Overhead line point cloud semantic segmentation optimization method and device, electronic equipment and storage medium

Abstract

The invention discloses an overhead line point cloud semantic segmentation optimization method and apparatus, an electronic device and a storage medium. The method comprises the steps of obtaining a to-be-segmented original point cloud; performing feature extraction and down-sampling on the original point cloud to generate a depth point cloud; clustering the depth point clouds to generate a plurality of clusters, and taking the cluster centers of the clusters as planar poles; if the number of the planar poles is not zero, calculating the distance from each point in the original point cloud to each planar pole, and dividing the original point cloud according to the distances to generate a plurality of sub-point clouds; updating each sub-point cloud into a current point cloud in sequence, and executing a point cloud semantic segmentation operation to generate a corresponding segmentation result; taking the segmentation result of each sub-point cloud as the segmentation result of the original point cloud; and if the number of the plane poles is 0, updating the original point cloud into the current point cloud, and executing a point cloud semantic segmentation operation to generate a segmentation result of the original point cloud. According to the invention, the precision of the point cloud semantic segmentation of the overhead line can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid operation and maintenance, and in particular to an optimization method, device, electronic equipment and storage medium for semantic segmentation of overhead line point clouds. Background Art

[0002] With the rapid development of LiDAR technology, point cloud data is increasingly used in overhead line inspection. Point cloud semantic segmentation is to assign specific semantic categories to each point in the point cloud data, such as ground, tower, conductor, etc., for subsequent analysis and processing. In overhead line inspection, accurate segmentation of ground point clouds is of great significance for tasks such as calculating line safety distances and detecting tower inclination. This segmentation can provide reliable data support for the management and maintenance of power facilities, ensuring the safety and efficiency of the lines.

[0003] However, when processing point cloud data, existing technologies often fail to effectively distinguish between planar point clouds and non-planar point clouds, but instead perform mixed processing on them, which significantly reduces the accuracy of point cloud semantic segmentation. Planar point clouds usually have more regular and uniform features, while non-planar point clouds may contain more noise and complex structures. If these two types of point clouds are not clearly distinguished, the segmentation algorithm will be disturbed by the mixed features, making it difficult to accurately identify different objects and features. In addition, the noise and outliers in non-planar point clouds will further affect the quality of the overall data, resulting in unreliable segmentation results. Summary of the invention

[0004] The embodiments of the present invention provide an optimization method, device, electronic device and storage medium for semantic segmentation of overhead line point clouds. By implementing the present invention, planar point clouds and non-planar point clouds can be effectively distinguished, thereby improving the accuracy of semantic segmentation of overhead line point clouds.

[0005] An embodiment of the present invention provides an optimization method for semantic segmentation of overhead line point clouds, comprising: obtaining an original point cloud to be segmented of the overhead line.

[0006] Perform feature extraction and downsampling operations on the original point cloud to generate a depth point cloud.

[0007] The depth point cloud is clustered to generate a plurality of clusters, and the cluster center of each cluster is set as a plane pole.

[0008] Determine the number of plane poles. If the number of plane poles is not 0, calculate the distance from each point in the original point cloud to each plane pole, divide the original point cloud according to the distance, and generate several sub-point clouds; update each sub-point cloud to the current point cloud in turn, perform the point cloud semantic segmentation operation, and generate the corresponding segmentation result; use the segmentation result of each sub-point cloud as the segmentation result of the original point cloud.

[0009] If the number of plane poles is 0, the original point cloud is updated to the current point cloud, and the point cloud semantic segmentation operation is performed to generate the segmentation result of the original point cloud.

[0010] The point cloud semantic segmentation operation includes:

[0011] Perform plane fitting on the current point cloud to generate the point cloud plane equation.

[0012] The current point cloud is divided by the point cloud plane equation to generate a planar point cloud and a non-planar point cloud; the planar point cloud is segmented by the point cloud plane equation to obtain a segmentation result of the planar point cloud.

[0013] The non-planar point cloud is input into a preset point cloud semantic segmentation neural network for segmentation to obtain the segmentation result of the non-planar point cloud.

[0014] The segmentation result of the planar point cloud and the segmentation result of the non-planar point cloud are used as the segmentation result of the current point cloud.

[0015] Furthermore, the feature extraction and downsampling operations are performed on the original point cloud to generate a depth point cloud, including:

[0016] Input the original point cloud into the multi-layer perceptron to extract features and generate the first feature point cloud;

[0017] Performing a downsampling operation on the first feature point cloud to generate a first downsampled point cloud;

[0018] Input the first down-sampled point cloud into a multi-layer perceptron to extract features and generate a second feature point cloud;

[0019] The second feature point cloud is downsampled to generate a second downsampled point cloud; and the second downsampled point cloud is used as a depth point cloud.

[0020] Furthermore, the original point cloud is divided according to the distance to generate a plurality of sub-point clouds, including:

[0021] For each point in the original point cloud, select the plane pole with the smallest distance to it as the target pole of the point;

[0022] The original point cloud is divided according to the target poles to generate several sub-point clouds.

[0023] Furthermore, performing plane fitting on the current point cloud to generate a point cloud plane equation includes:

[0024] Repeat the point cloud plane fitting operation until the proportion of internal points is less than the preset convergence threshold, and generate the point cloud plane equation;

[0025] The point cloud plane fitting operation includes:

[0026] Randomly select several points from the current point cloud as fitting points;

[0027] According to the fitting points, the parameters of the preset plane equation are solved to generate a specific plane equation;

[0028] According to the specific plane equation, generate the first point cloud plane;

[0029] Calculate the distances from all points of the current point cloud except the fitting point to the first point cloud plane, and compare each distance with a preset first distance threshold;

[0030] Recording points whose distance from the first point cloud plane is less than a preset first distance threshold as internal points;

[0031] Calculate the ratio of the number of inliers to the total number of points in the current point cloud to get the inliers ratio;

[0032] The proportion of inliers is compared with the preset convergence threshold;

[0033] When the proportion of internal points is less than a preset convergence threshold, the specific plane equation is used as the point cloud plane equation.

[0034] When the proportion of inliers exceeds the preset convergence threshold, several new points are randomly selected as fitting points.

[0035] Furthermore, dividing the current point cloud by the point cloud plane equation to generate a planar point cloud and a non-planar point cloud includes:

[0036] Generate a second point cloud plane according to the point cloud plane equation;

[0037] Calculate the distance between each point of the current point cloud and the second point cloud plane, and compare each distance with a preset first distance threshold;

[0038] The points whose distances from the second point cloud plane are less than a preset first distance threshold are combined as a plane point cloud;

[0039] The points whose distance from the second point cloud plane is not less than a preset first distance threshold are combined as a non-planar point cloud.

[0040] Based on the above method embodiment, the present invention provides a corresponding device embodiment.

[0041] An embodiment of the present invention provides an optimization device for semantic segmentation of overhead line point clouds, comprising: an original point cloud acquisition module, a depth point cloud generation module, a plane pole generation module, a sub-point cloud generation module, a sub-point cloud segmentation module, an original point cloud segmentation module and a point cloud semantic segmentation operation module.

[0042] The original point cloud acquisition module is used to acquire the original point cloud to be segmented of the overhead line.

[0043] The depth point cloud generation module is used to perform feature extraction and down-sampling operations on the original point cloud to generate a depth point cloud.

[0044] The plane pole generation module is used to cluster the depth point cloud to generate a plurality of clusters, and set the cluster center of each cluster as a plane pole.

[0045] The sub-point cloud generation module is used to calculate the distance from each point in the original point cloud to each plane pole when the number of plane poles is not 0, divide the original point cloud according to the distance, and generate a plurality of sub-point clouds.

[0046] The sub-point cloud segmentation module is used to update each sub-point cloud to the current point cloud in turn when generating several sub-point clouds, perform point cloud semantic segmentation operations, and generate corresponding segmentation results; and use the segmentation result of each sub-point cloud as the segmentation result of the original point cloud.

[0047] The original point cloud segmentation module is used to update the original point cloud to the current point cloud if the number of plane poles is 0, perform the point cloud semantic segmentation operation, and generate the segmentation result of the original point cloud.

[0048] The point cloud semantic segmentation operation module is used to perform plane fitting on the current point cloud to generate a point cloud plane equation; divide the current point cloud by the point cloud plane equation to generate a planar point cloud and a non-planar point cloud; segment the planar point cloud by the point cloud plane equation to obtain a segmentation result of the planar point cloud; input the non-planar point cloud into a preset point cloud semantic segmentation neural network for segmentation to obtain a segmentation result of the non-planar point cloud; and use the segmentation results of the planar point cloud and the non-planar point cloud as the segmentation results of the current point cloud.

[0049] Based on the above method embodiment, the present invention provides a corresponding electronic device embodiment.

[0050] An embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the optimization method for semantic segmentation of overhead line point clouds described in any one of the above-mentioned method embodiments can be implemented.

[0051] Based on the above method item embodiments, the present invention provides a corresponding storage medium item embodiment.

[0052] An embodiment of the present invention provides a storage medium having a computer program stored thereon. When the computer program is executed by a processor, the optimization method for semantic segmentation of overhead line point clouds described in any one of the above method embodiments can be implemented.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] The embodiment of the present invention provides an optimization method, device, electronic device and storage medium for semantic segmentation of overhead line point cloud, which obtains the original point cloud to be segmented of the overhead line; performs feature extraction and downsampling operations on the original point cloud to generate a deep point cloud; clusters the deep point cloud to generate a number of clusters, and sets the cluster center of each cluster as a plane pole; the plane pole is used as a reference point to help the subsequent distance calculation and area division, ensuring that each sub-point cloud is concentrated in a specific feature area. The distance from the original point cloud to the plane pole is calculated, and the original point cloud is divided into multiple sub-point clouds, so as to ensure that each sub-point cloud is more concentrated in a specific feature area. For each sub-point cloud, plane fitting is further used to extract its plane and non-plane parts, so that the plane point cloud and the non-plane point cloud can use different segmentation methods respectively. Among them, the plane point cloud is segmented according to the plane equation, and the non-plane point cloud is segmented according to the preset point cloud semantic segmentation neural network, which ensures the segmentation accuracy when processing complex structures, makes the data processing of different characteristic areas more targeted, and avoids the decrease in segmentation accuracy caused by mixed processing of different characteristic point clouds. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flow chart of an optimization method for semantic segmentation of overhead line point clouds provided by one embodiment of the present invention.

[0056] Figure 2 It is a flowchart of a point cloud semantic segmentation operation provided by an embodiment of the present invention.

[0057] Figure 3 It is a structural schematic diagram of an optimization device for semantic segmentation of overhead line point clouds provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe 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.

[0059] like Figure 1As shown, an embodiment of the present invention provides an optimization method for semantic segmentation of overhead line point clouds, which comprises at least the following steps:

[0060] Step S1, obtaining the original point cloud of the overhead line to be segmented;

[0061] Specifically, the overhead line is scanned by a laser scanner or other sensor to capture the three-dimensional spatial information of its surrounding environment. This process usually involves a comprehensive scan of a specific area to ensure that complete point cloud data is obtained, including the line itself and its surrounding supporting structures, vegetation, and other related objects. The acquired raw point cloud data has the characteristics of high density and high precision, which can truly reflect the spatial position and shape of the overhead line, and provide a basis for subsequent instance segmentation.

[0062] Step S2: perform feature extraction and downsampling operations on the original point cloud to generate a depth point cloud.

[0063] In a preferred embodiment, the feature extraction and downsampling operations are performed on the original point cloud to generate a depth point cloud, including:

[0064] The original point cloud is input into the multi-layer perceptron for feature extraction to generate the first feature point cloud.

[0065] The first feature point cloud is downsampled to generate a first downsampled point cloud.

[0066] The first down-sampled point cloud is input into a multi-layer perceptron for feature extraction to generate a second feature point cloud.

[0067] The second feature point cloud is downsampled to generate a second downsampled point cloud; and the second downsampled point cloud is used as a depth point cloud.

[0068] In the specific implementation, the original point cloud data is input into a multi-layer perceptron (MLP) for processing. The multi-layer perceptron can effectively capture the spatial relationship and feature information of each point in the point cloud through multiple hidden layers, thereby realizing feature extraction. Under the processing of the multi-layer perceptron, a first feature point cloud is generated. This feature point cloud contains richer geometric and topological information, reflecting the feature distribution of each point in the original point cloud and their mutual relationship. The first feature point cloud is downsampled to reduce the density and computational complexity of the point cloud. The downsampling method can use random sampling, uniform sampling, or importance sampling to generate the first downsampled point cloud. This step helps to reduce the computational complexity while maintaining important information. The first downsampled point cloud is input into the multi-layer perceptron again for feature extraction to generate a second feature point cloud. By extracting features again, the representation ability of the point cloud can be further enhanced and higher-level feature information can be captured. The second feature point cloud is downsampled to generate a second downsampled point cloud. This process is also intended to reduce the amount of data, making subsequent processing more efficient. The second downsampled point cloud is used as the final depth point cloud. The deep point cloud not only contains data that has undergone multiple feature extractions and optimizations, but also can effectively retain important structural information in the original point cloud, providing a good foundation for subsequent instance segmentation and analysis.

[0069] Step S3, clustering the depth point cloud to generate a number of clusters, and setting the cluster center of each cluster as a plane pole;

[0070] First, select a suitable clustering algorithm based on the characteristics and application requirements of the deep point cloud. Optional clustering algorithms include K-means, DBSCAN, and Mean Shift. Input the deep point cloud data into the selected clustering algorithm for processing. During the clustering process, the algorithm divides the points into several clusters based on the distance or similarity between the points in the point cloud. Each cluster represents a group of points that are close to each other in space and have similar features. After clustering is completed, for each cluster, calculate its cluster center. The cluster center can be obtained by averaging the coordinates of all points in the cluster. The cluster center represents the main feature position of the cluster and can effectively summarize the spatial distribution of points in the cluster. Set the cluster center of each cluster as a plane pole. These plane poles will serve as key reference points in subsequent processing to provide support for the subsequent point cloud semantic segmentation.

[0071] Step S4, determine the number of plane poles. If the number of plane poles is not 0, calculate the distance from each point in the original point cloud to each plane pole, divide the original point cloud according to the distance, and generate several sub-point clouds; update each sub-point cloud to the current point cloud in turn, perform the point cloud semantic segmentation operation, and generate the corresponding segmentation result; use the segmentation result of each sub-point cloud as the segmentation result of the original point cloud. If the number of plane poles is 0, update the original point cloud to the current point cloud, perform the point cloud semantic segmentation operation, and generate the segmentation result of the original point cloud;

[0072] First, check the number of currently acquired plane poles to determine the subsequent processing flow. If the number of detected plane poles is not 0, it means that the input point cloud contains undulating ground. At this time, first calculate the distance from each point in the original point cloud to each plane pole, divide the original point cloud according to these distances, and generate several sub-point clouds to process the detailed features in each area separately. Take each sub-point cloud as the currently processed point cloud in turn, and perform point cloud semantic segmentation operations on it to obtain the corresponding segmentation results; finally, integrate the segmentation results of each sub-point cloud as the overall segmentation result of the original point cloud.

[0073] If the number of plane poles is 0, it means that the input point cloud ground is a plane. At this time, the original point cloud is directly updated to the current point cloud, and a one-time point cloud semantic segmentation operation is performed to obtain the complete segmentation result of the original point cloud.

[0074] In an optional embodiment, dividing the original point cloud according to the distance to generate a plurality of sub-point clouds includes:

[0075] For each point in the original point cloud, select the plane pole with the smallest distance to it as the target pole of the point;

[0076] The original point cloud is divided according to the target poles to generate several sub-point clouds.

[0077] In actual operation, for each point in the original point cloud, the distance between it and all plane poles is calculated. Usually the Euclidean distance is used as the metric, and the calculation formula is:

[0078]

[0079] in, p is a point in the original point cloud, c is a certain plane pole;

[0080] By comparing the distances between all plane poles and the current point, the plane pole with the smallest distance is selected as the target pole of the point. For efficient processing, spatial data structures (such as KD trees) can be used to accelerate the nearest neighbor search and reduce calculation time. Each point and its corresponding target pole are associated to form a one-to-one mapping relationship. Each point in the original point cloud has a certain target pole, so that the information in the point cloud can be structured for subsequent processing. The original point cloud is divided according to the target pole, and the points with the same target pole are classified into the same sub-point cloud, thereby generating several sub-point clouds.

[0081] like Figure 2 As shown, in a preferred embodiment, the point cloud semantic segmentation operation includes:

[0082] Step S4.1, perform plane fitting on the current point cloud to generate the point cloud plane equation.

[0083] Step S4.2, dividing the current point cloud by the point cloud plane equation to generate a planar point cloud and a non-planar point cloud; segmenting the planar point cloud by the point cloud plane equation to obtain a segmentation result of the planar point cloud.

[0084] Step S4.3: input the non-planar point cloud into a preset point cloud semantic segmentation neural network for segmentation to obtain a segmentation result of the non-planar point cloud.

[0085] Step S4.4: Use the segmentation result of the planar point cloud and the segmentation result of the non-planar point cloud as the segmentation result of the current point cloud.

[0086] It should be noted that the plane fitting of the current point cloud to generate the point cloud plane equation includes:

[0087] Repeat the point cloud plane fitting operation until the proportion of internal points is less than the preset convergence threshold, and generate the point cloud plane equation;

[0088] The point cloud plane fitting operation includes:

[0089] Randomly select several points from the current point cloud as fitting points;

[0090] According to the fitting points, the parameters of the preset plane equation are solved to generate a specific plane equation;

[0091] According to the specific plane equation, generate the first point cloud plane;

[0092] Calculate the distances from all points of the current point cloud except the fitting point to the first point cloud plane, and compare each distance with a preset first distance threshold;

[0093] Recording points whose distance from the first point cloud plane is less than a preset first distance threshold as internal points;

[0094] Calculate the ratio of the number of inliers to the total number of points in the current point cloud to get the inliers ratio;

[0095] The proportion of inliers is compared with the preset convergence threshold;

[0096] When the proportion of internal points is less than a preset convergence threshold, the specific plane equation is used as the point cloud plane equation.

[0097] When the proportion of inliers exceeds the preset convergence threshold, several new points are randomly selected as fitting points.

[0098] In one embodiment, dividing the current point cloud by the point cloud plane equation to generate a planar point cloud and a non-planar point cloud includes:

[0099] Generate a second point cloud plane according to the point cloud plane equation;

[0100] Calculate the distance between each point of the current point cloud and the second point cloud plane, and compare each distance with a preset first distance threshold;

[0101] The points whose distances from the second point cloud plane are less than a preset first distance threshold are combined as a plane point cloud;

[0102] The points whose distance from the second point cloud plane is not less than a preset first distance threshold are combined as a non-planar point cloud.

[0103] Select a suitable plane model according to the characteristics of the point cloud. In one embodiment, the plane model can be in the form of a parameterized plane equation:

[0104] Ax+By+Cz+D=0

[0105] Among them, A, B, C and D are the parameters to be fitted, and x, y, z are the coordinates of the point cloud.

[0106] The parameters of the plane equation are estimated using a randomly selected set of points through an iterative algorithm. The process continuously adjusts the model parameters to minimize the sum of the distances from the points to the fitted plane, thereby generating an accurate plane equation. For example, three points (x1, y1, z1), (x2, y2, z2), (x3, y3, z3) are randomly selected, and the plane parameters A, B, C, and D are solved by the following set of equations:

[0107] Then, the distance from other points to the plane is calculated. If the distance is less than the set threshold, the point is considered to conform to the plane model. The threshold value of this solution is selected as 0.3 meters.

[0108]

[0109] Then, the distance from other points to the plane is calculated. If the distance is less than the set threshold, the point is considered to conform to the plane model. The threshold value of this solution is selected as 0.3 meters.

[0110] Specifically, according to the point cloud plane equation, for each point, calculate its distance to the plane. If the distance is less than the set threshold, the point is considered a plane point; otherwise, it is considered a non-plane point. According to the above classification, the current point cloud is divided into two subsets: plane point cloud and non-plane point cloud. The plane point cloud contains all points that match the plane equation, while the non-plane point cloud contains all points that fail to conform to the plane model. The plane point cloud is further processed using the point cloud plane equation to generate the segmentation result of the plane point cloud. For non-plane point clouds, the non-plane point clouds are used as input and sent to the pre-trained point cloud semantic segmentation neural network. The network will output the segmentation result of the non-plane point cloud.

[0111] Based on the above method embodiment, the present invention provides a corresponding device embodiment.

[0112] like Figure 3 As shown, an embodiment of the present invention provides an optimization device for semantic segmentation of overhead line point clouds, including: an original point cloud acquisition module 101, a depth point cloud generation module 102, a plane pole generation module 103, a sub-point cloud generation module 104, a sub-point cloud segmentation module 105, an original point cloud segmentation module 106 and a point cloud semantic segmentation operation module 107.

[0113] The original point cloud acquisition module 101 is used to acquire the original point cloud to be segmented of the overhead line;

[0114] The depth point cloud generation module 102 is used to perform feature extraction and downsampling operations on the original point cloud to generate a depth point cloud;

[0115] The plane pole generating module 103 is used to cluster the depth point cloud to generate a plurality of clusters, and set the cluster center of each cluster as a plane pole;

[0116] The sub-point cloud generation module 104 is used to calculate the distance from each point in the original point cloud to each plane pole when the number of plane poles is not 0, and divide the original point cloud according to the distance to generate a plurality of sub-point clouds;

[0117] The sub-point cloud segmentation module 105 is used to update each sub-point cloud to the current point cloud in turn when generating a plurality of sub-point clouds, perform a point cloud semantic segmentation operation, and generate a corresponding segmentation result; and use the segmentation result of each sub-point cloud as the segmentation result of the original point cloud;

[0118] The original point cloud segmentation module 106 is used to update the original point cloud to the current point cloud if the number of plane poles is 0, perform the point cloud semantic segmentation operation, and generate the segmentation result of the original point cloud;

[0119] The point cloud semantic segmentation operation module 107 is used to perform plane fitting on the current point cloud to generate a point cloud plane equation; divide the current point cloud by the point cloud plane equation to generate a planar point cloud and a non-planar point cloud; segment the planar point cloud by the point cloud plane equation to obtain a segmentation result of the planar point cloud; input the non-planar point cloud into a preset point cloud semantic segmentation neural network for segmentation to obtain a segmentation result of the non-planar point cloud; and use the segmentation results of the planar point cloud and the non-planar point cloud as the segmentation results of the current point cloud.

[0120] In a preferred embodiment, the depth point cloud generation module includes: a first feature point cloud generation unit, a first down-sampling point cloud generation unit, a second feature point cloud generation unit, and a second down-sampling point cloud generation unit;

[0121] The first feature point cloud generating unit is used to input the original point cloud into the multi-layer perceptron to perform feature extraction and generate a first feature point cloud;

[0122] The first down-sampling point cloud generating unit is used to perform a down-sampling operation on the first feature point cloud to generate a first down-sampling point cloud;

[0123] The second feature point cloud generating unit is used to input the first down-sampled point cloud into a multi-layer perceptron to perform feature extraction and generate a second feature point cloud;

[0124] The second down-sampling point cloud generating unit is used to perform a down-sampling operation on the second feature point cloud to generate a second down-sampling point cloud; and use the second down-sampling point cloud as a depth point cloud.

[0125] It should be noted that the embodiments of the device described above correspond to the above embodiments of the present invention, and can implement any of the methods described above in the present invention. In addition, the embodiments of the above-mentioned device are merely schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which can be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0126] Based on the above method embodiment of the present invention, a corresponding electronic device embodiment is provided.

[0127] An embodiment of the present invention provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the optimization method for semantic segmentation of overhead line point clouds described in any one of the present invention is implemented, or, when the processor executes the computer program, the functions of the modules in the above-mentioned device embodiments are implemented.

[0128] Exemplarily, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the terminal device.

[0129] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0130] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0131] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the terminal device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0132] Based on the above method embodiment, the present invention provides a storage medium embodiment;

[0133] Another embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute any one of the above-mentioned optimization methods for semantic segmentation of overhead line point clouds of the present invention.

[0134] The storage medium is a computer-readable storage medium, and the computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0135] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0136] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. An optimization method for semantic segmentation of overhead line point clouds, characterized in that: include: Obtaining the original point cloud of the overhead line to be segmented; Perform feature extraction and downsampling operations on the original point cloud to generate a depth point cloud; Clustering the depth point cloud to generate a plurality of clusters, and setting the cluster center of each cluster as a plane pole; Determine the number of plane poles. If the number of plane poles is not 0, calculate the distance from each point in the original point cloud to each plane pole, divide the original point cloud according to the distance, and generate several sub-point clouds; update each sub-point cloud to the current point cloud in turn, perform the point cloud semantic segmentation operation, and generate the corresponding segmentation result; use the segmentation result of each sub-point cloud as the segmentation result of the original point cloud; If the number of plane poles is 0, update the original point cloud to the current point cloud, perform the point cloud semantic segmentation operation, and generate the segmentation result of the original point cloud; The point cloud semantic segmentation operation includes: Perform plane fitting on the current point cloud to generate the point cloud plane equation; The current point cloud is divided by the point cloud plane equation to generate a planar point cloud and a non-planar point cloud; the planar point cloud is segmented by the point cloud plane equation to obtain a segmentation result of the planar point cloud; The non-planar point cloud is input into a preset point cloud semantic segmentation neural network for segmentation to obtain the segmentation result of the non-planar point cloud; The segmentation result of the planar point cloud and the segmentation result of the non-planar point cloud are used as the segmentation result of the current point cloud.

2. The optimization method for semantic segmentation of overhead line point cloud according to claim 1, characterized in that: The feature extraction and downsampling operations are performed on the original point cloud to generate a depth point cloud, including: Input the original point cloud into the multi-layer perceptron to extract features and generate the first feature point cloud; Performing a downsampling operation on the first feature point cloud to generate a first downsampled point cloud; Input the first down-sampled point cloud into a multi-layer perceptron to extract features and generate a second feature point cloud; The second feature point cloud is downsampled to generate a second downsampled point cloud; and the second downsampled point cloud is used as a depth point cloud.

3. The optimization method for semantic segmentation of overhead line point cloud according to claim 2, characterized in that: The dividing the original point cloud according to the distance to generate a plurality of sub-point clouds includes: For each point in the original point cloud, select the plane pole with the smallest distance to it as the target pole of the point; The original point cloud is divided according to the target poles to generate several sub-point clouds.

4. The optimization method for semantic segmentation of overhead line point cloud according to claim 3, characterized in that: The performing plane fitting on the current point cloud to generate the point cloud plane equation includes: Repeat the point cloud plane fitting operation until the proportion of internal points is less than the preset convergence threshold, and generate the point cloud plane equation; The point cloud plane fitting operation includes: Randomly select several points from the current point cloud as fitting points; According to the fitting points, the parameters of the preset plane equation are solved to generate a specific plane equation; According to the specific plane equation, generate the first point cloud plane; Calculate the distances from all points of the current point cloud except the fitting point to the first point cloud plane, and compare each distance with a preset first distance threshold; Recording points whose distance from the first point cloud plane is less than a preset first distance threshold as internal points; Calculate the ratio of the number of inliers to the total number of points in the current point cloud to get the inliers ratio; The proportion of inliers is compared with the preset convergence threshold; When the proportion of the inner points is less than a preset convergence threshold, the specific plane equation is used as the point cloud plane equation; When the proportion of inliers exceeds the preset convergence threshold, several new points are randomly selected as fitting points.

5. The optimization method for semantic segmentation of overhead line point cloud according to claim 4, characterized in that: The step of dividing the current point cloud by the point cloud plane equation to generate a planar point cloud and a non-planar point cloud includes: Generate a second point cloud plane according to the point cloud plane equation; Calculate the distance between each point of the current point cloud and the second point cloud plane, and compare each distance with a preset first distance threshold; The points whose distance from the second point cloud plane is less than a preset first distance threshold are combined as a plane point cloud; The points whose distance from the second point cloud plane is not less than a preset first distance threshold are combined as a non-planar point cloud.

6. An optimization device for semantic segmentation of overhead line point clouds, characterized in that: include: Original point cloud acquisition module, depth point cloud generation module, plane pole generation module, sub-point cloud generation module, sub-point cloud segmentation module, original point cloud segmentation module and point cloud semantic segmentation operation module; The original point cloud acquisition module is used to acquire the original point cloud to be segmented of the overhead line; The depth point cloud generation module is used to perform feature extraction and downsampling operations on the original point cloud to generate a depth point cloud; The plane pole generation module is used to cluster the depth point cloud to generate a plurality of clusters, and set the cluster center of each cluster as a plane pole; The sub-point cloud generation module is used to calculate the distance from each point in the original point cloud to each plane pole when the number of plane poles is not 0, divide the original point cloud according to the distance, and generate a plurality of sub-point clouds; The sub-point cloud segmentation module is used to update each sub-point cloud to the current point cloud in turn when generating a plurality of sub-point clouds, perform a point cloud semantic segmentation operation, and generate a corresponding segmentation result; and use the segmentation result of each sub-point cloud as the segmentation result of the original point cloud; The original point cloud segmentation module is used to update the original point cloud to the current point cloud if the number of plane poles is 0, perform the point cloud semantic segmentation operation, and generate the segmentation result of the original point cloud; The point cloud semantic segmentation operation module is used to perform plane fitting on the current point cloud to generate a point cloud plane equation; divide the current point cloud by the point cloud plane equation to generate a planar point cloud and a non-planar point cloud; The plane point cloud is segmented by the point cloud plane equation to obtain the segmentation result of the plane point cloud; the non-planar point cloud is input into the preset point cloud semantic segmentation neural network for segmentation to obtain the segmentation result of the non-planar point cloud; The segmentation result of the planar point cloud and the segmentation result of the non-planar point cloud are used as the segmentation result of the current point cloud.

7. The optimization device for semantic segmentation of overhead line point cloud according to claim 6, characterized in that: The depth point cloud generation module includes: a first feature point cloud generation unit, a first down-sampling point cloud generation unit, a second feature point cloud generation unit, and a second down-sampling point cloud generation unit; The first feature point cloud generating unit is used to input the original point cloud into the multi-layer perceptron to perform feature extraction and generate a first feature point cloud; The first down-sampling point cloud generating unit is used to perform a down-sampling operation on the first feature point cloud to generate a first down-sampling point cloud; The second feature point cloud generating unit is used to input the first down-sampled point cloud into a multi-layer perceptron to perform feature extraction and generate a second feature point cloud; The second down-sampling point cloud generating unit is used to perform a down-sampling operation on the second feature point cloud to generate a second down-sampling point cloud; and use the second down-sampling point cloud as a depth point cloud.

8. The optimization device for semantic segmentation of overhead line point cloud according to claim 7, characterized in that: The sub-point cloud generation module divides the original point cloud according to the distance to generate a plurality of sub-point clouds, including: For each point in the original point cloud, select the plane pole with the smallest distance to it as the target pole of the point; The original point cloud is divided according to the target poles to generate several sub-point clouds.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, it can implement the optimization method for semantic segmentation of overhead line point cloud as described in any one of claims 1 to 5.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it can implement the optimization method for semantic segmentation of overhead line point clouds as described in any one of claims 1 to 5.

Images