A method for identifying and locating insulator segments on a suspension insulator string based on solid-state lidar point cloud

Abstract

The present invention discloses a method for identifying and locating an insulator segment group on a suspension insulator string based on a solid-state laser radar point cloud, comprising the following steps: solid-state radar point cloud preprocessing; point cloud clustering; and point cloud cluster Q1 for each cluster of point cloud q1. i Perform feature calculations, including fitting the circle center coordinates qc i , fitting circle radius qr i , the plane normal vector qn i Insulator point cloud screening. This invention uses feature extraction technology to screen out suspension insulator segments from the point cloud and calculate their position in space. This solves the problem of traditional template matching methods being unable to identify and locate suspension insulators when the point cloud shape has low similarity with the template, providing an important perception solution for subsequent operations.

Description

Technical Field

[0001] The present invention belongs to the field of computer vision technology, and in particular relates to a method for identifying and locating an insulator segment group on a suspension insulator string based on a solid-state laser radar point cloud. Background Art

[0002] With the continuous development of LiDAR technology, the technology for identifying and locating specific target objects from 3D point cloud data collected by LiDAR has found widespread application in fields such as robotic perception and autonomous driving. Common methods for identifying objects from 3D point clouds include template matching based on feature descriptions and iterative closest point algorithms. Both methods rely heavily on the degree of similarity between the template and the target object in the actual point cloud to be detected. However, due to factors such as the external environment, the scanning position and angle, and the limitations of the scanning equipment itself, it is impossible to obtain a complete surface point cloud of the circular insulator segment in the 2.5D point cloud to be detected. Moreover, the missing parts have different shapes, making it impossible to use a unified template for matching. Summary of the Invention

[0003] Purpose of the invention: In order to solve the technical problems raised in the background technology, the present invention discloses a method for identifying and locating an insulator segment group on a suspension insulator string based on a solid-state laser radar point cloud.

[0004] Technical solution: The present invention discloses a method for identifying and locating an insulator segment group on a suspension insulator string based on a solid-state laser radar point cloud, comprising the following steps:

[0005] S1, solid-state radar point cloud preprocessing;

[0006] S2. Point cloud clustering

[0007] S2-1, retain the dense entity point cloud through Euclidean clustering, remove non-entity invalid noise points, and obtain point cloud P1;

[0008] S2-2, calculate the normals of all points in the point cloud P1, N1 = {n i |i=1,2…s1}, where s1 is the size of point cloud P1;

[0009] S2-3, through normal filtering, retain the solid surface point cloud facing the radar, that is, the normal vector n i (n ix , n iy , n iz ) in the X-direction component n ix <0;

[0010] S2-4, obtain the point cloud set again through Euclidean clustering;

[0011] S2-5. By calculating the centroid of the point cloud clusters in the set, sort them in order of distance from the radar from near to far, and merge the point cloud clusters whose distance in the radar scanning direction is less than the threshold and whose centroid distance is less than the diameter of the insulator sheet into the same point cloud cluster, and finally obtain the point cloud cluster set to be screened Q1 = {q i |i=1,2…s q1}, where s q1 is the size of the point cloud cluster set;

[0012] S3, for each cluster of point cloud q in point cloud cluster Q1 i Perform feature calculations, including fitting the circle center coordinates qc i , fitting circle radius qr i , the plane normal vector qn i ;

[0013] S4. Insulator point cloud screening

[0014] S4-1. Traverse all point cloud clusters q i , filter out the point cloud cluster set that meets the characteristics of the insulator piece to be identified, and sort them in order from near to far from the radar, and obtain the insulator piece point cloud cluster set Q v ;

[0015] S4-2. Calculate the insulator point cloud cluster set Q v Adjacent insulator segments (q i ,q i+1 ) center (qc i ,qc i+1 ) between the Euclidean distance, according to the threshold d p Merge the point cloud clusters belonging to the same insulator group and record the position of the first insulator (qc i ,qn i ) is the position of the insulator group, where d p The distance between the two pieces of the reference target insulator is set, and finally all the insulator piece groups that meet the conditions are obtained.

[0016] Specifically, in S1, point clouds outside the target range are removed by point cloud cropping; voxel filtering and downsampling are performed to remove duplicate points within the voxel.

[0017] Furthermore, S3 includes the following steps:

[0018] S3-1. Calculate the point cloud normal N q ={n m |m=1,2…s qi}, where s qi is the point cloud cluster q i size;

[0019] S3-2, extract the entity surface boundary point cloud P by the normal angle threshold b , when the point normal vector n m With k b The normal vectors of the adjacent points {n1…n k} are all greater than the threshold value of 150°, then the point is considered to be a boundary point, where k b Adjustable parameters for the boundary extraction algorithm;

[0020] S3-3, the boundary point cloud P b Projected to the YZ plane, the two-dimensional boundary point cloud P on the YZ plane is obtained b ';

[0021] S3-4, using the two-dimensional boundary point cloud P b 'The highest point P in the Z direction zmax As the starting point, find the adjacent points through KD-Tree, grow the upper boundary point cloud in the Y+ direction and Y- direction respectively, and extract the original point cloud P b The corresponding points on the solid surface are obtained by the boundary point cloud P bu ;

[0022] S3-5. Fitting the upper boundary point cloud P by random sampling method bu The plane in which it is located, the normal vector qn is obtained i , and project the upper boundary point cloud onto the plane, and then rotate it to the XY plane to obtain the 2D point cloud P bupr ;

[0023] S3-6, point cloud P bupr By fitting the circle with least squares, we can get the center qc pr and the insulator segment radius qr i ;

[0024] S3-7, the center of the projection plane qc pr After converting back to the plane where the upper boundary point cloud is located, the center position qc of the insulator segment is obtained i .

[0025] Furthermore, qc in S3-7 i =T i *qc pr , where T i From the unit vector (0,0,1) to the normal vector qn i The transformation matrix.

[0026] Furthermore, in S4-1, the radius of the fitting circle is close to the target insulator segment radius, i.e., qr i ∈[0.8R,R], R is the actual radius of the target insulator, and the point cloud fitting coefficient n f / n q >0.9, where nq is the point cloud cluster q i The number of points in n f In order to meet the target insulator segment size point cloud number, the calculation method is to use the insulator segment diameter 2R as the length and width, and the insulator segment thickness D as the thickness of the bounding box to cut the point cloud cluster q i Get point cloud cluster q if ={p m |m=1,2…n f};

[0027] Insulator sheet point cloud cluster set Q v ={q i |i=1,2…s v}, where s v is the number of point cloud clusters that meet the filtering conditions.

[0028] Beneficial effects: Compared with the existing technology: The present invention uses feature extraction technology to screen out the suspension insulator segment group in the point cloud and calculate its position in space, which solves the problem that the traditional template matching method cannot be used to identify and locate the suspension insulator when the similarity between the point cloud shape and the template is low, and provides an important perception solution for the subsequent operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Flowchart for the implementation of the present invention;

[0030] Figure 2 This is the point cloud image after cropping and downsampling in the present invention;

[0031] Figure 3 This is a schematic diagram of a point cloud cluster set according to the present invention;

[0032] Figure 4 It is the point cloud cluster and the upper boundary map of the point cloud of the present invention;

[0033] Figure 5 This is a schematic diagram of point cloud screening of insulator segments according to the present invention. DETAILED DESCRIPTION

[0034] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0035] The present invention uses a synthetic point cloud obtained by integrating a solid-state laser radar for 3 seconds as input point cloud. The radar scanning direction is consistent with the direction of the insulator string to be detected, and the radar coordinate system is the reference coordinate system.

[0036] like Figure 1 A method for identifying and locating an insulator segment group on a suspension insulator string based on a solid-state laser radar point cloud is shown, comprising the following steps:

[0037] S1. Solid-state radar point cloud preprocessing

[0038] Specifically, point clouds outside the recognition target range are removed through point cloud clipping;

[0039] Voxel filtering and downsampling removes duplicate points within a voxel, such as Figure 2 Shown is the point cloud P0 to be identified.

[0040] S2. Point cloud clustering

[0041] S2-1, retain the dense entity point cloud through Euclidean clustering, remove non-entity invalid noise points, and obtain point cloud P1;

[0042] S2-2, calculate the normals of all points in the point cloud P1, N1 = {n i |i=1,2…s1}, where s1 is the size of point cloud P1;

[0043] S2-3, through normal filtering, retain the solid surface point cloud facing the radar, that is, the normal vector n i (n ix , n iy , n iz ) in the X-direction component n ix <0;

[0044] S2-4, obtain the point cloud set again through Euclidean clustering;

[0045] S2-5. By calculating the centroid of the point cloud clusters in the set, sort them in order of distance from the radar from near to far, and merge the point cloud clusters whose distance in the radar scanning direction is less than the threshold and whose centroid distance is less than the diameter of the insulator sheet into the same point cloud cluster, and finally obtain the point cloud cluster set to be screened Q1 = {q i |i=1,2…s q1}, where s q1 is the size of the point cloud cluster set, such as Figure 3 shown.

[0046] S3, for each cluster of point cloud q in point cloud cluster Q1 i Perform feature calculations, including fitting the circle center coordinates qc i , fitting circle radius qr i , the plane normal vector qn i ;

[0047] The following steps are involved:

[0048] S3-1. Calculate the point cloud normal N q ={n m |m=1,2…s qi}, where s qi is the point cloud cluster q i size;

[0049] S3-2, extract the entity surface boundary point cloud P by the normal angle threshold b , when the point normal vector n m With k b The normal vectors of the adjacent points {n1…n k} are all greater than the threshold value of 150°, then the point is considered to be a boundary point, where k b Adjustable parameters for the boundary extraction algorithm;

[0050] S3-3, the boundary point cloud P b Projected to the YZ plane, the two-dimensional boundary point cloud P on the YZ plane is obtained b ';

[0051] S3-4, using the two-dimensional boundary point cloud P b 'The highest point P in the Z direction zmax As the starting point, find the adjacent points through KD-Tree, grow the upper boundary point cloud in the Y+ direction and Y- direction respectively, and extract the original point cloud P b The corresponding points on the solid surface are obtained by the boundary point cloud P bu ,like Figure 4 As shown;

[0052] S3-5. Fitting the upper boundary point cloud P by random sampling method bu The plane in which it is located, the normal vector qn is obtained i , and project the upper boundary point cloud onto the plane, and then rotate it to the XY plane to obtain the 2D point cloud P bupr ;

[0053] S3-6, point cloud P bupr By fitting the circle with least squares, we can get the center qc pr and the insulator segment radius qr i ;

[0054] S3-7, the center of the projection plane qc pr After converting back to the plane where the upper boundary point cloud is located, the center position qc of the insulator segment is obtained i ,qc i =T i *qc pr , where T i From the unit vector (0,0,1) to the normal vector qn i The transformation matrix.

[0055] S4. Insulator point cloud screening

[0056] S4-1. Traverse all point cloud clusters q i , filter out the point cloud cluster set that meets the characteristics of the insulator piece to be identified. First, the radius of the fitting circle needs to be close to the radius of the target insulator piece, that is, qr i∈[0.8R,R], R is the actual radius of the target insulator, and the point cloud fitting coefficient n f / n q >0.9, where n q is the point cloud cluster q i The number of points in n f In order to meet the target insulator segment size point cloud number, the calculation method is to use the insulator segment diameter 2R as the length and width, and the insulator segment thickness D as the thickness of the bounding box to cut the point cloud cluster q i Get point cloud cluster q if ={p m |m=1,2…n f},like Figure 5 And sort them in order from near to far from the radar to obtain the insulator point cloud cluster set Q v , insulator sheet point cloud cluster set Q v ={q i |i=1,2…s v}, where s v The number of point cloud clusters that meet the filtering conditions

[0057] S4-2. Calculate the insulator point cloud cluster set Q v Adjacent insulator segments (q i ,q i+1 ) center (qc i ,qc i+1 ) between the Euclidean distance, according to the threshold d p Merge the point cloud clusters belonging to the same insulator group and record the position of the first insulator (qc i ,qn i ) is the position of the insulator group, where d p The distance between the two pieces of the reference target insulator is set, and finally all the insulator piece groups that meet the conditions are obtained.

Claims

1. A method for identifying and locating insulator segments on a suspension insulator string based on a solid-state laser radar point cloud, characterized in that: The following steps are involved: S1, solid-state radar point cloud preprocessing; S2. Point cloud clustering S2-1, retain the dense entity point cloud through Euclidean clustering, remove non-entity invalid noise points, and obtain point cloud P1; S2-2, calculate the normals of all points in the point cloud P1, N1 = {n i |i=1,2…s1}, where s1 is the size of point cloud P1; S2-3, through normal filtering, retain the solid surface point cloud facing the radar, that is, the normal vector n i (n ix , n iy , n iz ) in the X-direction component n ix <0; S2-4, obtain the point cloud set again through Euclidean clustering; S2-5. By calculating the centroid of the point cloud clusters in the set, sort them in order of distance from the radar from near to far, and merge the point cloud clusters whose distance in the radar scanning direction is less than the threshold and whose centroid distance is less than the diameter of the insulator sheet into the same point cloud cluster, and finally obtain the point cloud cluster set to be screened Q1 = {q i |i=1,2·s q1 }, where s q1 is the size of the point cloud cluster set; S3, for each cluster of point cloud q in point cloud cluster Q1 i Perform feature calculations, including fitting the circle center coordinates qc i , fitting circle radius qr i , the plane normal vector qn i ; S4. Insulator point cloud screening S4-1. Traverse all point cloud clusters q i , filter out the point cloud cluster set that meets the characteristics of the insulator piece to be identified, and sort them in order from near to far from the radar, and obtain the insulator piece point cloud cluster set Q v ; S4-2. Calculate the insulator point cloud cluster set Q v Adjacent insulator segments (q i ,q i+1 ) center (qc i ,qc i+1 ) between the Euclidean distance, according to the threshold d p Merge the point cloud clusters belonging to the same insulator group and record the position of the first insulator (qc i ,qn i ) is the position of the insulator group, where d p The distance between the two pieces of the reference target insulator is set, and finally all the insulator piece groups that meet the conditions are obtained.

2. The method for identifying and locating an insulator segment group on a suspension insulator string based on a solid-state laser radar point cloud according to claim 1, characterized in that: In S1, point clouds outside the target range are removed by point cloud cropping; voxel filtering and downsampling are performed to remove duplicate points within the voxel.

3. The method for identifying and locating an insulator segment group on a suspension insulator string based on a solid-state laser radar point cloud according to claim 1, characterized in that: S3 includes the following steps: S3-1. Calculate the point cloud normal N q ={n m |m=1,2·s qi }, where s qi is the point cloud cluster q i size; S3-2, extract the entity surface boundary point cloud P by the normal angle threshold b , when the point normal vector n m With k b The normal vectors of the adjacent points {n1…n k } are all greater than the threshold value of 150°, then the point is considered to be a boundary point, where k b Adjustable parameters for the boundary extraction algorithm; S3-3, the boundary point cloud P b Projected to the YZ plane, the two-dimensional boundary point cloud P on the YZ plane is obtained b '; S3-4, using the two-dimensional boundary point cloud P b 'The highest point P in the Z direction zmax As the starting point, find the adjacent points through KD-Tree, grow the upper boundary point cloud in the Y+ direction and Y- direction respectively, and extract the original point cloud P b The corresponding points on the solid surface are obtained by the boundary point cloud P bu ; S3-5. Fitting the upper boundary point cloud P by random sampling method bu The plane in which it is located, the normal vector qn is obtained i , and project the upper boundary point cloud onto the plane, and then rotate it to the XY plane to obtain the 2D point cloud P bupr ; S3-6, point cloud P bupr By fitting the circle with least squares, we can get the center qc pr and the insulator segment radius qr i ; S3-7, the center of the projection plane qc pr After converting back to the plane where the upper boundary point cloud is located, the center position qc of the insulator segment is obtained i .

4. The method for identifying and locating insulator segments on a suspension insulator string based on a solid-state laser radar point cloud according to claim 3 is characterized in that: S3-7 in qc i =T i *qc pr , where T i From the unit vector (0,0,1) to the normal vector qn i The transformation matrix.

5. The method for identifying and locating an insulator segment group on a suspension insulator string based on a solid-state laser radar point cloud according to claim 1, characterized in that: In S4-1, the radius of the fitting circle is close to the target insulator segment radius, i.e., qr i ∈[0.8R,R], R is the actual radius of the target insulator, and the point cloud fitting coefficient n f / n q >0.9, where n q is the point cloud cluster q i The number of points in n f In order to meet the target insulator segment size point cloud number, the calculation method is to use the insulator segment diameter 2R as the length and width, and the insulator segment thickness D as the thickness of the bounding box to cut the point cloud cluster q i Get point cloud cluster q if ={p m |m=1,2…n f }; Insulator sheet point cloud cluster set Q v ={q i |i=1,2…s v }, where s v is the number of point cloud clusters that meet the filtering conditions.

Images