Power transmission line laser radar point cloud splicing method and system based on feature point matching
By automatically extracting feature points through a deep learning model and combining it with an optimization algorithm, the problems of low precision, low efficiency, and high noise impact in point cloud stitching are solved, achieving high-precision and fast point cloud data stitching, which is suitable for three-dimensional modeling of transmission lines in complex environments.
Patent Information
- Application Number
- CN202510673266.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-23
- Publication Date
- 2025-10-17
AI Technical Summary
Existing point cloud stitching technology has shortcomings in feature point selection and matching accuracy, computational efficiency, and data noise robustness, especially in complex environments where it is difficult to meet high precision and real-time requirements.
A deep learning model is used to automatically extract feature points, and an optimization algorithm is combined to perform point cloud matching and stitching, including preprocessing, feature point extraction, matching and alignment optimization. The feature descriptor of the deep learning model is used to measure the similarity of point clouds, and the stitching process is optimized through an iterative closest point algorithm.
It improves the accuracy and computational efficiency of point cloud stitching, enhances the robustness to noise and data missing, and adapts to point cloud data processing in complex environments.
Smart Images
Figure CN120807855A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of electric power inspection and three-dimensional modeling, and particularly relates to a power transmission line laser radar point cloud splicing method and system based on feature point matching. BACKGROUND
[0002] With the increasing demand for automatic inspection of power transmission lines in the power industry, laser radar (LiDAR) technology has been widely concerned in the application of power transmission line inspection. The point cloud data collected by laser radar can provide high-precision spatial information for detailed three-dimensional modeling of power transmission lines and their surrounding environment. Point cloud splicing technology is a key link to effectively utilize laser radar data, and its main purpose is to effectively align and fuse point cloud data collected at different angles and different times to generate a complete three-dimensional model.
[0003] Currently, the point cloud splicing method based on feature point matching has become one of the mainstream technologies. Feature points play a key role in point cloud splicing because they can provide reliable registration basis in different point cloud segments. Common feature points include edge points, plane points, etc., which have strong geometric characteristics and can effectively assist in aligning point cloud data. However, existing point cloud splicing technologies still face some challenges, mainly in the following aspects:
[0004] (1) Selection and matching accuracy of feature points: Traditional feature point matching methods often rely on specific geometric features, which may not be able to effectively extract stable feature points in some complex environments, resulting in increased splicing errors and affecting the accuracy of the final results.
[0005] (2) Computational efficiency: When dealing with large amounts of point cloud data, existing splicing methods often require high computational matching processes, resulting in low computational efficiency and failing to meet real-time requirements, especially in large-scale power transmission line three-dimensional modeling, which often leads to long processing times.
[0006] (3) Incompleteness and noise of point cloud data: In actual applications, due to factors such as sensor accuracy and environmental interference, the collected point cloud data often has noise and holes, which can seriously affect the accuracy and reliability of splicing. SUMMARY
[0007] In view of the above problems, the present application is proposed.
[0008] Therefore, the application aims to improve the selection and matching strategy of feature points, improve the splicing accuracy, reduce the calculation complexity, and enhance the robustness to noise and data loss. Specifically, through the innovative feature point extraction and matching algorithm, the application can stably select effective feature points under different environmental conditions, accurately complete point cloud splicing, and thus solve the problems of poor splicing accuracy, low calculation efficiency and large data noise in the prior art.
[0009] To solve the above technical problems, the application provides the following technical solutions: a power transmission line laser radar point cloud splicing method based on feature point matching, comprising the following steps,
[0010] Collecting original laser radar point cloud data and preprocessing; training a feature point extraction model using a deep learning model to automatically extract feature points from the point cloud data; performing feature point matching based on the feature points extracted by the deep learning model; after completing the feature point matching, aligning and optimizing the matched point cloud data using an optimization algorithm; and merging two or more point cloud segments into a three-dimensional model through a splicing algorithm.
[0011] As a preferred scheme of the power transmission line laser radar point cloud splicing method based on feature point matching, the preprocessing includes denoising, downsampling, and standardization.
[0012] Noise points are removed by a filtering algorithm, and when the distance is less than a preset threshold, the current is considered to be an outlier and is removed.
[0013] The point cloud is downsampled using a voxel grid method.
[0014] The differences in the coordinate system are eliminated through standardization processing.
[0015] As a preferred scheme of the power transmission line laser radar point cloud splicing method based on feature point matching, the training of the feature point extraction model includes training the feature point extraction model using a preset deep learning model, constructing a feature extraction network structure, and using labeled power transmission line point cloud data as training samples.
[0016] The loss function is optimized through supervised learning, the model is trained using a GPU, and the model is solidified as an extraction model after training.
[0017] As a preferred scheme of the power transmission line laser radar point cloud splicing method based on feature point matching, the extraction of feature points includes inputting the preprocessed three-dimensional point cloud data into the trained extraction model, outputting the feature score of each point and the local descriptor of each feature point, and selecting the points with the highest feature response as candidate feature points by setting a threshold.
[0018] As a preferred scheme of the power transmission line laser radar point cloud splicing method based on feature point matching, the feature point matching comprises: pairing the extracted candidate feature points, using a deep learning model to predict the similarity between all feature points, and considering that the current two feature points are allowed to be matched when the similarity is greater than a preset similarity threshold.
[0019] As a preferred scheme of the power transmission line laser radar point cloud splicing method based on feature point matching, the alignment and optimization comprises: after the feature point matching is completed, two or more point cloud segments are combined into a three-dimensional model through a splicing algorithm, and an iterative closest point algorithm is adopted to finely align and optimize the matched point cloud.
[0020] As a preferred scheme of the power transmission line laser radar point cloud splicing method based on feature point matching, the iterative closest point algorithm comprises:
[0021] calculating the closest point pair between the current point cloud and the target point cloud;
[0022] calculating the optimal rigid transformation, calculating the optimal rotation matrix and translation vector, and making the target function of the registration error between the point clouds reach a minimum value;
[0023] updating the position of the current point cloud according to the target function, judging whether the current point cloud error converges, and when the point cloud error reaches a set threshold, the convergence ends, otherwise the iteration continues.
[0024] Another object of the present application is to provide a power transmission line laser radar point cloud splicing system based on feature point matching.
[0025] To solve the above technical problems, the present application provides the following technical scheme: a power transmission line laser radar point cloud splicing system based on feature point matching, comprising: an original data acquisition module, a feature point extraction and training module, a data optimization module and a data splicing module.
[0026] The original data acquisition module acquires and pre-processes the original laser radar point cloud data.
[0027] The feature point extraction and training module trains a feature point extraction model using a deep learning model to automatically extract feature points from the point cloud data; and performs feature point matching based on the feature points extracted by the deep learning model.
[0028] The data optimization module aligns and optimizes the matched point cloud data using an optimization algorithm after the feature point matching is completed.
[0029] The data splicing module combines two or more point cloud segments into a three-dimensional model through a splicing algorithm.
[0030] The application provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the power transmission line laser radar point cloud splicing method based on feature point matching when executing the computer program.
[0031] The application provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the power transmission line laser radar point cloud splicing method based on feature point matching when executed by a processor.
[0032] The application has the following beneficial effects: The key point of the application is to automatically extract the best feature points in the point cloud data by using a deep learning model. This innovative method makes the feature point extraction no longer rely on manually set rules or traditional geometric features, but learns the intrinsic features of the point cloud through a deep neural network, and can automatically identify feature points with high recognition and stability in complex environments. Traditional point cloud matching methods usually match by calculating the distance or angle difference between feature points, but this method is prone to mismatch when dealing with a large amount of noisy data and complex environments. The application optimizes the matching process of feature points through deep learning, uses feature descriptors to measure the similarity between point clouds, avoids the limitations of traditional methods, and can adapt to point cloud data in different environments.
[0033] The application not only improves the accuracy of point cloud splicing, but also optimizes the computing efficiency. The combination of high-quality feature points extracted by the deep learning model and the matching algorithm can effectively process large-scale point cloud data, and the optimization algorithm used ensures the accuracy of the splicing result. In addition, the training and inference process of the deep learning model can be accelerated by GPU, thereby improving the speed of large-scale data processing.
[0034] The application can effectively splice the laser radar point cloud of the power transmission line in different environments, especially in complex power inspection environments, and the method has strong robustness. The deep learning model fully considers various disturbances and noise factors during training, thereby effectively adapting to various application scenarios. BRIEF DESCRIPTION OF DRAWINGS
[0035] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.
[0036] Figure 1A feature point matching-based power transmission line laser radar point cloud splicing method is provided for an embodiment of the present application.
[0037] Figure 2 A system scheme module diagram of a feature point matching-based power transmission line laser radar point cloud splicing system is provided for an embodiment of the present application. DETAILED DESCRIPTION
[0038] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.
[0039] Embodiment 1, with reference to Figure 1 For the first embodiment of the present application, the embodiment provides a feature point matching-based power transmission line laser radar point cloud splicing method, comprising:
[0040] The present application automatically extracts feature points with recognition from point cloud data by training a deep neural network, and matches and splices the point cloud based on these feature points. Compared with the traditional feature point matching method based on geometric rules, the method of the present application can adapt to point cloud data in complex environments, especially in the case of large data noise and large view angle change, and still maintains high matching accuracy and low computational complexity. The overall flowchart is as shown in Figure 1
[0041] S1, collect original laser radar point cloud data and perform preprocessing.
[0042] Preprocessing includes denoising, downsampling, and standardization to ensure data quality and improve the processing effect of subsequent steps. Point cloud data preprocessing is the basis for improving point cloud splicing accuracy. In an embodiment of the present application,
[0043] S101, the original point cloud data may contain inaccurate data points caused by noise of the laser radar sensor. These noise points are removed by a filtering algorithm. In the present application, a statistical outlier removal algorithm is used to improve the quality of the data.
[0044]
[0045] wherein p i =(x i ,y i ,z i ) and p j =(x j ,yj , z j ) are points in two point clouds. If the distance is less than a preset threshold, p j is considered an outlier and is removed.
[0046] S102, Point cloud data usually contains a large number of points, and direct processing will increase the computational burden. A voxel grid method is used to downsample the point cloud. The downsample formula is:
[0047]
[0048] where p new is the downsampled point, p i is each point in the original point cloud, and n is the number of points in the grid.
[0049] S103, by standardizing the process, the point cloud data is translated to the origin and the scale is unified to eliminate the difference of the coordinate system, and ensure that different point cloud segments can be unified into the same coordinate system.
[0050] In an optional embodiment, the preprocessing can be to directly train a convolutional neural network (CNN) model and input this CNN network into S202, input the local area of the point cloud, and output the denoised point cloud data, to construct a labeled point cloud data set, train the deep learning model by comparing the real point cloud and the noise point cloud, and apply the model to denoise the input point cloud data.
[0051] Random down-sampling method is used to randomly select a part of points from the original point cloud to reduce the computational complexity. The down-sampling ratio can be flexibly adjusted according to the application scenario (such as the detail requirement of the power line).
[0052] The point cloud data is translated to the origin and the scale is unified to eliminate the difference of the coordinate system.
[0053] In another optional embodiment, the preprocessing can also be to use a mean filter or a median filter to smooth the point cloud data and remove noise points, and calculate the mean or median in the neighborhood of each point to replace the value of the current point.
[0054] The grid down-sampling method is used to reduce the number of points by using a larger grid size, and the grid size is gradually adjusted to determine the optimal grid size through experiments.
[0055] The point cloud data is normalized in the same way to ensure the use of the same coordinate system.
[0056] S2, a preset deep learning model is used to train a feature point extraction model to automatically extract feature points from the point cloud data: in the present application, a convolutional neural network CNN is used for feature point matching, and in the embodiments of the present application,
[0057] S201, training a feature point extraction model using a preset deep learning model:
[0058] A feature extraction network structure based on KPConv is constructed;
[0059] A large amount of annotated power line point cloud data is used as training samples;
[0060] Through supervised learning, the loss function such as feature point positioning error or matching accuracy index is optimized;
[0061] GPU is used for model training to improve computing efficiency;
[0062] After the model training is completed, the extraction module is solidified for subsequent splicing process calling.
[0063] S202, input the preprocessed three-dimensional point cloud data into the trained extraction model, output the feature score of each point and the local descriptor of each feature point, and select the points with the highest feature response as candidate feature points by setting a threshold.
[0064] In the embodiment of the application, in the feature point extraction stage, the input is the preprocessed three-dimensional point cloud data, and the model output is the feature score of each point. A batch of points with the strongest feature response are selected as candidate feature points by setting a threshold.
[0065] At the same time, the model also outputs the local descriptor of each feature point, which is used for subsequent matching calculation. The extraction process does not require human intervention and is completely automatic, automatically extracting representative feature points.
[0066] The training data can come from point cloud data in different environments, covering different types of power line structures; the feature points f i are extracted by a convolutional neural network CNN:
[0067] f i =CNN(p i )
[0068] S203, extracting feature points with stability and recognition from point cloud data using the trained feature point extraction model;
[0069] Through pairing of the extracted candidate feature points, based on the feature descriptor extracted by the deep learning model, efficient and accurate feature point matching is performed.
[0070] The similarity between each feature point is predicted using a deep learning model, avoiding the simple matching strategy based on distance or angle in traditional methods, thereby enhancing the robustness of the algorithm. The feature point matching formula is as follows:
[0071]
[0072] where d i and d j are the feature descriptors of points p i and q j , and the dot product formula represents the similarity between two feature point descriptors.
[0073] When the similarity is greater than a preset similarity threshold, it is considered that the current two feature points are allowed to match.
[0074] In an optional embodiment, the feature point extraction and matching can use a method based on normal vector features, extract feature points by calculating the normal vector of each point, determine the feature points according to the degree of change of the normal vector, select the area with large change of the normal vector as the feature point, generate the descriptor of each feature point, for example, use the point cloud density or normal vector direction of the local area as the descriptor, match based on the field description of the feature points, and match by Euclidean distance. When the distance is less than a preset threshold, it is considered to be a successful match.
[0075] In another optional embodiment, the feature point extraction and matching can also be based on the traditional method SIFT (Scale Invariant Feature Transform) to extract feature points, identify the key points in the point cloud and calculate their feature descriptors, match the known template data with the extracted feature points, calculate the similarity between the template and the candidate feature points, compare the information of the template feature points through the descriptor, and select the most similar point for matching.
[0076] S3, after completing the feature point matching, aligning and optimizing the matched point cloud data by using an optimization algorithm.
[0077] S301, after completing the feature point matching, finely adjusting the matched point cloud by using an optimization algorithm (such as ICP algorithm, Bundle Adjustment, etc.), to ensure the accuracy and integrity of the point cloud data splicing. After the feature point matching is completed, two or more point cloud segments need to be combined into a complete three-dimensional model through a splicing algorithm. The core problem of point cloud splicing is how to handle the rotation and translation transformation between point clouds.
[0078] In an embodiment of the present application, the iterative closest point (ICP) algorithm is used to finely align and optimize the matched point cloud, to eliminate the errors caused by inaccurate feature point matching. The ICP optimization formula is:
[0079]
[0080] where p i and q i are points in the current point cloud and the target point cloud respectively, R and t are the rotation matrix and the translation vector, and N is the number of points.
[0081] The basic process of the ICP algorithm is to repeatedly perform the following steps:
[0082] Calculate the closest point pair between the current point cloud and the target point cloud. The calculation process is:
[0083] Let the source point cloud be P = {p1,p2,…,p N}, the target point cloud is Q = {q1,q2,…,q M}, then every source point p i The closest point to q j (i) is given by:
[0084]
[0085] Calculate the optimal rigid transformation (rotation and displacement). The calculation process is to find the optimal rotation matrix in order to minimize the registration error between the two point clouds. and translation vectors in, is a set of real numbers such that the following objective function reaches its minimum value:
[0086]
[0087] Apply the transformation, update the position of the current point cloud, and determine whether the current point cloud error has converged. When the point cloud error reaches the set threshold, the convergence is completed, otherwise continue to iterate.
[0088] The point cloud error is:
[0089]
[0090] Among them, p i and q i are the points from different point clouds in the matching point pair, and N is the number of matching point pairs. This error measures the accuracy of the stitching alignment.
[0091] In an optional embodiment, the point cloud data can be aligned and optimized by using a basic rigid transformation (such as the least squares method) to determine the rotation and translation parameters between the point clouds based on the matched feature points, and applying an iterative closest point (ICP) algorithm to further optimize the alignment results to reduce residual errors and achieve better registration effects.
[0092] In another optional embodiment, the point cloud data alignment and optimization can also reserve the noise removal and downsampling process in step S1. The statistical outlier removal algorithm is also used, but more emphasis is placed on merging information from different perspectives to reduce manual intervention, using the overlapping area of each perspective for rough matching, using the basic geometric method of feature point matching (such as RANSAC) to preliminarily register the local point cloud, using the iterative closest point (ICP) algorithm to fine-tune the rough matching result, gradually merging multiple point clouds, and ensuring the continuity and integrity of the final spliced model.
[0093] S4, merging two or more point cloud segments into a three-dimensional model by a splicing algorithm.
[0094] Finally, the complete point cloud data after splicing is output, generating a three-dimensional point cloud model of the power transmission line for subsequent application.
[0095] In order to verify the superiority of the method of the application, a comparative experiment is designed to compare the method of the application with two existing methods. Taking the point cloud splicing of a certain power transmission line as the scene, the experimental results are shown in Table 1:
[0096] Table 1 Comparison of various indexes of three methods
[0097]
[0098] The experimental results show that the method of the application is superior to the other two methods in terms of model response time, precision and operation and maintenance efficiency.
[0099] Embodiment 2 is a second embodiment of the application, which is different from the previous embodiment in that:
[0100] If the function is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the application or the part of the prior art that essentially contributes or the part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device) to execute all or part of the steps of the method described in the embodiments of the application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.
[0101] The logic and / or steps represented in flow diagrams or otherwise described herein, for example, can be considered as a sequence of executable instructions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For purposes of this specification, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber (optical), and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that can be later executed by a computer. In this context, a "computer-readable medium" can be any means that can store the program for use by or in connection with the instruction execution system, apparatus, or device.
[0102] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber (optical), and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that can be later executed by a computer.
[0103] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, various steps or methods can be implemented, for example, using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any of the following technologies, known in the art and combinations thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions upon an application of data signals, application-specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field-programmable gate arrays (FPGA), and so forth.
[0104] Embodiment 3, which is a third embodiment of the present application, provides a power transmission line laser radar point cloud splicing system based on feature point matching, comprising: an original data acquisition module, a feature point extraction and training module, a data optimization module, and a data splicing module;
[0105] The original data acquisition module acquires original laser radar point cloud data and performs preprocessing;
[0106] The feature point extraction and training module adopts a deep learning model to train a feature point extraction model to automatically extract feature points from point cloud data; and performs feature point matching based on the feature points extracted by the deep learning model;
[0107] The data optimization module aligns and optimizes the matched point cloud data by using an optimization algorithm after the feature point matching is completed.
[0108] The data splicing module combines two or more point cloud segments into a three-dimensional model by using a splicing algorithm.
[0109] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, and they should be covered in the scope of the claims of the present application.
Claims
1. A method for stitching power transmission line lidar point clouds based on feature point matching, characterized by: include, Collect raw lidar point cloud data and perform preprocessing; Use deep learning models to train feature point extraction models and automatically extract feature points from point cloud data; Perform feature point matching based on feature points extracted by deep learning model; After completing the feature point matching, the optimization algorithm is used to align and optimize the matched point cloud data; Merge two or more point cloud segments into a 3D model using a stitching algorithm.
2. The method for stitching power transmission line laser radar point clouds based on feature point matching according to claim 1, characterized in that: The preprocessing includes denoising, downsampling, and standardization; Noise points are removed through filtering algorithms. When the distance is less than a preset threshold, the point is considered an outlier and is removed. The point cloud is downsampled using a voxel grid method; The differences in coordinate systems are eliminated through standardization.
3. The method for stitching power transmission line laser radar point clouds based on feature point matching according to claim 2, characterized in that: The training of the feature point extraction model includes using a preset deep learning model to train the feature point extraction model, constructing a feature extraction network structure, and using the labeled transmission line point cloud data as a training sample; Through supervised learning, the loss function is optimized and the model is trained using GPU. After the model training is completed, it is solidified into an extraction model.
4. The method for stitching power transmission line laser radar point clouds based on feature point matching according to claim 3, characterized in that: The feature point extraction includes inputting the preprocessed three-dimensional point cloud data into a trained extraction model, outputting the feature score of each point and the local descriptor of each feature point, and screening out the point with the highest feature response as a candidate feature point by setting a threshold.
5. The method for stitching power line laser radar point clouds based on feature point matching according to claim 4, characterized in that: The feature point matching includes pairing the extracted candidate feature points and using a deep learning model to predict the similarity between all feature points. When the similarity is greater than a preset similarity threshold, it is considered that the current two feature points are allowed to match.
6. The method for stitching power transmission line laser radar point clouds based on feature point matching according to claim 5, characterized in that: The alignment and optimization include, after the feature point matching is completed, merging two or more point cloud segments into a three-dimensional model through a splicing algorithm, and using an iterative closest point algorithm to finely align and optimize the matched point clouds.
7. The method for stitching power transmission line laser radar point clouds based on feature point matching according to claim 6, characterized in that: The iterative closest point algorithm includes: Calculate the closest point pair between the current point cloud and the target point cloud; Calculate the optimal rigid transformation, calculate the optimal rotation matrix and translation vector, so that the objective function of the registration error between point clouds reaches the minimum value; The position of the current point cloud is updated according to the objective function to determine whether the current point cloud error has converged. When the point cloud error reaches the set threshold, the convergence ends, otherwise the iteration continues.
8. A power transmission line laser radar point cloud stitching system based on feature point matching, applying a power transmission line laser radar point cloud stitching method based on feature point matching according to any one of claims 1 to 7, characterized in that: include: Raw data acquisition module, feature point extraction and training module, data optimization module and data splicing module; The raw data acquisition module collects raw lidar point cloud data and performs preprocessing; The feature point extraction and training module uses a deep learning model to train a feature point extraction model to automatically extract feature points from point cloud data; Perform feature point matching based on feature points extracted by deep learning model; The data optimization module aligns and optimizes the matched point cloud data using an optimization algorithm after completing feature point matching; The data stitching module combines two or more point cloud segments into a three-dimensional model through a stitching algorithm.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the processor implements the steps of a power transmission line lidar point cloud splicing method based on feature point matching according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a power transmission line lidar point cloud splicing method based on feature point matching according to any one of claims 1 to 7 are implemented.