A robust unmanned aerial vehicle laser radar wire-like tower identification method and device

Abstract

The application discloses a robust unmanned aerial vehicle laser radar line-simulating-pole-tower identification method and device, which scans a power transmission line scene by using an unmanned aerial vehicle laser radar to obtain line and pole tower laser point clouds; first, power line channels and pole towers of each frame of data are extracted in the flight process, the extracted pole tower point clouds are filtered in different ranges by using a power line straight line equation of a current channel to obtain pole tower seed points and candidate points; then, the filtered pole tower seed points and candidate point clouds are accumulated, sampled and clustered frame by frame respectively, whether the clustered seed points reach a seed point quantity threshold is judged, and whether the pole tower exists is determined; finally, the shape of the pole tower point cloud is restored by using an overlapping relationship of the seed points and the candidate points, so that the pole tower is identified, and finally the coordinates of the pole tower and the center height of the pole tower are calculated.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of unmanned aerial vehicle laser radar simulation line flight, and particularly relates to a robust unmanned aerial vehicle laser radar simulation line tower identification method and device. BACKGROUND

[0002] In the process of automatic patrol inspection by means of unmanned aerial vehicle laser radar system simulation line flight, the flight position and posture of the unmanned aerial vehicle may cause the laser to scan incompletely in some time periods, accidental flight shaking may cause partial tower data to be lost, or the unmanned aerial vehicle gradually approaches the tower, and the tower data gradually becomes complete, instead of being able to observe the whole tower at once.

[0003] The main technical defects of the current simulation line flight are as follows: 1. The scanning range of the laser is wide, and there may be multiple lines below, so multiple towers may be observed at the same time, at which time it is necessary to distinguish which line the currently observed tower belongs to; 2. How to effectively distinguish irrelevant towers or ground objects from columnar ground objects is also a problem to be solved. Therefore, a method is needed to comprehensively process multiple frames of observation data, quickly and continuously and stably identify towers, and the earlier the tower in front is identified, the more conducive it is to the unmanned aerial vehicle simulation line system to make corresponding decisions. Otherwise, due to the lack of completeness of single-frame data or the existence of other interference lines, the tower identification is mistaken, and the tower appears and disappears, which is not conducive to making flight decisions. 3. When the unmanned aerial vehicle simulation line flies, the laser radar may scan the tower that has been flown over, and the tower that has been flown over has been recorded in the historical tower information, so the tower that has been flown over needs to be filtered. SUMMARY

[0004] The present application aims to provide a robust unmanned aerial vehicle laser radar simulation line tower identification method and device, which solves the above technical problems in the prior art.

[0005] The present application provides a robust unmanned aerial vehicle laser radar simulation line tower identification device, which comprises a simulation line flight unmanned aerial vehicle;

[0006] The simulation line flight unmanned aerial vehicle comprises a laser radar module, a central processor and a GPS positioning module.

[0007] The central processor is connected with the laser radar module and the GPS positioning module respectively.

[0008] The laser radar module is used to acquire all power line channel data of a current frame and tower point cloud of the current frame in real time.

[0009] The current frame power line channel data includes line straight line equations of all power line channels in the current frame; and the tower point cloud includes coordinates of the tower point cloud and height of the tower point cloud.

[0010] The GPS positioning module is configured to acquire a coordinate position of the line-simulating flight unmanned aerial vehicle.

[0011] The central processor is configured to acquire, according to all power line channel data of a current frame and a tower point cloud of the current frame acquired by the laser radar module in real time, the tower point cloud of the current frame from all tower point clouds of the current frame, and determine that a power line channel where the tower point cloud of the current frame is located is the power line channel where the tower point cloud of the current frame is located.

[0012] The central processor is further configured to screen a plurality of seed points and candidate points from the tower point cloud of the current frame.

[0013] The central processor is further configured to acquire a plurality of candidate points and a plurality of seed points in a plurality of continuous frames, accumulate and merge the plurality of candidate points and the plurality of seed points to obtain a plurality of merged candidate points and a plurality of merged seed points, filter the plurality of merged candidate points and the plurality of merged seed points through a voxel grid to obtain sampled seed points and candidate points, and perform clustering processing on the sampled candidate points and seed points by using a DBSCAN density clustering method, and preset a minimum threshold N of a point cloud number in a seed point cloud cluster. s The central processor is further configured to acquire a plurality of candidate point cloud clusters and seed point cloud clusters.

[0014] The central processor is further configured to merge seed point cloud cluster and candidate point cloud cluster clustering results to obtain a bounding box of the seed point cloud cluster and a bounding box of the candidate point cloud cluster according to a minimum threshold k of an overlap rate.

[0015] The central processor is further configured to calculate a bounding box volume of the seed point cloud cluster and the candidate point cloud cluster respectively, and determine whether the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster have an overlap.

[0016] The central processor is further configured to determine that, if the overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster is greater than the minimum threshold k of the overlap rate, the candidate point cloud cluster corresponding to the bounding box of the candidate point cloud cluster is an effective tower point cloud cluster.

[0017] The central processor is further configured to calculate a coincidence plane area of each effective tower point cloud cluster and a seed point cloud cluster, and determine that a tower point cloud cluster with a maximum coincidence plane area is a target tower point cloud cluster.

[0018] The central processor is further configured to traverse all tower point clouds in the target tower point cloud cluster, calculate a tower bounding box volume and a tower coordinate according to coordinates of the tower point cloud, and extract a highest point of the target tower point cloud cluster as a tower center highest point.

[0019] Correspondingly, the application further provides a robust unmanned aerial vehicle laser radar simulation wire tower identification method, comprising the following operation steps:

[0020] Real-time acquisition of all power line channel data of a current frame and all tower point clouds of the current frame, and acquisition of tower point clouds of the current frame from all tower point clouds of the current frame;

[0021] Screening determines that a power line channel where the tower point clouds of the current frame are located is the power line channel where the tower point clouds of the current frame are located;

[0022] Screening of a plurality of seed points and candidate points from the tower point clouds of the current frame;

[0023] Continuous multi-frame acquisition of a plurality of candidate points and a plurality of seed points, accumulation and merging of the plurality of candidate points and the plurality of seed points to obtain a plurality of merged candidate points and a plurality of seed points, voxel grid filtering of the merged candidate points and the merged seed points to obtain sampled seed points and candidate points, clustering of the sampled candidate points and the sampled seed points by using a DBSCAN density clustering method, and presetting a minimum threshold N of the number of point clouds in a seed point cloud cluster s , acquisition of a plurality of candidate point cloud clusters and seed point cloud clusters;

[0024] Presetting a minimum threshold k of an overlap rate, merging the clustering results of the seed point cloud cluster and the candidate point cloud cluster to obtain a bounding box of the seed point cloud cluster and a bounding box of the candidate point cloud cluster;

[0025] Respectively calculating the bounding box volumes of the seed point cloud cluster and the candidate point cloud cluster, and judging whether the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster have an overlap, if there is an overlap, calculating the overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster;

[0026] Determining that the overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster is greater than the minimum threshold k of the overlap rate, and determining that the candidate point cloud cluster corresponding to the bounding box of the candidate point cloud cluster is an effective tower point cloud cluster;

[0027] Calculating the overlapping plane area of each effective tower point cloud cluster and the seed point cloud cluster, and determining that the tower point cloud cluster with the maximum overlapping plane area is a target tower point cloud cluster;

[0028] Traversing all point clouds in the target tower point cloud cluster, calculating the tower bounding box volume and the tower coordinates, and extracting the highest point of the target tower point cloud cluster as the tower center highest point.

[0029] Compared with the prior art, the embodiment of the application has at least the following technical advantages:

[0030] It can be known from the above-mentioned robust unmanned aerial vehicle laser radar simulation wire tower identification method and device that in specific application, the power line channel and the tower are extracted in each frame of data in the flight process, all power line channel data of the current frame and the current frame tower point cloud are obtained according to the laser radar module, and the current frame tower point cloud is obtained from all the current frame tower point cloud. A power line channel where the current frame tower point cloud is located is determined as the power line channel where the current frame tower point cloud is located. Through the improved operation mode, the influence of other power line channels on tower identification can be screened out, and the power line channel where the current frame tower point cloud is located can be determined. On the basis of determining the power line channel where the current frame tower point cloud is located, a plurality of seed points and candidate points are screened out from the current frame tower point cloud. Other object point clouds (tree point clouds, column-shaped ground object point clouds similar to towers) are screened out, and the tower point cloud of the identified tower is screened out, to ensure the accuracy of the current tower identification.

[0031] Subsequently, a plurality of candidate points and a plurality of seed points are obtained in a plurality of continuous frames, and the plurality of candidate points and the plurality of seed points are accumulated and merged to obtain a plurality of merged plurality of candidate points and plurality of seed points. The plurality of merged plurality of candidate points and plurality of seed points are filtered through a voxel grid to obtain sampled seed points and candidate points. In the prior art, the laser radar scans the same position multiple times to obtain a plurality of repeated tower point clouds, causing point cloud aggregation. The technical solution provided in the embodiment of the application samples the obtained plurality of candidate points and plurality of seed points to obtain a small but insufficient amount of tower point clouds, thereby improving the identification accuracy of the tower.

[0032] The DBSCAN density clustering method is used to cluster the sampled candidate points and seed points, and a minimum threshold N of the number of point clouds in a seed point cloud cluster is preset s to obtain a plurality of candidate point cloud clusters and seed point cloud clusters. In the prior art, when the laser radar scans the tower point cloud, the point clouds of the column-shaped ground object similar to the tower and the branches beside the tower are also scanned, but generally, there is a spatial isolation between the tower and the column-shaped ground object similar to the tower and the branches beside the tower. After a value is given, DBSCAN can gather the point clouds to obtain a point cloud cluster.

[0033] According to the minimum overlap rate threshold k, the seed point cloud cluster and the candidate point cloud cluster clustering result are merged to obtain a seed point cloud cluster bounding box and a candidate point cloud cluster bounding box; the volumes of the seed point cloud cluster and the candidate point cloud cluster are calculated respectively, and it is judged whether the seed point cloud cluster bounding box and the candidate point cloud cluster bounding box have an overlap; if there is an overlap, the overlap rate of the seed point cloud cluster bounding box and the candidate point cloud cluster bounding box is calculated; it is determined that the overlap rate of the seed point cloud cluster bounding box and the candidate point cloud cluster bounding box is greater than the minimum overlap rate threshold k, and then it is determined that the candidate point cloud cluster corresponding to the candidate point cloud cluster bounding box is an effective tower point cloud cluster; the overlapping plane area of each effective tower point cloud cluster and the seed point cloud cluster is calculated, and the tower point cloud cluster with the maximum overlapping plane area is determined as a target tower point cloud cluster; in the embodiment of the application, the seed point cloud cluster is a subset of the candidate point cloud cluster, so the overlap rate of the seed point cloud cluster bounding box and the candidate point cloud cluster bounding box is relatively large, and multiple effective tower point cloud clusters can be obtained by using the technical scheme provided in the embodiment of the application.

[0034] Finally, all tower points in the target tower point cloud cluster are traversed, the tower bounding box volume and the tower coordinates are calculated according to the coordinates of the tower points, and the highest point of the target tower point cloud cluster is extracted as the tower center highest point. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the specific embodiments of the present application or the technical schemes in the prior art, the drawings needed in the following specific embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0036] Figure 1 A robust unmanned aerial vehicle laser radar line-imitating tower identification device architecture schematic diagram is provided for the first embodiment of the present application.

[0037] Figure 2 A robust unmanned aerial vehicle laser radar line-imitating tower identification method operation flowchart schematic diagram is provided for the second embodiment of the present application.

[0038] Figure 3 An operation flowchart schematic diagram for screening and determining a power line channel in which the current frame tower point cloud is located as the power line channel in which the current frame tower point cloud is located in the robust unmanned aerial vehicle laser radar line-imitating tower identification method provided for the second embodiment of the present application.

[0039] Figure 4 An operation flowchart schematic diagram for screening and determining a power line channel in which the current frame tower point cloud is located as the power line channel in which the current frame tower point cloud is located in the robust unmanned aerial vehicle laser radar line-imitating tower identification method provided for the second embodiment of the present application.

[0040] Figure 5 An operation flow diagram for screening multiple seed points and candidate points in a robust unmanned aerial vehicle laser radar simulated wire tower identification method provided for embodiment two of the application;

[0041] Figure 6 An operation flow diagram for obtaining multiple candidate point cloud clusters and seed point cloud clusters in a robust unmanned aerial vehicle laser radar simulated wire tower identification method provided for embodiment two of the application.

[0042] Label: simulated wire flying unmanned aerial vehicle 10; laser radar module 11; central processing unit 12; GPS positioning module 13. DETAILED DESCRIPTION

[0043] The technical solutions of the application will be described in detail below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the application.

[0044] The application will be described in further detail below with reference to specific embodiments and in conjunction with the drawings.

[0045] Embodiment one

[0046] Reference Figure 1 , the application provides a robust unmanned aerial vehicle laser radar simulated wire tower identification device, which comprises a simulated wire flying unmanned aerial vehicle 10;

[0047] Among them, the simulated wire flying unmanned aerial vehicle comprises a laser radar module 11, a central processing unit 12 and a GPS positioning module 13.

[0048] The central processing unit 12 is connected with the laser radar module 11 and the GPS positioning module 13 respectively.

[0049] Among them, the laser radar module 11 is used for acquiring all power line channel data of the current frame and tower point cloud of the current frame in real time.

[0050] Among them, the current frame power line channel data comprises line straight line equation of all power line channels of the current frame; the tower point cloud comprises coordinates of the tower point cloud and height of the tower point cloud.

[0051] The GPS positioning module 13 is used for acquiring coordinate position of the simulated wire flying unmanned aerial vehicle in real time.

[0052] The central processor 12 is configured to acquire all power line channel data and current frame tower point cloud of a current frame from the laser radar module in real time, acquire current frame tower point cloud from all current frame tower point cloud, and determine a power line channel where the current frame tower point cloud is located as the power line channel where the current frame tower point cloud is located.

[0053] The central processor is further configured to screen a plurality of seed points and candidate points from the current frame tower point cloud.

[0054] The central processor is further configured to acquire a plurality of candidate points and a plurality of seed points in a plurality of frames in succession, accumulate and merge the plurality of candidate points and the plurality of seed points to obtain a plurality of merged candidate points and a plurality of merged seed points, filter the plurality of merged candidate points and the plurality of merged seed points through a voxel grid to obtain sampled seed points and candidate points, and perform clustering processing on the sampled candidate points and seed points by using a DBSCAN density clustering method, and preset a minimum threshold N of the number of point clouds in a seed point cloud cluster. s The central processor is further configured to acquire a plurality of candidate point cloud clusters and seed point cloud clusters.

[0055] The central processor is further configured to merge the clustering results of the seed point cloud cluster and the candidate point cloud cluster to obtain a bounding box of the seed point cloud cluster and a bounding box of the candidate point cloud cluster according to a minimum threshold k of an overlap rate.

[0056] The central processor is further configured to calculate the volumes of the bounding boxes of the seed point cloud cluster and the candidate point cloud cluster respectively, determine whether the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster overlap, and if there is an overlap, calculate the overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster.

[0057] The central processor is further configured to determine that the candidate point cloud cluster corresponding to the bounding box of the candidate point cloud cluster is an effective tower point cloud cluster if the overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster is greater than the minimum threshold k of the overlap rate.

[0058] The central processor is further configured to calculate the overlapping plane area of each effective tower point cloud cluster and the seed point cloud cluster, and determine a tower point cloud cluster with the maximum overlapping plane area as a target tower point cloud cluster.

[0059] The central processor is further configured to traverse all tower point clouds in the target tower point cloud cluster, calculate the tower bounding box volume and the tower coordinates according to the coordinates of the tower point clouds, and extract the highest point of the target tower point cloud cluster as the tower center highest point.

[0060] In summary, the robust unmanned aerial vehicle laser radar simulation line tower identification device provided by the application obtains all power line channel data of a current frame and tower point cloud of the current frame through the laser radar module; coordinates of the simulation line flight unmanned aerial vehicle are obtained in real time through the GPS positioning module; the tower point cloud of the current frame is obtained from all tower point clouds of the current frame; a power line channel in which the tower point cloud of the current frame is located is determined by screening the tower point cloud of the current frame; then a plurality of seed points and candidate points are screened from the tower point cloud of the current frame;

[0061] A plurality of candidate points and a plurality of seed points are obtained in a plurality of continuous frames, and the plurality of candidate points and the plurality of seed points are accumulated and merged to obtain a plurality of merged candidate points and a plurality of seed points; the merged candidate points and the merged seed points are filtered through a voxel grid to obtain sampled seed points and candidate points; the sampled candidate points and the sampled seed points are clustered by using a DBSCAN density clustering method, and a minimum threshold N of the number of point clouds in a seed point cloud cluster is preset s , a plurality of candidate point cloud clusters and a plurality of seed point cloud clusters are obtained;

[0062] According to a minimum threshold k of an overlap rate, the seed point cloud cluster and the candidate point cloud cluster clustering result are merged to obtain a bounding box of the seed point cloud cluster and a bounding box of the candidate point cloud cluster;

[0063] Volumes of the bounding boxes of the seed point cloud cluster and the candidate point cloud cluster are calculated respectively, and whether the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster overlap is judged; if the bounding boxes overlap, an overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster is calculated;

[0064] If the overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster is greater than the minimum threshold k of the overlap rate, it is determined that the candidate point cloud cluster corresponding to the bounding box of the candidate point cloud cluster is an effective tower point cloud cluster;

[0065] The overlapping plane area of each effective tower point cloud cluster and the seed point cloud cluster is calculated, and the tower point cloud cluster with the maximum overlapping plane area is determined as a target tower point cloud cluster;

[0066] Finally, all tower point clouds in the target tower point cloud cluster are traversed, the tower bounding box volume and the tower coordinates are calculated according to the coordinates of the tower point clouds, and the highest point of the target tower point cloud cluster is extracted as the tower center highest point.

[0067] Embodiment two

[0068] As Figure 2 shown, accordingly, the application further provides a robust unmanned aerial vehicle laser radar simulation line tower identification method, comprising the following operation steps:

[0069] Step S10: acquiring all power line channel data of a current frame and all tower point clouds of the current frame in real time, and acquiring a tower point cloud of the current frame from all tower point clouds of the current frame; (The tower point cloud refers to a tower point cloud on a power line channel of the current frame; generally, the current frame contains many power line channels, numerous tower point clouds and other object point clouds; the tower point cloud includes coordinates of the tower point cloud and height of the tower point cloud)

[0070] It should be noted that when the unmanned aerial vehicle performs laser radar simulation flight, the unmanned aerial vehicle acquires power line channel data and tower point clouds in real time through the laser radar, and extracts power line channels and point clouds (including tower point clouds and the like) of each frame; the power line channel data and the tower point clouds of each frame are analyzed respectively, and details are shown in subsequent operations.

[0071] The current frame power line channel data includes line straight line equations of all power line channels of the current frame (generally, each power line channel has multiple straight lines, and each straight line can be represented by a line straight line equation).

[0072] The tower point cloud includes a tower point cloud on a power line channel of the current frame and a columnar ground object point cloud (the columnar ground object point cloud refers to a ground object point cloud similar to a tower on the power line channel, and the columnar ground object point cloud is also called a similar tower body point cloud in the industry).

[0073] Step S20: screening and determining a power line channel in which the tower point cloud of the current frame is located as the power line channel in which the tower point cloud of the current frame is located.

[0074] Specifically, referring to Figure 3 In step S20, the screening and determination of the power line channel in which the tower point cloud of the current frame is located as the power line channel in which the tower point cloud of the current frame is located includes the following operation steps:

[0075] Step S21: acquiring a line straight line equation in all power line channel data of the current frame.

[0076] Step S22: saving the height of the line straight line equation to a container and sorting to obtain a power line height sequence table; obtaining minimum height H min and maximum height H max of a power line channel line through the power line height sequence table.

[0077] Step S23: presetting minimum vertical spacing threshold δ min and maximum vertical spacing threshold δ max , calculating a line vertical spacing average value according to the power line height sequence table; and calculating minimum vertical spacing threshold δ min and maximum vertical spacing threshold δ maxCalculate and determine the 2D distance threshold D between each power line channel. 2d and the 3D distance threshold D between each power line channel 3d ;

[0078] Obtain the current frame's tower point cloud from all tower point clouds in the current frame. Calculate the 2D and 3D distances from the current frame's tower point cloud to each of the current frame's power line channels. Filter the current frame's tower point cloud so that the 2D distance to each of the current frame's power line channels is less than a 2D distance threshold D between each power line channel. 2d Furthermore, the 3D distances from the current frame tower point cloud to all power line channels in the current frame are less than the 3D distance threshold D between each power line channel. 3d The current frame tower point cloud is the filtered current frame tower point cloud; the power line channel where the filtered current frame tower point cloud is located is determined as the power line channel where the current frame tower point cloud is located.

[0079] It should be noted that each power line channel has a line straightness equation, and the line straightness equation can reflect the height of each power line channel in each frame.

[0080] Specifically, see Figure 4 In step S23, the preset minimum vertical spacing threshold δ min With the maximum vertical spacing threshold δ max Based on the aforementioned power line height sequence table, calculate the average vertical spacing of the lines; based on the average vertical spacing and the minimum vertical spacing threshold δ... min With the maximum vertical spacing threshold δ max Calculate and determine the 2D distance threshold D between each power line channel. 2d and the 3D distance threshold D between each power line channel 3d Calculate the 2D and 3D distances from the current frame's tower point cloud to all power line channels in the current frame, and filter those where the 2D distance from the current frame's tower point cloud to all power line channels in the current frame is less than a 2D distance threshold D between each power line channel. 2d Furthermore, the 3D distances from the current frame tower point cloud to all power line channels in the current frame are less than the 3D distance threshold D between each power line channel. 3d The current frame tower point cloud is the filtered current frame tower point cloud; determining the power line channel where the filtered current frame tower point cloud is located is the power line channel where the current frame tower point cloud is located, including the following operation steps:

[0081] Step S231: Preset minimum vertical spacing threshold δ min; traverse the height of all line straight-line equations in the above power line height sequence list, calculate the absolute height difference of the height of two adjacent line straight-line equations in the container, if the absolute height difference is greater than the minimum vertical spacing threshold δ min , determine that the absolute height difference is an effective height difference; calculate the average value of a plurality of effective height differences (i.e. a plurality of (absolute height differences of the height of two adjacent line straight-line equations)) to obtain a line vertical spacing average value;

[0082] Step S232: preset a maximum vertical spacing threshold δ max ; determine the relationship between the line vertical spacing average value and the maximum vertical spacing threshold δ max , if the line vertical spacing average value is greater than the maximum vertical spacing threshold δ max , determine that the 2D distance threshold between each power line channel and the 3D distance threshold between each power line channel is δ max (i.e. D 2d = δ max , D 3d = δ max ); if the current line vertical spacing average value is less than the maximum vertical spacing threshold δ max , the 2D distance threshold between each power line channel and the 3D distance threshold between each power line channel is δ min (i.e. D 2d = δ min , D 3d = δ min );

[0083] Calculate the 2D distance and 3D distance of the current frame tower point cloud to all power line channels in the current frame, and select the current frame tower point cloud whose 2D distance to all power line channels in the current frame is less than the 2D distance threshold D 2d between each power line channel and whose 3D distance to all power line channels in the current frame is less than the 3D distance threshold D 3d between each power line channel as the selected current frame tower point cloud; determine that the power line channel where the selected current frame tower point cloud is located is the power line channel where the current frame tower point cloud is located.

[0084] The 2D distance threshold D 2d between each power line channel and the 3D distance threshold D 3d between each power line channel are adjusted according to the spacing between lines.

[0085] For example, there is one tower point cloud in the current frame, and there are three power line channels in the current frame, which are respectively represented as power line channel 1, power line channel 2 and power line channel 3. The line straight line equation heights of the three power line channels are 1 meter, 5 meters and 11 meters respectively, and are sorted according to the power line height sequence table. The preset minimum vertical spacing threshold is 1 meter, and the absolute height difference between the line straight line equation heights of the three power line channels is calculated according to the adjacent two line straight line equation heights in the power line height sequence table, that is, the absolute height difference is 4 meters and 6 meters respectively. At this time, the absolute height difference is greater than the minimum vertical spacing threshold of 1 meter, so both absolute height differences are effective height differences.

[0086] The average value of the two effective height differences is calculated to obtain the line vertical spacing average value of 5 meters ((4 meters + 6 meters) / 2 = 5 meters).

[0087] The preset maximum vertical spacing threshold is 3 meters, and at this time the line vertical spacing average value is greater than the maximum vertical spacing threshold of 3 meters, so the 2D and 3D distance thresholds of the tower point cloud to the three power line channels are both the maximum vertical spacing threshold of 3 meters. It is determined that one of the three power line channels whose 2D and 3D distances from the current frame tower point cloud are both less than 3 meters is the power line channel where the current frame tower point cloud is located.

[0088] If the line vertical spacing average value is calculated to be 2 meters, which is less than the maximum vertical spacing threshold of 3 meters, it is determined that the 2D and 3D distance thresholds of the tower point cloud to the three power line channels are both the line vertical spacing average value of 2 meters. It is determined that one of the three power line channels whose 2D and 3D distances from the current frame tower point cloud are both less than 2 meters is the power line channel where the current frame tower point cloud is located.

[0089] Due to the different sizes of lines, the line spacing and the line height difference are large, and when filtering using the line straight line equation (using the current frame line straight line equation), the distance threshold parameter needs to be dynamically adjusted according to the actual line situation, otherwise it is easy to cause incomplete tower filtering. For distribution network (or distribution network line), the tower of the distribution network line is relatively small, and the distance between the adjacent lines in the distribution network is small, so two values that meet the actual line specification can be directly specified, so that the tower after filtering is still complete and no additional calculation is required. For the main network, the tower range is large, and the line distance itself needs to be used for adaptive estimation. The 2D and 3D distances from the set point to the line are respectively D 2d 、D 3d ;

[0090] The "point" is a coordinate point in space (or a coordinate point of the point cloud) or a power line channel line; the 2D distance from the point to the line is the projection onto the two-dimensional plane XY, and the vertical distance from the point to the line. The 3D distance from the point to the line is the vertical distance from the spatial three-dimensional point to the spatial straight line. The above "line" is a straight line segment obtained by fitting the power line point cloud, which can be represented by a spatial straight line equation.

[0091] It should be noted that if the absolute height difference is less than the given minimum distance threshold δ min , the height difference is not considered as an effective value, and there may be split conductors or lines with close height on the other side in the line; therefore, direct filtering is not involved in the calculation.

[0092] If the absolute height difference is greater than the given minimum distance threshold δ min , the current height difference is considered as an effective value and is used for average value calculation; all effective height differences are averaged to obtain the average value of the vertical distance of the line (i.e., the estimated average value of the vertical distance of the line).

[0093] It should be noted that some lines may have estimation deviation, such as the line itself may not be complete or the algorithm may lose lines when extracting the line (with a very small probability). According to the design of some lines, we can know a maximum range, if the estimation exceeds the range, it means that the estimation has a problem, and the maximum value is used as the estimation.

[0094] Step S30: screening a plurality of seed points and candidate points from the current frame tower point cloud;

[0095] Specifically, in step S30, a plurality of seed points and candidate points are screened from the current frame tower point cloud, including the following operation steps:

[0096] Step S31: judging whether a tower has been identified before the current frame tower point cloud is acquired, according to the preset height threshold Δ h , the minimum height H min and the maximum height H max of the power line channel line, the angle between the direction of the coordinate point of the current line-following flight UAV to the current tower point cloud and the flight direction of the current line-following flight UAV, the 2D distance and 3D distance from the tower point cloud to the power line, the 2D distance threshold and 3D distance threshold between lines, and the seed point distance threshold D s to screen a plurality of seed points and candidate points;

[0097] Specifically, referring to Figure 5 In step S31, it is judged whether a tower has been identified before the current frame tower point cloud is acquired, according to the preset height threshold Δ h , the minimum height H min and the maximum height H max, an angle between a direction of a coordinate point of the current line-imitating flight unmanned aerial vehicle to the current tower point cloud and a flight direction of the current line-imitating flight unmanned aerial vehicle, a 2D distance and a 3D distance from the tower point cloud to the power line, a 2D distance threshold and a 3D distance threshold between lines, a seed point distance threshold D s The plurality of seed points and the candidate points are screened out, including the following operation steps:

[0098] Step S311: randomly selecting any one point in the tower point cloud on the current frame power line channel as a primary target point cloud;

[0099] Step S312: judging whether there is a recognized tower before the tower point cloud on the current frame power line channel is acquired, if there is a recognized tower before the tower point cloud on the current frame power line channel is acquired, acquiring the bounding box information of the last tower, judging whether the primary target point cloud is in the bounding box, if the primary target point cloud is in the bounding box, filtering the primary target point cloud and reselecting the primary target point cloud, if the primary target point cloud is not in the current bounding box, determining the current primary target point cloud as a secondary target point cloud, if there is no recognized tower before the tower point cloud on the current frame power line channel is acquired, determining the primary target point cloud as a secondary target point cloud;

[0100] For example, there are four tower point clouds A, B, C and D in the bounding box of the last tower, and the current primary target point cloud is E, then the current primary target point cloud E is determined as a secondary target point cloud and the next operation is performed, if the current primary target point cloud is D, the current primary target point cloud is filtered and the next operation is not performed.

[0101] It should be noted that when the unmanned aerial vehicle flies along the power line, the laser radar will scan the tower that has been flown over, and the tower that has been flown over has been recorded in the historical tower information, and the tower that has been flown over needs to be filtered.

[0102] Step S313: presetting a height threshold Δ h , if the height of the secondary target point cloud is greater than H max + Δ h or less than H min - Δ h , the secondary target point cloud is filtered and reselected, if the height of the secondary target point cloud is less than H max + Δ h and greater than H min - Δ h , the secondary target point cloud is determined as a tertiary target point cloud.

[0103] For example, the preset height threshold is 1 meter, the minimum height of the line straight line equation of the power line channel is 3 meters, and the maximum height of the line straight line equation of the power line channel is 10 meters. The height of the secondary target point cloud (the tower point cloud) needs to be in H min -Δ h The above (i.e., the height of the secondary target point cloud is greater than H min -Δ h ), so as to screen out the point cloud of the tree or the column-like object similar to the tower under the power line channel; the height of the secondary target point cloud needs to be in H max +Δ h The following (i.e., the height of the secondary target point cloud is less than H max +Δ h ), so as to screen out the tower point cloud on the other power line channel above the current point line channel; that is, if the height of the secondary target point cloud is less than H max +Δ h and the height of the secondary target point cloud is greater than H min -Δ h , it is determined that the secondary target point cloud is a tertiary target point cloud and the next step is performed.

[0104] It should be noted that H min and H max are the minimum and maximum values of the line channel height. Here, it can be understood that the algorithm only needs points within a certain range vertically above and below the line channel. Points outside the range are likely to be other interference, such as other lines above or trees below.

[0105] Step S314: Obtain a direction from the current unmanned aerial vehicle position coordinate to the tertiary target point cloud coordinate, and determine whether an included angle between the direction and the current unmanned aerial vehicle line simulation flight direction (the "included angle between the direction and the current unmanned aerial vehicle line simulation flight direction" refers to an included angle formed by mapping the direction and the current unmanned aerial vehicle line simulation flight direction on a plane) exceeds 90 degrees. If the included angle between the direction and the current unmanned aerial vehicle line simulation flight direction does not exceed 90 degrees, it is determined that the tertiary target point cloud is a quaternary target point cloud.

[0106] For example, taking the current unmanned aerial vehicle position coordinate as the starting point a and the tertiary target point cloud as the ending point b, the direction is represented as Calculate the degree of the included angle ∠ab between the direction and the current unmanned aerial vehicle line simulation flight direction. If the degree of the included angle ∠ab is greater than 90 degrees, the tertiary target point cloud is behind the unmanned aerial vehicle flight direction, and the tertiary target point cloud is screened out and not processed further. If the degree of the included angle ∠ab is less than 90 degrees, the tertiary target point cloud is in front of the current unmanned aerial vehicle flight direction, and it is determined as a quaternary target point cloud and the next step is performed.

[0107] It should be noted that the direction filtering is only effective for the points in the front direction of the flight direction; when the unmanned aerial vehicle is flying along the contour line, the laser radar will scan the tower that has been flown over, and the tower that has been flown over has been recorded in the historical tower information, and the tower that has been flown over needs to be filtered; the direction is calculated by using the current unmanned aerial vehicle position coordinates and the point coordinates, to determine whether the direction of the unmanned aerial vehicle to the tower is consistent with the flight direction, if consistent, then judge whether the direct included angle is more than 90 degrees, if more than, it means that the tower point is behind the unmanned aerial vehicle, otherwise in front of the unmanned aerial vehicle; we can extract the tower center as a coordinate point, the current position of the unmanned aerial vehicle as a coordinate point, and the two coordinates can form a vector, in addition, the direction of flight is known, which is also a vector, then according to the included angle cosine formula of two vectors, the included angle size can be calculated;

[0108] Here we generally only calculate two-dimensional plane, without considering the vertical direction, after all, it is only to determine whether the aircraft is currently in front of the tower or behind the tower. We will complete the tower identification work before passing through the tower, if the aircraft has flown over the tower, then at this time, even if the laser can still scan the point cloud of this tower, this tower has been recorded in the historical tower information, and this point cloud interference should be excluded.

[0109] Step S315: determining the 2D distance D of the tower point cloud to the power line 2d 2D distance threshold value of the alternative point, 3D distance D of the tower point cloud to the power line 3d 3D distance threshold value of the alternative point, and preset seed point distance threshold value D s ; calculating the 2D distance and 3D distance of the four-level target point cloud to the power line, if the 2D distance of the four-level target point cloud to the power line is less than or equal to the 2D distance threshold value D of the alternative point 2d And the 3D distance of the four-level target point cloud to the power line is less than or equal to the 3D distance threshold value D of the alternative point 3d , then determine that the four-level target point cloud is an alternative point; if the 2D and 3D distances of the four-level target point cloud to the power line are less than or equal to the seed point distance threshold value D s , then determine that the four-level target point cloud is a seed point;

[0110] For example, the preset seed point distance threshold is 0.5 meters, the 2D distance threshold of the candidate point is 3 meters, and the 3D distance threshold of the candidate point is 5 meters; the four-level target point cloud has five points, which are respectively represented as points Q, W, E, R and T; the 2D and 3D distances of the points Q, W, E, R and T to the current power line channel are calculated respectively, which are represented as points Q(0.1 meters, 0.3 meters), W(0.4 meters, 0.5 meters), E(6 meters, 7 meters), R(1 meter, 4 meters) and T(3 meters, 6 meters); because 0.1 meters is less than the seed point distance threshold 0.5 meters and 0.3 meters is less than the seed point distance threshold 0.5 meters, the point Q is determined as a seed point; 0.4 meters is less than the seed point distance threshold 0.5 meters and 0.5 meters is equal to the seed point distance threshold 0.5 meters, so the point W is determined as a seed point; 6 meters is greater than the candidate point 2D distance threshold 3 meters and 7 meters is greater than the candidate point 3D distance threshold, so the point E is screened out and is not processed again; 1 meter is less than the candidate point 2D distance threshold 3 meters and 4 meters is less than the candidate point 3D distance threshold 5 meters, so the point R is determined as a candidate point; 3 meters is equal to the candidate point 2D distance threshold but 6 meters is greater than the candidate point 3D threshold, so the point T is screened out and is not processed again; the above Q(0.1 meters, 0.3 meters) means that the 2D distance of the four-level target point cloud Q to the current power line channel is 0.1 meters and the 3D distance of the four-level target point cloud Q to the current power line channel is 0.3 meters; the W(0.4 meters, 0.5 meters), E(6 meters, 7 meters), R(1 meter, 4 meters) and T(3 meters, 6 meters) have the same meaning as Q(0.1 meters, 0.3 meters) and will not be described again.

[0111] It should be noted that the point-to-line shortest distance filtering; setting different filtering thresholds divides the tower point cloud into tower points, candidate points and invalid points. The estimated distance D 2d and D 3d are used as candidate point thresholds, and a smaller distance threshold D s is set as a seed point threshold, only the points less than the threshold can be used as seed points. All line equations in the channel are traversed, the 2D and 3D distances of the points to all lines are calculated, and the minimum distance is used for judgment; the points with the minimum distance within D 2d and D 3d are used as candidate points, and the points within D s are used as seed points.

[0112] The seed point is a point that is more consistent with the tower relative to the alternative point. If some points are very close to the line on the channel, such as less than 0.5 meters, it means that the point is basically on the line, and the actual line tower must be connected to the line, and it is almost impossible to be other interference points, such as tower points near the line, trees, street lamp poles and other interference; the threshold of the alternative point is set to be larger, for example, 5 meters, so that the point cloud on the surrounding line, trees or other interference may be added; if the seed point exists, it means that the tower really exists, and the alternative point is more for the comprehensive recovery of the shape of the tower. A smaller threshold may result in an incomplete tower, and in addition, the size of the tower in the distribution network line varies greatly, some have a radius of only 1 meter, and some may have 4 meters. This real situation also needs a large range of filtering; if both exist, the seed point and the alternative point must overlap, so as to discover whether the tower exists by using the seed point, and recover the real shape of the tower through the overlapping relationship between the seed point and the alternative point.

[0113] Step S40: a plurality of alternative points and a plurality of seed points are obtained continuously in multiple frames, and the plurality of alternative points and the plurality of seed points are accumulated and merged to obtain a plurality of merged plurality of alternative points and a plurality of seed points; the plurality of merged plurality of alternative points and the plurality of seed points are filtered through a voxel grid to obtain sampled seed points and alternative points; DBSCAN density clustering method is used for clustering processing on the sampled alternative points and seed points, and a minimum threshold N s of the number of point clouds in a seed point cloud cluster is preset, a plurality of alternative point cloud clusters and seed point cloud clusters are obtained;

[0114] Specifically, referring to Figure 6 In step S40, a plurality of alternative points and a plurality of seed points are obtained continuously in multiple frames, and the plurality of alternative points and the plurality of seed points are accumulated and merged to obtain a plurality of merged plurality of alternative points and a plurality of seed points; the plurality of merged plurality of alternative points and the plurality of seed points are filtered through a voxel grid to obtain sampled seed points and alternative points; DBSCAN density clustering method is used for clustering processing on the sampled alternative points and seed points, and a minimum threshold N s of the number of point clouds in a seed point cloud cluster is preset, a plurality of alternative point cloud clusters and seed point cloud clusters are obtained, including the following operation steps:

[0115] Step S41: the plurality of alternative points and the plurality of seed points obtained continuously in multiple frames are added to a historical point set for accumulation and merging to obtain merged plurality of alternative points and merged plurality of seed points; the historical point set is a set in which the alternative points and the seed points obtained in each frame are accumulated together, that is, the accumulation and merging are put into the historical point set, and the merging also means putting without other operation meanings;

[0116] Step S42: using a voxel grid filter (or a kdtree filter) to sample the merged multiple candidate points and the merged multiple seed points respectively to obtain different multiple candidate points and multiple seed points, and to filter out repeated candidate points and seed points to obtain sampled seed points and candidate points;

[0117] It should be noted that the sampling processing of the merged seed points and the merged candidate points can be processed by using a commonly used method such as a voxel grid or a kdtree (the voxel grid or the kdtree is a prior art, and will not be described again) to obtain the sampled seed points and the candidate points.

[0118] The voxel grid or the kdtree can prevent the situation that some points are too dense from causing a judgment error; because we determine whether the tower exists by the number of seed points, and the number still reaches the quantity after sampling, it indicates that the actual volume of the point cloud is large, and the volume of the tower is large, and therefore it is reasonable; on the contrary, if no sampling is performed, the laser may scan the same position multiple times, causing the point cloud to be gathered, a large number of repeated points are gathered in a very small range, and sometimes a small branch next to the line may be identified as a tower; after sampling, a number of points also represent the actual size of the space occupied by the target, and the interference can be reduced.

[0119] Step S43: using a DBSCAN density clustering method to cluster the sampled candidate points to obtain multiple candidate point cloud clusters; and using the DBSCAN density clustering method to cluster the sampled seed points to obtain multiple seed point cloud clusters to be determined;

[0120] It should be noted that the DBSCAN density clustering method is a prior art, and will not be described again;

[0121] The DBSCAN is a conventional density clustering method; because the range of the candidate points is large, the internal points may be the tower points of the current channel, the tree points, or the tower points of other lines; because the current line always has a clearance distance, that is, the tower of the current line is necessarily isolated from the point cloud of the surrounding environment or the point cloud of the adjacent line in space; after a small and appropriate spatial distance threshold is given, the DBSCAN can automatically cluster the surrounding point cloud groups to obtain multiple point cloud clusters; because the parameter setting of the seed points is small, the distance between the line is wide, and multiple seed clusters exist, but this does not affect the final result, and we only need to select the largest one; the seed points are a subset of the candidate points; if the seed points and the candidate points overlap in a cluster, it indicates that the cluster is the required tower point cloud cluster, and the remaining interference clusters are necessarily tree, street lamp, or tower point cloud clusters of adjacent lines.

[0122] Step S44: presetting a minimum threshold N of the number of point clouds in the seed point cloud clusters determining whether the number of point clouds in the seed point cloud cluster to be determined is greater than a minimum threshold N of the number of point clouds in the seed point cloud cluster s If yes, the seed point cloud cluster to be determined is determined as a target seed point cloud cluster to be selected, and the target seed point cloud cluster to be selected with the maximum number of point clouds is selected as the seed point cloud cluster

[0123] It should be noted that if the seed points are clustered without gathering at least one seed point cloud cluster, or the number of point clouds in the largest point cloud cluster is less than a threshold N s , it is considered that the current frame still does not exist the tower, and the current seed point cloud cluster can be determined as the point cloud of the column-like tower-like ground object, which is screened out and not processed any more; otherwise, the point cloud cluster with the maximum number of points is taken as the final seed point cloud cluster; being available as a seed point means being in a position very close to the line, because each channel has multiple lines, and the distance between the lines may be too small due to the setting, while the line spacing is relatively large, for example, the distance between the set seed point and the line is 1 meter, and the actual line spacing is 5 meters, so there may be a seed point set for each line; no matter which one is selected, as long as the set overlaps with the current tower to reach a threshold, it is considered to be successfully identified, and we can preferentially select the largest overlap (i.e., the above-mentioned "selecting the target seed point cloud cluster to be selected with the maximum number of point clouds as the seed point cloud cluster"), which is more stable; especially near the boundary, it may be close to other towers, and the calculated overlap rate will be relatively small.

[0124] Step S50: a minimum threshold k of overlap rate is preset, and the seed point cloud cluster and the clustering result of the candidate point cloud cluster are combined to obtain a bounding box of the seed point cloud cluster and a bounding box of the candidate point cloud cluster;

[0125] The volumes (bounding box coverage) of the bounding boxes of the seed point cloud cluster and the candidate point cloud cluster are calculated respectively, and it is judged whether the bounding box of the seed point cloud cluster overlaps with the bounding box of the candidate point cloud cluster, if there is an overlap (the bounding box of the seed point cloud cluster overlaps with the bounding box of the candidate point cloud cluster), the overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster is calculated;

[0126] If the overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster is greater than the minimum threshold k of the overlap rate, it is determined that the candidate point cloud cluster corresponding to the bounding box of the candidate point cloud cluster is an effective tower point cloud cluster;

[0127] It should be noted that if there is an overlap (the bounding box of the seed point cloud cluster overlaps with the bounding box of the candidate point cloud cluster), it indicates that the candidate point and the seed point have a coincident part, which indicates that the current candidate point cloud cluster can be used to restore the shape of the tower, and is taken as an effective tower point cloud cluster; a size κ of the overlap rate is set, if the planar overlap rate of the candidate point cloud cluster and the seed point cloud cluster is less than κ, the candidate point cloud cluster is invalid, otherwise, it is added to the effective tower point cloud cluster.

[0128] Step S60: calculate the overlapping plane area of each of the effective tower point cloud cluster and the seed point cloud cluster (the overlapping plane area refers to the area of the plane formed by the corresponding candidate point cloud cluster of the effective tower point cloud cluster and the seed point cloud cluster mapped onto the plane), and determine the tower point cloud cluster with the maximum overlapping plane area as the target tower point cloud cluster;

[0129] Traverse all point clouds in the target tower point cloud cluster, calculate the tower bounding box volume (volume) and tower coordinates, and extract the highest point of the target tower point cloud cluster as the tower center highest point ("calculate the tower bounding box volume (volume) and tower coordinates, and extract the highest point of the target tower point cloud cluster" is to determine the volume, position and highest point of the to-be-determined tower).

[0130] As can be seen from the robust unmanned aerial vehicle laser radar line-imitating tower identification method and device provided in the examples of the present application, in specific application, first, the power line channel and the tower of each frame of data are extracted in the flight process, the current frame of all power line channel data and the current frame of tower point cloud are obtained in real time according to the laser radar module, and the current frame of tower point cloud is obtained from the current frame of all tower point cloud; a power line channel in which the current frame of tower point cloud is located is determined as the power line channel in which the current frame of tower point cloud is located, through the improved operation mode, the influence of other power line channels on tower identification can be screened out, and then the power line channel in which the current frame of tower point cloud is located can be determined; on the basis of the determination of the power line channel in which the current frame of tower point cloud is located, a plurality of seed points and candidate points are screened out from the current frame of tower point cloud; other object point clouds (tree point clouds, columnar ground object point clouds similar to towers) are screened out, and the tower point cloud of the identified tower is screened out, so as to ensure the accuracy of the identification of the current tower;

[0131] Then, a plurality of candidate points and a plurality of seed points are obtained in a plurality of continuous frames, and the plurality of candidate points and the plurality of seed points are accumulated and merged to obtain a plurality of merged plurality of candidate points and a plurality of seed points; the plurality of merged plurality of candidate points and the plurality of seed points are filtered through a voxel grid to obtain sampled seed points and candidate points; in the prior art, the laser radar scans the same position multiple times to obtain a plurality of repeated tower point clouds, causing point cloud aggregation, the technical scheme provided in the embodiment of the present application samples the obtained plurality of candidate points and plurality of seed points to obtain a small but insufficient amount of tower point clouds, thereby improving the identification accuracy of the tower.

[0132] The sampled candidate points and seed points are clustered by using a DBSCAN density clustering method, and a minimum threshold N of the number of point clouds in a seed point cloud cluster is preset s, obtain a plurality of candidate point cloud clusters and a seed point cloud cluster; in the prior art, when a laser radar scans a tower point cloud, it will also scan the point cloud of a columnar ground object similar to the tower beside the tower and the point cloud of a branch, but generally, there is a spatial isolation between the tower and the columnar ground object similar to the tower beside the tower and the branch, and after a value is given, DBSCAN can obtain point cloud clusters by clustering point clouds;

[0133] According to a minimum overlap rate threshold k, the clustering result of the seed point cloud cluster and the candidate point cloud cluster is merged to obtain a bounding box of the seed point cloud cluster and a bounding box of the candidate point cloud cluster; the volume of the bounding box of the seed point cloud cluster and the volume of the bounding box of the candidate point cloud cluster are calculated respectively, and it is judged whether the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster overlap; if they overlap, the overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster is calculated; if the overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster is greater than the minimum overlap rate threshold k, it is determined that the candidate point cloud cluster corresponding to the bounding box of the candidate point cloud cluster is an effective tower point cloud cluster; the overlapping plane area of each effective tower point cloud cluster and the seed point cloud cluster is calculated, and the tower point cloud cluster with the maximum overlapping plane area is determined as a target tower point cloud cluster; in the embodiment of the application, the seed point cloud cluster is a subset of the candidate point cloud cluster, so the overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster is relatively large, and a plurality of effective tower point cloud clusters can be obtained by using the technical solution provided in the embodiment of the application.

[0134] Finally, all tower points in the target tower point cloud cluster are traversed, the volume of the tower bounding box and the tower coordinates are calculated according to the coordinates of the tower points, and the highest point of the target tower point cloud cluster is extracted as the center highest point of the tower.

[0135] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the application, but not to limit them; a person skilled in the art can modify the technical solutions described in the above embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the application.

Claims

1. A robust unmanned aerial vehicle (UAV) laser-radar (ladar) catenary wire emulation tower identification method, comprising: The method comprises the following operation steps: Real-time acquisition of all power line channel data of a current frame and all tower point clouds of the current frame, and acquisition of tower point clouds of the current frame from all tower point clouds of the current frame; Screening and determining a power line channel where the tower point clouds of the current frame are located as the power line channel where the tower point clouds of the current frame are located; Screening a plurality of seed points and candidate points from the tower point clouds of the current frame; Continuous multi-frame acquisition of a plurality of candidate points and a plurality of seed points, and accumulation and merging of the plurality of candidate points and the plurality of seed points to obtain a plurality of merged candidate points and seed points; Filter the plurality of candidate points and the plurality of seed points through a voxel grid to obtain sampled seed points and candidate points; cluster the sampled seed points and candidate points by using a DBSCAN density clustering method, and preset a minimum threshold N of the number of point clouds in a seed point cloud cluster s , obtain a plurality of candidate point cloud clusters and seed point cloud clusters; Predefining a minimum overlap rate threshold k, and merging the seed point cloud cluster and the candidate point cloud cluster to obtain a bounding box of the seed point cloud cluster and a bounding box of the candidate point cloud cluster; Respectively calculating the volumes of the bounding boxes of the seed point cloud cluster and the candidate point cloud cluster, and judging whether the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster overlap, if they overlap, calculating the overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster; Determining that the overlap rate of the bounding box of the seed point cloud cluster and the bounding box of the candidate point cloud cluster is greater than the minimum overlap rate threshold k, and determining that the candidate point cloud cluster corresponding to the bounding box of the candidate point cloud cluster is an effective tower point cloud cluster; Calculating the overlapping plane area of each effective tower point cloud cluster and the seed point cloud cluster, and determining the tower point cloud cluster with the maximum overlapping plane area as a target tower point cloud cluster; Traversing all point clouds in the target tower point cloud cluster, calculating the tower bounding box volume and the tower coordinates, and extracting the highest point of the target tower point cloud cluster as the tower center highest point.

2. The robust unmanned aerial vehicle lidar catenary wire mimicking pole tower identification method of claim 1, wherein, The screening and determining of the power line channel where the tower point clouds of the current frame are located as the power line channel where the tower point clouds of the current frame are located comprises the following steps: Acquiring the line straight line equation in the current frame all power line channel data; The height of the line linear equation is saved into a container and sorted to obtain a power line height sequence table; the minimum height H of the power line channel line is obtained through the power line height sequence table min and the maximum height H max ; preset minimum vertical spacing threshold δ min with maximum vertical spacing threshold δ max , according to the above power line height sequence table, calculating the line vertical spacing average value; according to the line vertical spacing average value, the minimum vertical spacing threshold δ min with maximum vertical spacing threshold δ max , calculating the 2D distance threshold D 2d between each power line channel and the 3D distance threshold D 3d between each power line channel; Obtain a current frame tower point cloud from all tower point clouds in the current frame, calculate 2D distances and 3D distances of the current frame tower point cloud to all power line channels in the current frame respectively, and screen the current frame tower point cloud whose 2D distance to all power line channels in the current frame is less than a 2D distance threshold D between each power line channel 2d and whose 3D distance to all power line channels in the current frame is less than a 3D distance threshold D between each power line channel 3d The current frame tower point cloud after screening is the screened current frame tower point cloud; and one power line channel where the screened current frame tower point cloud is located is determined as a power line channel where the current frame tower point cloud is located.

3. The robust unmanned aerial vehicle lidar catenary wire mimicking pylon identification method of claim 2, wherein, The screening of a plurality of seed points and candidate points from the tower point clouds of the current frame comprises the following steps: determining whether a tower has been identified before acquiring the current frame tower point cloud, according to a preset height threshold Δ h , minimum height H of the power line channel line min , and maximum height H max , the angle between the direction of the coordinate point of the current line-imitating flight unmanned aerial vehicle to the current tower point cloud and the flight direction of the current line-imitating flight unmanned aerial vehicle, the 2D distance and 3D distance from the tower point cloud to the power line, the 2D distance threshold and 3D distance threshold between lines, the seed point distance threshold D s Screening multiple seed points and candidate points.

4. The robust unmanned aerial vehicle lidar catenary wire mimicking pylon identification method of claim 3, wherein, The preset minimum vertical spacing threshold δ min The maximum vertical spacing threshold δ max According to the above power line height sequence table, the line vertical spacing average value is calculated; according to the line vertical spacing average value, the minimum vertical spacing threshold δ min The maximum vertical spacing threshold δ max The 2D distance threshold D 2d And the 3D distance threshold D 3d Between each power line channel is calculated; the 2D distance and the 3D distance of the current frame tower point cloud to all power line channels in the current frame are calculated, and the current frame tower point cloud whose 2D distance to all power line channels in the current frame is less than the 2D distance threshold D 2d And the 3D distance of the current frame tower point cloud to all power line channels in the current frame is less than the 3D distance threshold D 3d Between each power line channel is screened; the current frame tower point cloud after screening is the current frame tower point cloud after screening; the power line channel where the current frame tower point cloud after screening is located is determined as the power line channel where the current frame tower point cloud is located, including the following operation steps: a preset minimum vertical spacing threshold δ min ; traversing the heights of all line straight-line equations in the above power line height sequence list, calculating the absolute height difference between the heights of two adjacent line straight-line equations in the container, if the absolute height difference is greater than the minimum vertical spacing threshold δ min , determining that the absolute height difference is an effective height difference; calculating the average value of a plurality of the effective height differences to obtain a line vertical spacing average value; a preset maximum vertical distance threshold δ max ; judging the relationship between the average value of the vertical distance of the lines and the maximum vertical distance threshold δ max , if the average value of the vertical distance of the lines is greater than the maximum vertical distance threshold δ max , then determining that the 2D distance threshold between each power line channel and the 3D distance threshold between each power line channel are δ max ; if the current average value of the vertical distance of the lines is less than the maximum vertical distance threshold δ max , then the 2D distance threshold between each power line channel and the 3D distance threshold between each power line channel are δ min ; The 2D distance and 3D distance of the current frame tower point cloud to all power line channels in the current frame are calculated respectively, and the 2D distance of the current frame tower point cloud to all power line channels in the current frame is screened to be less than the 2D distance threshold D between each power line channel 2d And the 3D distance of the current frame tower point cloud to all power line channels in the current frame is less than the 3D distance threshold D between each power line channel 3d The current frame tower point cloud after screening is the current frame tower point cloud after screening, and the power line channel where the current frame tower point cloud after screening is located is determined as the power line channel where the current frame tower point cloud is located.

5. The robust unmanned aerial vehicle lidar catenary wire mimicking pylon identification method of claim 4, wherein, The judgment is whether a tower has been identified before acquiring a current frame tower point cloud, according to a preset height threshold Δ h , the minimum height H of the power line channel line min , and the maximum height H max , the angle between the direction of the coordinate point of the current line-imitating flight unmanned aerial vehicle to the current tower point cloud and the flight direction of the current line-imitating flight unmanned aerial vehicle, the 2D distance and 3D distance of the tower point cloud to the power line, the 2D distance threshold and 3D distance threshold between lines, the seed point distance threshold D s The plurality of seed points and candidate points are screened out, including the following operation steps: Randomly selecting any one point in the tower point cloud on the current frame power line channel as a first target point cloud; Judging whether there is a recognized tower before acquiring the tower point cloud on the current frame power line channel, if there is a recognized tower before acquiring the tower point cloud on the current frame power line channel, acquiring the bounding box information of the last tower, judging whether the first target point cloud is in the bounding box, if the first target point cloud is in the bounding box, filtering it and reselecting the first target point cloud, if the first target point cloud is not in the current bounding box, determining the current first target point cloud as a second target point cloud, if there is no recognized tower before acquiring the tower point cloud on the current frame power line, determining the first target point cloud as a second target point cloud; a preset height threshold value Δ h ; if the height of the secondary target point cloud is greater than H max + Δ h or less than H min - Δ h , the secondary target point cloud is filtered and reselected. if the height of the secondary target point cloud is less than H max + Δ h and greater than H min - Δ h then the secondary target point cloud is determined to be a tertiary target point cloud; Acquiring the direction from the current UAV position coordinates to the third target point cloud coordinates, judging whether the angle between the direction and the current UAV line simulation flight direction exceeds 90 degrees, if the angle between the direction and the current UAV line simulation flight direction does not exceed 90 degrees, determining the third target point cloud as a fourth target point cloud; Determine the 2D distance D of the tower point cloud to the power line 2d Determine the 3D distance D of the tower point cloud to the power line 3d Determine the 3D distance threshold D of the candidate point, and preset the seed point distance threshold D s Calculate the 2D distance and 3D distance of the four-level target point cloud to the power line, if the 2D distance of the four-level target point cloud to the power line is less than or equal to the 2D distance threshold D of the candidate point 2d And the 3D distance of the four-level target point cloud to the power line is less than or equal to the 3D distance threshold D of the candidate point 3d Then determine the four-level target point cloud as the candidate point; if the 2D and 3D distances of the four-level target point cloud to the power line are less than or equal to the seed point distance threshold D s Then determine the four-level target point cloud as the seed point.

6. The robust unmanned aerial vehicle lidar catenary wire mimicking pole tower identification method of claim 5, wherein, The continuous multi-frame acquisition of a plurality of candidate points and a plurality of seed points, and the accumulation and merging of the plurality of candidate points and the plurality of seed points to obtain a plurality of merged candidate points and seed points; Filter the combined plurality of candidate points and the plurality of seed points through a voxel grid to obtain sampled seed points and candidate points; cluster the sampled candidate points and seed points using a DBSCAN density clustering method, and preset a minimum threshold N of the number of point clouds in a seed point cloud cluster s , obtain a plurality of candidate point cloud clusters and seed point cloud clusters, including the following operation steps: The plurality of candidate points and the plurality of seed points obtained by continuous frames are added to the historical point set for accumulation and merging, and a plurality of merged candidate points and a plurality of merged seed points are obtained; The plurality of candidate points and the plurality of seed points obtained by continuous frames are added to the historical point set for accumulation and merging, and a plurality of merged candidate points and a plurality of merged seed points are obtained; The plurality of candidate points and the plurality of seed points obtained by continuous frames are added to the historical point set for accumulation and merging, and a plurality of merged candidate points and a plurality of merged seed points are obtained; A preset minimum threshold N of a point cloud number in a seed point cloud cluster s , determining whether the point cloud number in the to-be-determined seed point cloud cluster is greater than the minimum threshold N of the point cloud number in the seed point cloud cluster s , if yes, determining that the to-be-determined seed point cloud cluster is a to-be-selected target seed point cloud cluster; and screening a to-be-selected target seed point cloud cluster with the largest point cloud number as a seed point cloud cluster.

7. The robust unmanned aerial vehicle lidar catenary wire mimicking pylon identification method of claim 6, wherein, The bounding box information of the tower includes all tower point clouds constituting the tower.

8. The robust unmanned aerial vehicle lidar catenary wire mimicking pylon identification method of claim 6, wherein, said minimum vertical spacing threshold δ min is a constant; said maximum vertical spacing threshold δ max is a constant.

9. The robust unmanned aerial vehicle lidar catenary wire mimicking pylon identification method of claim 6, wherein, a 2D distance threshold D between each power line channel 2d and a 3D distance threshold D between each power line channel 3d is adjusted according to the spacing variation between lines, the 2D distance threshold D between each power line channel 2d is a minimum vertical spacing threshold δ min or a maximum vertical spacing threshold δ max ; the 3D distance threshold D between each power line channel 3d is a minimum vertical spacing threshold δ min or a maximum vertical spacing threshold δ max ; the seed point distance threshold D s is a constant; the height threshold Δ h is a constant; the minimum threshold N of the number of point clouds in the seed point cloud cluster s is a constant; and the minimum threshold k of the overlap rate is a constant.

10. A robust unmanned aerial vehicle lidar wire-like pole tower detection device, characterized in that, The unmanned aerial vehicle comprises a laser radar module, a central processing unit and a GPS positioning module. The central processing unit is connected with the laser radar module and the GPS positioning module. The laser radar module is used for acquiring all power line channel data of a current frame and tower point clouds of the current frame in real time. The current frame power line channel data includes line straight line equations of all power line channels of the current frame, and the tower point clouds include coordinates of tower point clouds and heights of the tower point clouds. The central processing unit is further used for merging seed point cloud cluster and candidate point cloud cluster clustering results to obtain a seed point cloud cluster bounding box and a candidate point cloud cluster bounding box according to a minimum overlap rate threshold k. The GPS positioning module is used to acquire the coordinate position of the line-imitating flight unmanned aerial vehicle; the central processor is used to acquire all power line channel data and current frame tower point cloud in real time according to the laser radar module, acquire the line height sequence table of the power line through the straight line equation of the current power line channel, and further acquire the minimum height H of the power line channel line min and the maximum height H max of the power line channel line; the 2D distance threshold and the 3D distance threshold between the circuit lines are determined through the circuit line height sequence table; and the power line channel in which the current frame tower point cloud is located is determined as the power line channel in which the current frame tower point cloud is located. determining whether a tower has been identified before acquiring the current frame tower point cloud, according to a preset height threshold Δ h , the minimum height H of the power line channel line min , and the maximum height H max , the angle between the direction of the coordinate point of the current line-imitating flight unmanned aerial vehicle to the current tower point cloud and the flight direction of the current line-imitating flight unmanned aerial vehicle, the 2D distance and 3D distance from the tower point cloud to the power line, the 2D distance threshold D 2d and the 3D distance threshold D 3d between the lines s , the seed point distance threshold D screening a plurality of seed points and candidate points; The central processing unit is further configured to continuously acquire a plurality of candidate points and a plurality of seed points in a plurality of frames, accumulate and merge the plurality of candidate points and the plurality of seed points to obtain a plurality of merged candidate points and seed points, filter the plurality of merged candidate points and seed points through a voxel grid to obtain sampled seed points and candidate points, and perform clustering processing on the sampled candidate points and seed points by using a DBSCAN density clustering method, and preset a minimum threshold N of a number of point clouds in a seed point cloud cluster s , and acquire a plurality of candidate point cloud clusters and seed point cloud clusters. The volumes of the seed point cloud cluster bounding box and the candidate point cloud cluster bounding box are calculated, and it is judged whether the seed point cloud cluster bounding box and the candidate point cloud cluster bounding box have an overlap. If the seed point cloud cluster bounding box and the candidate point cloud cluster bounding box have an overlap, the overlap rate of the seed point cloud cluster bounding box and the candidate point cloud cluster bounding box is calculated. If the overlap rate of the seed point cloud cluster bounding box and the candidate point cloud cluster bounding box is greater than the minimum overlap rate threshold k, it is determined that the candidate point cloud cluster corresponding to the candidate point cloud cluster bounding box is an effective tower point cloud cluster. The overlapping plane area of each effective tower point cloud cluster and the seed point cloud cluster is calculated, and a tower point cloud cluster with the maximum overlapping plane area is determined as a target tower point cloud cluster. The central processing unit is further used for traversing all tower point clouds in the target tower point cloud cluster, calculating a tower bounding box volume and a tower coordinate according to the coordinates of the tower point clouds, and extracting a highest point of the target tower point cloud cluster as a tower center highest point.

Images