Laser radar-based cable icing detection method, device, equipment and medium

Abstract

The application discloses a cable icing detection method and device based on a laser radar, equipment and a medium. For two frames of point clouds to be registered, key points are extracted from each frame of point clouds to obtain key point clouds, the two frames of key point clouds are input into a registration network for processing, the registration network adaptively learns an overlapping area of the two frames of point clouds based on a spatial attention mechanism, predicts a relative pose of the two frames of point clouds, registers the two frames of point clouds based on the relative pose, obtains fused point clouds, extracts icing cable point clouds representing icing cables from the fused point clouds, determines three-dimensional sizes of the icing cables based on the icing cable point clouds, and calculates icing thickness based on the three-dimensional sizes of the icing cables. The attention weight mechanism enables the network to adaptively learn the overlapping area of continuous point cloud frames, thereby realizing fast and effective end-to-end registration of large-scale point cloud scenes, providing a good foundation for a fine registration algorithm, improving registration efficiency and accuracy, and further improving icing detection accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cable monitoring, and particularly relates to a cable icing detection method and device based on laser radar, equipment and medium. BACKGROUND

[0002] Icing monitoring of distribution networks is crucial for ensuring the stability and security of power systems. Snow and ice weather can cause an increase in the load of conductors and equipment, leading to power outages and equipment damage. By monitoring the icing situation in real time, preventive measures can be taken in time to reduce failures and ensure the normal operation of power supply and economic activities. Therefore, establishing an efficient icing monitoring system can provide reliable data support under extreme weather conditions and ensure the safe operation of distribution networks.

[0003] Traditional icing monitoring methods often require disassembly or direct contact with equipment, increasing safety hazards and operational complexity. Non-invasive monitoring, on the other hand, uses advanced sensors and data analysis techniques to obtain the icing state of conductors and equipment in real time, avoiding interference and damage to equipment. This method is particularly important in snow and ice weather, as it can identify potential risks in time and ensure the continuity and reliability of power supply, thereby effectively protecting residents' lives and economic activities and improving the intelligent management level of distribution networks. Laser radar scanning as a new method of non-invasive icing monitoring uses laser beam emission and reception to quickly obtain three-dimensional data of the surface of distribution network equipment, accurately measuring the thickness and distribution of icing, with high precision and efficiency. Compared with traditional methods, laser radar does not need to contact equipment or lines, reducing human intervention and safety risks. At the same time, laser radar can monitor under various weather conditions, providing real-time data to help operators quickly assess risks and take appropriate measures. The application of this technology provides an efficient and safe monitoring means for the power industry, which will significantly improve the monitoring ability of distribution networks under extreme weather conditions and enhance the safety and reliability of power systems.

[0004] In the process of acquiring point cloud data by laser radar scanning, due to the limited field of view of the sensor, a registration algorithm is needed to splice the local point cloud to generate a complete 3D scene point cloud of the large-scale power distribution network. The shape of the icing of the power distribution network is complex, and the laser radar point cloud is an effective measurement tool for evaluating the thickness and shape of the icing. However, for the laser radar point cloud in a large-scale space, accurate registration is crucial. Point cloud registration mainly involves estimating the Euclidean transformation matrix between two frames of scanned point clouds. After obtaining the transformation matrix, the point cloud data captured by different sensors can be converted to a unified world coordinate system for combination. With the rapid development of deep learning technology, learning-based methods have gradually replaced traditional handcrafted features, enabling faster and more accurate point cloud registration tasks. However, existing point cloud registration methods are mostly applied to the splicing of small-scale 3D scenes, such as indoor or object-level point clouds. In the large-scale power distribution network scene, due to the ranging error and noise in the laser radar scanning data, as well as the sparsity of the point cloud, traditional registration algorithms often fail to meet the actual performance requirements, which may result in excessive registration time or failure. SUMMARY

[0005] The present application provides a laser radar-based cable icing detection method, device, equipment and medium to improve registration efficiency and accuracy, thereby improving icing detection accuracy.

[0006] In a first aspect, the present application provides a laser radar-based cable icing detection method, comprising:

[0007] Obtaining two frames of point clouds to be registered;

[0008] Extracting key points from each frame of the point cloud to obtain a key point cloud of each frame of the point cloud;

[0009] Inputting two frames of the key point cloud into a registration network for processing, the registration network adaptively learns the overlapping area of two frames of the point cloud based on a spatial attention mechanism, and predicts the relative pose of two frames of the point cloud;

[0010] Registering two frames of the point cloud based on the relative pose of two frames of the point cloud to obtain a fused point cloud;

[0011] Extracting an icing cable point cloud representing an icing cable from the fused point cloud;

[0012] Determining the three-dimensional size of the icing cable based on the icing cable point cloud;

[0013] Calculating the icing thickness based on the three-dimensional size of the icing cable.

[0014] Optionally, extracting key points from each frame of the point cloud to obtain a key point cloud of each frame of the point cloud comprises:

[0015] Performing a rotation compensation transformation on the point cloud of each frame to obtain a compensated transformed point cloud;

[0016] Extracting key points from the transformed point cloud to obtain an initial key point cloud composed of key points;

[0017] Performing a rotation-compensated inverse transformation on the key points in the initial key point cloud to obtain a key point cloud for each frame of the point cloud.

[0018] Optionally, performing a rotation compensation transformation on the point cloud of each frame to obtain a compensated transformed point cloud includes:

[0019] Downsampling the point cloud of each frame to obtain a sparse point cloud;

[0020] extracting a first point cloud representing the ice-covered cable from the sparse point cloud;

[0021] Calculating an angle between the first point cloud and a normal vector of a horizontal plane;

[0022] Determine a rotation compensation transformation matrix based on the angle;

[0023] The point cloud is multiplied by the rotation compensation transformation matrix to obtain a compensated transformed point cloud.

[0024] Optionally, each frame of the key point cloud includes a surface key point cloud and a corner point cloud, and the registration network includes two feature encoders and a feature decoder with a spatial attention weighting mechanism. The two frames of the key point clouds are input into the registration network for processing. The registration network adaptively learns the overlapping area of ​​the two frames of point clouds based on the spatial attention mechanism and predicts the relative pose of the two frames of point clouds, including:

[0025] Inputting the surface key point cloud of the first frame key point cloud and the surface key point cloud of the second frame key point cloud into the first feature encoder for processing, obtaining the surface spatial attention features of the first frame key point cloud and the surface spatial attention features of the second frame key point cloud;

[0026] Input the corner point cloud of the key point cloud of the first frame and the corner point cloud of the key point cloud of the second frame into the second feature encoder for processing, thereby obtaining the corner point spatial attention features of the key point cloud of the first frame and the corner point spatial attention features of the key point cloud of the second frame;

[0027] Fusing the surface spatial attention feature and the corner spatial attention feature to obtain a spatial attention fusion feature;

[0028] The spatial attention fusion feature is input into the feature decoder for decoding to obtain the relative pose of the point cloud of the two frames.

[0029] Optionally, the surface key point cloud of the first frame of key point cloud and the surface key point cloud of the second frame of key point cloud are input into the first feature encoder for processing to obtain the surface spatial attention feature of the first frame of key point cloud and the surface spatial attention feature of the second frame of key point cloud, including:

[0030] For each of the surface key point cloud, a multi-layer perception is used to extract the global feature of the surface key point cloud;

[0031] The global feature is input into a spatial attention module for processing to obtain a spatial attention score matrix composed of spatial attention scores of elements in the global feature;

[0032] The spatial attention score matrix is normalized to obtain a spatial attention weight matrix;

[0033] The product of the global feature and the spatial attention weight matrix is calculated to obtain a surface spatial attention feature.

[0034] Optionally, the feature decoder includes a fully connected layer, and the spatial attention fusion feature is input into the feature decoder for decoding to obtain the relative pose of the two frames of point cloud, including:

[0035] The fully connected layer performs fully connected mapping on the spatial attention fusion feature to obtain a relative pose vector representing the relative pose of the two frames of point cloud.

[0036] Optionally, before obtaining the two frames of point cloud to be registered, further comprising:

[0037] Extracting a key point cloud of a training sample, the training sample including a source frame point cloud and a target point cloud;

[0038] Inputting the key point cloud of the training sample into a registration network for processing to obtain the relative pose of the source frame point cloud and the target point cloud;

[0039] Extracting a non-ground target point cloud in the source frame point cloud and a non-ground target point cloud in the target point cloud;

[0040] Converting the non-ground target point cloud in the source frame point cloud to the coordinate system of the target point cloud based on the relative pose of the source frame point cloud and the target point cloud to obtain a converted point cloud;

[0041] Calculating the chamfer distance loss of the non-ground target point cloud in the target point cloud and the converted point cloud;

[0042] Updating the network parameters of the registration network based on the chamfer distance loss until the chamfer distance loss is less than a preset value.

[0043] In a second aspect, the present application further provides a cable icing detection device based on a laser radar, comprising:

[0044] a point cloud acquisition module, configured to acquire two frames of point clouds to be registered;

[0045] a key point extraction module, configured to extract key points from each frame of the point clouds to obtain key point clouds of each frame of the point clouds;

[0046] a relative pose prediction module, configured to input two frames of the key point clouds into a registration network for processing, the registration network being based on a spatial attention mechanism to adaptively learn an overlapping area of two frames of the point clouds and predict a relative pose of two frames of the point clouds;

[0047] a point cloud registration module, configured to register two frames of the point clouds based on the relative pose of two frames of the point clouds to obtain a fused point cloud;

[0048] an icing cable point cloud extraction module, configured to extract an icing cable point cloud representing an icing cable from the fused point cloud;

[0049] a three-dimensional size determination module, configured to determine a three-dimensional size of the icing cable based on the icing cable point cloud;

[0050] an icing thickness calculation module, configured to calculate an icing thickness based on the three-dimensional size of the icing cable.

[0051] In a third aspect, the present application further provides an electronic device, comprising:

[0052] one or more processors;

[0053] a storage device, configured to store one or more programs;

[0054] when the one or more programs are executed by the one or more processors, the one or more processors implement the cable icing detection method based on a laser radar as provided in the first aspect of the present application.

[0055] In a fourth aspect, the present application further provides a computer readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the cable icing detection method based on a laser radar as provided in the first aspect of the present application.

[0056] The laser radar-based cable icing detection method provided by the application comprises the following steps: extracting key points from each frame of point cloud to obtain key point clouds of each frame of point cloud, inputting two frames of key point clouds into a registration network for processing, adaptively learning an overlapping area of the two frames of point cloud by the registration network based on a spatial attention mechanism, predicting a relative pose of the two frames of point cloud, registering the two frames of point cloud based on the relative pose of the two frames of point cloud to obtain fused point cloud, extracting icing cable point cloud representing icing cable from the fused point cloud, determining a three-dimensional size of the icing cable based on the icing cable point cloud, and calculating an icing thickness based on the three-dimensional size of the icing cable. The attention weight mechanism enables the network to adaptively learn the overlapping area of continuous point cloud frames, thereby realizing fast and effective end-to-end registration of a large-scale point cloud scene, providing a good foundation for a fine registration algorithm, improving registration efficiency and accuracy, and further improving icing detection accuracy.

[0057] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the application, nor is it intended to limit the scope of the application. Other features of the application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments description. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can also be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0059] Figure 1 A flowchart of a laser radar-based cable icing detection method provided by the application;

[0060] Figure 2 A structural schematic diagram of a registration network provided by the application;

[0061] Figure 3 A structural schematic diagram of a feature encoder provided by the application;

[0062] Figure 4 A structural schematic diagram of a laser radar-based cable icing detection device provided by the application;

[0063] Figure 5 A structural schematic diagram of an electronic device provided by the application.

[0064] The above drawings have shown the specific embodiments of the application, and will be described in more detail hereinafter. These drawings and textual descriptions are not intended to limit the scope of the concept of the application by any means, but to illustrate the concept of the application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0065] In order to better understand the technical scheme of the present application, the technical scheme in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts should fall within the protection scope of the present application.

[0066] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0067] Figure 1 A flowchart of a cable icing detection method based on a laser radar is provided for the present application. The embodiment can be applicable to the registration of large-scale space point clouds and the detection of cable icing. The method can be performed by a cable icing detection device based on a laser radar provided by the embodiment of the present application. The device can be implemented in the form of software and / or hardware, and is usually configured in an electronic device, such as a computer. Figure 1 As shown in the figure, the cable icing detection method based on a laser radar includes the following steps:

[0068] S101, acquiring two frames of point clouds to be registered.

[0069] In the embodiment of the present application, the laser radar continuously scans the space to be detected to obtain continuous multiple frames of point clouds. In the embodiment of the present application, the laser radar can be a multi-line laser radar. The two frames of point clouds to be registered can be two consecutive frames of point clouds.

[0070] S102, extracting key points from each frame of point cloud to obtain a key point cloud of each frame of point cloud.

[0071] For each frame of point cloud, key points are extracted from the frame of point cloud to obtain a key point cloud for each frame of point cloud. Point cloud key point extraction refers to extracting some important points from the point cloud dataset for use in subsequent point cloud analysis and processing. These key points usually include high curvature points, edge points, corner points, etc., which have important feature information and can be used to describe the local shape and structure of the point cloud. For example, algorithms such as SIFT, Harris, NARF, ISS and SUSAN can be used to extract key points of the point cloud, which is not limited in the embodiments of the present invention.

[0072] During the radar scanning process, if the radar itself moves in the opposite direction of rotation, the radar laser rotation angle will be smaller than the actual angle. This phenomenon is called rotational distortion. Specifically, when the radar is rotating, if the radar itself is also rotating, the points in the collected point cloud data will produce position deviations due to the rotation, resulting in inaccurate or even failed subsequent key point extraction. In some embodiments of the present invention, in response to the problem of low robustness of existing point cloud key point extraction methods in distribution network cable linear strip scenarios, an improved key point extraction method based on rotation compensation is proposed to solve the problem of inaccurate or even failed key point extraction when there is an inclination angle between the point cloud frame and the XOY plane (horizontal plane) of the laser radar. Rotational compensation refers to a technology that compensates for the distortion of point cloud data caused by the movement of the radar carrier during the laser radar data acquisition process. Specifically, point cloud rotation compensation mainly corrects the distortion generated by the radar during the rotation process to ensure the accuracy of subsequent key point extraction.

[0073] Specifically, the key point cloud extraction process is as follows:

[0074] 1. Perform rotation compensation transformation on each frame of point cloud to obtain the compensated transformed point cloud.

[0075] Exemplarily, a rotation compensation transformation is performed on each frame of point cloud to correct the distortion generated by the radar during the rotation process, thereby obtaining a compensated transformed point cloud.

[0076] For example, for each frame point cloud P i Downsampling is performed to reduce the amount of data processing and obtain a sparse point cloud. The first point cloud φ representing the ice-covered cable is extracted from the sparse point cloud. R For example, a clustering algorithm or other algorithm can be used to extract the first point cloud φ representing the ice-covered cable. R . Calculate the first point cloud φ R The angle θ with the normal vector of the XOY plane (horizontal plane) of the laser radar, and the rotation compensation transformation matrix is ​​determined based on the angle θ Then the point cloud P i Multiply by the rotation compensation transformation matrix Get the transformed point cloud P after compensationi′ .

[0077] For example, after obtaining the first point cloud φ R After the angle θ with the normal vector of the XOY plane (horizontal plane) of the laser radar, the rotation compensation transformation matrix It can be expressed as:

[0078]

[0079] 2. Extract key points from the transformed point cloud to obtain an initial key point cloud composed of key points.

[0080] For example, SIFT, Harris, NARF, ISS and SUSAN algorithms can be used to transform the point cloud P i′ Extract key points and obtain the initial key point cloud composed of key points The embodiments of the present invention are not limited here.

[0081] 3. Perform rotation compensation inverse transformation on the key points in the initial key point cloud to obtain the key point cloud of each frame point cloud.

[0082] For example, the initial key point cloud The key points in the rotation compensation inverse transformation are performed to obtain the key point cloud of each frame point cloud

[0083] S103: Input the two frames of key point clouds into the registration network for processing. The registration network adaptively learns the overlapping area of ​​the two frames of point clouds based on the spatial attention mechanism and predicts the relative pose of the two frames of point clouds.

[0084] In an embodiment of the present invention, two frames of key point clouds are input into a pre-trained registration network for processing. The registration network adaptively learns the overlapping area of ​​the two frames of point clouds based on the spatial attention mechanism and predicts the relative pose of the two frames of point clouds.

[0085] Figure 2 A structural diagram of a registration network provided by the present invention is shown in FIG. Figure 2 As shown, for example, each frame of keypoint cloud includes a surface keypoint cloud Surf Keypoint Cloud and a corner keypoint cloud Corner Keypoint Cloud. The registration network includes two feature encoders with a spatial attention weighting mechanism and a feature decoder Decoder. The two frames of keypoint clouds are input into the registration network for processing. The registration network adaptively learns the overlapping area of ​​the two frames of point clouds based on the spatial attention mechanism and predicts the relative pose of the two frames of point clouds, including:

[0086] 1、input the surface keypoint cloud Surf Keypoint Cloud1 of the first frame keypoint cloud and the surface keypoint cloud Surf Keypoint Cloud2 of the second frame keypoint cloud into the first feature encoder Encoder1 for processing, to obtain the surface spatial attention feature Surf Spatial Feature1 of the first frame keypoint cloud and the surface spatial attention feature Surf Spatial Feature2 of the second frame keypoint cloud.

[0087] Figure 3 The structural diagram of a feature encoder provided by the application is shown in Figure 3 The feature encoder includes a multi-layer perception machine MLP and a spatial attention module Spatial Attention Module, and the processing process of the feature encoder is as follows:

[0088] 1.1, for each surface keypoint cloud Surf Keypoint Cloud, the global feature of the surface keypoint cloud is extracted by using a multi-layer perception machine MLP.

[0089] The MLP is similar to the PointNet architecture, and by inputting the surface keypoint cloud Surf Keypoint Cloud into the MLP, the global feature of the surface keypoint cloud Surf Keypoint Cloud can be output.

[0090] 1.2, input the global feature into the spatial attention module for processing, to obtain a spatial attention score matrix composed of spatial attention scores of each element in the global feature.

[0091] For example, the global feature is input into a global maximum pooling layer (Globle Max-Pooling, GMP) and a global average pooling layer (Globle Average-Pooling, GAP) respectively for processing, to obtain global maximum pooling features and global average pooling features. The global maximum pooling features and the global average pooling features are spliced in the channel dimension to obtain spliced features. The spliced features are input into a convolution layer for convolution processing, to obtain a spatial attention score matrix composed of spatial attention scores of each element in the global feature.

[0092] 1.3, normalize the spatial attention score matrix to obtain a spatial attention weight matrix.

[0093] The spatial attention score matrix is activated nonlinearly to obtain a spatial attention weight matrix of the global feature, and each element in the spatial attention weight matrix is a spatial attention weight of the corresponding element in the global feature.

[0094] 1.4, calculate the product of the global feature and the spatial attention weight matrix to obtain the surface spatial attention feature.

[0095] Calculate the product of the global feature and the spatial attention weight matrix, that is, multiply the global feature and the corresponding elements of the spatial attention weight matrix to obtain the surface spatial attention feature Surf Spatial Feature.

[0096] 2, input the corner point cloud Corner Keypoint Cloud1 of the first frame key point cloud and the corner point cloud Corner Keypoint Cloud2 of the second frame key point cloud into the second feature encoder Encoder2 for processing to obtain the corner point spatial attention feature Corner Spatial Feature1 of the first frame key point cloud and the corner point spatial attention feature Corner Spatial Feature2 of the second frame key point cloud.

[0097] The structure and processing process of the second feature encoder and the first feature encoder are similar, and the structure and processing process of the first feature encoder can be referred to the foregoing embodiments, which will not be described here in the embodiments of the application.

[0098] 3, fuse the surface spatial attention feature and the corner point spatial attention feature to obtain the spatial attention fusion feature.

[0099] As shown in Figure 2 , the two surface spatial attention features Surf Spatial Feature1 and Surf Spatial Feature1, and the two corner point spatial attention features Corner Spatial Feature1 and Corner Spatial Feature2 are spliced in dimension, and then the spliced features are subjected to global maximum pooling (Globle Max-Pooling) processing to obtain the spatial attention fusion feature Aggregated Feature.

[0100] 4, input the spatial attention fusion feature into the feature decoder for decoding to obtain the relative pose of the two frames of point clouds.

[0101] In the embodiments of the application, as shown in Figure 2 , the spatial attention fusion feature is input into the feature decoder for decoding to obtain the relative pose of the two frames of point clouds. For example, the feature decoder includes a fully connected layer with 7 hidden layers, and the fully connected layer performs fully connected mapping on the spatial attention fusion feature to obtain a relative pose vector T representing the relative pose of the two frames of point clouds. final(1×7) , wherein the first four quantities represent the rotation quaternion q Tq = 1, the last three quantities represent the translation vector

[0102] S104, register two frames of point clouds based on the relative pose of the two frames of point clouds, and obtain a fusion point cloud.

[0103] After obtaining the relative pose of the two frames of point clouds, the translation transformation matrix of the two frames of point clouds is determined according to the relative pose of the two frames of point clouds, and then the world coordinate system of the two frames of point clouds is combined based on the translation transformation matrix. Similarly, in the above registration process of the present application, all point clouds collected by the laser radar are combined in the world coordinate system, and a fusion point cloud is obtained.

[0104] S105, extract an icing cable point cloud representing an icing cable from the fusion point cloud.

[0105] In the embodiment of the present application, a clustering algorithm or a semantic segmentation algorithm is used to extract an icing cable point cloud representing an icing cable from the fusion point cloud, and the algorithm for extracting the icing cable point cloud representing the icing cable is not limited in the embodiment of the present application.

[0106] S106, determine the three-dimensional size of the icing cable based on the icing cable point cloud.

[0107] In the embodiment of the present application, a three-dimensional incremental modeling method can be used to model the icing cable point cloud and determine the three-dimensional size of the icing cable.

[0108] S107, calculate the icing thickness based on the three-dimensional size of the icing cable.

[0109] In the embodiment of the present application, after determining the three-dimensional size of the icing cable, the icing thickness is calculated in combination with the prior knowledge of the power distribution network cable (such as the cable diameter, etc.). For example, the diameter of the icing cable is subtracted from the cable diameter and divided by 2 to obtain the icing thickness.

[0110] The cable icing detection method based on the laser radar provided by the present application extracts key points from each frame of point cloud to obtain a key point cloud of each frame of point cloud, inputs two frames of key point clouds into a registration network for processing, the registration network adaptively learns the overlapping area of two frames of point clouds based on a spatial attention mechanism, predicts the relative pose of two frames of point clouds, registers two frames of point clouds based on the relative pose of two frames of point clouds, obtains a fusion point cloud, extracts an icing cable point cloud representing an icing cable from the fusion point cloud, determines the three-dimensional size of the icing cable based on the icing cable point cloud, and calculates the icing thickness based on the three-dimensional size of the icing cable. The attention weight mechanism enables the network to adaptively learn the overlapping area of continuous point cloud frames, thereby realizing fast and effective end-to-end registration of large-scale point cloud scenes, providing a good foundation for the fine registration algorithm, improving the registration efficiency and accuracy, and further improving the icing detection accuracy.

[0111] In some embodiments of the present invention, the registration network needs to be trained before it is applied. The existing training process usually requires a large amount of pose truth label data. In the actual scenario of distribution network icing monitoring, obtaining real ground trajectory data is both complex and labor-intensive. To address the problem of lack of pose truth label data for large-scale distribution network lidar scanning point cloud registration, the present invention proposes an unsupervised training method. The specific training process is as follows:

[0112] 1. Prepare multiple training samples in advance. The training samples include source frame point cloud and target point cloud.

[0113] 2. Extract key point clouds of training samples. The process of extracting key point clouds can refer to the above embodiments of the present invention and will not be described in detail here.

[0114] 3. Input the key point cloud of the training sample into the registration network for processing to obtain the relative pose of the source frame point cloud and the target point cloud. The processing process of the registration network can refer to the above embodiment of the present invention and will not be repeated here.

[0115] 4. Extract the non-ground target point cloud in the source frame point cloud and the non-ground target point cloud in the target point cloud.

[0116] That is, the ground point cloud is removed from the source frame point cloud and the target point cloud, and the non-ground target point cloud is retained.

[0117] 5. Based on the relative pose of the source frame point cloud and the target point cloud, the non-ground target point cloud in the source frame point cloud is converted to the target point cloud coordinate system to obtain the converted point cloud.

[0118] Exemplarily, the non-ground target point cloud of the source frame point cloud is multiplied by the relative pose vector output by the registration network to obtain the converted point cloud.

[0119] 6. Calculate the chamfer distance loss between the non-ground target point cloud and the converted point cloud in the target point cloud.

[0120] Chamfer distance loss is a metric used to evaluate the similarity between two sets of point clouds. In this embodiment of the present invention, in order to make the registration network more robust and converge faster, a dynamic overlap coefficient γ is used to perform real-time correction on the chamfer distance loss. For example, the calculation formula of the chamfer distance loss is:

[0121]

[0122] in, is the non-ground target point cloud in the target point cloud, is the converted point cloud, γ is the dynamic overlap coefficient, x is the midpoint of the converted point cloud, For the points in the target point cloud, size() represents the number of points in the point cloud in the parentheses, mean() represents the average value in the parentheses, std() represents the standard deviation in the parentheses, and min() represents the minimum value.

[0123] 7. Update the network parameters of the registration network based on the chamfer distance loss until the chamfer distance loss is less than a preset value.

[0124] For example, the chamfer distance loss corresponding to the sample is compared with the loss threshold, and if the chamfer distance loss is greater than the loss threshold, the network parameters of the registration network are updated, and new training samples are obtained again to repeat the above training process until the chamfer distance loss is less than or equal to the loss threshold, or the number of iterations reaches a preset number.

[0125] Figure 4 The structure of a cable icing detection device based on a laser radar provided by the application is shown in Figure 4 As shown in the figure, the cable icing detection device based on a laser radar comprises:

[0126] The point cloud acquisition module 201 is configured to acquire two frames of point clouds to be registered.

[0127] The key point extraction module 202 is configured to extract key points from each frame of the point cloud to obtain a key point cloud of each frame of the point cloud.

[0128] The relative pose prediction module 203 is configured to input two frames of the key point cloud into a registration network for processing, wherein the registration network is based on a spatial attention mechanism to adaptively learn an overlapping area of two frames of the point cloud and predict a relative pose of two frames of the point cloud.

[0129] The point cloud registration module 204 is configured to register two frames of the point cloud based on the relative pose of two frames of the point cloud to obtain a fused point cloud.

[0130] The icing cable point cloud extraction module 205 is configured to extract an icing cable point cloud representing an icing cable from the fused point cloud.

[0131] The three-dimensional size determination module 206 is configured to determine a three-dimensional size of the icing cable based on the icing cable point cloud.

[0132] The icing thickness calculation module 207 is configured to calculate an icing thickness based on the three-dimensional size of the icing cable.

[0133] In some embodiments of the application, the key point extraction module 202 comprises:

[0134] The rotation compensation submodule is configured to perform a rotation compensation transformation on each frame of the point cloud to obtain a compensated transformed point cloud.

[0135] The key point extraction submodule is configured to extract key points from the transformed point cloud to obtain an initial key point cloud composed of the key points.

[0136] The inverse transformation submodule is configured to perform inverse rotation compensation transformation on the key points in the initial key point cloud to obtain a key point cloud of each frame of the point cloud.

[0137] In some embodiments of the present application, the rotation compensation submodule comprises:

[0138] The downsampling unit is configured to downsample each frame of the point cloud to obtain a sparse point cloud.

[0139] The first point cloud extraction unit is configured to extract a first point cloud representing the iced cable from the sparse point cloud.

[0140] The included angle calculation unit is configured to calculate an included angle between the first point cloud and a normal vector of a horizontal plane.

[0141] The transformation matrix determination unit is configured to determine a rotation compensation transformation matrix based on the included angle.

[0142] The compensation transformation unit is configured to multiply the point cloud by the rotation compensation transformation matrix to obtain a compensated transformed point cloud.

[0143] In some embodiments of the present application, each frame of the key point cloud comprises a surface key point cloud and a corner point cloud, the registration network comprises two feature encoders with a spatial attention weighting mechanism and a feature decoder, and the relative pose prediction module 203 comprises:

[0144] The surface spatial attention feature extraction submodule is configured to input the surface key point cloud of the first frame of key point cloud and the surface key point cloud of the second frame of key point cloud into the first feature encoder for processing to obtain surface spatial attention features of the first frame of key point cloud and the second frame of key point cloud.

[0145] The corner point spatial attention feature extraction submodule is configured to input the corner point cloud of the first frame of key point cloud and the corner point cloud of the second frame of key point cloud into the second feature encoder for processing to obtain corner point spatial attention features of the first frame of key point cloud and the second frame of key point cloud.

[0146] The feature fusion submodule is configured to fuse the surface spatial attention features and the corner point spatial attention features to obtain spatial attention fusion features.

[0147] The decoding submodule is configured to input the spatial attention fusion features into the feature decoder for decoding to obtain relative poses of the two frames of point cloud.

[0148] In some embodiments of the present application, the surface spatial attention feature extraction submodule comprises:

[0149] a global feature extraction unit configured to extract global features of the surface key point cloud using a multi-layer perception network for each of the surface key point clouds;

[0150] an attention processing unit configured to input the global features into a spatial attention module for processing to obtain a spatial attention score matrix composed of spatial attention scores of elements in the global features;

[0151] a normalization unit configured to normalize the spatial attention score matrix to obtain a spatial attention weight matrix;

[0152] a product unit configured to calculate the product of the global features and the spatial attention weight matrix to obtain surface spatial attention features.

[0153] In some embodiments of the present application, the feature decoder comprises a fully connected layer, and the decoding submodule comprises:

[0154] a fully connected mapping unit configured to perform fully connected mapping of the spatial attention fusion features by the fully connected layer to obtain a relative pose vector representing the relative pose of the two frames of point clouds.

[0155] In some embodiments of the present application, the cable icing detection device based on the laser radar further comprises:

[0156] a sample key point cloud extraction module configured to extract key point clouds of training samples before obtaining two frames of point clouds to be registered, the training samples comprising a source frame point cloud and a target point cloud;

[0157] a sample relative pose prediction module configured to input the key point clouds of the training samples into a registration network for processing to obtain the relative pose of the source frame point cloud and the target point cloud;

[0158] a non-ground target point cloud extraction module configured to extract non-ground target point clouds in the source frame point cloud and non-ground target point clouds in the target point cloud;

[0159] a point cloud conversion module configured to convert the non-ground target point clouds in the source frame point cloud to the target point cloud coordinate system based on the relative pose of the source frame point cloud and the target point cloud to obtain a converted point cloud;

[0160] a loss calculation module configured to calculate the chamfer distance loss of the non-ground target point clouds in the target point cloud and the converted point cloud;

[0161] a parameter updating module configured to update the network parameters of the registration network based on the chamfer distance loss until the chamfer distance loss is less than a preset value.

[0162] The above laser radar-based cable icing detection device can execute the laser radar-based cable icing detection method provided by the foregoing embodiments of the present application, and has corresponding functional modules and beneficial effects of the laser radar-based cable icing detection method.

[0163] Figure 5 A block diagram of an electronic device is provided. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices (e.g., headgear, eyewear, watches, etc.), and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the applications described and / or claimed in this document.

[0164] As shown in Figure 5 The electronic device includes at least one processor 11, and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., connected to the at least one processor 11 in communication, where the memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer programs stored in the read-only memory (ROM) 12 or loaded into the random access memory (RAM) 13 from the storage unit 18. In the RAM 13, various programs and data required for operation of the electronic device can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0165] A plurality of components in the electronic device are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunications networks.

[0166] The processor 11 can be various general and / or special purpose processing components having processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 performs various methods and processes described above, such as the laser radar-based cable icing detection method.

[0167] In some embodiments, the laser radar-based cable icing detection method can be implemented as a computer program tangibly embodied in a computer readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device via the ROM 12 and / or the communication unit 19. When the computer program is loaded onto the RAM 13 and executed by the processor 11, one or more steps of the laser radar-based cable icing detection method described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to perform the laser radar-based cable icing detection method by any other appropriate means, such as by means of firmware.

[0168] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (PLD), a computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0169] Computer programs used to implement the methods of the application can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the computer program, when executed by the processor of the machine, implements the functions / acts specified in the flowcharts and / or block diagrams. The computer program can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0170] In the context of the present application, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. A computer-readable storage medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of a machine-readable storage medium will include one or more lines of a program of instructions in a transitory signal, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0171] To provide for interaction with a user, the systems and techniques described here can be implemented on an electronic device having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0172] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), blockchain network, and the Internet.

[0173] The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. Servers can be cloud servers, also known as cloud computing servers or cloud hosts, which are a host product in the cloud computing service system to solve the defects of great management difficulty and weak business scalability in traditional physical hosts and VPS services.

[0174] The embodiment of the present application also provides a computer program product comprising a computer program which, when executed by a processor, implements the laser radar-based cable icing detection method provided by any of the embodiments of the present application.

[0175] The computer program product can be written in one or more programming languages or combinations of languages including object-oriented languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0176] It should be understood that the various forms of flow shown above can be re-ordered, added to, or deleted from without departing from the scope of the present application. For example, the steps recited in the present application can be performed in parallel, in series, or in a different order, and the present application is not limited in this regard.

[0177] The specific embodiments described above are not intended to limit the scope of the present application. Those skilled in the art will understand that various modifications, combinations, sub-combinations, and alternatives of the present application can be made based on the technical idea of the present application. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present application shall fall within the scope of the present application.

Claims

1. A laser radar-based cable icing detection method, characterized by, The method comprises the following steps: obtaining two frames of point clouds to be registered; extracting key points from each frame of the point clouds to obtain key point clouds of each frame of the point clouds; inputting the two frames of the key point clouds into a registration network for processing, wherein the registration network adaptively learns an overlapping area of the two frames of the point clouds based on a spatial attention mechanism and predicts a relative pose of the two frames of the point clouds; registering the two frames of the point clouds based on the relative pose of the two frames of the point clouds to obtain a fused point cloud; extracting an icing cable point cloud representing an icing cable from the fused point cloud; determining a three-dimensional size of the icing cable based on the icing cable point cloud; calculating an icing thickness based on the three-dimensional size of the icing cable; extracting key points from each frame of the point clouds to obtain key point clouds of each frame of the point clouds, comprising: performing a rotation compensation transformation on each frame of the point clouds to obtain a compensated transformed point cloud; performing a rotation compensation transformation on each frame of the point clouds to obtain a compensated transformed point cloud, comprising: down-sampling each frame of the point clouds to obtain a sparse point cloud; extracting a first point cloud representing an icing cable from the sparse point cloud; calculating an angle between the first point cloud and a normal vector of a horizontal plane; determining a rotation compensation transformation matrix based on the angle; multiplying the point cloud by the rotation compensation transformation matrix to obtain a compensated transformed point cloud; Each frame of the key point cloud comprises a surface key point cloud and a corner point cloud, and the registration network comprises two feature encoders with a spatial attention weighting mechanism and a feature decoder. The two frames of the key point cloud are input into the registration network for processing. The registration network adaptively learns an overlapping area of the two frames of the point clouds based on a spatial attention mechanism and predicts a relative pose of the two frames of the point clouds, comprising: inputting the surface key point cloud of the first frame of the key point cloud and the surface key point cloud of the second frame of the key point cloud into the first feature encoder for processing to obtain surface spatial attention features of the first frame of the key point cloud and the second frame of the key point cloud; inputting the corner point cloud of the first frame of the key point cloud and the corner point cloud of the second frame of the key point cloud into the second feature encoder for processing to obtain corner spatial attention features of the first frame of the key point cloud and the second frame of the key point cloud; fusing the surface spatial attention features and the corner spatial attention features to obtain spatial attention fusion features; inputting the spatial attention fusion features into the feature decoder for decoding to obtain the relative pose of the two frames of the point clouds.

2. The laser radar-based cable icing detection method according to claim 1, characterized by, Extracting key points from each frame of the point clouds to obtain key point clouds of each frame of the point clouds also comprises: extracting key points from the transformed point cloud to obtain an initial key point cloud composed of key points; performing a rotation compensation inverse transformation on the key points in the initial key point cloud to obtain a key point cloud of each frame of the point clouds.

3. The laser radar-based cable icing detection method according to claim 1, characterized by, Inputting the surface key point cloud of the first frame of the key point cloud and the surface key point cloud of the second frame of the key point cloud into the first feature encoder for processing to obtain surface spatial attention features of the first frame of the key point cloud and the second frame of the key point cloud, comprising: for each of the surface key point clouds, extracting global features of the surface key point cloud using a multi-layer perception; The global feature is input into a spatial attention module for processing to obtain a spatial attention score matrix composed of spatial attention scores of elements in the global feature; The spatial attention score matrix is normalized to obtain a spatial attention weight matrix; The product of the global feature and the spatial attention weight matrix is calculated to obtain a surface spatial attention feature.

4. The laser radar-based cable icing detection method according to claim 1, characterized by, The feature decoder includes a fully connected layer, and the spatial attention fusion feature is input into the feature decoder for decoding to obtain a relative pose of the two frames of point clouds, including: The fully connected layer performs fully connected mapping on the spatial attention fusion feature to obtain a relative pose vector representing the relative pose of the two frames of point clouds.

5. The laser radar-based cable icing detection method according to claim 1, characterized by, Before obtaining the two frames of point clouds to be registered, further comprising: extracting key point clouds of training samples, the training samples including a source frame point cloud and a target point cloud; inputting the key point clouds of the training samples into a registration network for processing to obtain a relative pose of the source frame point cloud and the target point cloud; extracting non-ground target point clouds in the source frame point cloud and non-ground target point clouds in the target point cloud; converting the non-ground target point clouds in the source frame point cloud to the coordinate system of the target point cloud based on the relative pose of the source frame point cloud and the target point cloud to obtain converted point clouds; calculating an inverse corner distance loss of the non-ground target point clouds in the target point cloud and the converted point clouds; updating network parameters of the registration network based on the inverse corner distance loss until the inverse corner distance loss is less than a preset value. 6.A laser radar-based cable icing detection device, characterized by A method for performing laser radar-based cable icing detection according to any one of claims 1-5, comprising: a point cloud acquisition module for acquiring two frames of point clouds to be registered; a key point extraction module for extracting key points from each frame of the point clouds to obtain key point clouds of each frame of the point clouds; a relative pose prediction module for inputting two frames of the key point clouds into a registration network for processing, the registration network based on a spatial attention mechanism to adaptively learn an overlapping region of the two frames of point clouds and predict a relative pose of the two frames of point clouds; a point cloud registration module for registering two frames of point clouds based on the relative pose of the two frames of point clouds to obtain a fusion point cloud; an iced cable point cloud extraction module for extracting an iced cable point cloud representing an iced cable from the fusion point cloud; a three-dimensional size determination module for determining a three-dimensional size of the iced cable based on the iced cable point cloud; an icing thickness calculation module for calculating an icing thickness based on the three-dimensional size of the iced cable.

7. An electronic device, comprising: comprising: one or more processors; a storage device for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement a laser radar-based cable icing detection method according to any one of claims 1-5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement a laser radar-based cable icing detection method according to any one of claims 1-5.

Images