Power line real-time recognition and spatial positioning method based on depth vision assistance

Abstract

The application relates to a power line real-time recognition and spatial positioning method based on depth vision assistance and relates to a power line real-time recognition and spatial positioning method.The application aims to solve the problems of poor real-time performance, low accuracy and poor adaptability to dynamic environments in the recognition and three-dimensional position estimation of power lines by using existing image processing technologies.The process is as follows: 1. power line image acquisition; 2. pre-processing of the collected power line image; 3. identifying the power line contour in the pre-processed RGB image based on a pre-trained YOLOv8 model; 4. obtaining a segmented depth map, wherein each segmented depth map only contains a power line contour; 5. obtaining a comprehensive depth map of all power line depth information; 6. user interaction and power line selection; 7. three-dimensional position estimation of the power line contour; and 8. output of the three-dimensional position of the power line contour.The application is used in the field of power line real-time recognition and spatial positioning.

Description

Technical Field

[0001] This invention relates to a method for real-time identification and spatial positioning of power lines. Background Technology

[0002] As a critical component of the power system, the regular inspection of power lines is essential for ensuring the stable operation of the power system. Currently, power line inspection methods mainly include traditional manual inspection, helicopter inspection, robotic inspection, satellite inspection, and drone inspection. With the rapid development of drone technology and high-resolution airborne cameras, drone-based power line inspection has gradually become a research hotspot. Drones can not only perform routine inspections but also execute more complex tasks such as foreign object removal, connection work, and pulley loading and unloading, which increases the necessity of obtaining accurate, real-time three-dimensional spatial information of power transmission lines.

[0003] Currently, LiDAR-based methods for extracting power lines have been extensively studied. In narrow power corridors, ground points are removed using filtering methods, retaining non-ground point cloud data. The spatial geometry of the non-ground point cloud can be utilized to distinguish power line points from vegetation points and tower points. However, this method has limitations in handling complex environments and achieving real-time processing.

[0004] In contrast, research on image-based electric field line extraction is relatively limited. While image processing techniques have made significant progress in object recognition, major challenges remain in accurately identifying and estimating the three-dimensional location of electric field lines, especially in dynamic and complex environments. Existing techniques need improvement in terms of real-time performance, accuracy, and adaptability to dynamic environments. Summary of the Invention

[0005] The purpose of this invention is to address the problems of poor real-time performance, low accuracy, and poor adaptability to dynamic environments in existing image processing technologies for electric line recognition and three-dimensional position estimation. Therefore, this invention proposes a real-time electric line recognition and spatial positioning method based on depth vision assistance.

[0006] The specific process of the method for real-time power line identification and spatial localization based on depth vision assistance is as follows:

[0007] Step 1: Power line image acquisition;

[0008] Step 2: Preprocess the RGB images in the acquired power line images;

[0009] Step 3: Identify the electric field line contours in the pre-trained YOLOv8 model;

[0010] Step 4: Based on the electric field line contours in the RGB image identified by the YOLOv8 model in Step 3, segment each electric field line contour in the simultaneously acquired depth map, remove the area outside the electric field line contour, and obtain the segmented depth map. Each segmented depth map contains only one electric field line contour.

[0011] Step 5: Perform histogram analysis to denoise each segmented depth map to obtain a comprehensive depth map containing all power line depth information;

[0012] Step 6: Based on the electric field line contours in the RGB image identified by the YOLOv8 model in Step 3, perform user interaction and electric field line selection;

[0013] Step 7: Based on Steps 3, 5, and 6, perform three-dimensional position estimation of the power line profile:

[0014] Step 8: Output the three-dimensional position of the power line profile.

[0015] The beneficial effects of this invention are as follows:

[0016] This invention relates to the application of machine vision and deep learning in the field of power system monitoring. Specifically, it belongs to the technical field of object recognition and spatial positioning using computer vision technology, particularly utilizing depth cameras and deep learning algorithms for real-time identification and 3D position estimation of power cables. This technology is suitable for automated monitoring, smart grid management, transmission line maintenance, and other applications requiring high-precision and real-time power line tracking and analysis.

[0017] To address the aforementioned challenges, this invention proposes a real-time power line identification and spatial positioning system that combines deep learning and depth camera technology. This system aims to overcome the limitations of existing technologies through innovative image processing and depth estimation methods, achieving efficient, accurate, and real-time identification and 3D position estimation of power lines.

[0018] This invention introduces YOLOv8 for power line image segmentation: compared with existing technologies, YOLOv8 has higher accuracy and robustness in handling complex backgrounds and changing lighting conditions.

[0019] Histogram Analysis Denoising Technology: The histogram denoising method of this invention is optimized for the characteristics of power line depth data, effectively improving the accuracy of 3D positioning, which is difficult to achieve in traditional methods. Histogram analysis further purifies the depth data, identifying and discarding outlier depth values, improving the accuracy of power line depth information. This process greatly reduces false data and outliers, producing clearer depth maps. Each processed depth map is combined into a composite depth map for easy subsequent use.

[0020] 3D Power Line Fitting Optimization: The proposed power line fitting algorithm simplifies the computation process, ensuring processing speed and making it suitable for real-time processing on mobile platforms, greatly enhancing the application potential of UAVs in practical operations. This operation simplifies the 3D fitting of large-scale point clouds to a 1D fitting of a small number of sampled points. Experimental data shows that our method is 90% faster than performing principal component analysis (PCA) on the entire point cloud.

[0021] This invention improves recognition accuracy and efficiency: By combining the YOLO v8 image segmentation model based on deep learning with traditional image processing techniques, this system can quickly and accurately identify power lines in complex environments. The TensorRT-optimized model accelerates the processing flow, improving the overall efficiency of the system.

[0022] Enhancing the accuracy of 3D position estimation: By utilizing a depth camera and refined image processing methods, the system can accurately calculate the 3D coordinates of the transmission line, improving the accuracy of position estimation. Through histogram analysis and comprehensive processing of depth data, errors are effectively reduced, and the accuracy of depth estimation is improved.

[0023] Real-time performance optimization: Through efficient algorithm design and optimized ROS nodes, the system can process image and depth data in real time, making it suitable for applications in dynamic environments. Compared with LiDAR-based or other more complex systems, this system has significant advantages in real-time data processing.

[0024] User interaction and ease of operation: The mouse selection function provides an intuitive and simple way for users to easily select specific transmission lines for analysis. Compared to systems requiring complex settings, this system's user interaction design is more concise and user-friendly.

[0025] Energy and cost savings:

[0026] Compared to systems requiring significant computing resources, this invention reduces the demand for computing resources through optimized algorithms, thereby saving energy. By utilizing common area-array lidar and standard hardware, the overall system cost is reduced.

[0027] Overall, this invention has achieved significant results in improving the accuracy, efficiency, and real-time performance of power transmission line identification and three-dimensional position estimation. It also has the advantages of simple operation and high cost-effectiveness, which make the application of the system in complex environments more feasible and efficient. Attached Figure Description

[0028] Figure 1 This is a program flowchart;

[0029] Figure 2 It is an RGB image;

[0030] Figure 3 This is a depth map;

[0031] Figure 4 The detection results after inputting the RGB image into the neural network show the location of the power transmission line using bounding boxes and straight lines on the power line.

[0032] Figure 5 The image was histogram filtered to remove noise, based on the depth map selected by the rectangle.

[0033] Figure 6 For the selected depth map, select the nearest transmission line contour based on the XY of the mouse selection point, and then use this contour to select the selected depth map from the filtered depth map.

[0034] Figure 7 The fitted wire diagram is displayed as a line in space in the RVIZ software. Detailed Implementation

[0035] Specific Implementation Method 1: The specific process of the real-time power line identification and spatial positioning method based on depth vision assistance in this implementation method is as follows:

[0036] The core of this invention lies in using deep learning and image processing technology to accurately identify and estimate the three-dimensional location of power transmission lines.

[0037] Step 1: Power line image acquisition;

[0038] Step 2: Preprocess the RGB images in the acquired power line images;

[0039] Step 3: Identify the electric field line contours in the pre-trained YOLOv8 model;

[0040] Step 4: Based on the electric field line contours in the RGB image identified by the YOLOv8 model in Step 3, segment each electric field line contour in the simultaneously acquired depth map (at the same time and location), remove the area outside the electric field line contour, and obtain the segmented depth map. Each segmented depth map contains only one electric field line contour.

[0041] After removing depth information outside the contour, these segmented depth maps only contain the depth information of the corresponding single power line, although they may still contain some noise.

[0042] Step 5: Perform histogram analysis to denoise each segmented depth map to obtain a comprehensive depth map containing all power line depth information;

[0043] Step 6: Based on the electric field line contours in the RGB image identified by the YOLOv8 model in Step 3, perform user interaction and electric field line selection;

[0044] Step 7: Based on Steps 3, 5, and 6, perform three-dimensional position estimation of the power line profile:

[0045] Step 8: Output the three-dimensional position of the power line profile.

[0046] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that the power line image acquisition in step one is performed; the specific process is as follows:

[0047] Images of power lines were captured from multiple angles using a Realsense D455 depth camera. The images include RGB images of the power lines and depth maps of the power lines.

[0048] The other steps and parameters are the same as in Specific Implementation Method 1.

[0049] Specific Implementation Method Three: This implementation method differs from Specific Implementation Method One or Two in that step two involves preprocessing the RGB image in the acquired power line image; the specific process is as follows:

[0050] The acquired RGB images of power lines are standardized and adjusted to a uniform size (for images input into the neural network, the size needs to be scaled to a uniform size (e.g., 640*640)). The brightness and contrast of the images are also adjusted to adapt to different lighting conditions.

[0051] Other steps and parameters are the same as in specific implementation method one or two.

[0052] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that step three involves recognizing the electric field line contours in the pre-processed RGB image based on a pre-trained YOLOv8 model; the specific process is as follows:

[0053] Through a ROS node, the standardized RGB image data is read, and the pre-processed RGB image data of the power line is input into the pre-trained YOLOv8 model for power line contour recognition.

[0054] The other steps and parameters are the same as those in one of the specific implementation methods one to three.

[0055] Specific Implementation Method Five: This implementation method differs from Specific Implementation Methods One to Four in that the process of obtaining the pre-trained YOLOv8 model is as follows:

[0056] The RGB image of the electric field line is used as the input to the YOLOv8 model, and the outline of the electric field line in the RGB image is used as the output of the YOLOv8 model. The YOLOv8 model is trained until it converges to obtain a pre-trained YOLOv8 model.

[0057] The YOLOv8 model, after specific training, can accurately identify the location and shape of power lines in complex backgrounds.

[0058] The other steps and parameters are the same as those in one of the specific implementation methods one to four.

[0059] Specific Implementation Method Six: This implementation method differs from Specific Implementation Methods One to Five in that, in step four, based on the electric field line contours in the RGB image identified by the YOLOv8 model in step three, each electric field line contour is segmented in the simultaneously acquired depth map (at the same time and position), and the area outside the electric field line contour is removed to obtain the segmented depth map. Each segmented depth map contains only one electric field line contour.

[0060] After removing depth information outside the contour, these segmented depth maps contain only the depth information of the corresponding single power line, although they may still contain some noise.

[0061] The other steps and parameters are the same as those in one of the specific implementation methods one to five.

[0062] Specific Implementation Method Seven: This implementation method differs from Specific Implementation Methods One through Six in that, in step five, histogram analysis is performed on each segmented depth map to remove noise, thereby obtaining a comprehensive depth map containing all power line depth information; the specific process is as follows:

[0063] Step 51: Remove depth values ​​greater than 2 meters from the outline of only one power line contained in each segmented depth map to obtain the processed depth map for each segment.

[0064] The purpose of this invention is to serve the robotic arm on a drone, which is less than 2 meters long, so a depth value that is too far is meaningless.

[0065] Step 52: Perform histogram analysis to denoise each segmented depth map processed in Step 51 to obtain a comprehensive depth map containing all power line depth information; the specific process is as follows:

[0066] Step 521: Draw histograms for each segmented depth map after processing in Step 51 and identify peak values;

[0067] Step 522: Set a dynamic threshold. The dynamic threshold is 1 / N, where N = 1, 2, ..., 100 (N takes the value of 1, 2, or 100).

[0068] Step 523: If any depth value in the histogram other than the peak value is less than or equal to 1 / N of the peak depth value, then the corresponding depth value is marked as invalid.

[0069] If any depth value in the histogram other than the peak value is greater than 1 / N of the peak depth value, then the corresponding depth value is marked as a valid value.

[0070] Centered on the peak value in the histogram;

[0071] Determine if there are M consecutive invalid depth values ​​to the left of the center. If there are no M consecutive invalid depth values ​​to the left of the center, continue the search on the left. If there are M consecutive invalid depth values ​​to the left of the center, stop the search on the left and set the last valid value as the left boundary. M = 1, 2, ..., 1000.

[0072] Determine if there are M consecutive invalid depth values ​​to the right of the center. If there are no M consecutive invalid depth values ​​to the right of the center, continue the search to the right. If there are M consecutive invalid depth values ​​to the right of the center, stop the search to the right and set the last valid value as the right boundary.

[0073] Set the depth values ​​(pixels) outside the left and right boundaries to zero;

[0074] Eliminate noise in non-peak regions to obtain a well-filtered depth map containing only the depth information of a single power line;

[0075] Step 524: Draw a blank depth map;

[0076] Each well-filtered depth map obtained in steps 5, 2, and 3, containing only the depth information of a single power line, is merged into a blank depth map to obtain a comprehensive depth map containing the depth information of all power lines (the final result is a single map).

[0077] Analysis of the histogram revealed that most pixels in the image are real pixels that need to be retained, with only a small portion being noise. The histogram plotted on the logarithmic y-axis shows a spike in the power line parallel to the camera, while the diagonal lines have multiple plateau-like peaks.

[0078] The method for selecting necessary pixels has evolved from simply selecting a fixed width around the histogram peak to an analysis-based dynamic approach. This involves identifying the peak and then expanding outwards, marking values ​​exceeding one-tenth of the peak as valid. If 50 consecutive values ​​fall below this threshold, the search in that direction ends, and the last valid value is set as the boundary. This adjustment results in two boundary values, outside which pixels are set to zero.

[0079] To further improve the accuracy of the data, the system draws histograms of the effective pixel portions of these depth maps, selects the highest peak in the histogram and its surrounding area as the true depth of the power line, and deletes the rest.

[0080] The other steps and parameters are the same as those in one of the specific implementation methods one to six.

[0081] Specific Implementation Method Eight: This implementation method differs from Specific Implementation Methods One through Seven in that, in step six, user interaction and electric field selection are performed based on the electric field line contours in the RGB image identified by the YOLOv8 model in step three; the specific process is as follows:

[0082] Another ROS node is responsible for implementing the mouse click function.

[0083] In a multi-power line environment, the RGB image of the power line contours output by the YOLOv8 model in step three is embedded into the OpenCV platform. Users can select a specific power line by clicking on the window of the RGB image of the power line contours output by the YOLOv8 model in step three and output the (x,y) coordinates of the mouse click position (the (x,y) coordinates of the mouse click position are a built-in function of the OpenCV platform).

[0084] The system's third ROS node features mouse selection functionality. When multiple power lines are present in the field of view, users can select a specific power line by clicking with the mouse and obtain related information. This functionality relies on receiving RGB images from a depth camera and displaying them in a window using OpenCV. When the user clicks on a location in the image, a mouse callback function is triggered, and the ROS node then publishes the x and y coordinates of the clicked location. Thus, the system publishes three output ROS topics: the outlines of all power lines, a cleaned depth map containing all power lines, and the x and y coordinates of the clicked location.

[0085] The other steps and parameters are the same as those in any of the specific implementation methods one to seven.

[0086] Specific Implementation Method Nine: This implementation method differs from Specific Implementation Methods One to Eight in that, in step seven, a three-dimensional position estimation of the power line profile is performed based on steps three, five, and six.

[0087] Step 71: Based on the (x,y) coordinates of the mouse click position output in Step 6, select the nearest contour as the target from all the electric field line contours output by the YOLOv8 model in Step 3.

[0088] Step 72: Calculate the minimum rotation bounding rectangle of the target and determine the two central axes in the minimum rotation bounding rectangle. Take the longer of the two central axes as the final selected central axis, which represents the position of the electric field line in the two-dimensional image.

[0089] Each electric field line has one and only one minimum bounding rectangle of rotation. Each minimum bounding rectangle of rotation has two central axes. For a relatively slender object like an electric field line, one axis is longer and the other is shorter. The longer axis can be considered as the representative of the electric field line.

[0090] Step 73

[0091] On the final central axis selected in step 72, select L points uniformly, where L = 1, 2, ..., 10;

[0092] With the first selected point as the center, draw a circle A with the shortest central axis of the target's smallest rotational bounding rectangle as the diameter. Take the average depth value of all non-zero depth points within circle A as the depth value of the first selected point.

[0093] With the selected second point as the center, draw a circle A with the shortest central axis of the target's smallest rotational circumscribed rectangle as the diameter. Take the average depth value of all non-zero depth points within circle A as the depth value of the selected second point.

[0094] And so on,

[0095] With the selected Lth point as the center, draw a circle A with the shortest central axis of the target's minimum rotation circumscribed rectangle as the diameter. Take the average depth value of all non-zero depth points within circle A as the depth value of the selected Lth point.

[0096] Obtain the final depth values ​​for the L points;

[0097] In fact, only two points are needed, that is, two endpoints. We could directly obtain the depth values ​​(Z) of these two points, but we did not do so because there are two possibilities. First, the depth map may be disturbed or fluctuating, and these two values ​​may be inaccurate. Second, the positions of these two points may be completely empty, with no depth values ​​available. This would affect the robustness of our program. Therefore, we used ten points.

[0098] Step 74: Remove the points with a depth value of 0 from the final L points obtained in Step 73 to obtain the final P points;

[0099] P points are represented as (x, y), such as (1, 0.125), (2, 0.134), ..., (10, 0.26);

[0100] Where x is the number and y is the depth value (Z value);

[0101] Step 75: Perform least-squares fitting on the final P points obtained in Step 74 to obtain the expression y′=Ax′+B.

[0102] Where y′ is the depth value (Z value), x′ is the point number, and A and B are the parameters of the fitted line;

[0103] Step 76: Substitute x′=1 and x′=P into the expression y′=Ax′+B to obtain two y′ values. These two y′ values ​​represent the depth values ​​of the two endpoints of the target, that is, the Z values ​​of the two endpoints of the target. This depth value is a robust depth value.

[0104] First, it is not easily fluctuated; second, it is not easily affected by holes in the depth map.

[0105] We now have the depth values ​​of the two endpoints, which we add to the two endpoints x and y we obtained earlier;

[0106] Two sets of (x, y, Z) were obtained, where (x, y) are pixel coordinates and Z is the real coordinate system.

[0107] Step 77: Calculate each set of (x,y,Z) and camera intrinsic parameters (fx,fy,cx,cy) to obtain the position coordinates (XYZ) of the two endpoints of the target in the real coordinate system, that is, the three-dimensional position of the two endpoints of the electric field profile.

[0108] The other steps and parameters are the same as those in one of the specific implementation methods one to eight.

[0109] Specific Implementation Method Ten: This implementation method differs from Specific Implementation Methods One to Nine in that the three-dimensional position of the power line profile is output in step eight:

[0110] One method is to directly publish the three-dimensional location of the power line outline;

[0111] Second, it publishes the three-dimensional coordinates of one endpoint of the power line profile and a vector pointing to the other endpoint, represented by XYZ coordinates or quaternions.

[0112] Depth maps may contain holes, resulting in zero values ​​among the depth values. We remove zero values ​​and perform a least-squares fit on the depth values. The fitted expression is then used to update the depth values ​​at the two endpoints, yielding more robust depth values. This operation simplifies 3D fitting of large-scale point clouds to 1D fitting of a small number of sampled points. Experimental data shows that our method is 90% faster than performing principal component analysis (PCA) on the entire point cloud.

[0113] The other steps and parameters are the same as those in any of the specific implementation methods one to nine.

[0114] This invention may have other embodiments. Without departing from the spirit and essence of this invention, those skilled in the art can make various corresponding changes and modifications according to this invention, but these corresponding changes and modifications should all fall within the protection scope of the appended claims.

Claims

1. A method for real-time identification and spatial positioning of power lines based on depth vision assistance, characterized in that: The specific process of the method is as follows: Step 1: Power line image acquisition; Step 2: Preprocess the RGB images in the acquired power line images; Step 3: Identify the electric field line contours in the pre-trained YOLOv8 model; Step 4: Based on the electric field line contours in the RGB image identified by the YOLOv8 model in Step 3, segment each electric field line contour in the simultaneously acquired depth map, remove the area outside the electric field line contour, and obtain the segmented depth map. Each segmented depth map contains only one electric field line contour. Step 5: Perform histogram analysis to denoise each segmented depth map to obtain a comprehensive depth map containing all power line depth information; Step 6: Based on the electric field line contours in the RGB image identified by the YOLOv8 model in Step 3, perform user interaction and electric field line selection; Step 7: Based on Steps 3, 5, and 6, perform three-dimensional position estimation of the power line profile; the specific process is as follows: Step 71: Based on the mouse click position output in Step 6 Coordinates, select the nearest contour as the target from all power line contours output by the YOLOv8 model in step three; Step 72: Calculate the minimum rotation bounding rectangle of the target, and determine the two central axes of the minimum rotation bounding rectangle. Take the longer of the two central axes as the final selected central axis. Step 73 Uniformly select along the final central axis chosen in step 72 One point, ; With the first selected point as the center, draw a circle A with the shortest central axis of the target's smallest rotational bounding rectangle as the diameter. Take the average depth value of all non-zero depth points within circle A as the depth value of the first selected point. With the selected second point as the center, draw a circle A with the shortest central axis of the target's smallest rotational circumscribed rectangle as the diameter. Take the average depth value of all non-zero depth points within circle A as the depth value of the selected second point. And so on, The selected number Using the nth point as the center, draw a circle A with the shorter central axis of the minimum rotation circumscribed rectangle of the target as the diameter. Take the average depth value of all non-zero depth points within circle A as the selected nth point. Depth value of each point; To obtain the final Depth value of each point; Step 74: Remove the final result obtained in Step 73. From the points with a depth of 0, obtain the final P points; Step 75: Perform least-squares fitting on the final P points obtained in Step 74 to obtain the expression. , in, It is the depth value (Z value). These are the point numbers, and A and B are the parameters of the fitted line; Step 76, and Substitution expression , obtained two Value, two The value represents the depth of the two endpoints of the target, that is, the Z value of the two endpoints of the target; Two groups were obtained ; Step 77: For each group By performing calculations with the camera intrinsic parameters (fx, fy, cx, cy), the position coordinates (XYZ) of the two endpoints of the target in the real coordinate system can be obtained, that is, the three-dimensional position of the two endpoints of the electric field profile; Step 8: Output the three-dimensional position of the power line profile.

2. The method for real-time identification and spatial positioning of power lines based on depth vision assistance according to claim 1, characterized in that: The power line image acquisition in step one is as follows: Images of power lines were captured from multiple angles using a Realsense D455 depth camera. The images include RGB images of the power lines and depth maps of the power lines.

3. The method for real-time identification and spatial positioning of power lines based on depth vision assistance according to claim 2, characterized in that: In step two, the RGB image in the acquired power line image is preprocessed; the specific process is as follows: The acquired RGB images of power lines were standardized and adjusted to a uniform size.

4. The method for real-time identification and spatial positioning of power lines based on depth vision assistance according to claim 3, characterized in that: In step three, the electric field line contours in the pre-processed RGB image are identified based on the pre-trained YOLOv8 model; the specific process is as follows: The preprocessed RGB image data of the power lines is input into the pre-trained YOLOv8 model for power line contour recognition.

5. The method for real-time identification and spatial positioning of power lines based on depth vision assistance according to claim 4, characterized in that: The process of obtaining the pre-trained YOLOv8 model is as follows: The RGB image of the electric field line is used as the input to the YOLOv8 model, and the outline of the electric field line in the RGB image is used as the output of the YOLOv8 model. The YOLOv8 model is trained until it converges to obtain a pre-trained YOLOv8 model.

6. The method for real-time identification and spatial positioning of power lines based on depth vision assistance according to claim 5, characterized in that: In step four, based on the electric field line contours in the RGB image identified by the YOLOv8 model in step three, each electric field line contour is segmented in the simultaneously acquired depth map, and the area outside the electric field line contour is removed to obtain a segmented depth map. Each segmented depth map contains only one electric field line contour.

7. The method for real-time identification and spatial positioning of power lines based on depth vision assistance according to claim 6, characterized in that: In step five, histogram analysis is performed on each segmented depth map to remove noise, resulting in a comprehensive depth map containing all power line depth information. The specific process is as follows: Step 51: Remove depth values ​​greater than 2 meters from the outline of only one power line contained in each segmented depth map to obtain the processed depth map for each segment. Step 52: Perform histogram analysis to denoise each segmented depth map processed in Step 51 to obtain a comprehensive depth map containing all power line depth information; the specific process is as follows: Step 521: Draw histograms for each segmented depth map after processing in Step 51 and identify peak values; Step 522: Set the dynamic threshold. The dynamic threshold is... ; Step 523: If the depth values ​​in the histogram other than the peak value are less than or equal to the peak depth value... If so, the corresponding depth value is marked as invalid; If the depth values ​​in the histogram other than the peak value are greater than the peak depth value If so, the corresponding depth value is marked as a valid value; Centered on the peak value in the histogram; Determine if there is a continuous line to the left of the center. The depth value is invalid if there is no continuous depth to the left of the center. If the depth value is invalid, continue the search to the left; if there is a continuous value to the left of the center... If a depth value is invalid, stop the search on the left and set the last valid value as the left boundary. ; Determine if there is a continuity to the right of the center. The depth value is invalid if there is no continuous depth to the right of the center. If the depth value is invalid, continue the search to the right; if there is a continuous value to the right of the center... If a depth value is invalid, stop the search on the right; and set the last valid value as the right boundary. Set the depth values ​​outside the left and right boundaries to zero; Obtain a depth map containing only the depth information of a single power line; Step 524: Draw a blank depth map; Each depth map obtained in steps 5, 2, and 3, which contains only the depth information of a single power line, is merged into a blank depth map to obtain a comprehensive depth map containing the depth information of all power lines.

8. The method for real-time identification and spatial positioning of power lines based on depth vision assistance according to claim 7, characterized in that: In step six, based on the electric field line contours in the RGB image identified by the YOLOv8 model in step three, user interaction and electric field line selection are performed; the specific process is as follows: The RGB image of the identified electric field lines, output by the YOLOv8 model in step three, is embedded into the OpenCV platform. Users can select specific electric field lines and display the mouse click position by clicking on the window containing the RGB image of the identified electric field lines output by the YOLOv8 model in step three. coordinate.

9. The method for real-time identification and spatial positioning of power lines based on depth vision assistance according to claim 8, characterized in that: The three-dimensional position of the electric field line profile is output in step eight; the specific process is as follows: One method is to directly publish the three-dimensional location of the power line outline; Second, it publishes the three-dimensional coordinates of one endpoint of the power line profile and a vector pointing to the other endpoint, represented by XYZ coordinates or quaternions.

Images