High-voltage transmission line inspection data processing method and platform based on UAV

Abstract

The present invention discloses a method and platform for processing patrol data of a high-voltage transmission line based on an unmanned aerial vehicle. The method comprises: obtaining point cloud data of the high-voltage transmission line; processing the point cloud data of the high-voltage transmission line by a processing platform to obtain a high-precision map of the high-voltage transmission line; obtaining the position and posture of the patrol object from the high-precision map, and determining the position and posture of the patrol point according to the position and posture of the patrol object, the high-precision map of the area where the patrol object is located, the model data of the unmanned aerial vehicle, and the camera parameters; generating a patrol path of the unmanned aerial vehicle according to the position and posture of the patrol point, sending the patrol path and the high-precision map of the unmanned aerial vehicle to the unmanned aerial vehicle, so that the unmanned aerial vehicle takes patrol images according to the corresponding position and posture at each patrol point based on the high-precision map; receiving the patrol image returned by the unmanned aerial vehicle, and using a target detection model to perform defect recognition on the patrol image to obtain the patrol result, which can not only ensure that there is no collision between the unmanned aerial vehicle and the patrol object, but also can take a clearer photo of the patrol object.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and platform for processing high-voltage transmission line inspection data based on an unmanned aerial vehicle. Background Art

[0002] In the field of power inspection, the inspection and maintenance of high-voltage transmission lines is an important part of ensuring the safe and stable operation of the power grid. The traditional manual inspection method is not only time-consuming and laborious, but also limited by factors such as terrain and weather, making it difficult to achieve comprehensive, efficient and timely monitoring of transmission lines. In recent years, with the rapid development of drone technology, drone inspection, as an emerging technical means, has gradually shown its unique advantages in the inspection of high-voltage transmission lines.

[0003] Drone inspections are equipped with advanced equipment such as high-definition cameras and infrared thermal imagers, which can shoot and monitor the lines and their surrounding environment from all directions and angles without touching the transmission lines. This method not only greatly improves the inspection efficiency, but also reduces the workload and risks of inspectors. The process of using drones to inspect high-voltage transmission lines specifically includes: the pilot arrives at the site, surveys the terrain, checks whether the high-definition cameras, infrared thermal imagers and other equipment on the drone are normal, and whether the drone battery is sufficient. After the preparations are completed, the pilot controls the remote control device to control the drone to fly to the vicinity of the required inspection parts according to the parts required to be inspected in the inspection list for shooting.

[0004] However, in order to take clear images, the pilot must control the drone as close to the power line as possible. If the drone is too close to the power line, it is easy for the drone to collide with the power line, causing damage to the drone, which requires the pilot to have high flying experience. Summary of the invention

[0005] The embodiments of the present invention provide a high-voltage transmission line inspection data processing method and platform based on an unmanned aerial vehicle, so as to achieve the effect of clearly capturing images of the inspected object while reducing the risk of collision of the unmanned aerial vehicle.

[0006] Some embodiments of the present invention provide a method for processing high-voltage transmission line inspection data based on an unmanned aerial vehicle, the method being applied to a processing platform, and the method comprising: Acquire a survey area of ​​the high-voltage transmission line, generate a survey instruction according to the survey area, and send the survey instruction to the UAV so that the UAV performs survey in the survey area of ​​the high-voltage transmission line to obtain point cloud data of the high-voltage transmission line; Receive the point cloud data of the high-voltage transmission line returned by the UAV, process the point cloud data of the high-voltage transmission line, and obtain a high-precision map of the high-voltage transmission line; obtain the position and posture of the patrol object from the high-precision map, and determine the position and posture of the patrol point according to the position and posture of the patrol object, the high-precision map of the area where the patrol object is located, the model data of the UAV, and the camera parameters; Generate a patrol path of the UAV according to the positions and postures of the patrol points, send the patrol path and high-precision map of the UAV to the UAV, and make the UAV take patrol images at each patrol point according to the corresponding positions and postures based on the high-precision map; wherein the patrol path includes the positions and postures of the patrol points and the order of the patrol points; Receive the inspection images returned by the UAV, and use the target detection model to identify defects in the inspection images to obtain the inspection results.

[0007] In the above technical solution, the processing platform sends a survey instruction to the UAV, and the UAV surveys the high-voltage transmission line in the survey area circled by the survey instruction, obtains point cloud data of the high-voltage transmission line, and transmits the point cloud data of the high-voltage transmission line back, so that the processing platform generates high-precision map data of the high-voltage transmission line based on the point cloud data of the high-voltage transmission line. In this way, the processing platform can generate the posture of the UAV at each patrol point based on the posture of each patrol object in the high-precision map data, the high-precision map of the area where the patrol object is located, the camera parameters and the model data of the UAV, and control the UAV to capture images at each patrol point according to the corresponding posture, which can not only ensure that there is no collision between the UAV and the patrol object, but also can capture the patrol object more clearly, without the need for the pilot to control the flight of the UAV based on personal experience, thus avoiding damage to the UAV due to insufficient pilot experience or inability to capture clear images.

[0008] In a possible implementation, a survey area of ​​a high-voltage transmission line is obtained, a survey instruction is generated according to the survey area, and the survey instruction is sent to a drone so that the drone performs a survey in the survey area of ​​the high-voltage transmission line to obtain point cloud data of the high-voltage transmission line, specifically including: Generate survey and measurement points of high-voltage transmission lines, generate survey instructions according to the survey and measurement points, and the processing platform sends the survey instructions to the UAV; so that the UAV surveys the survey and measurement points according to the survey instructions to obtain point cloud data of the survey and measurement points; Accordingly, the point cloud data of the high-voltage transmission line returned by the drone is received, and the point cloud data of the high-voltage transmission line is processed to obtain a high-precision map of the high-voltage transmission line, specifically including: Receive the point cloud data of the surveyed location returned by the drone, match the point cloud data of the surveyed location with the model data of the high-voltage transmission line, and generate a high-precision map of the high-voltage transmission line.

[0009] In the above technical scheme, since the high-voltage transmission line is mainly composed of conductors, ground wires and towers, all of which are long strips with relatively small reflection surfaces, relatively few reflected waves are received when surveying with drones, which is not conducive to building map data. By using the existing model data of the high-voltage transmission line, some positioning points on the conductors, ground wires and towers are selected for survey, and the measured three-dimensional data of the surveyed positioning points are matched with the surface model data of the high-voltage transmission line to obtain a high-precision map of the high-voltage transmission line, the amount of survey data of the drone can be reduced.

[0010] In a possible implementation, the posture of the patrol point is determined according to the posture of the patrol object, the high-precision map of the area where the patrol object is located, the model data of the drone, and the camera parameters, specifically including: For each patrol object, an observation point of the patrol object is generated according to the posture of the patrol object, the high-precision map of the area where the patrol object is located, the model data of the drone, and the camera parameters. The observation point includes the position and posture. According to the positions of multiple observation points and the camera optical axis coordinates under the multiple observation points, it is determined whether the multiple observation points meet the fusion requirements. If they do, the observation points are fused to obtain patrol points, and the patrol objects corresponding to the observation points are used as the patrol objects of the patrol points.

[0011] In the above technical solution, after calculating the observation point of the patrol object, multiple observation points are fused according to the observation point of the patrol object and the camera optical axis coordinates at each observation point. In this way, multiple patrol objects can be patrolled at one patrol point, thereby improving the patrol efficiency of the drone.

[0012] In a possible implementation, according to the position of the observation point of the patrol object and the camera optical axis coordinates at each observation point, the observation points of the patrol object are fused to obtain the patrol point, and the patrol object corresponding to the observation point is used as the patrol object of the patrol point, specifically including: Step a, grouping multiple observation points to obtain an observation point group; Step b: for each group of observation points, a core observation point is selected from the observation point group, and the spatial distance between each observation point and the core observation point, as well as the angle between each observation point and the core observation point, are calculated; Step c, obtaining a target observation point that meets the fusion requirements from multiple observation points, wherein the target observation point is an observation point whose spatial distance is less than a preset distance threshold and whose optical axis angle is less than a preset angle threshold, Step d: if the number of target observation points is greater than half of the number of observation points in the observation point group, the average position and average posture of the multiple target observation points are calculated to obtain patrol points, and the patrol objects corresponding to the multiple target observation points are used as patrol objects corresponding to the patrol points; Step e: if the number of patrol points obtained is greater than half of the number of observation point groups, stop the loop; if the number of patrol points is less than or equal to half of the number of observation point groups, return to step a.

[0013] In the above technical solution, the spatial distance between the two observation points and the angle between the camera optical axes corresponding to the two observation points are used as fusion parameters. If the spatial distance is less than the preset distance threshold and the angle between the camera optical axes is less than the preset angle threshold, it means that the two observation points are close and the camera postures at the observation points are also similar. The average value of the two observation points is used as the posture of the patrol point. This ensures that when observing multiple patrol objects at the patrol point, the patrol image can be clearly captured.

[0014] In a possible implementation, for each patrol object, an observation point of the patrol object is generated according to the posture of the patrol object, a high-precision map of the area where the patrol object is located, the model data of the drone, and the camera parameters, specifically including: Obtain the patrol surface coordinates of the patrol object, and generate multiple candidate points according to the patrol surface coordinates of the patrol object and camera parameters, wherein the candidate points include positions and postures; For each candidate point, determine whether the drone will collide at the candidate point based on the drone's model data and the high-precision map of the area where the patrol object is located. If no collision occurs, the candidate point is retained; A candidate point farthest from an obstacle is selected from a plurality of reserved candidate points as an observation point, wherein the obstacle is an object in the area where the patrol object is located.

[0015] In the above technical solution, multiple candidate points are generated based on the patrol surface coordinates and camera parameters of the patrol object. For each candidate point, the candidate point that collides with the surrounding objects is first deleted, and the candidate point farthest from the obstacle is selected from the retained candidate points as the observation point. This ensures that the drone can clearly photograph the patrol object without colliding with obstacles.

[0016] In a possible implementation, multiple candidate points are generated according to the patrol surface coordinates and camera parameters of the patrol object, specifically including: The normal angle of the patrol surface is taken as the optical axis angle to obtain the projection of the patrol object on the patrol surface; Select multiple points in the projection center area; for each point, obtain the coordinates of the straight line where the camera optical axis is located according to the angle between a point in the projection center area and the optical axis; For the coordinates of the straight line where the optical axis of each camera is located, an alternative point is calculated according to the coordinates of the straight line where the optical axis of the camera is located, the installation parameters of the camera and the focal length.

[0017] In the above technical solution, the normal angle of the patrol surface coordinates is used as the optical axis angle, so that the camera optical axis passes through the projection center area on the patrol surface coordinates, ensuring that the patrol object is located in the middle of the patrol image and can be clearly photographed. By generating multiple camera postures, one can select one from multiple postures that is farthest from obstacles to ensure that no collision occurs when the drone is shooting.

[0018] In a possible implementation manner, if the patrol object is a pin, the patrol surface is a plane passing through the pin axis and two end surfaces perpendicular to the pin axis; If the patrol object is an insulator, the patrol surface is a plane passing through the axis of the insulator and two end surfaces perpendicular to the axis of the insulator; In the above technical solution, by selecting a plane passing through the axis and a plane perpendicular to the axis, images of important areas of the inspected object can be clearly captured, and defects of the inspected object can be accurately identified.

[0019] In a possible implementation, the processing platform uses the target detection model to perform defect recognition on the inspection image to obtain the inspection result, which specifically includes: The processing platform receives the patrol image and the patrol object corresponding to the patrol image; Obtain the target detection model corresponding to the patrol object according to the patrol object and model mapping relationship table; the model mapping relationship table is the corresponding relationship between the patrol object category and the target detection model; The object detection model corresponding to the inspected object is used to identify the inspected image and obtain the defect type of the inspected object.

[0020] In the above technical solution, different target detection models are trained for different patrol objects. After the drone takes patrol images at each patrol point, the patrol images and patrol object identifiers corresponding to the patrol points are sent to the background platform. In this way, the background platform can select the corresponding target detection model to identify the patrol images, thereby improving the recognition accuracy.

[0021] In a possible implementation, the method further includes: When erecting a high-voltage transmission line, a monitoring command is generated and sent to a drone, so that the drone can take monitoring images of the erection site; Receive monitoring images, and use the target detection model to process the monitoring images to obtain monitoring results; When the monitoring results indicate that a person has violated a rule, the drone is controlled to play a warning voice, and the processing platform generates an inspection report after the drone inspection is completed.

[0022] In the above technical solution, during the installation of high-voltage transmission lines, the processing platform generates monitoring instructions and controls the drone to take monitoring images of the installation site. The processing platform uses the target detection model to identify the monitoring images and determine whether there are any violations at the installation site.

[0023] The embodiment of the present invention provides a high-voltage transmission line inspection data processing platform based on an unmanned aerial vehicle, comprising: An acquisition module is used to acquire a survey area of ​​a high-voltage transmission line, generate a survey instruction according to the survey area, and send the survey instruction to a drone so that the drone performs a survey in the survey area of ​​the high-voltage transmission line to obtain point cloud data of the high-voltage transmission line; A receiving module, used for receiving the point cloud data of the high-voltage transmission line returned by the drone, processing the point cloud data of the high-voltage transmission line, and obtaining a high-precision map of the high-voltage transmission line; A processing module, used to obtain the position and posture of the patrol object from the high-precision map, and determine the position and posture of the patrol point according to the position and posture of the patrol object, the high-precision map of the area where the patrol object is located, the model data of the drone, and the camera parameters; A sending module, used for generating a patrol path of the UAV according to the positions and postures of the patrol points, and sending the patrol path of the UAV and the high-precision map to the UAV, so that the UAV takes patrol images at each patrol point according to the corresponding positions and postures based on the high-precision map; wherein the patrol path includes the positions and postures of the patrol points and the order of the patrol points; The receiving module is also used to receive the patrol image returned by the drone, and use the target detection model to perform defect recognition on the patrol image to obtain the patrol result.

[0024] The embodiment of the present invention provides a high-voltage transmission line inspection data processing platform based on an unmanned aerial vehicle, comprising: a memory, a processor; Memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor executes various possible implementations as described above.

[0025] An embodiment of the present invention provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the above various possible implementation methods.

[0026] An embodiment of the present invention provides a computer program product, including a computer program, which implements the above various possible implementation modes when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0028] Figure 1 A schematic diagram of a high-voltage overhead transmission line in the present invention; Figure 2 A system architecture diagram provided for some embodiments of the present invention; Figure 3 A schematic diagram of the flow of a high-voltage transmission line inspection data processing method based on an unmanned aerial vehicle provided by the present invention; Figure 4 A schematic diagram of generating the optical axis pose of a camera provided in some embodiments of the present invention; Figure 5 A schematic diagram of the structure of a processing platform provided in some embodiments of the present invention.

[0029] Reference numerals: 1. Conductor; 2. Ground wire; 3. Anti-vibration hammer; 4. Wire clamp; 5. Insulator; 6. Pole tower; 7. Grounding device; 8. Grounding down conductor; 10. Processing platform; 20. UAV remote control; 30. UAV; 11. Processor; 12. Memory; 13. Communication component; 14. Bus.

[0030] The above drawings have shown clear embodiments of the present invention, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0031] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0032] To ensure the safety of high-voltage transmission line operations and improve the level of safety supervision. The present invention combines drones to collect data from high-voltage transmission lines, and then transmits it to the processing platform for intelligent analysis and judgment, and outputs warning information in a timely manner when abnormalities occur in the analysis and judgment results. For example: output warning information to the manager's smart terminal. Among them, intelligent analysis includes using target detection models to identify transmission line defects during the power grid operation and maintenance stage, and whether the operators comply with the operating regulations during the power grid construction stage.

[0033] Figure 1is a schematic diagram of a high voltage overhead transmission line, such as Figure 1 As shown, the structure of the high-voltage overhead transmission line includes a pole tower 6, and a conductor 1 and a ground wire 2 erected on the pole tower 6. When erecting the conductor 1 and the ground wire 2, an insulator 5 is required to connect the conductor 1 and the pole tower 6. A wire clamp 4 is also installed near the end of the insulator 5 close to the conductor 1. The wire clamp 4 wraps the conductor 1 and is fixed to the end of the insulator 5. An anti-vibration hammer 3 is also installed near the conductor 1 close to the insulator 5. An anti-vibration hammer 3 is also provided on the ground wire 2, and a wire rack 4 is also used to fix the ground wire 2 to the pole tower 6. A grounding device 7 is also provided below the underground part of the pole tower 6.

[0034] After the high-voltage transmission lines are erected, the maintenance of high-voltage transmission lines is extremely important. In recent years, with the rapid development of drone technology, the use of drones to patrol high-voltage transmission lines has become a new trend.

[0035] The specific process of using drones to inspect high-voltage transmission lines includes: the pilot arrives at the site, surveys the site terrain, and checks whether the high-definition camera, infrared thermal imager and other equipment on the drone are normal, and whether the drone battery is sufficient. After the preparation work is completed, the pilot controls the remote control device to control the drone to fly near the parts required to be inspected according to the inspection list to take pictures.

[0036] However, in order to take clear images, the pilot must control the drone as close to the power line as possible. If the drone is too close to the power line, it is easy for the drone to collide with the power line, causing damage to the drone, which requires the pilot to have high flying experience.

[0037] The present invention provides a high-voltage transmission line inspection data processing method and platform based on drones. The purpose is to ensure the safety of maintenance and inspection operations after the high-voltage transmission line is erected and improve the level of safety supervision. In addition, during the high-voltage transmission line erection stage, the construction site can be photographed by drones, and the images of the construction site can be identified to determine whether there are any violations in the construction scene.

[0038] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0039] Figure 2 A system architecture diagram provided for some embodiments of the present invention, such as Figure 2As shown, the high-voltage transmission line inspection data processing platform 10 (hereinafter referred to as the processing platform 10) is an edge computing terminal, which is directly connected to the drone remote controller 20 via WiFi. The edge computing terminal generates a high-precision map of the high-voltage transmission line, plans the drone inspection path based on the high-precision map, and sends the high-precision map and the drone inspection path to the drone 30 through the drone remote controller 20, so that the drone 30 can patrol and shoot according to the drone inspection path based on the high-precision map. Since there is no need for pilot control, it can be guaranteed that the drone 30 can shoot clear images within a safe distance. The drone remote controller 20 pushes the video stream to the edge computing terminal via WiFi, and the target detection model running on the edge computing terminal analyzes the video stream in real time to determine whether there are any violations in the construction scene or whether there are defects in the transmission line. It should be noted here that the communication between the drone 30 and the processing platform 10 described below refers to communication using the drone remote controller 20 as an intermediate medium.

[0040] More specifically, during the power grid construction phase, drones are used for high-altitude patrols to capture images of construction site operations. The target detection model determines in real time whether personnel are engaging in the following five types of violations: safety helmet wearing violations, work clothes wearing violations, safety belt wearing violations, standing under the tower, and using mobile phones. If any of the above violations occur, a voice warning will be issued through the loudspeaker mounted on the drone, and the system will generate an inspection report after the patrol is completed.

[0041] After the power grid is built and put into operation, drones are used for high-altitude patrols to capture patrol images of high-voltage transmission lines. The target detection model identifies transmission line defects in real time to determine whether the following defects exist: missing pins, pin withdrawal, pin opening, loose nuts, missing nuts, missing anti-vibration hammer nails, self-explosion of glass insulators, dirty insulators, loose conductors, tilted equalizing rings, flooding at the bottom of the tower, damaged tower base, vegetation coverage, unclear signboard images and text, damaged signboards, etc.

[0042] Figure 3 A schematic diagram of a high-voltage transmission line inspection data processing method based on an unmanned aerial vehicle is provided in some embodiments of the present invention, such as Figure 3 As shown, the method is applied to a processing platform and a UAV, and the method specifically includes: S101. The processing platform obtains a survey area of ​​the high-voltage transmission line, generates a survey instruction according to the survey area, and sends the survey instruction to the UAV.

[0043] The survey area refers to the geographical area where the high-voltage transmission line is built. The location data of the geographical area includes the latitude, longitude and elevation information of the key locations in the survey area, as well as the shape of the survey area. The survey area can be characterized by the above information.

[0044] The processing platform encodes the location data of the geographic area to generate a survey instruction, and sends the survey instruction to the drone. The drone performs survey based on the survey instruction.

[0045] Specifically, first, the processing platform needs to obtain the survey area of ​​the high-voltage transmission line. This step usually involves collecting the geographical location information of the high-voltage transmission line, including key data such as longitude, latitude, and elevation. With this data, the processing platform is able to build a complete survey area. Then, based on the data of the survey area, the processing platform generates corresponding survey instructions. These instructions include detailed information such as the flight path of the drone and shooting requirements. Subsequently, the processing platform sends the survey instructions to the drone via wireless communication.

[0046] S102. The UAV conducts a survey in a survey area of ​​the high-voltage transmission line, obtains point cloud data of the high-voltage transmission line, and sends the point cloud data of the high-voltage transmission line to a processing platform.

[0047] Among them, the drone is equipped with a laser radar. The drone traverses the high-voltage transmission line, uses the laser radar to scan the high-voltage transmission line, constructs its three-dimensional point cloud model, and sends the point cloud data of the high-voltage transmission line to the processing platform.

[0048] Specifically, after receiving the survey command, the drone conducts a flight survey in the survey area of ​​the high-voltage transmission line according to the command requirements. The laser radar equipment on the drone scans the survey area and generates point cloud data of the high-voltage transmission line. After completing the survey, the drone transmits the point cloud data back to the processing platform via wireless communication.

[0049] S103. The processing platform processes the point cloud data of the high-voltage transmission line to obtain a high-precision map of the high-voltage transmission line.

[0050] Among them, the processing platform vectorizes and objectifies the point cloud data of the high-voltage transmission lines to obtain a high-precision map of the high-voltage transmission lines.

[0051] Specifically, after receiving the point cloud data returned by the drone, the processing platform processes the data. The processing process includes steps such as data cleaning, vectorization, and objectification, and finally generates a high-precision map of the high-voltage transmission line. The high-precision map contains not only the location information of the line, but also detailed data such as the shape and height of the line.

[0052] Optionally, point cloud data often contains noise points introduced by equipment errors, environmental factors, etc. The processing platform first uses statistical methods and filtering algorithms, such as radius filtering and statistical filtering, to remove these noise points to improve the accuracy and efficiency of subsequent processing.

[0053] In addition, since drones may be equipped with different types of lidar equipment, the format of the returned point cloud data may be different. The processing platform will unify the format of these data and convert them into an internal unified point cloud data format for subsequent processing.

[0054] Among them, the processing platform will mesh the cleaned point cloud data and convert discrete three-dimensional points into a continuous mesh model. This process usually uses methods such as Delaunay triangulation algorithm or Poisson surface reconstruction to convert point cloud data into triangular meshes or quadrilateral meshes. The generated mesh model may have problems such as unevenness and overlap. The processing platform will optimize the mesh, such as smoothing and mesh simplification, to improve the accuracy and rendering efficiency of the model.

[0055] On the vectorized mesh model, the processing platform will perform feature extraction operations. This includes extracting geometric features such as line edges, corners, planes, and visual features such as color and texture. These features will be used for subsequent object recognition and classification.

[0056] Based on the extracted features, the processing platform uses machine learning or deep learning algorithms to identify and classify objects in the grid model. This includes identifying key components of high-voltage transmission lines such as conductors, ground wires, poles, insulators, and classifying and labeling them. For the identified and classified objects, the processing platform will give them detailed attribute information, such as location, shape, height, etc. This information will be used to generate high-precision maps and provide data support for subsequent patrol route planning, defect identification and other tasks.

[0057] To ensure the consistency and accuracy of the HD map, the processing platform will uniformly process the coordinates of all objects. This includes converting the local coordinates of the object into global coordinates, and correcting and aligning the coordinate system. Based on the objectified results and the unified coordinate system, the processing platform will build a HD map. The HD map contains not only the location information of the line, but also detailed data such as the shape, height, and direction of the line. At the same time, the HD map also records the attribute information of each object, such as type, material, size, etc.

[0058] After the high-precision map is generated, the processing platform will conduct a quality check to ensure the accuracy and completeness of the map. For any errors or problems, the processing platform will perform corresponding optimization processing, such as interpolation and smoothing, to improve the quality of the map.

[0059] S104: The processing platform obtains the position and posture of the patrol object from the high-precision map, and determines the position and posture of the patrol point according to the position and posture of the patrol object, the high-precision map of the area where the patrol object is located, the camera parameters and the model data of the UAV.

[0060] The high-precision map includes the shape data, position data and posture data of each component in the high-voltage transmission line, such as conductors, ground wires, poles, insulators, various hardware, etc. The inspection objects are extracted from the inspection requirement operation file.

[0061] The position and posture of the patrol object and the high-precision map of the area where the patrol object is located are obtained from the high-precision map. The position and posture of the patrol point are calculated based on the position and posture of the patrol object, the camera parameters, the high-precision map of the area where the patrol object is located, and the model data of the drone. The position and posture of the patrol object and the camera parameters are used to ensure that the camera can clearly capture the patrol image when the drone is located at the patrol point and in the required posture. The high-precision map of the area where the patrol object is located and the model data of the drone are used to ensure that the drone will not collide with other objects when it is located at the patrol point and in the required posture.

[0062] There are multiple inspection objects, and each inspection object corresponds to multiple inspection points. The inspection object can be observed from multiple angles at different inspection points to inspect all parts of the inspection object and improve the reliability of the inspection.

[0063] Specifically, in the high-precision map, the processing platform can accurately obtain the position and posture information of each patrol object. The patrol objects usually include key components such as conductors, ground wires, poles, insulators, etc. Through the high-precision map, the processing platform can understand the precise position and posture of these components.

[0064] Then, based on the posture of the patrol object, the high-precision map of the area where the patrol object is located, the model data of the drone, and the camera parameters, the processing platform can calculate the posture of the patrol point. The posture of the patrol point includes position and attitude information, ensuring that the drone can clearly capture the patrol object at the patrol point.

[0065] Optionally, the processing platform first applies a deep learning-based target detection algorithm to automatically identify key components such as conductors, ground wires, poles, insulators, etc. in the HD map. These algorithms have been extensively trained to efficiently and accurately identify target objects. After identifying the target objects, the processing platform extracts the attribute information of these components from the HD map, including location coordinates, size, direction, etc.

[0066] For each identified patrol object, the processing platform will calculate its posture information based on its geometric shape and relative position in the high-precision map. For example, for a pole tower, its verticality and tilt angle will be calculated; for a conductor and ground wire, its direction and suspension status will be calculated. The calculated posture information will undergo a rigorous verification process to ensure the accuracy and reliability of the data. Verification methods may include comparison with known data, internal logic consistency check, etc.

[0067] After determining the position and posture of the patrol object, the processing platform will calculate the position and posture of the patrol point based on this information, as well as the high-precision map of the patrol area, the model data of the drone, and the camera parameters. The following is a further specific implementation of this step: Step 1: Generate observation points Alternative point generation: For each patrol object, the processing platform generates multiple alternative observation points based on its position, high-precision map, drone model data, and camera parameters. These observation points represent potential locations where the drone may photograph the patrol object.

[0068] Collision detection: For each candidate point, the processing platform will perform collision detection to ensure that the drone will not collide with surrounding obstacles when shooting at these points. This usually involves complex spatial geometry calculations and drone flight simulation.

[0069] Step 2: Observation point fusion Grouping and evaluation: The processing platform groups the generated observation points and evaluates the spatial distance between the observation points and the camera optical axis angle within each group. If the distance between two observation points is less than the preset distance threshold and the camera optical axis angle is less than the preset angle threshold, the two observation points are considered to be fused.

[0070] Fusion calculation: For observation points that meet the fusion conditions, the processing platform will calculate their average position and average attitude to determine the final patrol point position and attitude. This process will be iterated until no more observation points that meet the fusion conditions can be found.

[0071] Step 3: Optimization and verification Path optimization: After determining the positions of all patrol points, the processing platform will apply the path optimization algorithm to optimize the entire patrol path to reduce the flight time and energy consumption of the drone.

[0072] Simulation verification: Finally, the processing platform will use high-precision maps and drone models to perform simulated flight verification to ensure the feasibility and safety of the patrol path. This includes checking whether the drone will collide with obstacles during flight and whether it can capture clear images of the patrol object at each patrol point.

[0073] S105. The processing platform generates a patrol path for the UAV according to the position and posture of the patrol points, and sends the patrol path and high-precision map of the UAV to the UAV.

[0074] Among them, the processing platform uses the existing path planning algorithm to process the posture of the patrol points and generate the patrol path of the drone. The patrol path of the drone includes the posture of the patrol points and the order of the patrol points. The order of the patrol points is sorted among the patrol points.

[0075] Specifically, the processing platform generates the patrol path of the UAV using a path planning algorithm based on the position information of the patrol points. The patrol path includes the position information of the patrol points and the order of the patrol points. Figure 1 And send it to the drone. After receiving the patrol route and high-precision map, the drone flies according to the route planning. At each patrol point, the drone takes pictures according to the preset posture to ensure the clarity and accuracy of the patrol image. After shooting, the drone sends the patrol image back to the processing platform.

[0076] Optionally, the processing platform selects a suitable path planning algorithm based on the position information of the patrol point. Commonly used path planning algorithms include A* algorithm, Dijkstra algorithm, RRT (Rapidly-exploring Random Trees) algorithm, etc. These algorithms can quickly generate the flight path of the drone while ensuring the optimal or suboptimal path.

[0077] When applying the path planning algorithm, the processing platform needs to set the algorithm parameters according to the actual situation, such as the weight of the heuristic function, the expansion method of the node, etc., to ensure that the generated path meets the actual needs. The processing platform uses the selected path planning algorithm to calculate the flight path of the drone based on the posture information of the patrol points. The path includes the location coordinates, flight altitude, flight speed and other information of each patrol point. To ensure the efficiency and safety of the patrol path, the processing platform will optimize the initially generated path. The optimization goals may include shortening the flight time, reducing energy consumption, avoiding obstacles, etc.

[0078] The processing platform integrates the position information of the patrol points and the path planning results into complete patrol path information, including the position information of the patrol points, flight path, flight altitude, flight speed, etc. The processing platform transmits the patrol path information and high-precision Figure 1 And send it to the drone. After receiving this information, the drone can fly and shoot according to the planned path.

[0079] After receiving the patrol route and high-precision map, the drone uses the built-in flight control system and navigation system to fly along the planned route. During the flight, the drone will update its position in real time and compare it with the planned path to ensure the accuracy of the flight trajectory. During the flight, the drone will automatically adjust according to the planned flight altitude and speed to ensure flight stability and shooting effect. When the drone arrives at each patrol point, it will adjust its attitude according to the preset posture information. This includes the drone's flight altitude, heading, pitch angle and other parameters to ensure the optimization of shooting angle and distance. After adjusting the posture, the high-definition camera on the drone will start the shooting function to take clear and accurate photos of the patrol object. During the shooting process, the drone will ensure the stability and clarity of the image to avoid jitter and blur.

[0080] After shooting, the drone will transmit the patrol images back to the processing platform via wireless communication. During the transmission process, the drone will compress and encrypt the image data to ensure the security and efficiency of data transmission.

[0081] After receiving the inspection images, the processing platform will perform further image processing and analysis. This includes steps such as image decompression, enhancement, and correction to ensure image quality and usability. Subsequently, the processing platform will use the object detection model to identify and analyze defects in the inspection images and generate the final inspection report.

[0082] For example, after obtaining the positions of three patrol points, the patrol path is obtained through the path planning algorithm: patrol point 3, patrol point 1, patrol point 2. Then the order of patrol point 3 is 1, the order of patrol point 1 is 2, and the order of patrol point 2 is 3.

[0083] S106. The UAV takes patrol images at each patrol point according to the corresponding posture based on the high-precision map, and returns the patrol images to the processing platform.

[0084] Among them, the UAV uses high-precision maps for flight navigation, passes through the inspection points in the order of each inspection point in the inspection route, takes inspection images at each inspection point according to the corresponding posture, and returns the inspection images to the processing platform.

[0085] For example, the position of patrol point 3 is (x, y, z), and the posture of patrol point 3 is (α, β, γ). When the drone reaches the position (x, y, z), it adjusts its posture to (α, β, γ), and the camera on the drone takes a picture to obtain the patrol image of patrol point 3.

[0086] S107: The processing platform uses the target detection model to perform defect recognition on the inspection image to obtain the inspection result.

[0087] Among them, by training the target detection model, the target detection model can identify various defects in the high-voltage transmission line.

[0088] Specifically, after receiving the inspection images returned by the drone, the processing platform uses the pre-trained target detection model to identify defects in the images. The target detection model can identify various defects in high-voltage transmission lines, such as missing pins and dirty insulators. The processing platform generates an inspection report based on the recognition results, providing an important reference for subsequent maintenance and repair work.

[0089] Optionally, the processing platform receives the patrol images sent back by the drone through a stable wireless communication link. To ensure the integrity and accuracy of the data, an error check and retransmission mechanism will be implemented during the reception process. The received images will go through a series of preprocessing steps, including decoding, denoising, color correction, etc., to eliminate distortion and interference that may be introduced during the transmission process, improve image quality, and lay a good foundation for subsequent target detection.

[0090] The processing platform loads pre-trained target detection models from storage. These models are usually built based on deep learning frameworks (such as TensorFlow, PyTorch), and are trained on a large amount of high-voltage transmission line image data to have the ability to identify various types of defects. After loading the model, the processing platform will perform necessary initialization operations, such as setting model parameters, loading weight files, and configuring computing resources, to ensure that the model can run efficiently.

[0091] The preprocessed image is input into the target detection model. The model analyzes the image pixel by pixel, extracts features, and identifies the target objects (such as pins, insulators, etc.) and their status in the image. Based on the learned knowledge, the model makes defect judgments on the identified target objects. For example, by comparing the appearance of the target object with the normal state, defects such as missing pins and dirty insulators can be identified. The model outputs the identified defect information (including defect type, location, severity, etc.) in a structured format for subsequent processing and analysis.

[0092] The processing platform integrates the defect information output by the model with the metadata of the image (such as shooting time, location, drone number, etc.) to form a complete inspection record. Based on the inspection record, the processing platform automatically generates an inspection report. The report content includes defect summary, defect distribution map, defect severity assessment, recommended maintenance measures, etc., providing intuitive and comprehensive information support for operation and maintenance personnel. The generated inspection report will be distributed to the relevant operation and maintenance team through email, internal system notification, etc., so that they can understand the line status in a timely manner and formulate and implement maintenance and repair plans.

[0093] During the actual maintenance process, the operation and maintenance team will conduct on-site verification of the defects in the inspection reports and feedback the verification results to the processing platform. This feedback data is crucial to improving the accuracy of the model. The processing platform will regularly collect new inspection images and feedback data for retraining and updating the model. Through continuous learning, the model can continuously adapt to changes in the line environment and improve the accuracy and robustness of defect identification.

[0094] For example: by training the target detection model, the target detection model can identify defects such as missing pins, pin withdrawal, pin opening, loose nuts, missing nuts, missing anti-vibration hammer nails, self-explosion of glass insulators, dirty insulators, loose conductors, tilted equalizing rings, flooding at the bottom of the tower, damaged tower base, vegetation coverage, unclear signboard images and text, and damaged signboards.

[0095] After receiving the inspection images sent back by the UAV, the processing platform uses the target detection model to identify defects in the inspection images and obtain the inspection results.

[0096] In the above technical solution, the processing platform sends a survey instruction to the UAV, and the UAV surveys the high-voltage transmission line in the survey area circled by the survey instruction, obtains point cloud data of the high-voltage transmission line, and transmits the point cloud data of the high-voltage transmission line back, so that the processing platform generates high-precision map data of the high-voltage transmission line based on the point cloud data of the high-voltage transmission line. In this way, the processing platform can generate the posture of the UAV at each patrol point based on the posture of each patrol object in the high-precision map data, the high-precision map of the area where the patrol object is located, the camera parameters and the model data of the UAV, and control the UAV to capture images at each patrol point according to the corresponding posture, which can not only ensure that the UAV will not collide with the patrol object, but also capture the patrol object more clearly, without the need for the pilot to control the flight of the UAV based on personal experience, thus avoiding damage to the UAV or inability to capture clear images due to insufficient pilot experience.

[0097] Specifically, the drone conducts detailed surveys in the designated area according to the survey instructions issued by the processing platform, and the collected point cloud data of high-voltage transmission lines are characterized by high precision and high density. The processing platform uses these point cloud data to generate accurate high-precision maps of high-voltage transmission lines. This map not only contains the spatial location information of the line, but also accurately reflects the detailed shape and size of the line and its ancillary facilities (such as towers, insulators, conductors, etc.).

[0098] Based on the generated high-precision map, the processing platform can automatically identify the key components and areas that need to be inspected, namely the inspection objects. By comprehensively considering the position of the inspection object, the topography of the area, the flight capability of the drone and the camera parameters, the processing platform can intelligently plan the optimal inspection path and the specific position of each inspection point. This intelligent path planning method ensures that the drone can avoid collisions with the inspection object during the inspection process and can capture clear images at the best angle and distance.

[0099] The drone flies and shoots according to the planned patrol path and posture, without relying on the pilot's personal experience and operating skills. This automated and standardized shooting method greatly improves the clarity and accuracy of the captured images, providing a reliable data basis for subsequent defect identification and analysis. At the same time, by avoiding the uncertainty caused by the pilot's operation, it also significantly improves the safety of drone flight and reduces flight accidents and equipment damage caused by improper operation.

[0100] Traditional transmission line inspection methods often require a lot of manpower and are greatly affected by external factors such as weather and terrain. However, the use of this technical solution, through the automated inspection of drones, can greatly reduce the need for manual inspections and reduce operation and maintenance costs. At the same time, since the flight speed and shooting efficiency of drones are much higher than those of manual labor, the overall efficiency of inspection work is significantly improved.

[0101] Some embodiments of the present invention provide a method for processing high-voltage transmission line inspection data based on an unmanned aerial vehicle, the method being applied to a processing platform and an unmanned aerial vehicle, and the method specifically comprising: S201. The processing platform generates a survey location of the high-voltage transmission line, generates a survey instruction according to the survey location, and sends the survey instruction to the UAV.

[0102] Among them, the high-voltage transmission line includes conductors, ground wires, poles, insulators, wire clamps, and anti-vibration hammers. At least one surveying and measuring point can be selected on each component. In particular, since there is usually vegetation coverage near the bottom of the pole, multiple surveying and measuring points can be set at the bottom of the pole to survey the point cloud data near the tower foundation. The geographical area where each surveying and measuring point is located is determined according to the erection data of the high-voltage transmission line, and the geographical area where each surveying and measuring point is located is encoded to generate a survey instruction, and the processing platform sends the survey instruction to the drone.

[0103] Optionally, the area to be surveyed can be clearly defined based on the geographical location, length, direction, and patrol requirements of the high-voltage transmission line. Within the determined survey scope, key nodes of the high-voltage transmission line are selected as survey and determination points, such as substation exits, line turning points, important crossing points (such as rivers, railways, roads, etc.), tower locations, etc. Considering the importance and complexity of the line, the density of positioning points can be appropriately increased to ensure that the subsequently generated high-precision map can fully reflect the line status. Then, use the geographic information system (GIS) or line design drawings to accurately calculate the geographic coordinates (longitude, latitude, altitude) of each survey and determination point.

[0104] Next, the coordinates, flight altitude, speed, shooting angle, shooting distance and other parameters of each surveying location are encoded into a command format that the drone can recognize. The command can also include a safe obstacle avoidance strategy during the drone flight to ensure that the drone does not collide with the line or obstacles during the survey. The survey command is sent to the drone through the drone ground control station or remote communication link. Ensure that the drone receives the command and interprets it correctly, and is ready to perform the survey mission.

[0105] S202: The UAV surveys the surveying location point according to the surveying instruction to obtain point cloud data of the surveying location point.

[0106] Among them, the survey and determination points include pole tower positioning points, conductor positioning points, ground wire positioning points, insulator positioning points, wire clamp positioning points and anti-vibration hammer positioning points.

[0107] The drone flies to the geographical area where each survey point is located and uses a lidar to scan the geographical area to obtain point cloud data of the survey point.

[0108] Optionally, the drone automatically plans a flight path based on the survey instructions received, and flies along the specified route and altitude. During the flight, the drone should adjust its flight attitude and speed in real time to ensure that the survey mission is performed in accordance with the instructions.

[0109] The laser radar or photogrammetry equipment carried by the drone continuously scans and photographs the survey location and the surrounding line facilities during the flight. The collected point cloud data should contain the three-dimensional coordinates, shape, size and other information of the positioning point, and be stored in real time on the drone.

[0110] After the survey mission is completed, the drone will transmit the collected point cloud data back to the ground processing platform through a wireless communication link to ensure the integrity and accuracy of the data transmission for subsequent point cloud data processing and high-precision map generation.

[0111] S203: The processing platform receives the point cloud data of the surveyed and measured locations returned by the drone, matches the point cloud data of the surveyed and measured locations with the model data of the high-voltage transmission line, and generates a high-precision map of the high-voltage transmission line.

[0112] Among them, the processing platform obtains the model data of the high-voltage transmission line and obtains the initial point cloud data by discretizing the model data. After receiving the point cloud data of the surveyed measurement points, the processing platform performs sparse and clustering processing on the point cloud data, and then matches the point cloud data of the surveyed measurement points with the initial point cloud data of the high-voltage transmission line for the same-name points to obtain the data difference between the two same-name points. The initial point cloud data of the high-voltage transmission line is corrected using the data difference of the same-name points to generate the corrected point cloud data, and the corrected point cloud data is vectorized to obtain a high-precision map.

[0113] Optionally, the ground processing platform receives the point cloud data sent back by the drone and performs pre-processing operations such as denoising, filtering, and registration to improve data quality. Load the model data of the high-voltage transmission line, including the three-dimensional model of the line, the tower model, the insulator model, etc.

[0114] Using the point cloud processing algorithm, the pre-processed point cloud data is matched and aligned with the model data to ensure that the point cloud data and the model data are consistent in space and form.

[0115] Based on the matched and aligned point cloud data and model data, a high-precision map of the high-voltage transmission line is constructed. The map can contain the spatial location information, morphological information, size information and detailed information of auxiliary facilities of the line, such as tower number, insulator type, conductor specification, etc. The generated high-precision map is optimized, such as smoothing lines, correcting errors, adding annotations, etc., to improve the accuracy and readability of the map.

[0116] The generated high-precision map is verified on the spot to ensure its accuracy and practicality. According to the actual changes of the route and the patrol requirements, the high-precision map is updated and maintained regularly to maintain the timeliness and accuracy of the map.

[0117] S204: The processing platform obtains the position and posture of the patrol object from the high-precision map, and determines the position and posture of the patrol point according to the position and posture of the patrol object, the high-precision map of the area where the patrol object is located, the camera parameters and the model data of the UAV.

[0118] S205. The processing platform generates a patrol path for the UAV according to the positions and postures of the patrol points, and sends the patrol path and high-precision map of the UAV to the UAV.

[0119] S206: The processing platform uses the target detection model to perform defect recognition on the inspection image to obtain the inspection result.

[0120] Among them, S204 to S206 have been described in detail in the above embodiment and will not be repeated here.

[0121] In the above technical scheme, since the high-voltage transmission line is mainly composed of conductors, ground wires and towers, all of which are long strips with relatively small reflection surfaces, relatively few reflected waves are received when surveying with drones, which is not conducive to building map data. By using the existing model data of the high-voltage transmission line, some positioning points on the conductors, ground wires and towers are selected for survey, and the measured three-dimensional data of the surveyed positioning points are matched with the surface model data of the high-voltage transmission line to obtain a high-precision map of the high-voltage transmission line, the amount of survey data of the drone can be reduced.

[0122] Specifically, when facing targets such as high-voltage transmission lines that are long and have small reflective surfaces, traditional drone survey methods often require the collection of a large amount of point cloud data to ensure the accuracy and integrity of the map. However, this method not only increases the flight time and data processing burden of the drone, but may also result in low data quality due to the small number of reflected waves. The above technical solution significantly reduces the amount of point cloud data that needs to be collected by pre-generating survey and determination points of high-voltage transmission lines and only performing drone surveys on these locations. In this way, drones can perform survey tasks more efficiently while ensuring the accuracy and reliability of the data.

[0123] Using the existing model data of high-voltage transmission lines as a reference, the measured three-dimensional data of the positioning points obtained by drone survey can be accurately matched with the model data. This matching process not only takes into account the spatial position of the positioning points, but also the overall shape and direction of the line, thereby ensuring the accuracy and consistency of the high-precision map. The high-precision map generated in this way can more accurately reflect the actual situation of the high-voltage transmission line, providing reliable basic data for subsequent line maintenance, defect identification and drone inspections.

[0124] After the surveying and measuring points are generated, the optimal surveying path can be planned for the drone based on the location and distribution of these points. In this way, the drone can cover all areas that need to be surveyed more efficiently during the flight, avoiding unnecessary repeated flights and omissions. The optimized surveying path not only improves the surveying efficiency of the drone, but also reduces the energy consumption and risks during the flight, ensuring the safe and smooth progress of the survey mission.

[0125] By regularly updating survey locations and model data, we can promptly reflect changes in high-voltage transmission lines and maintain the timeliness and accuracy of high-precision maps. This dynamic update mechanism enables high-precision maps to better serve actual needs such as line maintenance, troubleshooting, and emergency response, and improve the operating efficiency and safety of the power system.

[0126] In a possible implementation, S104, determining the position and posture of the patrol point according to the position and posture of the patrol object, the high-precision map of the area where the patrol object is located, the model data of the drone, and the camera parameters, specifically includes: S301. For each patrol object, generate an observation point of the patrol object according to the position and posture of the patrol object, a high-precision map of the area where the patrol object is located, the model data of the UAV, and the camera parameters.

[0127] For each patrol object, the computational model is used to process the position and posture of the patrol object, the high-precision map of the area where the patrol object is located, the model data of the drone, and the camera parameters to generate the observation point of the patrol object. The observation point includes the position and posture, and the observation point is used to capture a patrol image of the patrol object.

[0128] Optionally, first, the precise position and posture information of the patrol object (such as conductor, ground wire, pole tower, insulator, etc.) is obtained from the high-precision map of the high-voltage transmission line. This information usually includes the three-dimensional coordinates, direction vector and possible size parameters of the patrol object.

[0129] Load the model data of the drone, including the size, shape, flight performance parameters, etc. This data is crucial to determine the range of patrol objects that the drone can observe at a specific location. Obtain detailed parameters of the drone's onboard camera, such as focal length, field of view, resolution, etc.

[0130] Based on the position and attitude of the patrol object, the model data of the drone, and the camera parameters, the position and attitude of the drone are calculated to best observe the patrol object. These positions and attitude combinations constitute observation points, and each observation point contains a specific position (3D coordinates) and attitude (direction vector).

[0131] S302, determining whether the multiple observation points meet the fusion requirements based on the positions of the multiple observation points and the camera optical axis coordinates at the multiple observation points, and if so, fusing the observation points to obtain patrol points, and using patrol objects corresponding to the observation points as patrol objects of the patrol points.

[0132] Among them, if multiple observation points are relatively close, a fusion algorithm can be used to fuse the multiple observation points into one patrol point. The patrol object corresponding to the observation point is used as the patrol object of the patrol point. If the multiple observation points are far apart, no fusion is performed. More specifically, it is determined whether the multiple observation points meet the fusion condition based on the positions of the multiple observation points and the camera optical axis coordinates under the multiple observation points, that is, whether the multiple observation points are relatively close.

[0133] Optionally, calculate the spatial distance between all observation points to evaluate their proximity. For each observation point, calculate the coordinates of the camera optical axis, that is, the direction of the center line of the camera when shooting. Analyze these optical axis coordinates to determine whether they point to the same or similar patrol object areas. Based on the distance between observation points and the similarity of camera optical axis coordinates, determine whether multiple observation points can be fused into one patrol point. Fusion conditions may include: the observation points are close enough, the optical axis coordinates point to similar or overlapping patrol object areas, and the fused patrol points can provide sufficient coverage and clarity.

[0134] For observation points that meet the fusion conditions, the fusion operation is performed. The patrol objects corresponding to each observation point before fusion are marked as patrol objects of the patrol point after fusion. In this way, the patrol point is associated with a specific patrol object such as a conductor, ground wire, pole tower or insulator. The fused patrol point (including its position, posture and patrol object) is added to the patrol mission plan. This information will be used to guide the actual flight and patrol operation of the drone.

[0135] For example, there are two patrol objects. For the first patrol object, three observation points of the patrol object are generated, and each observation point is used to capture a patrol image of the first patrol object.

[0136] For the second patrol object, three observation points of the patrol object are generated. Each observation point is used to capture a patrol image of the second patrol object.

[0137] In this way, there are 6 observation points for 2 patrol objects. It is directly determined whether the 6 observation points meet the fusion conditions. If they meet the conditions, the 6 observation points are fused into one and 1 patrol point is output. The patrol point is used to patrol the 2 patrol objects. If they do not meet the conditions, no fusion is performed and the 6 observation points are directly output as 6 patrol points in a one-to-one correspondence. The 6 observation points can also be divided into two groups to determine whether the observation points in each group can be fused. If they can be fused, they are fused. If they cannot be fused, the observation points are directly used as patrol points.

[0138] In the above technical solution, after calculating the observation point of the patrol object, multiple observation points are fused according to the observation point of the patrol object and the camera optical axis coordinates at each observation point. In this way, multiple patrol objects can be patrolled at one patrol point, thereby improving the patrol efficiency of the drone.

[0139] Specifically, observation points are generated for each patrol object by combining the precise position information of the patrol object, the high-precision map of the area, and the model data and camera parameters of the drone. These observation points not only contain the position and posture that the drone should be in, but also ensure that the camera can capture the details of the patrol object at the best angle. The generation and fusion process of observation points is actually an optimization of the drone patrol path. By fusing multiple observation points into one patrol point, the number of times the drone needs to stop and shoot is reduced, thereby shortening the patrol path and improving patrol efficiency. The fusion of observation points enables one patrol point to cover multiple patrol objects at the same time. This means that in the same flight time, the drone can inspect more lines and equipment, improving the coverage of the patrol.

[0140] Since the patrol points are obtained by fusing multiple observation points, the data between them has higher consistency, which helps to reduce errors in subsequent data processing and analysis and improve the accuracy and reliability of the data.

[0141] The optimization of patrol paths and the improvement of patrol efficiency directly reduce the energy consumption and operating costs of drones. Less flight time and fewer shots mean less battery consumption and less maintenance requirements.

[0142] By determining the position and posture of the patrol points and fusing the observation points, the flight trajectory of the drone can be better planned to avoid collisions with line equipment or other obstacles, thereby improving the safety of the patrol.

[0143] In a possible implementation, S302, determining whether the multiple observation points meet the fusion requirements according to the positions of the multiple observation points and the camera optical axis coordinates at the multiple observation points, and if so, fusing the observation points to obtain patrol points, and using patrol objects corresponding to the observation points as patrol objects of the patrol points, specifically includes: Step a: group multiple observation points to obtain an observation point group.

[0144] Optionally, the system will perform preliminary analysis and grouping of the collected multiple observation points. The grouping principle can be based on the geographical proximity of the observation points, the similarity of the observation angles, or the division according to the specific patrol task requirements. For example, observation points with adjacent geographical locations and similar observation directions can be grouped into the same group to facilitate subsequent fusion processing.

[0145] Here, taking 6 observation points as an example, observation point 1, observation point 2, and observation point 3 are grouped into one observation point group, and observation point 4, observation point 5, and observation point 6 are grouped into another observation point group.

[0146] Step b: for each group of observation points, a core observation point is selected from the observation point group, and the spatial distance between each observation point and the core observation point, as well as the angle between the optical axis of each observation point and the core observation point are calculated.

[0147] Optionally, for each group of observation points, the system will select one as the core observation point. The selection of the core observation point can be based on a variety of factors, such as the central location, the most representative observation angle, etc. The system will then calculate the spatial distance between each observation point in the group and the core observation point, as well as the angle between their respective camera optical axes and the optical axis of the core observation point camera. The calculation of spatial distance can use common methods such as Euclidean distance and Manhattan distance, while the calculation of the optical axis angle involves vector operations, which requires the use of the vector angle formula in three-dimensional space.

[0148] Among them, for the first group of observation points, observation point 1 is selected as the core observation point, the spatial distance between observation point 1 and observation point 2 is calculated, and the optical axis angle between the camera optical axis of observation point 1 and the camera optical axis of observation point 2 is calculated. The camera optical axis of the observation point refers to the camera optical axis posture when the drone is located at the observation point and is in the posture under the observation point.

[0149] For the second group of observation points, observation point 4 is selected as the core observation point, the spatial distance between observation point 4 and observation point 5 is calculated, and the optical axis angle between the camera optical axis of observation point 4 and the camera optical axis of observation point 5 is calculated.

[0150] The calculation process of observation point 3 and observation point 6 is similar and will not be repeated here.

[0151] Step c: obtaining a target observation point that meets the fusion requirements from multiple observation points, wherein the target observation point is an observation point whose spatial distance is less than a preset distance threshold and whose optical axis angle is less than a preset angle threshold.

[0152] Optionally, based on the preset distance threshold and angle threshold, the system will screen out observation points whose spatial distance is less than the distance threshold and whose optical axis angle is less than the angle threshold as target observation points that meet the fusion requirements. The setting of these two thresholds can be determined based on actual patrol requirements and camera performance to ensure that the patrol points after fusion have sufficient accuracy and reliability.

[0153] Among them, in the first group of observation points, observation point 1 is the core observation point, so only observation point 2 and observation point 3 need to be judged. For observation point 2, it is determined that the spatial distance between observation point 2 and observation point 1 is less than the preset distance threshold, and the optical axis angle between observation point 2 and observation point 1 is less than the preset angle threshold, then observation point 2 is the target observation point.

[0154] For observation point 3, it is determined that the spatial distance between observation point 3 and observation point 1 is less than the preset distance threshold, and the optical axis angle between observation point 3 and observation point 1 is less than the preset angle threshold. If so, observation point 3 is the target observation point.

[0155] In the second group of observation points, observation point 4 is the core observation point, so only observation points 5 and 6 need to be judged. For observation point 5, it is determined that the spatial distance between observation point 5 and observation point 4 is less than the preset distance threshold, but the optical axis angle between observation point 5 and observation point 4 is greater than the preset angle threshold, then observation point 5 is not the target observation point.

[0156] For observation point 6, if it is determined that the spatial distance between observation point 6 and observation point 4 is greater than a preset distance threshold, and the optical axis angle between observation point 6 and observation point 4 is greater than a preset angle threshold, then observation point 6 is not a target observation point.

[0157] Step d: If the number of target observation points is greater than or equal to half of the number of observation points in the observation point group, calculate the average position and average posture of multiple target observation points as the posture of the patrol point, and take the patrol objects corresponding to the multiple target observation points as the patrol objects corresponding to the patrol points.

[0158] Optionally, if the number of target observation points that meet the fusion requirements in a group of observation points is greater than or equal to half of the number of observation points in the group, the system will fuse these target observation points. The fusion method can be to calculate their average position and average posture, so as to obtain a more accurate and stable patrol point position and posture information. At the same time, the patrol objects corresponding to these target observation points will also be integrated into the patrol objects corresponding to the patrol points.

[0159] Among them, the number of target observation points in the first group of observation points is 2, which is greater than half of the number of observation points in the first group. Therefore, the position average and posture average of observation points 1, 2, and 3 are calculated, and the position average is taken as the position of the patrol point, and the posture average is taken as the posture of the patrol point.

[0160] Among them, the number of target observation points in the second group of observation points is 0, which is less than half of the number of observation points in the second group. Therefore, the second group of observation points is not fused.

[0161] Step e: if the number of patrol points obtained is greater than or equal to half of the number of observation point groups, stop the loop; if the number of patrol points is less than half of the number of observation point groups, return to step a.

[0162] Optionally, after completing a round of fusion, the system will count the number of patrol points that have been obtained. If the number of patrol points is greater than or equal to half of the number of initial observation point groups, it means that a sufficient number of patrol points have been obtained and the loop can be stopped. Otherwise, it means that further fusion is needed, and the system will return to step a to perform new grouping and fusion processing on the remaining observation points until the stop condition is met.

[0163] Among them, if one of the two groups of observation points completes fusion and the other group does not, the cycle stops, and the point after the fusion of observation points 1 to 3 is used as a patrol point, and the patrol point shoots the patrol objects observed by observation points 1 to 3. Observation point 4 is used as a patrol point, and the patrol point shoots the patrol object observed by observation point 4. Observation point 5 is used as a patrol point, and the patrol point shoots the patrol object observed by observation point 5. Observation point 6 is used as a patrol point, and the patrol point shoots the patrol object observed by observation point 6.

[0164] In the above technical scheme, the spatial distance between the two observation points and the angle between the camera optical axes corresponding to the two observation points are used as fusion parameters. If the spatial distance is less than the preset distance threshold and the angle between the camera optical axes is less than the preset angle threshold, it means that the two observation points are close and the camera postures at the observation points are also similar. The average value of the two observation points is used as the posture of the patrol point. This ensures that when observing two patrol objects at the patrol point, the patrol image can be clearly captured.

[0165] Specifically, the above scheme first groups multiple observation points through step a to form observation point groups, which helps to classify observation points with similar geographical locations and observation angles together, and facilitates subsequent fusion processing. In step b, a core observation point is selected from each group of observation points, and the spatial distance and optical axis angle between other observation points and the core observation point are calculated. This step provides key parameters for fusion judgment and ensures the accuracy and effectiveness of the fusion process. In step c, by setting preset thresholds for spatial distance and optical axis angle, the target observation points that meet the fusion requirements are accurately screened. This dual parameter control mechanism ensures that only those observation points that are very close in spatial position and observation angle will be selected for fusion, thereby improving the accuracy and reliability of the patrol points. In step d, when the number of target observation points reaches or exceeds half of the number of observation point groups, the position and posture of the patrol points are determined by calculating the position average and posture average of these target observation points. This method not only integrates the information of multiple observation points, but also optimizes the position selection of patrol points, so that the patrol points can more accurately reflect the actual patrol needs. At the same time, the patrol objects corresponding to multiple target observation points are integrated into patrol objects of patrol points, which improves the coverage and efficiency of patrol. Step e introduces a dynamic loop mechanism, which determines whether to continue the loop according to the ratio of the number of patrol points obtained to the number of observation point groups. This mechanism ensures that the generation of patrol points is neither too redundant nor too sparse, thereby controlling the computational cost and patrol time while ensuring the patrol quality. When the number of patrol points meets the requirements, the loop is stopped in time to avoid unnecessary computational waste; when the number of patrol points is insufficient, it returns to step a for regrouping and fusion processing until the requirements are met. Patrol points are obtained by fusing multiple observation points, and the position and posture of the patrol points are optimized, so that patrol images can be clearly captured when observing two patrol objects at the patrol points. This significantly improves the quality and efficiency of patrols and reduces the risk of patrol failure or image blur caused by improper selection of observation points or poor camera posture. The fused patrol point data are more consistent and reliable because they are calculated based on the position and posture information of multiple observation points. The consistency and reliability of this data are helpful for subsequent data processing and analysis, and improve the accuracy and credibility of patrol results.

[0166] In a possible implementation, S301, for each patrol object, generating an observation point of the patrol object according to the posture of the patrol object, a high-precision map of the area where the patrol object is located, the model data of the drone, and the camera parameters, specifically includes: S401, the processing platform obtains the patrol surface coordinates of the patrol object, and generates a plurality of candidate points according to the patrol surface coordinates of the patrol object and camera parameters, wherein the candidate points include positions and postures.

[0167] The processing platform obtains the patrol surface of the patrol object by extracting the patrol requirement file, and extracts the patrol surface coordinates of the patrol object from the high-precision map. The patrol surface coordinates and camera parameters of the patrol object are processed using the geometric relationship between the patrol surface and the camera to obtain multiple candidate points. The geometric relationship between the patrol surface and the camera includes the perpendicularity between the patrol surface and the camera optical axis. This ensures that the camera can clearly capture the patrol image of the patrol object.

[0168] Optionally, first, the specific location of the patrol object (such as the tower, insulator, wire, etc. of the high-voltage transmission line) is identified and determined through the drone's onboard sensors (such as GPS, laser radar, camera, etc.) or pre-stored information. Based on the identification results, the precise coordinates of the key parts of the patrol object (such as the tower foundation, insulator hanging point, etc.) in three-dimensional space are calculated, and these coordinates constitute the patrol surface coordinates. The coordinate calculation may need to consider the conversion between the geographic coordinate system and the drone's own coordinate system.

[0169] Based on the patrol surface coordinates, combined with camera parameters (e.g. focal length, viewing angle, resolution, etc.), multiple possible shooting positions (i.e. candidate points) are calculated. Each candidate point contains specific position coordinates (longitude, latitude, altitude) and posture (e.g. orientation, pitch angle). Optionally, these candidate points should ensure that the key parts of the patrol object can be fully and clearly photographed.

[0170] S402: For each candidate point, the processing platform determines whether the drone will collide at the candidate point based on the model data of the drone and the high-precision map of the area where the patrol object is located. If no collision occurs, the candidate point is retained.

[0171] Among them, at each candidate point, the space occupied by the drone when it is located at the candidate point and its flight posture is the posture of the candidate point is determined based on the model data of the drone. According to the high-precision map of the area where the patrol object is located and the occupied space, it is determined whether the drone will collide at the candidate point. If no collision occurs, the candidate point is retained. If it is determined that a collision occurs, the candidate point is removed.

[0172] Optionally, use the 3D model data of the drone, including its size, shape, propeller position, etc., to accurately simulate the drone's position in the air. Use a high-precision map of the area where the patrol object is located, which should contain the precise location and height information of obstacles such as terrain, buildings, and trees. For each candidate point, place the drone's model data at that location and check whether it overlaps with obstacles in the high-precision map or is close enough to collide. If it is judged that a collision will occur, exclude the candidate point.

[0173] For example, unoccupied space is extracted from the high-precision map of the area where the patrol object is located, and the unoccupied space is compared with the space occupied by the drone to determine whether the drone will collide at the candidate point. If no collision occurs, the candidate point is retained. If a collision is determined, the candidate point is removed.

[0174] S403: The processing platform selects a candidate point farthest from an obstacle from a plurality of reserved candidate points as an observation point, wherein the obstacle is an object in the area where the patrol object is located.

[0175] Among them, the space occupied by the drone when it is at the alternative point and its flight posture is the posture of the alternative point is obtained, the unoccupied space is extracted from the high-precision map of the area where the patrol object is located, and the shortest distance between the boundary of the space occupied by the drone and the boundary of the unoccupied space is calculated as the distance between the drone and the obstacle when it is at the alternative point.

[0176] From the distance between the drone and the obstacle at each reserved candidate point, the candidate point corresponding to the longest distance is selected as the observation point. This ensures that the drone will not collide with the patrol object when shooting at the observation point.

[0177] Optionally, for all the retained (i.e. non-collision) candidate points, calculate their respective distances to the nearest obstacle. Select the candidate point farthest from the obstacle as the final observation point. This selection is intended to maximize the safety distance of the drone and reduce the risk of accidental collisions. Optionally, in some special cases, such as requiring a clearer shooting angle or specific lighting conditions, it may be necessary to consider other factors to fine-tune the selection of observation points.

[0178] In the above technical solution, multiple candidate points are generated based on the patrol surface coordinates and camera parameters of the patrol object. For each candidate point, the candidate point that collides with the surrounding objects is first deleted, and the candidate point farthest from the obstacle is selected from the retained candidate points as the observation point. This ensures that the drone can clearly photograph the patrol object without colliding with obstacles.

[0179] Specifically, first, the patrol surface coordinates of the patrol object are obtained and multiple candidate points are generated in combination with the camera parameters. This step ensures that the generation of candidate points takes into account both the actual position of the patrol object and the technical requirements of camera shooting. The candidate points contain position and attitude information, providing the drone with multiple possible shooting angles and positions, thereby greatly improving the flexibility and accuracy of shooting patrol objects.

[0180] Secondly, for each candidate point, collision detection is performed using the model data of the drone and the high-precision map of the area where the patrol object is located, effectively eliminating those candidate points that are not feasible due to collisions with surrounding objects. This step not only ensures the safety of the drone during flight, but also avoids risks such as equipment damage and data loss caused by collisions, thereby improving the reliability and stability of the patrol mission.

[0181] Finally, the candidate point farthest from the obstacle is selected from multiple reserved candidate points as the observation point. This decision-making strategy further optimizes the shooting position of the drone. By maximizing the distance between the drone and the obstacle, it not only reduces the possibility of the drone being disturbed by obstacles during flight, but also ensures that the drone can clearly capture the patrol object in a safer and more stable environment. This not only helps to improve the quality of patrol data, but also provides a more accurate and reliable basis for subsequent fault detection and line maintenance.

[0182] In a possible implementation, S401, generating a plurality of candidate points according to the patrol surface coordinates and camera parameters of the patrol object, specifically includes: S501. Taking the normal angle of the patrol surface as the optical axis angle, obtaining the projection of the patrol object on the patrol surface.

[0183] The normal of the patrol surface is obtained by calculating the coordinates of the patrol surface, and the angle of the normal of the patrol surface is used as the angle of the optical axis. The patrol object is projected onto the patrol surface to obtain the projection on the patrol surface.

[0184] Optionally, first, according to the geometric shape and position information of the patrol object, the normal direction of the patrol surface (usually the surface of the patrol object or the plane where the key parts are located) is calculated. The normal is a straight line perpendicular to the patrol surface. Next, determine the angle between the camera optical axis (that is, the direction in which the camera lens points) and the normal of the patrol surface. This angle is called the optical axis angle. This angle can be calculated based on the installation angle of the camera and the orientation of the patrol surface. Finally, the patrol object is projected onto the patrol surface according to the optical axis angle to obtain a two-dimensional projection image. This projection image will be used for the subsequent selection of candidate points.

[0185] S502, selecting multiple points in the projection center area. For each point, according to the coordinates of a point in the projection center area and the angle of the optical axis, obtain the coordinates of the straight line where the camera optical axis is located.

[0186] The center area is circled in the projection so that the area of ​​the center area is smaller than the preset area. Multiple points are randomly selected in the center area. The coordinates of a straight line can be determined by an angle and the coordinates of a point. Based on the coordinates of a point and the angle of the optical axis, the coordinates of the straight line where the optical axis of the camera is located are obtained.

[0187] Optionally, on the projected image, a central area is defined according to the size and shape of the patrol object and the shooting range of the camera. This area can be the geometric center of the projected image or a specific area adjusted according to actual needs. In the projection center area, multiple points are selected according to certain rules (such as uniform distribution, priority of key parts, etc.). These points will serve as the basis for subsequent calculation of alternative points. And, for these determined points, they are processed in turn. According to the selected points, the optical axis angle, and the installation parameters of the camera (such as the installation position and direction of the camera on the drone), the three-dimensional coordinates of the straight line where the camera optical axis is located are calculated. This straight line will pass through the selected point and form an optical axis angle with the normal of the patrol surface.

[0188] Figure 4 A schematic diagram of generating the optical axis pose of a camera provided in some embodiments of the present invention is shown in FIG. Figure 4 As shown, the center area B is circled on the projection A so that the area of ​​the center area is less than 2 / 3 of the projection area. Multiple points are randomly selected in the projection center area B. The coordinates of the line where the optical axis of a camera is located can be obtained based on the coordinates of a point and the angle between the optical axes. By traversing each point, the coordinates of the lines where the optical axes of multiple cameras are located can be obtained.

[0189] S503 . For each camera optical axis line coordinate, calculate and obtain a candidate point according to the camera optical axis line coordinate, camera installation parameters and focal length.

[0190] The straight line where the camera optical axis is located cannot locate the camera, and the camera can be located at any point on the straight line. Therefore, for the coordinates of the straight line where the camera optical axis is located, the camera pose is calculated according to the coordinates of the straight line where the camera optical axis is located and the focal length, and the pose of a drone is calculated according to the camera pose and camera installation parameters, and the pose of the drone is used as the candidate point.

[0191] Optionally, for each straight line where the camera optical axis is located, combined with the focal length of the camera (i.e., the distance from the camera lens to the imaging plane), it is possible to calculate the key parts of the patrol object that can be clearly photographed when the camera is located at a certain position. This position is the position of the candidate point. Based on the coordinates of the straight line where the camera optical axis is located, the camera installation parameters (including the specific position and direction of the camera on the drone), and the focal length, the three-dimensional coordinates (including longitude, latitude, altitude) and attitude (such as orientation, pitch angle) of the candidate point can be accurately calculated through geometric calculation or simulation software.

[0192] In the above technical solution, the normal angle of the patrol surface coordinates is used as the optical axis angle, so that the camera optical axis passes through the projection center area on the patrol surface coordinates, ensuring that the patrol object is located in the middle of the patrol image and can be clearly photographed. By generating multiple camera postures, one can select one that is farthest from obstacles from multiple postures to ensure that no collision occurs when the drone is shooting.

[0193] Specifically, first, the normal angle of the patrol surface is determined as the optical axis angle. The purpose of this step is to ensure that the camera optical axis can pass through the projection center area of ​​the patrol object on the patrol surface, so that the patrol object is in the center position in the final patrol image and can be clearly photographed. To achieve this, the normal direction of the patrol surface can be calculated first, and the optical axis angle can be determined according to the installation angle of the camera and the orientation of the patrol surface.

[0194] Next, obtain the projection of the patrol object on the patrol surface, and select multiple points located in the center area of ​​the projection. The selection of these points should be based on the actual shape and size of the patrol object and the shooting range of the camera, ensuring that the selected points can cover the key parts of the patrol object and are located in the center area of ​​the projection. This ensures that no matter which candidate point the camera shoots from, the patrol object can be completely included in the image.

[0195] Then, for each point in the projection center area, the coordinates of the straight line where the camera optical axis is located are calculated based on the angle between the point and the previously determined optical axis. This step can be done by using 3D geometric coordinate transformation to convert the point coordinates and the angle between the optical axis and the straight line coordinates of the camera optical axis in 3D space. This straight line will serve as the basis for the subsequent calculation of the position of the candidate point.

[0196] Finally, for the coordinates of the straight line where the optical axis of each camera is located, combined with the camera's installation parameters (including the specific position and orientation of the camera on the drone) and focal length, the position and posture of an alternative point are determined through geometric calculation or simulation software. This alternative point is the position and orientation that the camera should be in when shooting the patrol object. By repeating this process, multiple alternative points can be generated, which correspond to different shooting positions and orientations.

[0197] In a possible implementation, if the patrol object is a pin, the patrol surface is a plane passing through the pin axis and two end surfaces perpendicular to the pin axis.

[0198] If the inspected object is an insulator, the inspected surface is a plane passing through the axis of the insulator and two end surfaces perpendicular to the axis of the insulator.

[0199] Optionally, first, identify and determine the axis of the pin, which is a straight line that passes through the center of the pin and is consistent with its length. Then, construct a plane that contains this axis, that is, this plane is coplanar with the pin axis. This plane is mainly used to capture the overall shape of the pin and the characteristics in the axis direction. Next, at both ends of the pin, construct two planes perpendicular to the pin axis. These two end faces are used to capture details of the pin end, such as wear, corrosion or deformation. After determining the patrol surface, combine the patrol surface coordinates of the pin and the camera parameters to generate multiple candidate points. Pay special attention to the axis and end face features of the pin. When generating alternative points, ensure that the camera optical axis can cover these key parts for clear shooting.

[0200] Similarly, first identify and determine the axis of the insulator, which is a straight line that passes through the center of the insulator and is consistent with its length. Then, construct a plane that contains this axis to capture the overall shape of the insulator and the features in the axis direction, such as cracks, dirt, etc. At both ends of the insulator, construct two planes perpendicular to the axis of the insulator. These two end faces are used to capture the details of the insulator end, such as the condition of the connection, the integrity of the shed, etc. After determining the inspection surface, also combine the inspection surface coordinates of the insulator and the camera parameters to generate multiple candidate points. Pay special attention to the axis and end face features of the insulator. When generating alternative points, ensure that the camera optical axis can cover these key parts, especially the cracks, dirt, and connection details of the insulator, so as to perform accurate fault detection and status assessment.

[0201] For each patrol object, two mutually perpendicular surfaces passing through the axis can be selected as patrol surfaces. Two end surfaces perpendicular to the axis can be selected. This makes it easy to observe whether the patrol object has defects from multiple directions.

[0202] In the above technical solution, by selecting a plane passing through the axis and a plane perpendicular to the axis, images of important areas of the inspected object can be clearly captured, defects of the inspected object can be accurately identified, and missed inspections due to insufficient observation angles can be avoided.

[0203] Specifically, by selecting the plane passing through the pin axis as one of the inspection surfaces, the camera is able to capture the overall shape of the pin and the features in the axial direction, such as length, diameter, and possible signs of wear or corrosion. At the same time, the two end faces perpendicular to the pin axis are used as the other two inspection surfaces, so that the camera can capture the end details of the pin at close range and with high resolution, including the connection, wear or cracks, etc. This inspection surface selection strategy ensures that the camera can fully observe the pin from multiple angles, thereby greatly improving the accuracy of defect identification. Clear images can be obtained for both the axial direction and the end details of the pin, avoiding missed inspections due to insufficient observation angles. By accurately selecting the inspection surface, the drone's hovering time in the air is reduced, improving the inspection efficiency. The camera can directly aim at the key areas of the pin for shooting without excessive position adjustments or repeated shooting.

[0204] If the inspection object is an insulator, the plane passing through the axis of the insulator is selected as the inspection surface, so that the camera can capture the overall shape of the insulator and important features in the axis direction, such as cracks, dirt distribution, and the integrity of the shed. At the same time, the two end faces perpendicular to the axis of the insulator are used to shoot the end details of the insulator, such as the condition of the connection, corrosion or damage of metal parts, etc. This inspection surface selection strategy ensures that every key part of the insulator can get a clear image, thereby improving the accuracy of defect detection. Whether it is surface cracks or internal damage of the insulator, it can be accurately identified through multi-angle shooting. Through comprehensive and multi-angle observation, the risk of missed detection due to insufficient observation angles is greatly reduced. This is crucial for the safe operation of high-voltage transmission lines, because any defects in the insulator may cause serious failures or accidents.

[0205] In a possible implementation, S107, the processing platform uses the target detection model to perform defect recognition on the inspection image to obtain the inspection result, specifically including: S601: The processing platform receives an inspection image and an inspection object corresponding to the inspection image.

[0206] Optionally, the processing platform first receives the patrol images taken and transmitted by the drone or other patrol equipment, which contain various patrol objects on the high-voltage transmission line, such as pins, insulators, etc. At the same time, the processing platform also receives patrol object information corresponding to these patrol images, including the type and location of the patrol object. This information is usually obtained through the positioning system and image recognition technology of the patrol equipment.

[0207] Among them, component defects in high-voltage transmission lines usually include: missing pins, pin withdrawal, pin opening, loose nuts, missing nuts, missing anti-vibration hammer nails, self-explosion of glass insulators, dirty insulators, loose conductors, tilted equalizing rings, flooding at the bottom of the tower, damaged tower base, vegetation coverage, unclear signboard images and text, damaged signboards and other defects.

[0208] From the above defects, we can see that there are many types of components in high-voltage transmission lines, mainly including pins, nuts, insulators, wires, poles, etc. Each component has multiple defects. In order to adapt to the identification of multiple types of defects under multiple component types, a target detection model is trained for each type of component, and the target detection model can identify all defects of the component.

[0209] The processing platform stores a model mapping relationship table, which is the mapping relationship between the target detection model and the inspection object category. For example, inspection object categories such as pins, screws and nuts. For inspection objects of the pin type, train a target detection model to identify defects such as pin missing, pin exit, and pin opening. For inspection objects of the screw and nut type, train a target detection model to identify defects such as nut loosening and nut missing.

[0210] Since each patrol point carries patrol object information, the drone can carry the patrol object information when taking patrol images, and the drone transmits the patrol images and patrol objects back to the background server.

[0211] S602, the processing platform obtains the target detection model corresponding to the patrol object according to the patrol object corresponding to the patrol image and the model mapping relationship table; the model mapping relationship table is the correspondence between the patrol object category and the target detection model.

[0212] Among them, the processing platform obtains the category information of the patrol object, uses the patrol object category to query the model mapping relationship table, and calls the target detection model corresponding to the patrol object.

[0213] Optionally, the processing platform maintains a model mapping relationship table internally, which records the correspondence between the patrol object category and the target detection model. For example, for pins, a model specially trained for detecting pin defects may be used; for insulators, another model specially trained for detecting insulator defects is used. Based on the received patrol object information, the processing platform queries and obtains the target detection model corresponding to the patrol object in the model mapping relationship table.

[0214] S603: The background platform uses the target detection model corresponding to the inspected object to identify the inspected image and obtain the defect type of the inspected object.

[0215] Among them, by running the target detection model, the inspection image is input into the target detection model as input data, and the target detection model outputs the defect type of the inspected object.

[0216] Optionally, the processing platform loads the acquired target detection model and prepares for image recognition. Before the inspection image is input into the model, some preprocessing operations may be required, such as adjusting the image size, normalization, denoising, etc., to improve the recognition effect of the model. The preprocessed inspection image is input into the target detection model, and the model extracts and classifies the image features to identify the defect type in the inspection object. After the model recognition is completed, the recognition results are output, including information such as the defect type and location. These results will be used for subsequent defect processing and analysis.

[0217] In the above technical solution, different target detection models are trained for different patrol objects. After the drone takes patrol images at each patrol point, the patrol images and patrol object identifiers corresponding to the patrol points are sent to the background platform. In this way, the background platform can select the corresponding target detection model to identify the patrol images, thereby improving the recognition accuracy.

[0218] Specifically, since the characteristics and defect forms of each patrol object (such as pins, insulators, etc.) are different, the use of a target detection model specially trained for it can more accurately capture its subtle features and defect patterns. Compared with general models, this customized model has significantly improved recognition accuracy and robustness. By selecting a model that matches the target patrol object, the backend platform can more accurately identify real defects, while reducing the false alarm and missed alarm rates caused by model mismatch, and improving the overall efficiency and quality of patrol work.

[0219] The backend platform dynamically selects and loads the corresponding target detection model based on the received patrol object identification. This mechanism avoids the waste of resources and inefficiency that may result from using a single model to process all patrol objects. In addition, in a high-performance computing environment, the backend platform can load and run multiple target detection models at the same time, realizing parallel processing of multiple patrol images, thereby improving overall processing efficiency.

[0220] When it is necessary to introduce new patrol objects or update the recognition algorithm of existing patrol objects, it is only necessary to add or update the corresponding entries in the model mapping relationship table and upload the new target detection model. This design makes the expansion and update of the system simple and fast. In addition, with the continuous development of high-voltage transmission line inspection technology, patrol objects and requirements may change. By maintaining a flexible model mapping relationship table, the background platform can quickly adapt to these changes and ensure the accuracy and timeliness of defect identification.

[0221] In addition, the target detection model of the background platform can realize automatic identification and reporting of defects in the inspection images. This greatly reduces the workload of operation and maintenance personnel and improves work efficiency and accuracy. The background platform can also count and analyze the recognition results to provide data-driven decision support for operation and maintenance personnel. For example, by analyzing the type, distribution and trend of defects, more scientific and reasonable inspection plans and maintenance strategies can be formulated.

[0222] Some embodiments of the present invention provide a method for processing high-voltage transmission line inspection data based on an unmanned aerial vehicle, the method further comprising: S701. The processing platform generates a monitoring instruction when erecting a high-voltage transmission line, and sends the monitoring instruction to the drone, so that the drone takes a monitoring image of the erection site.

[0223] Among them, the processing platform can also monitor the high-voltage transmission lines during the installation process, obtain the area where the high-voltage transmission lines are being installed, generate monitoring instructions according to the installation area, send the monitoring instructions to the drone, and the drone flies to the installation area to take monitoring images of the installation area.

[0224] When taking images, the drone can use its human recognition mechanism to capture areas where construction workers are working, thereby enabling targeted monitoring and shooting.

[0225] Specifically, in the early stages of the construction of high-voltage transmission lines, the background processing platform will intelligently generate a set of detailed monitoring instructions based on the project plan and safety specifications. These instructions include but are not limited to the flight path, shooting angle, shooting frequency and other parameters of the drone, aiming to fully capture all key links of the construction site. After generating the monitoring instructions, the background platform sends the instructions to the designated drone through the wireless communication network. After receiving the instructions, the drone will automatically adjust to the preset state and prepare to perform the monitoring task.

[0226] S702: The processing platform receives the monitoring image and uses the target detection model to process the monitoring image to obtain the monitoring result.

[0227] Optionally, the drone can fly and shoot at the installation site according to the received monitoring instructions to obtain real-time monitoring images. These images are transmitted back to the background processing platform in real time through the wireless communication network. After receiving the monitoring images, the background processing platform immediately starts the pre-trained target detection model to process the images. The model will perform intelligent analysis on the content in the image, identify elements such as personnel, equipment, and safety hazards, and determine whether there are any violations.

[0228] More specifically, the processing platform collects training data, for example: images of workers wearing safety helmets and images without safety helmets, images of workers wearing work clothes and images of workers without work clothes, images of workers wearing safety belts normally and images of workers not wearing safety belts, images of people under the tower and images of people not under the tower, images of people using mobile phones and images of people not using mobile phones, and uses these training data to identify the target detection model, so that the target detection model has the ability to identify the above problems.

[0229] Use drones for high-altitude patrols to capture surveillance images of the installation site, and transmit the surveillance images back to the processing platform. The processing platform runs a target detection model, which determines in real time whether there are any of the following five violations in the personnel's operations: violations in wearing safety helmets, work clothes, safety belts, standing under the tower, and using mobile phones.

[0230] S703. After the monitoring results indicate that the personnel have violated the rules, the processing platform controls the drone to play a warning voice, and after the drone patrol is completed, the processing platform generates an inspection report.

[0231] If the monitoring results output by the target detection model indicate that the operator has committed the above-mentioned violations, the processing platform sends a play command to the drone, and the drone plays a warning voice. The processing platform determines the identifier of the warning voice to be played based on the monitoring results indicating the violations. The broadcast command is generated based on the identifier of the warning voice to be played, so that the drone plays the corresponding warning voice.

[0232] Optionally, if the processing results of the target detection model show that there are violations of personnel at the installation site (such as not wearing a safety helmet, illegal operation, etc.), the background processing platform will immediately trigger the warning mechanism. The platform will send a warning playback command to the drone through the wireless communication network. After receiving the command, the drone will immediately play the preset warning voice to remind the on-site personnel to pay attention to standardized operations and ensure safety.

[0233] For example, there are multiple warning voices, including correctly wearing a helmet, correctly wearing work clothes, correctly wearing a helmet, etc. The target detection model can identify the specific violations of the operator. If the violation is not wearing a helmet, the drone can be controlled to play the warning voice for correct wearing of safety.

[0234] After the drone completes its inspection, an inspection report is generated based on the monitoring results of the target detection model.

[0235] Optionally, after the drone completes the inspection mission of the entire installation site, the background processing platform will integrate all monitoring images and processing results to generate a detailed inspection report. The report will list in detail the problems, violations, safety hazards and other information found during the inspection, and put forward corresponding improvement suggestions. This report will provide an important reference for subsequent line operation and safety management.

[0236] In the above technical solution, during the installation of high-voltage transmission lines, the processing platform generates monitoring instructions and controls the drone to take monitoring images of the installation site. The processing platform uses the target detection model to identify the monitoring images and determine whether there are any violations at the installation site. When there are violations, targeted voice warnings are played.

[0237] Specifically, the processing platform can generate monitoring instructions in real time and control the drone to take all-round and multi-angle photos of the installation site to ensure that there are no blind spots in monitoring. Once violations are found, such as not wearing safety equipment, illegal operations, etc., the target detection model can quickly identify and trigger a warning mechanism, so as to correct the violations in time and avoid safety accidents. In addition, through real-time monitoring and early warning, the above technical solution can significantly reduce safety hazards during the installation process, protect the lives of workers, and reduce accident losses caused by illegal operations.

[0238] The processing platform uses the target detection model to intelligently identify monitoring images and can automatically detect violations without human intervention. This automated processing method greatly improves the efficiency of patrol and monitoring and reduces the workload of operation and maintenance personnel. When the target detection model identifies a violation, the processing platform can accurately locate the location of the violator and control the drone to play a targeted warning voice. This precise positioning and warning method can effectively remind violators to pay attention to standardized operations and improve the effectiveness of patrol and monitoring.

[0239] The above technical solution seamlessly connects the inspection and monitoring processes, allowing the processing platform to automatically generate an inspection report containing monitoring results after the inspection is completed. This report not only covers the problems found during the inspection, but also includes monitoring data and records of violations during the installation process, providing comprehensive data analysis support for operation and maintenance personnel. By integrating inspection and monitoring data, the processing platform can have a more comprehensive understanding of the operating status and potential risks of the line. These data provide strong support for operation and maintenance personnel to formulate maintenance plans and optimize inspection strategies, improving data utilization and the pertinence of inspection work.

[0240] In addition, this method establishes a standardized monitoring process by processing the platform to generate monitoring instructions, drones to take monitoring images, and target detection models to identify violations. This process can ensure the consistency and standardization of monitoring work and improve the overall level of operation and maintenance management. Regular monitoring and warning playback can enhance the safety awareness and standardized operation awareness of staff. By constantly reminding and correcting violations, this method helps to cultivate good working habits of staff and reduce safety risks caused by human factors.

[0241] In summary, the introduction of drone real-time monitoring technology in the data processing method of high-voltage transmission line inspection significantly improves the safety of the erection process, the intelligence level of inspection and monitoring, optimizes the inspection data processing process, and enhances the standardization of operation and maintenance management by generating monitoring instructions through the processing platform, taking monitoring images by drones, using target detection models for image recognition, and playing warning voices. These technical effects together provide a strong guarantee for the safe operation of high-voltage transmission lines.

[0242] Figure 5 This is a schematic diagram of the structure of the processing platform provided by the present invention. Figure 5 As shown, the electronic device 10 provided in this embodiment includes: at least one processor 11 and a memory 12. Optionally, the device 10 also includes a communication component 13. The processor 11, the memory 12 and the communication component 13 are connected via a bus 14.

[0243] In a specific implementation process, at least one processor 11 executes the computer-executable instructions stored in the memory 12, so that at least one processor 11 executes the above method.

[0244] The specific implementation process of the processor 11 can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.

[0245] In the above embodiments, it should be understood that the processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the invention can be directly implemented as a hardware processor, or can be implemented by a combination of hardware and software modules in the processor.

[0246] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (NVM), such as at least one disk storage.

[0247] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present invention is not limited to only one bus or one type of bus.

[0248] The present invention also provides a computer program product, comprising a computer program, and the computer program implements the above method when executed by a processor.

[0249] The present invention also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0250] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0251] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0252] The division of units is only a logical function division, and there may be other divisions in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0253] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0254] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0255] If the function is implemented in the form of a software functional 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 solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0256] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0257] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for processing high-voltage transmission line inspection data based on drones, characterized in that: The method is applied to a high-voltage transmission line inspection data processing platform based on a drone, and the method comprises: Acquire a survey area of ​​the high-voltage transmission line, generate a survey instruction according to the survey area, and send the survey instruction to the drone, so that the drone performs a survey in the survey area of ​​the high-voltage transmission line to obtain point cloud data of the high-voltage transmission line; Receiving the point cloud data of the high-voltage transmission line returned by the drone, processing the point cloud data of the high-voltage transmission line, and obtaining a high-precision map of the high-voltage transmission line; Acquire the position and posture of the patrol object from the high-precision map, and determine the position and posture of the patrol point according to the position and posture of the patrol object, the high-precision map of the area where the patrol object is located, the model data of the drone, and the camera parameters; Generate a patrol path of the UAV according to the positions and postures of the patrol points, send the patrol path of the UAV and the high-precision map to the UAV, so that the UAV takes patrol images at each patrol point according to the corresponding positions and postures based on the high-precision map; wherein the patrol path includes the positions and postures of the patrol points and the order of the patrol points; The inspection image returned by the UAV is received, and a target detection model is used to perform defect recognition on the inspection image to obtain an inspection result.

2. The method for processing high-voltage transmission line inspection data based on an unmanned aerial vehicle according to claim 1 is characterized in that: Acquiring a survey area of ​​a high-voltage transmission line, generating a survey instruction according to the survey area, and sending the survey instruction to a drone so that the drone performs a survey in the survey area of ​​the high-voltage transmission line to obtain point cloud data of the high-voltage transmission line, specifically including: Generate a surveying and measuring point of the high-voltage transmission line, generate the surveying instruction according to the surveying and measuring point, and send the surveying instruction to the UAV; so that the UAV surveys the surveying and measuring point according to the surveying instruction to obtain point cloud data of the surveying and measuring point; Accordingly, receiving the point cloud data of the high-voltage transmission line returned by the drone, processing the point cloud data of the high-voltage transmission line, and obtaining a high-precision map of the high-voltage transmission line specifically includes: Receive the point cloud data of the surveyed location returned by the drone, match the point cloud data of the surveyed location with the model data of the high-voltage transmission line, and generate a high-precision map of the high-voltage transmission line.

3. The method for processing high-voltage transmission line inspection data based on an unmanned aerial vehicle according to claim 1 is characterized in that: The position and posture of the patrol point are determined according to the position and posture of the patrol object, the high-precision map of the area where the patrol object is located, the model data of the drone, and the camera parameters, including: For each patrol object, an observation point of the patrol object is generated according to the posture of the patrol object, a high-precision map of the area where the patrol object is located, the model data of the drone, and the camera parameters, wherein the observation point includes a position and a posture; According to the positions of the multiple observation points and the camera optical axis coordinates at the multiple observation points, it is determined whether the multiple observation points meet the fusion requirements. If they meet the requirements, the observation points are fused to obtain the patrol points, and the patrol objects corresponding to the observation points are used as the patrol objects of the patrol points.

4. The method for processing high-voltage transmission line inspection data based on an unmanned aerial vehicle according to claim 3 is characterized in that: For each patrol object, the observation point of the patrol object is generated according to the posture of the patrol object, the high-precision map of the area where the patrol object is located, the model data of the drone, and the camera parameters, specifically including: Acquire the patrol surface coordinates of the patrol object, and generate a plurality of candidate points according to the patrol surface coordinates of the patrol object and camera parameters, wherein the candidate points include positions and postures; For each candidate point, determine whether the drone will collide at the candidate point based on the model data of the drone and the high-precision map of the area where the patrol object is located. If no collision occurs, retain the candidate point; A candidate point farthest from an obstacle is selected from a plurality of reserved candidate points as the observation point, wherein the obstacle is an object in the area where the patrol object is located.

5. The method for processing high-voltage transmission line inspection data based on an unmanned aerial vehicle according to claim 4 is characterized in that; If the patrol object is a pin, the patrol surface is a plane passing through the pin axis and two end surfaces perpendicular to the pin axis; If the patrol object is an insulator, the patrol surface is a plane passing through the axis of the insulator and two end surfaces perpendicular to the axis of the insulator.

6. The method for processing high-voltage transmission line inspection data based on an unmanned aerial vehicle according to any one of claims 3 to 5, characterized in that: The processing platform uses the target detection model to perform defect recognition on the inspection image to obtain the inspection result, which specifically includes: The processing platform receives the patrol image and the patrol object corresponding to the patrol image; Obtaining the target detection model corresponding to the patrol object according to the patrol object and model mapping relationship table; the model mapping relationship table is the corresponding relationship between the patrol object category and the target detection model; The inspection image is identified using a target detection model corresponding to the inspection object to obtain the defect type of the inspection object.

7. The method for processing high-voltage transmission line inspection data based on an unmanned aerial vehicle according to any one of claims 1 to 5, characterized in that: The method further comprises: When erecting a high-voltage transmission line, a monitoring instruction is generated, and the monitoring instruction is sent to a drone, so that the drone takes a monitoring image of the erection site; Receiving the monitoring image, and processing the monitoring image using the target detection model to obtain a monitoring result; If the monitoring results indicate that a person has violated a rule, the drone is controlled to play a warning voice, and the processing platform generates an inspection report after the drone patrol is completed.

8. A high-voltage transmission line inspection data processing platform based on drones, characterized in that: Includes the following modules: An acquisition module is used to acquire a survey area of ​​a high-voltage transmission line, generate a survey instruction according to the survey area, and send the survey instruction to a drone so that the drone performs a survey in the survey area of ​​the high-voltage transmission line to obtain point cloud data of the high-voltage transmission line; A receiving module, used for receiving the point cloud data of the high-voltage transmission line returned by the drone, processing the point cloud data of the high-voltage transmission line, and obtaining a high-precision map of the high-voltage transmission line; A processing module, used to obtain the position and posture of the patrol object from the high-precision map, and determine the position and posture of the patrol point according to the position and posture of the patrol object, the high-precision map of the area where the patrol object is located, the model data of the drone, and the camera parameters; A sending module, used for generating a patrol path of the UAV according to the positions and postures of the patrol points, and sending the patrol path of the UAV and the high-precision map to the UAV, so that the UAV takes patrol images at each patrol point according to the corresponding positions and postures based on the high-precision map; wherein the patrol path includes the positions and postures of the patrol points and the order of the patrol points; The receiving module is also used to receive the patrol image returned by the drone, and use the target detection model to perform defect recognition on the patrol image to obtain the patrol result.

9. A high-voltage transmission line inspection data processing platform based on drones, characterized in that: include: processor; as well as, A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 7 by executing the executable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed by a processor.

Images