety distance diagnosis method of LiDAR point cloud for UAV power inspection[J]. Power System Technology, 2017, 41(08): 2723-2730. DOI: 10.13335 / j.1000-3673.pst.2016.3194. Summary of the invention

[0012] In response to the deficiencies and shortcomings of traditional power inspection LiDAR point cloud safety distance measurement methods, this patent proposes a real-time safety distance measurement method for UAV LiDAR data combined with shape analysis, which can automatically identify in real time whether various natural and artificial structures including buildings and vegetation under the line exceed the safety area and height.

[0013] In order to solve the above technical problems, the present invention adopts the following technical solution as a real-time safety distance diagnosis method for UAV power inspection laser point cloud. First, a multi-scale elevation distribution window fast ground filtering method is used to filter out ground points, obtain ground point clouds to construct DEM, and use DEM to further divide the ground point cloud to extract non-ground point clouds. Secondly, for the point cloud voxels marked as non-ground points, the covariance matrix eigenvalues ​​and point density of the point cloud in a single voxel are calculated, and the point cloud dimensional features are constructed; for non-ground voxels that are not classified as buildings or trees, the local shape eigenvalues ​​and point density eigenvalues ​​are calculated based on the voxel and the non-empty voxels in the neighboring neighborhood of the voxel, and it is divided into buildings or trees based on the eigenvalues. Finally, according to DL / T 741-2010 "Overhead Transmission Line Operation Regulations", the safety distance threshold is set, and the point cloud marked as trees and buildings is used to construct an octree and used as the query object. The point cloud marked as wire is thinned out by the octree and used as the query center point. Different categories of safety distance thresholds are used as query conditions, and the safety distance exceeding point is calculated and marked for warning. The specific steps include the following:

[0014] Step 1: Establish an octree index structure for the point cloud data of the classified transmission lines, and calculate the slope characteristic value of the voxel to be filtered through the voxel window, mark the point cloud in the voxel with a slope characteristic value less than the slope threshold as the ground category, and mark the others as non-ground point clouds, that is, perform rapid ground filtering of elevation distribution through the slope threshold;

[0015] Step 2: Based on the ground filtered point cloud, the voxel local shape and point density eigenvalues ​​of the non-ground point cloud are calculated, and the point clouds in the voxels that meet the constraints are marked as buildings, and other categories are marked as trees, so as to achieve multi-scale voxel shape analysis and rapid classification of objects;

[0016] Step 3: Set the safety distance threshold, construct an octree with the point cloud marked as trees and buildings and use it as the query object, perform octree thinning on the point cloud marked as transmission lines and use it as the query center point, use the safety distance thresholds of different categories as query conditions, calculate the safety distance exceeding points and mark them for warning.

[0017] Furthermore, the specific implementation of step 1 includes the following sub-steps:

[0018] Step 1.1: Create an octree index for the point cloud data and divide it into voxel storage. The voxel scale is denoted by R. g ; and record the lowest voxel at the same plane position, and save the point cloud in the voxel to obtain the ground surface point cloud and filter the transmission towers and transmission wires in the transmission corridor;

[0019] Step 1.2: Search the point cloud data with the constructed octree index from bottom to top in the elevation direction to find the first non-empty voxel as the first voxel V of the ground filter window. g , calculate the voxel center point Z g , and calculate the voxel center point Z of the two adjacent voxels perpendicular to the known power corridor direction g-2 ,Z g-1 ,Z g+1 ,Z g+2 ;

[0020] Step 1.3, calculate V g The slope characteristic value of is as shown in formula (1);

[0021]

[0022] Set the empirical slope threshold, take L slope The point clouds within the voxels with slope < the threshold are marked as ground points, and the others are marked as non-ground points.

[0023] Furthermore, step 1 also includes the following sub-steps:

[0024] Step 1.4, with different voxel scales R g Perform ground filtering operation, add and fuse the filtering results under multi-scale ground filtering, record the fusion result as non-ground point cloud, and mark the remaining point clouds as ground point clouds;

[0025] Step 1.5, using the ground point cloud data as a driver to build a ground DEM point cloud model, in the non-ground point cloud, the elevation value is subtracted from the elevation value interpolated in the DEM point cloud model. If the difference is less than the threshold, the non-ground point cloud is considered to be a misclassified point cloud and is further classified as ground.

[0026] Furthermore, the calculation of the voxel local shape feature value of the non-ground point cloud in step 2 includes the following sub-steps;

[0027] Step 2.1, calculate the covariance matrix eigenvalues ​​of the point cloud marked as non-ground category voxels in the point cloud with constructed octree, λ1, λ2, λ3, where λ1≥λ2≥λ3, and construct the point cloud dimension features, such as formula (2):

[0028]

[0029] Step 2.2, calculate the point density eigenvalue ρ of the point cloud in the voxel marked as the non-ground category in the point cloud of the constructed octree, where the number of point clouds in the voxel is P and the voxel side length is r;

[0030]

[0031] Step 2.3, calculate the voxel dimension feature V of the point cloud marked as non-ground point cloud in the constructed octree (U,V,C) (a 1D ,a 2D ,a 3D ), take V (U,V,C) The local shape eigenvalue L is calculated from the eight non-empty voxels in its three-dimensional space. shape , as shown in formula (4);

[0032]

[0033] In formula (4), n represents the voxel V (U,V,C) There are n non-empty point cloud voxels in the eight adjacent neighborhood non-empty voxels in its three-dimensional space; They represent the point cloud dimension features calculated in step 2.1 for the i-th neighborhood voxel respectively;

[0034] Step 2.4, calculate the sparse features V of the voxels marked as non-ground point clouds in the point cloud with the constructed octree (U,V,C) (ρ), take V (U,V,C) The eight neighboring non-empty voxels in its three-dimensional space calculate the local sparse eigenvalue L density , as shown in formula (5):

[0035]

[0036] In formula (5), n represents the voxel V (U,V,C) There are n non-empty point cloud voxels in the eight adjacent neighborhood non-empty voxels in its three-dimensional space; Represents the point density eigenvalue ρ in the i-th neighborhood voxel.

[0037] Furthermore, in the inspection point cloud collected by the UAV laser radar (LiDAR) system, the scanned parts of the building are mainly roofs, which have a flat spatial distribution feature, and its eigenvalue That is a 2D Larger and closer to 1, a 1D ,a 3D Smaller and closer to 0; set constraints: (1) L shape <L B , the empirical threshold L B =3; (2) Experience threshold L aD =0.2; (3) L density <L D , the empirical threshold L D =2.5; the point clouds in the voxels that meet the above conditions are marked as buildings, and other categories are marked as trees.

[0038] Furthermore, the specific implementation of step 3 includes the following sub-steps:

[0039] Step 3.1, perform octree thinning on the point cloud of the transmission line category to reduce the density of the LiDAR point cloud;

[0040] Step 3.2, construct an octree for the point cloud of non-ground objects classified as trees and buildings to achieve fast point cloud search and improve the search speed;

[0041] In step 3.3, the non-ground point cloud octree is used as the query object, the thinned conductor point cloud is used as the query center point, and the safety distance thresholds corresponding to different categories defined in DL / T 741-2010 "Overhead Transmission Line Operation Regulations" are used as query conditions to calculate the safety distance exceeding points and mark them for warning.

[0042] Furthermore, the slope threshold is 0.3.

[0043] Furthermore, the different voxel scales R in step 1.4 g Including R g =0.5m, 1.0m, 1.5m.

[0044] Furthermore, the threshold of the difference in step 1.5 is set to 2.8 meters.

[0045] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0046] The present invention takes the UAV power inspection laser radar (LiDAR) point cloud as the research object, formulates a three-step real-time safety distance measurement model application strategy according to its data characteristics, and concretizes the measurement methods of multi-scale rapid ground filtering, shape analysis rapid classification of objects, and rapid calculation of safety distance defined in the three-step safety distance measurement model of laser radar (LiDAR) point cloud to complete the safety distance detection of power inspection.

[0047] The present invention can better solve the problems of complex calculation steps and difficulty in real-time calculation of safety distance in traditional safety distance measurement methods. It can quickly calculate the safety distance and can be used in scenarios with high timeliness requirements (such as disaster emergency inspections, etc.). BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of the real-time safety distance diagnosis method of UAV power inspection laser radar (LiDAR) point cloud in an embodiment of the present invention.

[0049] Figure 2 It is a schematic diagram of obtaining the ground surface point cloud result in an embodiment of the present invention.

[0050] Figure 3 This is a schematic diagram of the result of multi-scale fusion combining slope and DEM ground filtering in an embodiment of the present invention.

[0051] Figure 4 Schematic diagram of point cloud classification detection results combined with shape analysis in an embodiment of the present invention.

[0052] Figure 5 Schematic diagram of the safety distance detection result in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The technical solution of the present invention is described below in conjunction with the accompanying drawings and embodiments.

[0054] The method proposed in the present invention is specifically described by selecting the point cloud data obtained by the UAV power inspection system. Figure 1 , the embodiment of the present invention comprises the following steps:

[0055] Step 1: Establish an octree index structure for the point cloud data of the classified transmission lines, and calculate the slope value of the voxel to be filtered through the voxel window, and then perform fast ground filtering of the elevation distribution through the slope threshold. This step is performed with voxel sizes of different scales, and the multi-scale filtering results are fused to classify the ground point cloud and the non-ground point cloud to achieve multi-scale fast ground filtering.

[0056] The above step 1 further comprises the following steps:

[0057] Step 1.1: Create an octree index for the point cloud data and divide it into voxel storage. The voxel scale is denoted by R. g ; and record the lowest voxel at the same plane position, and save the point cloud in the voxel to obtain the ground surface point cloud and filter the transmission towers, transmission wires and other ground objects in the transmission corridor ( Figure 2 );

[0058] Step 1.2: Search the point cloud data with the constructed octree index from bottom to top in the elevation direction to find the first non-empty voxel as the first voxel V of the ground filter window. g , calculate the voxel center point Z g , and calculate the voxel center point Z of the two adjacent voxels perpendicular to the known power corridor direction g-2 ,Z g-1 ,Z g+1 ,Z g+2 ;

[0059] Step 1.3, calculate V g The slope characteristic value of is as shown in formula (1).

[0060]

[0061] In the inspection point cloud collected by the UAV laser radar (LiDAR) system, the non-ground point cloud scanned parts are mainly roofs and tree crowns, which have the characteristics of voxel center elevation distribution significantly higher than the adjacent ground voxel center elevation. Set the empirical threshold to 0.3 and take L slope The point clouds within the voxels with a value <0.3 are marked as ground categories, and the others are marked as non-ground point clouds.

[0062] Step 1.4, with different voxel scales R g Perform ground filtering operations (such as: R g =0.5m, 1.0m, 1.5m), perform point cloud addition and fusion on the filtering results under multi-scale ground filtering, record the fusion result as non-ground point cloud, and mark the remaining point clouds as ground point clouds.

[0063] Step 1.5, driven by the ground point cloud data, use the grid interpolation method to build a ground DEM point cloud model, and subtract the elevation value in the non-ground point cloud from the elevation value interpolated in the DEM point cloud model. If the difference is too small, the non-ground point cloud is considered to be a misclassified point cloud and is further classified as ground. According to the relevant regulations on the height of residential buildings in the country, the threshold of the difference is set to 2.8 meters ( Figure 3 ).

[0064] Step 2: Based on the ground filtered point cloud, the voxel local shape and point density eigenvalues ​​of the non-ground point cloud are calculated to achieve rapid classification of objects by multi-scale voxel shape analysis.

[0065] The above step 2 further comprises the following steps:

[0066] Step 2.1, calculate the covariance matrix eigenvalues ​​of the point cloud marked as non-ground category voxels in the point cloud with constructed octree, λ1, λ2, λ3 (λ1≥λ2≥λ3), and construct the point cloud dimensional features, such as formula (2).

[0067]

[0068] Step 2.2, calculate the point density eigenvalue ρ of the point cloud in the voxel marked as the non-ground category in the point cloud of the constructed octree, denoting the number of point clouds in the voxel as P, and the voxel side length as r, and formula (3) is obtained.

[0069]

[0070] Step 2.3, calculate the voxel dimension feature V of the point cloud marked as non-ground point cloud in the constructed octree (U,V,C) (a 1D ,a 2D ,a 3D ), take V (U,V,C) The local shape eigenvalue L is calculated from the eight non-empty voxels in its three-dimensional space.shape , as shown in formula (4).

[0071]

[0072] In formula (4), n represents the voxel V (U,V,C) There are n non-empty point cloud voxels in the eight adjacent neighborhood non-empty voxels in its three-dimensional space; They respectively represent the point cloud dimensional features calculated in step 2.1 for the i-th neighborhood voxel.

[0073] Step 2.4, calculate the sparse features V of the voxels marked as non-ground point clouds in the point cloud with the constructed octree (U,V,C) (ρ), take V (U,V,C) The eight neighboring non-empty voxels in its three-dimensional space calculate the local sparse eigenvalue L density , as shown in formula (5).

[0074]

[0075] In formula (5), n represents the voxel V (U,V,C) There are n non-empty point cloud voxels in the eight adjacent neighborhood non-empty voxels in its three-dimensional space; Represents the point density eigenvalue ρ in the i-th neighborhood voxel.

[0076] In the inspection point cloud collected by the UAV laser radar (LiDAR) system, the scanned parts of the building are mainly roofs, which have a flat spatial distribution feature, and its eigenvalue That is a 2D Larger and closer to 1, a 1D ,a 3D The smaller the value, the closer it is to 0. Set the constraints: (1) L shape <L B , the empirical threshold L B =3; (2) Experience threshold L aD =0.2; (3) L density <L D , the empirical threshold L D = 2.5; the point clouds in the voxels that meet the above conditions are marked as buildings, and other categories are marked as trees ( Figure 4 ).

[0077] Step 3: Set the safety distance threshold according to DL / T 741-2010 "Operation Regulations for Overhead Transmission Lines", construct an octree with the point cloud marked as trees and buildings as the query object, perform octree thinning on the point cloud marked as transmission wires and use it as the query center point, use different categories of safety distance thresholds as query conditions, calculate the safety distance limit points and mark them for warning.

[0078] The above step 3 further comprises the following steps:

[0079] In step 3.1, the point cloud of the transmission line category is thinned by octree to reduce the density of the LiDAR point cloud.

[0080] In step 3.2, an octree is constructed for point clouds of non-ground objects such as trees and buildings to achieve fast search of point clouds and improve the search speed.

[0081] Step 3.3, take the non-ground point cloud octree as the query object, take the thinned conductor point cloud as the query center point, take the safety distance thresholds corresponding to different categories defined in DL / T 741-2010 "Overhead Transmission Line Operation Regulations" as the query condition, calculate the safety distance limit point and mark it for warning ( Figure 5 ).

[0082] The technical solution and beneficial effects of the present invention will be further explained below with reference to specific examples.

[0083] Using a certain section of power inspection data collected by the UAV power inspection system, after being processed by the method of the present invention, 16 safety distances exceeding the limit were detected in real time in the transmission corridor of about 23 km, which is consistent with the actual situation. This shows that the present invention can provide a high-precision safety distance detection result in real time efficiency.

[0084] In specific implementation, the process provided by the present invention can be automatically run using software technology.

[0085] The above contents are further detailed descriptions of the present invention in combination with specific implementation methods, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. UAV power inspection laser point cloud real-time safety distance diagnosis method, characterized by: The following steps are involved: Step 1: Establish an octree index structure for the point cloud data of the classified transmission lines, and calculate the slope characteristic value of the voxel to be filtered through the voxel window, mark the point cloud in the voxel with a slope characteristic value less than the slope threshold as the ground category, and mark the others as non-ground point clouds, that is, perform rapid ground filtering of elevation distribution through the slope threshold; The specific implementation of step 1 includes the following sub-steps: Step 1.1: Create an octree index for the point cloud data and divide it into voxel storage. The voxel scale is denoted by R. g ; and record the lowest voxel at the same plane position, and save the point cloud in the voxel to obtain the ground surface point cloud and filter the transmission towers and transmission wires in the transmission corridor; Step 1.2: Search the point cloud data with the constructed octree index from bottom to top in the elevation direction to find the first non-empty voxel as the first voxel V of the ground filter window. g , calculate the voxel center point Z g , and calculate the voxel center point Z of the two adjacent voxels perpendicular to the known power corridor direction g-2 ,Z g-1 ,Z g+1 ,Z g+2 ; Step 1.3, calculate V g The slope characteristic value of is as shown in formula (1); Set the empirical slope threshold, take L slope The point clouds within the voxels with slope < the threshold are marked as ground categories, and the others are marked as non-ground point clouds; Step 2: Based on the ground filtered point cloud, the voxel local shape and point density eigenvalues ​​of the non-ground point cloud are calculated, and the point clouds in the voxels that meet the constraints are marked as buildings, and other categories are marked as trees, so as to achieve multi-scale voxel shape analysis and rapid classification of objects; The calculation of the voxel local shape feature values ​​of the non-ground point cloud in step 2 includes the following sub-steps; Step 2.1, calculate the covariance matrix eigenvalues ​​of the point cloud marked as non-ground category voxels in the point cloud with constructed octree, λ1, λ2, λ3, where λ1≥λ2≥λ3, and construct the point cloud dimension features, such as formula (2): Step 2.2, calculate the point density eigenvalue ρ of the point cloud in the voxel marked as the non-ground category in the point cloud of the constructed octree, where the number of point clouds in the voxel is P and the voxel side length is r; Step 2.3, calculate the voxel dimension feature V of the point cloud marked as non-ground point cloud in the constructed octree (U,V,C) (a 1D ,a 2D ,a 3D ), take V (U,V,C) The local shape eigenvalue L is calculated from the eight non-empty voxels in its three-dimensional space. shape , as shown in formula (4); In formula (4), n represents the voxel V (U,V,C) There are n non-empty point cloud voxels in the eight adjacent neighborhood non-empty voxels in its three-dimensional space; They represent the point cloud dimension features calculated in step 2.1 for the i-th neighborhood voxel respectively; Step 2.4, calculate the sparse features V of the voxels marked as non-ground point clouds in the point cloud with the constructed octree (U,V,C) (ρ), take V (U,V,C) The eight neighboring non-empty voxels in its three-dimensional space calculate the local sparse eigenvalue L density , as shown in formula (5): In formula (5), n represents the voxel V (U,V,C) There are n non-empty point cloud voxels in the eight adjacent neighborhood non-empty voxels in its three-dimensional space; represents the point density eigenvalue ρ in the i-th neighborhood voxel; Step 3: Set the safety distance threshold, construct an octree with the point cloud marked as trees and buildings and use it as the query object, perform octree thinning on the point cloud marked as transmission lines and use it as the query center point, use the safety distance thresholds of different categories as query conditions, calculate the safety distance exceeding points and mark them for warning.

2. The UAV power inspection laser point cloud real-time safety distance diagnosis method according to claim 1, characterized in that: Step 1 also includes the following sub-steps: Step 1.4, with different voxel scales R g Perform ground filtering operation, add and fuse the filtering results under multi-scale ground filtering, record the fusion result as non-ground point cloud, and mark the remaining point clouds as ground point clouds; Step 1.5, using the ground point cloud data as a driver to build a ground DEM point cloud model, in the non-ground point cloud, the elevation value is subtracted from the elevation value interpolated in the DEM point cloud model. If the difference is less than the threshold, the non-ground point cloud is considered to be a misclassified point cloud and is further classified as ground.

3. The real-time safety distance diagnosis method for UAV power inspection laser point cloud according to claim 1, characterized in that: In the inspection point cloud collected by the UAV laser radar (LiDAR) system, the scanned parts of the building are mainly roofs, which have a flat spatial distribution feature, and its eigenvalue That is a 2D Larger and closer to 1, a 1D ,a 3D Smaller and closer to 0; set constraints: (1) L shape <L B , the empirical threshold L B =3; (2) Experience threshold L aD =0.2; (3) L density <L D , the empirical threshold L D =2.5; the point clouds in the voxels that meet the above conditions are marked as buildings, and other categories are marked as trees.

4. The UAV power inspection laser point cloud real-time safety distance diagnosis method according to claim 1, characterized in that: The specific implementation of step 3 includes the following sub-steps: Step 3.1, perform octree thinning on the point cloud of the transmission line category to reduce the density of the LiDAR point cloud; Step 3.2, construct an octree from the point cloud of non-ground objects of the categories of trees and buildings to achieve fast search of point clouds and improve the search speed; In step 3.3, the non-ground point cloud octree is used as the query object, the thinned conductor point cloud is used as the query center point, and the safety distance thresholds corresponding to different categories defined in DL / T 741-2010 "Overhead Transmission Line Operation Regulations" are used as query conditions to calculate the safety distance exceeding points and mark them for warning.

5. The UAV power inspection laser point cloud real-time safety distance diagnosis method according to claim 1, characterized in that: The slope threshold is 0.

3.

6. The real-time safety distance diagnosis method for UAV power inspection laser point cloud according to claim 2, characterized in that: Different voxel scales R in step 1.4 g Including R g =0.5m, 1.0m, 1.5m.

7. The real-time safety distance diagnosis method for UAV power inspection laser point cloud according to claim 2, characterized in that: The threshold of the difference in step 1.5 is set to 2.8 meters.

Images