AI vision-based overhead transmission line spatial distance measurement method

Abstract

The invention discloses an overhead transmission line spatial distance measurement method based on AI vision, and relates to the technical field of overhead transmission line measurement. According to the method, a visible light image and auxiliary data of a power transmission line are acquired through an unmanned aerial vehicle, intelligent material identification and edge segmentation are realized by adopting an improved YOLOv5 algorithm based on a constructed typical material database, three-dimensional reconstruction of a target area is completed in combination with a multi-baseline NCC matching method, and finally accurate measurement of a key spatial distance is realized based on a multi-coordinate system conversion model. According to the method, the problems of low precision, poor efficiency and high safety risk of a traditional measurement method are solved, the measurement error is controlled at the decimeter level, the method is suitable for overhead transmission line measurement under different voltage levels, complex terrains and weather conditions, and efficient and accurate technical support is provided for digital acceptance of transmission lines.

Description

Technical Field

[0001] This invention relates to the field of overhead transmission line measurement technology, specifically to a method for measuring the spatial distance of overhead transmission lines based on AI vision. Background Technology

[0002] With the acceleration of the global energy transition and the construction of new power systems, the large-scale integration of renewable energy and the increasing complexity of distributed energy systems have placed higher demands on the safe and stable operation of the power grid. As the core backbone of the power system, transmission lines undertake the critical task of power transmission, and their operating status directly affects the steady-state operation of regional power grids and even the entire power system.

[0003] We must accelerate the integration of information technology and the energy industry, promote the digital upgrade of the energy industry, and vigorously promote the application of new technologies such as artificial intelligence, cloud computing, the Internet of Things, and big data in the energy field. Overhead transmission lines, as a key carrier of energy transmission, are growing in length at a rate of tens of thousands of kilometers per year. Traditional manual acceptance and measurement methods are no longer sufficient to meet the needs of equipment operation and maintenance model transformation and upgrading, and the construction of new power systems.

[0004] The main problems with current overhead transmission line spatial distance measurement are as follows:

[0005] Low measurement accuracy. Traditional measurement methods mainly rely on handheld rangefinders for pile location re-survey and acceptance. Affected by factors such as the accuracy of the measuring equipment and weather conditions, the ranging error can reach the meter level in harsh environments such as strong sunlight. Moreover, it cannot fully diagnose the construction quality of key parts such as the top of the tower. Although lidar measurement can achieve centimeter-level accuracy, the hardware cost is high, requiring a dedicated high-performance server to process massive point cloud data. Report generation is slow and cannot meet the needs of real-time measurement.

[0006] Measurement is inefficient and carries high safety risks. Manual tower climbing is a traditional method of measurement, with acceptance measurements for a single tower typically taking up to 30 minutes. This is not only inefficient but also poses safety risks such as falls from heights. Furthermore, the power industry generally suffers from high staff shortages, with some units experiencing shortages approaching or exceeding 50%, leading to frequent instances of "acceptance without inspection" and a severe lack of manpower. In addition, post-processing of acceptance measurement data is difficult, and omissions in the statistical analysis of tower, equipment, and other material information further reduce measurement efficiency.

[0007] Existing technologies have limitations. With the development of drone technology, some power companies have begun to explore the application of drones in transmission line measurement, but existing technologies still have significant shortcomings: the drone acceptance used by State Grid Jiangsu and Anhui Electric Power companies is limited to photo archiving, defect identification, and tower coordinate recording, and does not carry out key spatial distance measurements such as tower height and crossarm spacing; although State Grid Sichuan Electric Power can roughly calculate tower height based on monocular visible light vision measurement technology, the measurement accuracy is insufficient due to differences in tower diameter, and it cannot measure key parameters such as crossarm spacing and transformer-to-tower spacing, and its application scope is limited to distribution towers; although State Grid Shanxi Company's drone AI autonomous acceptance can complete refined defect detection, it does not carry out three-dimensional spatial reconstruction and does not have the functions of accurate measurement of various spatial distances and material statistical analysis.

[0008] Compared to distribution lines, transmission lines have tower heights ranging from tens to hundreds of meters, insulator string lengths reaching tens of meters (ultra-high voltage lines), and conductor and ground wire heights of 20-50 meters. Spatial distance measurements are significantly more challenging in areas such as critical distance measurement, phase-to-phase distance measurement, cross-span distance measurement, and construction material statistics. Currently, in-depth research has not been conducted on precise spatial distance measurement technology for overhead transmission lines using drones based on visible light image AI recognition. There is an urgent need to develop an efficient, accurate, and safe measurement method to address the shortcomings of traditional methods and provide technical support for the digital acceptance of overhead transmission lines. Summary of the Invention

[0009] The purpose of this invention is to provide an AI vision-based method for measuring the spatial distance of overhead transmission lines, in order to solve the problems of low accuracy, poor efficiency, high safety risks and limitations of existing technologies in traditional measurement methods, and to achieve efficient and accurate measurement of key spatial distances of overhead transmission lines.

[0010] To achieve the above objectives, the present invention adopts the following technical solution:

[0011] The AI ​​vision-based method for measuring spatial distance of overhead transmission lines includes six steps: data acquisition, construction of a typical material database, intelligent material identification and edge segmentation, 3D reconstruction of the target area, spatial distance calculation, and result verification and output. The specific steps are as follows:

[0012] Step S1: Data Acquisition

[0013] This step involves using a drone equipped with a visible light camera, laser rangefinder, and POS positioning device to collect comprehensive data on overhead power transmission lines. The drone utilizes a multi-rotor flight platform with autonomous navigation, automatic target tracking and locking, and adaptive zoom capabilities, enabling multi-angle, close-range, and blind-spot-free photography to ensure the integrity and clarity of the collected data.

[0014] The visible light camera equipment uses high-definition cameras with a resolution of 4K or higher and a frame rate of 30fps or higher, which can clearly capture the detailed features of components such as towers, insulator strings, crossarms, and conductors; the laser rangefinder has a measurement range of 0.5-100m and a measurement accuracy of no less than ±2cm, and is used to obtain the imaging distance of the object from the lens plane; the POS positioning device integrates GNSS and IMU modules, with a positioning accuracy of ±1cm for plane and ±2cm for elevation, and a sampling frequency of 10Hz, which can acquire the position and attitude data of the UAV in real time, providing basic data support for subsequent 3D reconstruction.

[0015] During the data acquisition process, the UAV's flight altitude was controlled at 5-10 meters above the top of the tower, its flight speed was set to 5 m / s, and the overlap rate of adjacent images was set to 70% to ensure the continuity and coverage integrity of the image data. Simultaneously, the laser rangefinder measured the distance between the object and the lens, and the POS positioning device recorded the UAV's position and attitude data in real time. These three types of data were transmitted in real time to the ground control terminal for storage via a high-speed data interface.

[0016] Step S2: Construction of Typical Material Database

[0017] The typical material database is the foundation for achieving intelligent material identification and accurate measurement. Its construction process strictly follows the design specifications for transmission lines and DL / T741 "Operation Regulations for Overhead Transmission Lines", and specifically includes the following three stages:

[0018] Material Characteristics Review: This review summarizes the typical material list and component characteristics involved in the acceptance measurement of overhead transmission lines, covering 32 typical materials in 5 major categories: poles and towers (straight-line towers, tension towers, etc.), insulators (suspension insulators, post insulators, etc.), conductors and ground wires (steel-cored aluminum stranded wires, ground wires, etc.), hardware (line clamps, spacers, vibration dampers, etc.), and auxiliary facilities (pole number plates, warning signs, etc.). The review clarifies the structural characteristics, size range, and appearance attributes of each material.

[0019] Data Acquisition and Augmentation: Using the UAV data acquisition system from step S1, material image data of transmission lines with different voltage levels (110kV, 220kV, 500kV) were collected under various weather conditions (sunny, cloudy, partly cloudy) and terrain conditions (plains, mountains, hills), totaling over 15,000 images to ensure the database covers different application scenarios. Simultaneously, data augmentation techniques (including rotation, scaling, flipping, brightness adjustment, and noise addition) were used to augment the collected image data, improving the model's generalization ability.

[0020] Data annotation and database construction: Unified data compilation and annotation rules were established. The LabelImg annotation tool was used to standardize the annotation of collected image data, including material type, location coordinates, and defect type. For construction process defect samples, the location and extent of 12 typical defects, such as insulator damage, broken conductor strands, missing clamps, and blind spots, were highlighted. After annotation, the data underwent quality verification, and blurry or incorrectly labeled images were removed. Finally, a visible light image database containing typical material and construction process defect samples was constructed, providing high-quality data support for subsequent algorithm model training.

[0021] Step S3: Intelligent Material Recognition and Edge Segmentation

[0022] This step uses an improved YOLOv5 algorithm model to achieve intelligent identification of typical materials, and combines reinforcement learning and SAM model to complete the edge segmentation of target materials, providing accurate target area positioning for subsequent 3D reconstruction and distance measurement.

[0023] Improved YOLOv5 Algorithm Model Construction: To address the issues of weak geometric deformation adaptability and poor background interference resistance in the existing YOLOv5 algorithm for power transmission line material identification, the following three improvements are made:

[0024] Module Replacement: The traditional convolution in the C3 module of the YOLOv5n backbone is replaced with Variable Convolution v2 (DCNv2) to construct the C3-DCNv2 module. Variable Convolution v2 expands the convolution kernel by introducing an offset variable, allowing the kernel to flexibly sample near the current position. It also adds weights for each sampling point, reducing the influence of irrelevant factors on feature extraction and enhancing the model's adaptability to deformable targets.

[0025] Feature fusion optimization: A context enhancement module (CAM) is integrated between the backbone network and the neck network (PANet). This module extracts channel features through global average pooling and global max pooling, adaptively adjusts the feature map weights, enhances the feature expression of the target area, reduces background interference, and effectively solves the problem of low target-background distinction in complex environments of transmission lines.

[0026] Loss function improvement: The original GIoU loss function is replaced by the CIoU loss function. The CIoU loss function considers the overlap rate of the target bounding box, the distance between the center points and the aspect ratio at the same time, which can more accurately measure the bounding box regression error and improve the target bounding box regression accuracy.

[0027] Model Training and Inference: The constructed typical material database was divided into training, validation, and test sets in a 7:2:1 ratio to train the improved YOLOv5 algorithm model. During training, an adaptive learning rate adjustment strategy (CosineAnnealingLR) was adopted, with an initial learning rate of 0.01, a batch size of 16, and 100 training epochs. Model performance was monitored in real time using the validation set. Training was stopped and the optimal model parameters were saved when the model's mean accuracy (mAP) on the validation set showed no improvement for 10 consecutive epochs. The visible light images acquired in step S1 were input into the trained model for inference, outputting identification results such as material category, location coordinates, and defect type. After identification, a multi-image identification and deduplication algorithm was used to eliminate duplicate identification results and improve identification accuracy. Testing showed that the improved YOLOv5 algorithm model achieved a mean accuracy (mAP) of 92.3% in typical material identification tasks, a 5.7% improvement over the original YOLOv5n algorithm, with an inference speed of 35fps, meeting real-time identification requirements.

[0028] Edge segmentation of target materials: A reinforcement learning algorithm is used to optimize the segmentation strategy of the SegmentAnythingModel (SAM) model, performing edge detection and segmentation on key measuring equipment such as tower tops, crossarms, insulators, and conductors. The reinforcement learning algorithm, through a reward mechanism, guides the SAM model to focus on the critical edge regions of the target materials, improving segmentation accuracy. The SAM model, in turn, utilizes its powerful general segmentation capabilities to accurately segment the pixel range of the target materials, eliminating interference from irrelevant factors such as conductors and the background environment. The segmentation results are marked on the original image with white outlines, clearly defining the boundary range of the measured materials and providing precise target region localization for subsequent 3D reconstruction and distance measurement, achieving pixel-level segmentation accuracy.

[0029] Step S4: 3D Reconstruction of the Target Area

[0030] This step employs a multi-baseline NCC matching method to extract image depth maps. Combined with laser ranging data and POS positioning data, it performs 3D reconstruction of the segmented target region, generating a color point cloud map to provide 3D data support for spatial distance calculations. The 3D reconstruction technical route is as follows: Figure 1 As shown, the specific process is as follows:

[0031] Keyframe extraction and pose estimation: The image sequence acquired in step S1 is processed using the ORB-SLAM system to extract keyframe images and estimate their pose parameters (including position coordinates and attitude angles). Through algorithms such as feature point extraction, matching, and nonlinear optimization, the ORB-SLAM system can achieve stable camera localization and map construction in complex environments, providing accurate pose data for 3D reconstruction.

[0032] Depth Map Extraction: To address the issues of low stereo matching accuracy and the impact of illumination variations on block matching in multi-view wide-field images of transmission lines, a multi-baseline NCC (Normalized Cross-Correlation) matching method is employed to extract image depth maps. The specific process is as follows: Five preceding frames of the current keyframe are selected as observable frames to construct a multi-baseline matching window; the NCC matching cost between corresponding pixels in different observable frames and the current frame is calculated. The NCC matching cost effectively measures the correlation between pixels and resists the influence of illumination variations; using the WTA (Winner-Takes-All) algorithm, the depth with the lowest matching cost among all sampled depth values ​​for each pixel is selected as the optimal depth value to generate the initial depth map; a median filtering algorithm (with a filtering window size of 3×3) is used to smooth and correct the initial depth map, removing abnormal depth values ​​to obtain a dense and accurate depth map.

[0033] 3D Point Cloud Reconstruction of the Target Area: Considering the limited computing power of actual measurement platforms (drone-borne or handheld remote control edge devices), reconstructing the entire tower would result in an enormous amount of data processing and slow processing speed, failing to meet real-time measurement requirements. Therefore, this invention adopts a "local reconstruction of the target area" strategy, performing 3D reconstruction only on the target area (such as crossarms, insulator strings, conductors, etc.) segmented in step S3, without needing to reconstruct the entire tower. The specific process is as follows: Based on the depth map and camera intrinsic parameters, 2D image pixels are converted into 3D coordinate points; combined with the RGB color information of the visible light image acquired in step S1, the 3D coordinate points are colored to generate a color point cloud map; the number of point cloud data points for a single target area is controlled to 500-2000, significantly reducing the amount of data processing and meeting the real-time requirements of mobile devices. The color point cloud map can clearly restore the spatial structure of the target area, the relative positional relationships of each component, and its appearance features, providing accurate 3D data support for spatial distance calculation.

[0034] Step S5: Spatial distance calculation

[0035] This step, based on a multi-coordinate system transformation model and combined with the color point cloud data generated by 3D reconstruction, calculates the key spatial distance parameters of overhead transmission lines. The specific process is as follows:

[0036] Multi-coordinate system transformation model construction: Based on the principles of optical imaging, a transformation model is constructed between the real-world coordinate system, camera coordinate system, image coordinate system, and pixel coordinate system.

[0037] Key spatial distance calculation: Combining multi-coordinate system transformation models and 3D point cloud data, key spatial distance parameters of overhead transmission lines are calculated, including total tower height, nominal height, crossarm spacing, conductor spacing, crossarm-to-ground distance, and crossing distance. The specific calculation method is as follows:

[0038] Total tower height: Select the three-dimensional coordinates of the bottom and top of the tower in the three-dimensional point cloud map, calculate the straight-line distance between the two points, which is the total tower height.

[0039] Tower height: Select the 3D coordinates of the lower edge of the crossarm and the bottom of the tower in the 3D point cloud map, and calculate the vertical distance between the two points, which is the tower height.

[0040] Crossarm spacing: Select the three-dimensional coordinates of the corresponding points of two adjacent crossarms in the three-dimensional point cloud map, calculate the vertical distance between the two points, which is the crossarm spacing.

[0041] Conductor spacing: Select the three-dimensional coordinates of the corresponding points of the three-phase conductors in the three-dimensional point cloud map, and calculate the straight-line distance between each conductor, which is the conductor spacing.

[0042] Crossarm to ground distance: Select the three-dimensional coordinates of the lower edge of the crossarm and the ground in the three-dimensional point cloud map, and calculate the vertical distance between the two points, which is the crossarm to ground distance.

[0043] Crossing distance: Select the three-dimensional coordinates of the corresponding points of the guide line and the crossing object (such as highway, railway, river, other line, etc.) in the three-dimensional point cloud map, and calculate the minimum vertical distance between the two points, which is the crossing distance.

[0044] Tests have shown that the spatial distance measurement error of this invention is controlled within ±0.3m, which meets the accuracy requirements for acceptance measurement of overhead transmission lines.

[0045] Step S6: Result Verification and Output

[0046] This step compares the measurement results from step S5 with the requirements of the transmission line design standards and DL / T741 "Operation Regulations for Overhead Transmission Lines" to verify the conformity of the measurement results; generates a digital measurement report, establishes a defect database, and provides data support for the acceptance and subsequent operation and maintenance of transmission lines.

[0047] Measurement result comparison: The measurement results of key parameters such as total tower height, nominal height, crossarm spacing, and conductor spacing are compared with the design standards, and items exceeding the tolerance are marked. If the measurement results are within the allowable error range of the design standards, they are judged to meet the requirements; if they exceed the allowable error range, they are judged to not meet the requirements, and the reasons for the deviation are analyzed.

[0048] Digital Report Generation: A report generation module developed using Python integrates measurement data, identification results, and comparison results to generate a digital measurement report in HTML format. The report includes basic project information, material ledger statistics tables, defect distribution diagrams, key parameter measurement result comparison tables, and 3D point cloud reconstruction renderings, presenting the measurement results intuitively.

[0049] Defect Database Establishment: A MySQL database is used to store defect data. Database fields include defect ID, defect type, defect location coordinates, defect image, measurement results, out-of-tolerance status, and discovery time. The defect database supports searching and querying by defect type, voltage level, discovery time, and other criteria, providing accurate defect location and handling information for subsequent maintenance personnel.

[0050] The spatial distance measurement method for overhead transmission lines based on AI vision provided by this invention integrates multiple advanced technologies such as AI vision, laser ranging, BeiDou high-precision positioning, and 3D reconstruction, and has the following beneficial effects:

[0051] High measurement accuracy: The improved YOLOv5 algorithm is used to achieve accurate material identification. Combined with the multi-baseline NCC matching method, three-dimensional reconstruction is performed to build a multi-coordinate system transformation model. The measurement error is controlled within ±0.3m. Compared with the meter-level error of traditional handheld rangefinders and the coarse calculation of existing visual measurement technology, the measurement accuracy is greatly improved, meeting the accuracy requirements of acceptance measurement of overhead transmission lines.

[0052] High measurement efficiency: Through drone autonomous navigation, automatic shooting and front-end intelligent recognition technology, the measurement time of a single tower is greatly shortened, avoiding the tedious process of manual tower climbing; the "target area local reconstruction" strategy is adopted to reduce the amount of data processing and realize real-time measurement and analysis; digital reports are automatically generated without manual processing, solving the problems of low efficiency and "acceptance without verification" in traditional measurement.

[0053] High safety performance: No manual tower climbing is required, completely eliminating the safety risk of falling from heights. It is especially suitable for power transmission line measurement work in complex terrain conditions such as mountainous and hilly areas, ensuring the personal safety of operators.

[0054] High adaptability: The improved YOLOv5 algorithm enhances the adaptability to geometrically deformed targets and complex backgrounds. The multi-baseline NCC matching method effectively solves problems such as illumination changes and weak textures. It is suitable for overhead transmission line measurements under different voltage levels such as 110kV, 220kV, and 500kV, different terrains such as plains, mountains, and hills, and different weather conditions such as sunny, cloudy, and overcast days.

[0055] Comprehensive Functions: It can realize functions such as typical material identification, construction process defect detection, key spatial distance measurement, digital report generation and defect database establishment. Compared with the single function of existing UAV measurement technology, it meets the all-round needs of digital acceptance of overhead transmission lines and provides complete data support for transmission line acceptance and subsequent operation and maintenance. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a flowchart illustrating the steps of the AI ​​vision-based spatial distance measurement method for overhead transmission lines according to the present invention. Detailed Implementation

[0058] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0059] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0060] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0061] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0062] To make the technical problem to be solved, the technical solution and advantages of the present invention clearer, the following will be described in conjunction with the appendix. Figure 1 The specific embodiments are described in detail below.

[0063] Example: Spatial distance measurement of a 220kV overhead transmission line

[0064] A 220kV overhead transmission line project was selected as the measurement object. This project includes 86 towers, a line length of 32 kilometers, and terrain encompassing plains and low mountainous areas. The spatial distance measurement method of this invention was used for the measurement, and the specific implementation process is as follows:

[0065] Step S1: Data Acquisition. A multi-rotor UAV equipped with a 4K high-definition visible light camera, a laser rangefinder, and an RTK-POS positioning device was used for data acquisition. The UAV's flight altitude was set at 8m above the top of the tower, the flight speed at 5m / s, and the overlap rate of adjacent images was 70%. A total of over 8,600 visible light images, over 8,600 sets of laser rangefinder data, and over 17,200 sets of POS positioning data were acquired. All data was transmitted to the ground control terminal for storage in real time.

[0066] Step S2: Construction of a Typical Material Database. Based on DL / T741 "Operating Procedures for Overhead Transmission Lines", a list and characteristics of 32 typical materials in 5 categories were compiled; more than 15,000 material images were collected under different voltage levels, weather and terrain conditions, and data augmentation techniques were used to expand the data; the LabelImg tool was used to standardize the annotation of the images, and a typical material database was built, including 12 typical defect samples such as insulator damage and conductor strand breakage.

[0067] Step S3: Intelligent Material Identification and Edge Segmentation. Based on the improved YOLOv5 algorithm model, the acquired visible light images were processed, identifying 86 towers, 344 suspension insulator strings, 516 crossarms, 1376 wire clamps, 430 spacers, and 86 pole number plates. Three insulator damage defects, two conductor strand breakage defects, one missing wire clamp defect, and four blind spot defects were detected. Reinforcement learning combined with the SAM model was used to perform edge segmentation on the identified target materials such as crossarms, insulator strings, and conductors, with segmentation accuracy reaching the pixel level.

[0068] Step S4: 3D Reconstruction of the Target Area. Keyframe images and pose parameters are extracted using the ORB-SLAM system; depth maps are extracted using the multi-baseline NCC matching method, and after median filtering and smoothing correction, dense depth maps are generated; 3D reconstruction is performed on the segmented target areas such as crossarms and insulator strings to generate color point cloud maps. The average number of point cloud data points per target area is 800, and the average data processing time is 15 seconds.

[0069] Step S5: Spatial Distance Calculation. Construct a multi-coordinate system transformation model and calculate key spatial distance parameters based on 3D point cloud data. For example, the measurement results of a certain straight-line tower are: total height 32.4m, nominal height 24.1m, distance between upper and middle crossarms 5.2m, distance between middle and lower crossarms 4.8m, three-phase conductor spacing 4.5m, minimum conductor-to-ground distance 12.3m, and distance across a highway 10.8m.

[0070] Step S6: Result Verification and Output. The measurement results are compared with the design standards. The design standards for this straight-line tower require: total height 32.0±0.5m, nominal height 24.0±0.3m, crossarm spacing 5.0±0.2m, conductor spacing ≥4.0m, minimum conductor-to-ground distance ≥12.0m, and distance across highways ≥10.0m. The comparison results show that all measurement parameters meet the design standards. A digital measurement report is generated, including basic project information, material ledger, defect distribution, parameter comparison, etc. The defect data is stored in a database, supporting retrieval and querying by defect type.

[0071] The measurement work in this embodiment took a total of 2 days, which is significantly more efficient than the 15 days of traditional manual measurement. The measurement error was controlled within ±0.3m, meeting the acceptance accuracy requirements. No manual tower climbing was required throughout the process, ensuring the personal safety of the operators. The generated digital measurement report and defect database provided important data support for the subsequent operation and maintenance of the line.

[0072] In summary, this invention solves the problems of low accuracy, poor efficiency, high safety risks, and limitations of existing technologies in traditional methods for measuring spatial distance of overhead transmission lines. It provides an efficient, accurate, and safe technical solution for the digital acceptance of overhead transmission lines, and has broad engineering application value and prospects for promotion.

[0073] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for measuring spatial distance of overhead transmission lines based on AI vision, characterized in that, Includes the following steps: S1: Data acquisition, using a drone equipped with a visible light camera, laser rangefinder and POS positioning device to collect visible light images, laser rangefinder data and position and attitude data of overhead power transmission lines; S2: Typical material database construction. Based on the design specifications and industry standards of transmission lines, a list of typical materials and their characteristics are sorted out. The collected visible light images are standardized and labeled to build a typical material database containing samples of construction process defects. S3: Intelligent material identification and edge segmentation. Based on the improved YOLOv5 algorithm model, the visible light image collected in step S1 is processed to identify typical materials and construction process defects. Reinforcement learning combined with the SAM model is used to achieve edge segmentation of the target material. S4: 3D reconstruction of the target area. The multi-baseline NCC matching method is used to extract the image depth map. Combined with laser ranging data and POS positioning data, the target area segmented in step S3 is reconstructed in three dimensions to generate a color point cloud map. S5: Spatial distance calculation, establish a multi-coordinate system transformation model, and calculate key spatial parameters such as total tower height, nominal height, crossarm spacing, conductor spacing and crossover distance based on the three-dimensional reconstruction results of step S4; S6: Result verification and output. Compare the measurement results from step S5 with the design standards, generate a digital measurement report, establish a defect database, and support defect classification and retrieval.

2. The method according to claim 1, characterized in that, In step S1, the visible light camera device supports 4K and above resolution and 30fps and above frame rate shooting, the laser rangefinder has a measurement accuracy of not less than ±2cm, and the POS positioning device has a positioning accuracy of ±1cm in plane and ±2cm in elevation.

3. The method according to claim 1, characterized in that, In step S2, typical materials include poles and towers, insulators, conductors and ground wires, hardware and ancillary facilities, and construction defects include broken insulators, broken conductor strands, missing wire clamps and blind spots.

4. The method according to claim 1, characterized in that, The improved YOLOv5 algorithm model in step S3 includes: replacing the traditional convolution in the C3 module of the YOLOv5n backbone with variable convolution v2 to construct the C3-DCNv2 module; integrating a context enhancement module between the backbone and the neck network to optimize the feature fusion structure; and replacing the GIoU loss function with the CIoU loss function to improve the accuracy of target bounding box regression.

5. The method according to claim 1, characterized in that, In step S3, the specific process of edge segmentation is as follows: based on the reinforcement learning algorithm, the segmentation strategy of the SAM model is optimized, and edge detection is performed on the key measuring equipment of the tower top, crossarm, insulator, and conductor to clarify the pixel range of the material to be measured.

6. The method according to claim 1, characterized in that, In step S4, the specific process of 3D reconstruction includes: S41: acquiring the keyframe images and their pose parameters output by the ORB-SLAM system; S42: selecting forward multi-baseline observable frames and calculating the NCC matching cost of corresponding pixels between different observation frames and the frame to be measured; S43: determining the optimal depth value through the WTA algorithm, generating a dense depth map and performing smoothing correction; S44: converting the two-dimensional image pixels into three-dimensional coordinate points, performing RGB coloring processing, and generating a color point cloud map of the target area.

7. The method according to claim 1, characterized in that, In step S4, only the target area to be detected is reconstructed in three dimensions. There is no need to reconstruct the entire tower. The number of point cloud data points for a single target area is controlled between 500 and 2000.

8. The method according to claim 1, characterized in that, The multi-coordinate system transformation model in step S5 includes the real-world coordinate system, camera coordinate system, image coordinate system, and pixel coordinate system. The transformation between these coordinate systems is achieved through rotation matrices, translation vectors, and perspective projection matrices.

9. The method according to claim 8, characterized in that, The specific relationships of multi-coordinate system transformation are as follows: rigid body transformation is achieved from the real-world coordinate system to the camera coordinate system through rotation matrix R and translation vector T; projection transformation from 3D to 2D is achieved from the camera coordinate system to the image coordinate system through perspective projection matrix; and transformation from the image coordinate system to the pixel coordinate system is achieved through scaling and translation.

10. The method according to claim 1, characterized in that, In step S3, after material identification, a multi-image identification and deduplication algorithm is used to eliminate duplicate identification results and improve identification accuracy; in step S4, the depth map smoothing correction adopts a median filtering algorithm, and the filtering window size is set to 3×3; in step S5, the measurement error of key spatial parameters is controlled within ±0.3m.

Images