A wire laser point cloud classification method
Through coordinate transformation, segmentation, projection and clustering algorithms, the conductor laser point cloud is classified, which solves the problem of low classification efficiency of conductor laser point clouds in the existing technology, and realizes the accurate classification and analysis of each conductor.
Patent Information
- Application Number
- CN202211076519.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-09-05
- Publication Date
- 2025-05-13
- Estimated Expiration
- 2042-09-05
AI Technical Summary
The prior art lacks efficient wire laser point cloud classification methods, making it difficult to separate the wire from other objects and extract each wire separately.
Classification of conductor laser point clouds through coordinate transformation, segmentation, projection and clustering algorithms is realized. The specific steps include determining the transition matrix for coordinate transformation, segmenting according to the uniformity of point cloud density, projecting along the tz-axis direction, and using clustering algorithm to divide the point cloud into multiple categories, and finally assigning labels to each point.
The effective classification of each wire in the conductor laser point cloud is realized, and the analysis accuracy and efficiency of the conductor laser point cloud data is improved.
Smart Images
Figure CN115439632B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of point cloud processing, and in particular to a wire laser point cloud classification method. Background Art
[0002] The laser radar carried by the drone can collect laser point clouds of the transmission channel. The laser point cloud of the transmission channel contains the precise structural data of important facilities such as conductors and towers. In subsequent applications such as tree obstacle inspection and sag measurement, the conductors need to be analyzed separately. This not only requires separating the conductors from other objects, but also requires extracting each conductor separately, that is, classifying the conductor laser point cloud and giving each point more specific information. However, there is currently a lack of efficient conductor laser point cloud classification methods for extracting each conductor. Summary of the invention
[0003] In order to solve the above problems, the present invention provides a wire laser point cloud classification method, and the specific technical solution is as follows:
[0004] S1: coordinate transformation; determine the transition matrix T = [t x t y t z ] T , transform the coordinates of the points in the wire laser point cloud to obtain the transformed point coordinates
[0005] Among them, t x ,t y With t z Represent three vectors in the three-dimensional coordinate system, Indicates the coordinates of the point after coordinate transformation;
[0006] S2: Segment the wire laser point cloud after coordinate transformation; divide the wire laser point cloud into several segments according to the number of points in each segment or the length of the interval;
[0007] S3: For each point cloud segment along t z Axis projection is used to reduce the three-dimensional coordinates of the points to two dimensions. Then, clustering algorithms are used to divide the points in each point cloud into multiple categories, and the center points of each category are obtained.
[0008] S4: Assign a label to each point in the point cloud. The process is as follows:
[0009] S401: assigning labels to points in the first segment of the point cloud according to the classification results;
[0010] S402: Based on the labels assigned to the first segment of the point cloud, assign labels to points in the remaining segments of the point cloud.
[0011] Furthermore, the process of obtaining the transition matrix is as follows:
[0012] S101: Determine t by the direction of the wire z , as follows:
[0013] Using the three-dimensional space straight line equation, fitting the points of the wire laser point cloud, we can get the straight line direction vector t z =[a z b z c z ] T ;
[0014] S102: At t z An arbitrary vector on the plane normal to t x ;
[0015] S103: vector t z ,t x Normalized by t z With t x The cross product of t y .
[0016] Furthermore, in step S2, depending on whether the density of the wire laser point cloud is uniform, the wire laser point cloud is segmented according to the number of points or the length of the interval. If the density of the wire laser point cloud is uniform, the segmentation is performed according to the length of the interval; if the density of the wire laser point cloud is uneven, the segmentation is performed according to the number of points.
[0017] Furthermore, whether the density of the wire laser point cloud is uniform is determined as follows:
[0018] The wire laser point cloud is pre-segmented according to the length of the interval, and the variance of the number of points in each segment is calculated. The variance is compared to see whether it is greater than the preset threshold. If so, the density of the wire laser point cloud is considered to be uneven, otherwise, the density of the wire laser point cloud is considered to be relatively uniform.
[0019] Furthermore, the wire laser point cloud is segmented according to the number of points, and the number of points in each segment is equal, as follows:
[0020] Press the point of the wire laser point cloud z The values in the axis direction are sorted, and the point cloud is divided into several segments with an equal number of points in each segment according to the preset number of points in each segment.
[0021] Furthermore, the wire laser point cloud is segmented according to the interval length, and the length of each interval is equal, as follows:
[0022] In t z A number of equal-length intervals are marked in the axial direction. Each equal-length interval corresponds to a point cloud. For the points in the wire laser point cloud, it is determined whether each point is within the range of t z The value in the axial direction, depending on which equal-length interval the value belongs to, will assign the point to the corresponding point cloud segment.
[0023] Furthermore, in step S4, different labels are assigned to each category of the first segment of point cloud; and the same label is assigned to all point clouds of each category.
[0024] Furthermore, the specific process of step S402 is as follows:
[0025] Select the second segment of point cloud that is adjacent to the first segment, and assign the same label to one of the classes of the second segment of point cloud as the first segment that is closest to it; in this way, assign corresponding labels to all classes of the second segment of point cloud;
[0026] Repeat this process to assign labels to the next segment of point cloud adjacent to the current segment until label assignment is completed for the point clouds of all remaining segments.
[0027] The beneficial effects of the present invention are as follows:
[0028] The present invention transforms the wire laser point cloud in three-dimensional space into a new three-dimensional coordinate system by performing coordinate transformation on the wire laser point cloud data, which is convenient for subsequent segmentation and classification operations. The point cloud is segmented in different ways according to the uniformity of the density of the wire laser point cloud, and then the segmented point cloud is projected and classified, thereby realizing effective classification of each wire in the wire laser point cloud. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 It is a schematic diagram of the overall process of the method of the present invention;
[0030] Figure 2 It is a schematic diagram comparing the segmentation effects of the wire laser point cloud of the present invention. DETAILED DESCRIPTION
[0031] The following description clearly and completely describes the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0032] The purpose of wire laser point cloud classification is to classify the collected wire laser point cloud containing multiple wires into multiple categories, and each category of wire laser point cloud belongs to only one wire.
[0033] The operation of wire laser point cloud classification can be described as follows: The wire laser point cloud contains N points, each of which is denoted by P i (i=1,2,…,N),P i The three-dimensional coordinates are expressed as [x i y i z i ]T ; The wire laser point cloud contains M wires, and the category label of each wire is recorded as C j (j=1,2,…,M). The task of wire laser point cloud classification is to assign a category label C to each point j .
[0034] The embodiment of the present invention discloses a wire laser point cloud classification method, such as Figure 1 As shown, the specific steps are as follows:
[0035] S1: coordinate transformation; determine the transition matrix T = [t x t y t z ] T , the coordinates of the midpoints of the wire laser point cloud are transformed. The specific coordinate transformation method is: Get the transformed point coordinates
[0036] Among them, t x ,t y With t z Represent three vectors in the three-dimensional coordinate system, Indicates the coordinates of the point after coordinate transformation;
[0037] For the collected wire laser point cloud, the wire direction is not necessarily consistent with a certain coordinate axis, which is not conducive to subsequent segmentation and classification. Through coordinate transformation, the wire laser point cloud in three-dimensional space is transformed into a new three-dimensional coordinate system, so that the wire direction becomes consistent with a certain coordinate axis, which is more conducive to subsequent segmentation and classification.
[0038] In this embodiment, the specific process of determining the transition matrix is as follows:
[0039] S101: Determine t by the direction of the wire z , as follows:
[0040] Use the three-dimensional space straight line equation to fit the point P of the wire laser point cloud i , we can get the straight line direction vector t z =[a z b z c z ] T , the direction of the wire is t z express;
[0041] S102: At t z An arbitrary vector on the plane normal to t x , using any point [x o y o z o ] and the normal tz , the plane equation a can be determined z (xx o )+b z (yy o )+c z (zz o )=0, and the vector t can be determined by any two different points on the plane. x ;
[0042] S103: Using t z ,t x Determine t y , respectively for vector t z ,t x Normalized by t z With t x The cross product of t y ,Right now
[0043] S2: Segment the wire laser point cloud; divide the wire laser point cloud into multiple segments according to the number of points in each segment or the length of the interval;
[0044] In this embodiment, the conductor laser point cloud can be segmented according to the number of points or the length of the interval, depending on whether the density of the conductor laser point cloud is uniform.
[0045] When the density of the wire laser point cloud is relatively uniform, the segmentation based on the number of points and the length of the interval are equivalent. However, since the segmentation based on the number of points often requires sorting, the algorithm has a high time complexity.
[0046] For conductors with large height differences and large spans, the sag is large and the density of the conductor laser point cloud is often uneven. In this case, segmentation based on the number of points is more effective.
[0047] like Figure 2 As shown in the figure, the upper wire laser point cloud contains two wires, which are not segmented and directly projected to the plane on the left, resulting in points of different categories being mixed together, which is unclassifiable; the lower wire laser point cloud contains two wires, and a segment is cut out according to the dotted line and projected to the plane on the left. Points of different categories are not mixed together and can be well separated;
[0048] Therefore, if the density of the wire laser point cloud is uniform, segmentation is performed based on the interval length; if the density of the wire laser point cloud is uneven, segmentation is performed based on the number of points.
[0049] Specifically, if the wire laser point cloud is divided into segments of equal length, when the variance of the number of points in each segment Var(N k )>β, β>0, it is determined that the density of the wire laser point cloud is uneven, otherwise it is considered that the density of the wire laser point cloud is relatively uniform; variance Var(Nk ) is calculated as:
[0050]
[0051] in, K represents the number of segments, N k Indicates the number of points in the kth segment.
[0052] Specifically, the wire laser point cloud is segmented according to the number of points, and the wire laser point cloud is segmented at t z The axis direction is divided into K segments according to the number of points N, and the number of points in each segment is equal. The process is as follows:
[0053] Press the point of the wire laser point cloud z The values in the axis direction are sorted, and the points in the wire laser point cloud after sorting are recorded as The sorted wire laser point cloud is divided into each segment containing K segments of point clouds of points, each segment of point cloud can be recorded as a set Where k = 1, 2,…, K.
[0054] Segment the wire laser point cloud by interval length and divide the wire laser point cloud into segments according to t z The length of the interval occupied by the axis direction is L z , divided into K segments, each segment has the same length, The details are as follows:
[0055] In t z The axis direction marks K intervals of equal length, each interval is recorded as where r min Point Cloud In t z The starting point coordinates in the axis direction, k = 1, 2, ..., K. Each equal-length interval corresponds to a point cloud, and the number of points in each point cloud is recorded as N k For the N points in the wire laser point cloud, determine the z The value in the axis direction, which equal-length interval the value belongs to, will assign the point to the corresponding point cloud segment.
[0056] The K value is set according to the actual situation and is not specifically limited here. Generally, the larger the K value, the finer the segmentation and the better the classification effect, but the more complicated the calculation. If the K value is too small, the classification effect will be poor or even unclassifiable.
[0057] When the classification effect is poor, increase the value of K appropriately; when the classification speed is slow, decrease the value of K appropriately.
[0058] S3: For each point cloud projected, perform classification operations on the two-dimensional plane;
[0059] For each point cloud segment, the point is projected onto a two-dimensional plane. In this embodiment, each point cloud segment is projected onto a two-dimensional plane along t z Axis projection reduces the three-dimensional coordinates of the point to two dimensions. Since the coordinate transformation has been implemented, the third coordinate can be directly removed to realize the projection from the three-dimensional space coordinates to the two-dimensional plane coordinates.
[0060] Then, the points in each point cloud segment are divided into multiple categories using the classification method in two-dimensional space. In this embodiment, the points in each point cloud segment are divided into M categories by clustering algorithm, and the center point O of the pth category is obtained. p (p=1,2,…,M).
[0061] The role of point cloud projection is that for long strips of data in three-dimensional space, conventional clustering methods such as K-means cannot achieve good results; after projecting to a two-dimensional plane, K-means can easily achieve good results.
[0062] S4: Assign a label to each point in the point cloud;
[0063] First, labels are assigned to the points in the first segment of the point cloud according to the classification results; then, labels are assigned to the point clouds of the remaining segments with reference to the labels of the first segment of the point cloud.
[0064] The specific process is as follows:
[0065] For the pth class of the first segment of point cloud, the center point of the pth class of the first segment of point cloud is denoted as O 1,p (p=1,2,…,M), first assign label O to the center point 1,p ←C j , and then assign the same label as the center point to all points of this category;
[0066] Assign labels to the second segment of point cloud adjacent to the first segment as follows:
[0067] For the qth class of the second segment point cloud, its center point is O 2,q (q=1,2,…,M), all the center points O of the first segment of the point cloud 1,p , find the 2,q The nearest center point is determined, and the label of the center point is determined as the center point O 2, q assigns the label and assigns the label to all points of this category;
[0068] According to the same method as above, labels are assigned to the third segment of point cloud adjacent to the second segment until label assignment is completed for the point clouds of all remaining segments.
[0069] The present invention is not limited to the above-mentioned specific embodiments, but extends to any new features or any new combination disclosed in this specification, as well as any new method or process steps or any new combination disclosed.
Claims
1. A wire laser point cloud classification method, characterized in that: include: S1: coordinate transformation; Determine the transition matrix T = [t x t y t z ] T , transform the coordinates of the points in the wire laser point cloud to obtain the transformed point coordinates The process of obtaining the transition matrix is as follows: S101: Determine t by the direction of the wire z , as follows: Using the three-dimensional space straight line equation, fitting the points of the wire laser point cloud, we can get the straight line direction vector t z =[a z b z c z ] T ; S102: At t z An arbitrary vector on the plane normal to t x ; S103: vector t z ,t x Normalized by t z With t x The cross product of t y ; Among them, t x ,t y With t z Represent three vectors in the three-dimensional coordinate system, Indicates the coordinates of the point after coordinate transformation; S2: Segment the wire laser point cloud after coordinate transformation; divide the wire laser point cloud into several segments according to the number of points in each segment or the length of the interval; According to whether the density of the wire laser point cloud is uniform, choose to segment the wire laser point cloud according to the number of points or according to the length of the interval. If the density of the wire laser point cloud is uniform, segment it according to the length of the interval. If the density of the wire laser point cloud is uneven, segment it according to the number of points. S3: For each point cloud segment along t z Axis projection is used to reduce the three-dimensional coordinates of the points to two dimensions, and then clustering algorithms are used to divide the points in each point cloud into multiple categories. S4: Assign a label to each point in the point cloud. The process is as follows: S401: assigning labels to points in the first segment of the point cloud according to the classification results; S402: Based on the labels assigned to the first segment of the point cloud, assign labels to points in the remaining segments of the point cloud.
2. The wire laser point cloud classification method according to claim 1, characterized in that: The uniformity of the laser point cloud density of the conductor is determined as follows: The wire laser point cloud is pre-segmented according to the length of the interval, and the variance of the number of points in each segment is calculated. The variance is compared to see whether it is greater than the preset threshold. If so, the density of the wire laser point cloud is considered to be uneven, otherwise, the density of the wire laser point cloud is considered to be uniform.
3. The wire laser point cloud classification method according to any one of claims 1-2, characterized in that: The wire laser point cloud is segmented according to the number of points, and the number of points in each segment is equal, as follows: Press the point of the wire laser point cloud z The values in the axis direction are sorted, and the point cloud is divided into several segments with an equal number of points in each segment according to the preset number of points in each segment.
4. The wire laser point cloud classification method according to any one of claims 1-2, characterized in that: The wire laser point cloud is segmented according to the interval length, and the length of each interval is equal, as follows: In t z A number of equal-length intervals are marked in the axial direction. Each equal-length interval corresponds to a point cloud. For the points in the wire laser point cloud, it is determined whether each point is within the range of t z The point is assigned to the corresponding point cloud segment according to the equal-length interval to which the axis direction value belongs.
5. The wire laser point cloud classification method according to claim 1, characterized in that: In step S4, different labels are assigned to each category of the first segment of point cloud; and the same label is assigned to all point clouds of each category.
6. The wire laser point cloud classification method according to claim 5, characterized in that: The specific process of step S402 is as follows: Select the second segment of point cloud that is adjacent to the first segment, and assign the same label to one of the classes of the second segment of point cloud as the first segment that is closest to it; in this way, assign corresponding labels to all classes of the second segment of point cloud; Repeat this process to assign labels to the next segment of point cloud adjacent to the current segment until label assignment is completed for the point clouds of all remaining segments.