Unmanned aerial vehicle three-dimensional surveying and mapping method and system based on visual SLAM

Abstract

The invention provides an unmanned aerial vehicle three-dimensional surveying and mapping method and system based on visual SLAM, and relates to the technical field of unmanned aerial vehicle surveying and mapping, and the method comprises the steps: 1, laying and measuring three power transmission tower foot grounding bolt real objects with known spatial positions on the ground of a power transmission line corridor through an unmanned aerial vehicle carrying a visible light camera; 2, forming a basic space structure according to the space coordinates of the three power transmission tower foot grounding bolt objects with known space positions, and collecting continuous sequence images along a power transmission line corridor according to a preset overlapping rate; the method comprises the following steps: performing frame-by-frame processing on collected sequence images, and extracting an initial feature point set for each frame of image; according to the invention, high-precision three-dimensional modeling of power transmission line corridor terrains, ground features and electric power facilities is realized.

Description

Visual SLAM-based UAV 3D mapping method and system Technical Field

[0001] This invention relates to the field of UAV mapping technology, and in particular to a UAV 3D mapping method and system based on visual SLAM. Background Technology

[0002] With the increasing demand for refined management and control in the field of natural resources and planning, UAV 3D mapping technology based on visual SLAM has become a core supporting means for confirming the spatial rights of power transmission line corridors, verifying the compliance of pipelines, and delineating the ecological protection red line in the surrounding areas due to its advantages of not relying on GPS signals and being adaptable to complex terrain scenarios. It provides accurate spatial data for the Bureau of Natural Resources and Planning to carry out work such as connecting power transmission facilities with spatial planning and investigating potential hazard areas.

[0003] In the field of natural resources and planning, when conducting 3D mapping of 110kV transmission line corridors within its jurisdiction, the traditional visual SLAM UAV mapping scheme based on SIFT features is adopted. This involves using a UAV equipped with a visible light camera to collect sequential images along the line. A 3D model is constructed through feature point extraction, inter-frame matching, pose estimation, and point cloud fusion to clarify the spatial relationship between the transmission line and surrounding farmland, forest land, and buildings, providing data for line reconstruction and expansion planning and spatial use control. However, this technology has shortcomings in practical application. Transmission line corridors often traverse mountainous areas and barren slopes, with some sections having exposed rock and sparse vegetation, resulting in a lack of texture information. This leads to sparse distribution or local clustering of image feature points, increased mismatch rates in inter-frame matching, and difficulty in controlling the cumulative error of UAV pose estimation. Consequently, the constructed 3D model exhibits local distortion, and the positioning accuracy for transmission tower foundations, line alignment, and surrounding terrain inflection points is insufficient, failing to meet the accuracy requirements of the natural resources and planning bureau for the mapping of transmission line corridors. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a UAV 3D mapping method and system based on visual SLAM to achieve high-precision 3D modeling of the terrain, features and power facilities of power transmission line corridors.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a UAV-based 3D mapping method using visual SLAM, the method comprising: Step 1, using a UAV equipped with a visible light camera to deploy and measure three grounding bolts of transmission towers with known spatial locations on the ground along a transmission line corridor; Step 2, constructing a basic spatial structure based on the spatial coordinates of the three grounding bolts, and acquiring a continuous sequence of images along the transmission line corridor according to a preset overlap rate; processing the acquired sequence of images frame by frame, and extracting an initial feature point set for each frame; Step 3, filtering the initial feature point set, removing redundant feature points with clustered distributions, and supplementing with feature points extracted from sparse texture areas, to obtain... Step 4: Based on the optimized feature point set, perform feature point matching between adjacent frame images to obtain matching results; estimate the inter-frame motion pose of the UAV based on the matching results; Step 5: Based on the motion pose, combine the obtained uniformly distributed optimized feature point set to extract the topological association characteristics of the feature point set in space; based on the extracted spatial topological association characteristics of the feature point set, construct an energy optimization function with a robust kernel function, calculate an error adjustment value by minimizing the energy optimization function, and correct the motion pose using the error adjustment value to obtain the corrected motion pose; Step 6: Based on the corrected motion pose, fuse the optimized feature point set obtained from multiple frames of images to construct and optimize a three-dimensional point cloud model of the transmission line corridor to obtain the measured three-dimensional mapping results.

[0006] Furthermore, using a drone equipped with a visible light camera, the grounding bolts of three transmission towers with known spatial locations were deployed and measured on the ground along the transmission line corridor. This process included: controlling the drone to fly along a preset route, scanning the ground of the transmission line corridor with the visible light camera, identifying and locating the grounding bolt positions of the three transmission towers, and generating an initial image dataset containing the three grounding bolts; based on the initial image dataset, edge detection and center point extraction were performed on the contour features of the three grounding bolts to calculate the two-dimensional pixel coordinates of the three grounding bolts in the image coordinate system; based on the two-dimensional pixel coordinates, combined with the drone's flight altitude and camera intrinsic parameters, the two-dimensional pixel coordinates were converted into three-dimensional relative coordinates of the three grounding bolts in the drone's coordinate system; based on the real-time position data provided by the three-dimensional relative coordinates, the three-dimensional relative coordinates of the three grounding bolts were converted into absolute spatial coordinates in the geodetic coordinate system; based on the absolute spatial coordinates, error analysis and consistency verification were performed on the spatial position accuracy of the three grounding bolts; when the verification results met a preset accuracy threshold, the spatial coordinate data of the three grounding bolts with known spatial locations were output.

[0007] Furthermore, a basic spatial structure is constructed based on the spatial coordinates of the grounding bolts at the bases of three transmission towers, whose spatial locations are known. A continuous sequence of images is then acquired along the transmission line corridor according to a preset overlap rate. Frame-by-frame processing of the acquired images is performed, extracting an initial feature point set for each frame. This includes: constructing a basic spatial structure composed of three spatial points based on the spatial coordinate data of the grounding bolts at the bases of the three transmission towers; determining the spatial orientation and scale information of the basic spatial structure in the geodetic coordinate system; and, based on the spatial orientation and scale information of the basic spatial structure, combined with the geographic information data of the transmission line corridor, planning the drone's movement along the transmission line corridor. The system calculates the flight path and distance parameters between adjacent image acquisition points. Based on the flight path and distance parameters, it generates UAV flight control commands and sends them to the UAV to control it to fly along the planned path. Simultaneously, it triggers the visible light camera to continuously capture images at the calculated acquisition interval, acquiring a sequence of image data along the power transmission line corridor. Based on the sequence of image data, it sorts and preprocesses the image data in chronological order to eliminate image noise and distortion, obtaining a continuous frame image sequence. By analyzing the continuous frame image sequence frame by frame, it processes each frame to identify corners, edges, and texture regions in the image, and extracts the initial feature point set corresponding to each frame.

[0008] Furthermore, by filtering the initial feature point set, redundant feature points in clustered distribution are removed, and feature points in sparse texture regions are extracted to obtain a uniformly distributed optimized feature point set. This includes: receiving the initial feature point set of each frame of image; projecting the feature points onto a three-dimensional coordinate system based on the spatial coordinates of the foundation space structure formed by the grounding bolts of the three transmission towers to obtain three-dimensional spatial distribution data of the feature points; calculating the spatial distance between each feature point and its neighboring feature points based on the three-dimensional spatial distribution data of the feature points, constructing feature point density distribution data, identifying clustered regions with spatial distances less than a preset threshold, and marking redundant feature points within the clustered regions; and gradually removing redundant feature points within the clustered regions according to the order of feature point response values ​​from low to high based on the redundant feature points. Feature points are extracted until the feature point density in each cluster region drops below a preset density threshold, resulting in a filtered feature point set. Based on the filtered feature point set, the regions in the basic spatial structure not covered by feature points are analyzed, the number of feature points in each grid cell is calculated, and texture sparse regions with a number of feature points below a preset sparsity threshold are identified. For texture sparse regions, feature points are extracted again within the sparse regions, with a focus on enhancing feature point detection at edges, corners, and structured texture regions, and new feature points are extracted. The filtered feature point set is then fused with the newly extracted feature points, and the spatial uniformity of the fused feature point set is verified. When the verification results show that the feature point distribution meets the preset uniformity index, a uniformly distributed optimized feature point set is obtained.

[0009] Furthermore, based on the optimized feature point set, feature point matching is performed between adjacent frame images to obtain matching results. The inter-frame motion pose of the UAV is estimated using the matching results, including: receiving a uniformly distributed optimized feature point set, sorting the optimized feature point set according to the image acquisition time order, and constructing a feature point correspondence data structure between adjacent frame images; based on the feature point correspondence data structure, feature point matching is performed between the optimized feature point sets of adjacent frame images to obtain a preliminary set of feature point matching pairs; geometric consistency verification is performed on the preliminary set of feature point matching pairs, and mismatched point pairs that do not conform to geometric constraints are eliminated to obtain the feature point matching results; based on the feature point matching results, combined with the foundation spatial structure composed of the grounding bolts of three power transmission towers as a scale reference, the relative transformation matrix between adjacent frame images is calculated; the relative transformation matrix is ​​converted into the inter-frame motion pose parameters of the UAV, including rotation matrix and translation vector, and the motion pose parameters are smoothed and outlier detected to obtain the inter-frame motion pose of the UAV.

[0010] Furthermore, based on the motion pose and the obtained uniformly distributed optimized feature point set, the topological association characteristics of the feature point set in space are extracted. Based on the extracted spatial topological association characteristics of the feature point set, an energy optimization function of a robust kernel function is constructed. An error adjustment value is calculated by minimizing the energy optimization function, and the motion pose is corrected using the error adjustment value to obtain the corrected motion pose. This includes: mapping the optimized feature point set to the three-dimensional space corresponding to the motion pose based on the inter-frame motion pose of the UAV and the uniformly distributed optimized feature point set; analyzing the spatial connectivity and distribution pattern between feature points; and extracting the topological association characteristics of the feature point set in space. The system uses topological correlation characteristic data; based on this data and the foundation spatial structure formed by the grounding bolts of three power transmission towers as geometric constraints, a robust kernel function energy optimization function is constructed; based on this energy optimization function, the error adjustment value of the current motion pose is obtained by analyzing the geometric deviation between the topological correlation characteristics and the foundation spatial structure; the error adjustment value is applied to the inter-frame motion pose of the UAV, and the rotation and translation components of the motion pose are corrected to obtain the corrected motion pose data; the convergence of the corrected motion pose data is verified, and the corrected motion pose is obtained when the verification result meets the preset accuracy threshold requirement.

[0011] Furthermore, based on the corrected motion pose, an optimized feature point set obtained by fusing multiple frames of images is used to construct and optimize a 3D point cloud model of the transmission line corridor, resulting in the measured 3D mapping results. This includes: sorting and storing the corrected motion pose data according to a time series to construct a motion trajectory dataset of the UAV during its flight along the transmission line corridor; based on the motion trajectory dataset, and combining the optimized feature point set corresponding to each frame of image, projecting the feature points in the optimized feature point set onto the geodetic coordinate system to obtain an initial 3D point cloud dataset; and performing spatial filtering on the initial 3D point cloud dataset to remove outliers and noise points, while simultaneously registering and aligning the point cloud data according to the geometric constraints of the basic spatial structure. A preliminary registered 3D point cloud model is obtained. Based on the preliminary registered 3D point cloud model, the point cloud model is densified, and the positional accuracy of each 3D point in the point cloud is optimized by bundle adjustment, while maintaining the geometric structural characteristics of the transmission line corridor terrain and features, resulting in an optimized dense 3D point cloud model. Surface reconstruction processing is performed on the dense 3D point cloud model to obtain a complete 3D mesh model of the transmission line corridor terrain, vegetation, buildings, and transmission facilities. Based on the complete 3D mesh model, and using the spatial coordinates of the grounding bolts at the bases of three transmission towers as control points, the 3D mesh model is subjected to absolute coordinate correction and accuracy verification. When the verification results meet the preset mapping accuracy requirements, the 3D mapping results data of the transmission line corridor are obtained.

[0012] Secondly, a UAV-based 3D mapping system using visual SLAM includes: an acquisition module for deploying and measuring three grounding bolts at the base of power transmission towers with known spatial locations on the ground along a power transmission line corridor using a UAV equipped with a visible light camera; an extraction module for constructing a basic spatial structure based on the spatial coordinates of the three grounding bolts and acquiring a continuous sequence of images along the power transmission line corridor according to a preset overlap rate; extracting an initial feature point set for each frame of the acquired image sequence through frame-by-frame processing; and a filtering module for filtering the initial feature point set, removing redundant feature points with clustered distributions, and supplementing the extraction of feature points in sparse texture areas to obtain a uniformly distributed optimized feature point set; and matching... The system comprises four modules: a matching module for matching feature points between adjacent frames based on an optimized feature point set, and a correction module for extracting the topological correlation characteristics of the feature point set in space based on the motion pose and the uniformly distributed optimized feature point set; an energy optimization function with a robust kernel function for constructing a feature point set based on the extracted topological correlation characteristics; an error adjustment value for calculating an error adjustment value for correcting the motion pose; and a processing module for constructing and optimizing a 3D point cloud model of the transmission line corridor based on the corrected motion pose and the optimized feature point set obtained from multiple frames.

[0013] Thirdly, a computing device includes: one or more processors; and a storage device for storing one or more programs that, when executed by the one or more processors, cause the one or more processors to implement the method.

[0014] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0015] The above-mentioned solution of the present invention has at least the following beneficial effects: By adopting the collaborative technical means of constructing the basic spatial structure with the grounding bolts of the transmission tower feet, bidirectionally optimizing the distribution of feature points, combining mesh generation with robust kernel function energy optimization to correct pose, and multi-stage point cloud optimization, it effectively overcomes the technical problems of existing visual SLAM UAV mapping in the transmission line corridor scenario, such as the need for manual deployment of control points, uneven distribution of feature points in sparse texture areas leading to high matching mismatch rates, difficulty in controlling cumulative pose estimation errors, and insufficient accuracy of 3D models. Thus, it achieves the technical effect of improving the stability of feature point matching and the accuracy of pose estimation without the need for additional control points, reducing manpower and time costs, and ultimately constructing a 3D mapping result of the transmission line corridor that meets the measurement needs of transmission line operation and maintenance, has high integrity and high accuracy, is adaptable to complex terrain scenarios, and improves work efficiency. Attached Figure Description

[0016] Figure 1 is a flowchart illustrating the UAV 3D mapping method based on visual SLAM provided in an embodiment of the present invention.

[0017] Figure 2 is a schematic diagram of a UAV 3D mapping system based on visual SLAM provided in an embodiment of the present invention. Detailed Implementation

[0018] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0019] As shown in Figure 1, an embodiment of the present invention proposes a UAV-based 3D mapping method using visual SLAM. The method includes the following steps: Step 1, using a UAV equipped with a visible light camera, three grounding bolts of transmission towers with known spatial locations are deployed and measured on the ground along the transmission line corridor; Step 2, a basic spatial structure is constructed based on the spatial coordinates of the three grounding bolts, and a continuous sequence of images is acquired along the transmission line corridor according to a preset overlap rate; the acquired sequence of images is processed frame by frame, and an initial feature point set is extracted for each frame; Step 3, the initial feature point set is filtered to remove redundant feature points with clustered distribution, and feature points in sparse texture areas are extracted to obtain a uniformly distributed feature point set. Step 4: Based on the optimized feature point set, perform feature point matching between adjacent frame images to obtain matching results; estimate the inter-frame motion pose of the UAV based on the matching results; Step 5: Based on the motion pose, combine the obtained uniformly distributed optimized feature point set to extract the topological association characteristics of the feature point set in space; based on the extracted spatial topological association characteristics of the feature point set, construct an energy optimization function of robust kernel function, calculate an error adjustment value by minimizing the energy optimization function, and correct the motion pose using the error adjustment value to obtain the corrected motion pose; Step 6: Based on the corrected motion pose, fuse the optimized feature point set obtained from multiple frames of images to construct and optimize a three-dimensional point cloud model of the transmission line corridor to obtain the measured three-dimensional mapping results. In this embodiment of the invention, by employing the technical means of constructing the basic spatial structure using physical grounding bolts of three transmission towers at known spatial locations, optimizing the distribution of the initial feature point set by eliminating redundancy and supplementing sparse regions, constructing a robust kernel function energy optimization function based on the grid geometric characteristics to correct the pose, and then fusing and optimizing the feature point set to construct and optimize the 3D point cloud model, the technical problems of uneven feature point distribution, easy accumulation of pose estimation errors, and difficulty in meeting the measurement requirements of the 3D model in the traditional visual SLAM UAV mapping scenario of transmission line corridor are effectively overcome. Thus, the stability of feature point matching and the accuracy of pose estimation are improved, and high-precision 3D mapping results of transmission line corridor are constructed.

[0020] In a preferred embodiment of the present invention, step 1 above may include: Step 1.1, controlling the UAV to fly along a preset route, scanning the ground of the power transmission line corridor with a visible light camera, identifying and locating the positions of the grounding bolts at the bases of three power transmission towers, and generating an initial image dataset containing the three grounding bolts. Specifically, this includes: First, combining the geographical orientation of the power transmission line corridor, the distribution of the towers, and the terrain features, planning a preset route for the UAV. The preset route is centered on the power transmission line, with a flight path spacing of 10 meters, a flight altitude of 5 meters, and a flight speed of 2 meters per second, ensuring that the UAV's flight path completely covers the base area of ​​the target power transmission towers; After completing the route planning, starting the UAV and its onboard visible light camera, and controlling the UAV to fly along the preset route; The visible light camera is an industrial-grade camera with a resolution of 4096×3072 and a shooting angle of... The drone's altitude was adjusted to be perpendicular to the ground, the camera's frame rate was set to 2 frames per second, and the overlap rate of adjacent images was set to 80%. During the drone's flight, the visible light camera continuously scanned and photographed the ground of the power transmission line corridor, focusing on capturing image information of the base area of ​​the power transmission towers. During image acquisition, the onboard image processing unit performed preliminary identification on the acquired images in real time. Based on the appearance characteristics of the metal material and the shape characteristics of the hexagonal head of the grounding bolt, images containing the base grounding bolts of the power transmission towers were selected from the continuously acquired images. When three different base grounding bolts of power transmission towers were identified and each grounding bolt presented a clear and complete outline in the image, image acquisition for that section of the flight path was stopped. Subsequently, all images containing three grounding bolts were summarized, numbered and organized according to the acquisition time sequence, and an initial image dataset containing three grounding bolts was generated.

[0021] Step 1.2: Based on the initial image dataset, edge detection and center point extraction are performed on the contour features of the three grounding bolts to calculate the two-dimensional pixel coordinates of the three grounding bolts in the image coordinate system. Specifically, this includes: First, calling all images in the initial image dataset and performing grayscale processing on each image in sequence to convert the color image to a grayscale image, reducing the complexity of image data processing. Then, Gaussian filtering is performed on the grayscale image to filter out high-frequency noise in the image and improve the image clarity. After completing the image preprocessing, an edge detection algorithm based on gradient change is used to process the preprocessed image, analyzing the grayscale value change range of the image pixel by pixel to identify the edge pixels of the grounding bolt contours. Connect consecutive edge pixels to form a complete grounding bolt outline. Then, for each grounding bolt outline, first determine the boundary range of the outline, calculate the pixel extreme values ​​of the outline in the horizontal and vertical directions of the image coordinate system, and then take the average of the horizontal and vertical extreme values ​​as the two-dimensional pixel coordinates of the grounding bolt in the image coordinate system. Following the above operation, process all images containing grounding bolts in the initial image dataset in turn, and calculate the two-dimensional pixel coordinates of the three grounding bolts in different images.

[0022] Step 1.3: Based on the two-dimensional pixel coordinates, combined with the UAV's flight altitude and camera intrinsic parameters, the two-dimensional pixel coordinates are converted into three-dimensional relative coordinates of the three grounding bolts in the UAV coordinate system. Specifically, this includes: First, retrieving the flight altitude data recorded in real-time during the UAV's flight. This flight altitude data is obtained by the laser altimeter onboard the UAV; during this implementation, the UAV's flight altitude remained stable at 5 meters. Simultaneously, retrieving the intrinsic parameters of the visible light camera, including a focal length of 6 mm and a pixel size of 2 micrometers, which are fixed parameters calibrated at the camera's factory. Then, establishing the UAV coordinate system with the optical center of the visible light camera onboard the UAV as the origin, where the X-axis is perpendicular to the optical center of the visible light camera. The Y-axis is parallel to the ground plane, and the Z-axis is perpendicular to the ground and pointing upwards. Based on the mapping relationship between the image coordinate system and the UAV coordinate system, the obtained two-dimensional pixel coordinates are converted into imaging plane coordinates. In specific operation, the actual physical size corresponding to the pixel coordinates is calculated according to the pixel size, and then combined with the camera focal length, the relative position of the grounding bolts in the X-axis and Y-axis directions of the UAV coordinate system is derived through geometric projection relationship. Since the UAV's flight altitude is stable and the grounding bolts are located on the ground, the coordinate value of the grounding bolts in the Z-axis direction of the UAV coordinate system is the negative of the UAV's flight altitude. Combining the coordinate values ​​in the X-axis, Y-axis, and Z-axis directions, the three-dimensional relative coordinates of the three grounding bolts in the UAV coordinate system are calculated.

[0023] Step 1.4: Based on the real-time position data provided by the three-dimensional relative coordinates, convert the three-dimensional relative coordinates of the three grounding bolts into absolute spatial coordinates in the geodetic coordinate system. Specifically, this includes: First, retrieving the real-time positioning data of the UAV recorded by the RTK positioning module onboard the UAV at the time of image acquisition. This data includes the UAV's three-dimensional coordinates and attitude angles in the geodetic coordinate system, where the three-dimensional coordinates are geodetic longitude, geodetic latitude, and geodetic elevation, and the attitude angles are heading angle, pitch angle, and roll angle. Then, establishing the transformation relationship between the UAV coordinate system and the geodetic coordinate system. This transformation relationship includes coordinate translation transformation and... The coordinate transformation consists of two parts: coordinate translation transformation and coordinate rotation transformation. The coordinate translation transformation uses the three-dimensional coordinates of the UAV in the geodetic coordinate system as a reference to calculate the position of the origin of the UAV coordinate system in the geodetic coordinate system. The coordinate rotation transformation is based on the attitude angle of the UAV and calculates the rotational relationship between the two coordinate systems. After the transformation relationship is established, the three-dimensional relative coordinates of the three grounding bolts in the UAV coordinate system are first transformed into coordinates in the same direction as the geodetic coordinate system through rotation transformation, and then the position coordinates of the UAV in the geodetic coordinate system are superimposed through translation transformation to finally obtain the absolute spatial coordinates of the three grounding bolts in the geodetic coordinate system.

[0024] Step 1.5: Based on the absolute spatial coordinates, perform error analysis and consistency verification on the spatial position accuracy of the three grounding bolts. When the verification results meet the preset accuracy threshold, output the spatial coordinate data of the three grounding bolts at the tower feet of the transmission towers with known spatial positions. Specifically, this includes: First, performing error analysis on the absolute spatial coordinates of the three grounding bolts. Using a repeated measurement method, perform coordinate measurement calculations five times for each grounding bolt to obtain five sets of absolute spatial coordinate data for each grounding bolt. Calculate the average coordinate values ​​of each grounding bolt in the X, Y, and Z axes of the geodetic coordinate system. Then, calculate the deviation between each set of coordinate data and the average value, and statistically analyze the maximum deviation and the average deviation. Next, consistency verification is performed. Based on the actual installation positions of the three grounding bolts, the actual physical distance between the three grounding bolts is measured. At the same time, based on the calculated absolute spatial coordinates, the theoretical distance between the three grounding bolts is calculated. The difference between the actual physical distance and the theoretical distance is compared to determine whether the difference is within the allowable range. In this implementation, the preset position accuracy threshold is set to 2 mm. When the maximum deviation value obtained from the error analysis is less than 2 mm, and the difference between the actual distance and the theoretical distance obtained from the consistency verification is less than 2 mm, the verification result is determined to meet the preset accuracy threshold. At this time, the average absolute spatial coordinates of the three grounding bolts are used as the final spatial coordinate data.

[0025] In this embodiment of the invention, by employing a progressive technical approach—using a UAV to scan and locate the grounding bolt via a preset flight path, obtaining two-dimensional pixel coordinates through edge detection and center point extraction, converting the three-dimensional relative coordinates by combining flight altitude and camera intrinsic parameters, further converting them into absolute coordinates in the geodetic coordinate system, and ensuring accuracy through error analysis and consistency verification—the technical problems of low efficiency in manual measurement, easy deviations in coordinate conversion, and insufficient reliability of control points due to a lack of effective verification of position accuracy in traditional grounding bolt spatial positioning are effectively overcome. This achieves automated and precise positioning of the grounding bolt, ensuring the accuracy and stability of spatial coordinate data.

[0026] In a preferred embodiment of the present invention, step 2 above may include: Step 2.1, constructing a basic spatial structure composed of three spatial points based on the spatial coordinate data of the three transmission tower foot grounding bolts, and determining the spatial orientation and scale information of the basic spatial structure in the geodetic coordinate system. Specifically, this includes: First, using the absolute spatial coordinate data of the three transmission tower foot grounding bolts as a reference, and taking the absolute spatial coordinates of the three grounding bolts as three spatial vertices, constructing a triangular basic spatial structure formed by connecting three spatial line segments; then, performing parameter calculations on the basic spatial structure, first calculating the pairwise spatial straight-line distances between the three vertices to obtain the scale reference data of the basic spatial structure, and then calculating the centroid coordinates of the triangular structure and the azimuth angles of the three sides to determine the spatial orientation of the basic spatial structure in the geodetic coordinate system; finally, integrating and storing the calculated scale reference data and spatial orientation data to complete the construction of the basic spatial structure.

[0027] Step 2.2, based on the spatial orientation and scale information of the basic spatial structure, combined with the geographic information data of the power transmission line corridor, plans the flight path of the UAV along the power transmission line corridor and calculates the distance parameters between adjacent image acquisition points. Specifically, this includes: first, retrieving the spatial orientation and scale information of the basic spatial structure, and simultaneously retrieving the geographic information data of the power transmission line corridor. The geographic information data includes the direction data of the power transmission line, the elevation data of the terrain along the line, the distribution location data of the towers, and the vegetation cover area data; then, taking the center of gravity of the basic spatial structure as the starting point and the direction of the power transmission line as the extension direction, plans the flight path of the UAV along the power transmission line corridor. The flight path of the drone was planned; the flight path was set as a straight strip along the power transmission line corridor, the flight altitude was set at 50 meters, taking into account both the mapping coverage and image resolution, and the strip width was set at 40 meters to ensure effective image connection between adjacent strips; then, the distance parameters between adjacent image acquisition points were calculated based on the preset image overlap rate. In this implementation, the preset image overlap rate was 70%, the image width of the visible light camera was 30 meters, and the distance parameter between adjacent image acquisition points was determined to be 9 meters through calculation. The distance parameter ensures that the sequence of continuously acquired images meets the preset overlap rate requirement.

[0028] Step 2.3: Based on the flight path and distance parameters, generate UAV flight control commands and send them to the UAV to control the UAV to fly along the planned path. Simultaneously, trigger the visible light camera to continuously capture images at the calculated acquisition interval, acquiring a sequential image data stream along the power transmission line corridor. Specifically, this includes: generating UAV flight control commands based on the planned flight path and the distance parameters between adjacent image acquisition points. The commands include the UAV's takeoff point coordinates, flight direction, flight speed, flight altitude, and image acquisition trigger commands; the flight speed is set to 3 meters per second to ensure stable image acquisition by the UAV; then, the generated flight control commands are sent to the UAV's flight control unit via a wireless communication link. After receiving the commands, the UAV flight control unit controls the UAV to take off from the takeoff point and fly along the power transmission line corridor according to the preset flight path, flight altitude, and flight speed; simultaneously, the UAV flight control unit calculates the image acquisition interval as 3 seconds based on the distance parameters between adjacent image acquisition points and the flight speed, and triggers the visible light camera to continuously capture images at this time interval. The visible light camera acquires images at a resolution of 4096×3072, and the acquired image data is transmitted to the onboard storage unit in real time, forming a sequential image data stream along the power transmission line corridor.

[0029] Step 2.4: Based on the sequence image data stream, the image data is sorted and preprocessed in chronological order to eliminate image noise and distortion, resulting in a continuous frame image sequence. Specifically, this includes: First, retrieving the sequence image data stream from the UAV's onboard storage unit, reading the acquisition timestamp of each image, and sorting all image data according to the acquisition time to form a time-series image set. Then, preprocessing the sorted image set involves noise removal using a Gaussian filter algorithm with a 3×3 kernel size to effectively remove high-frequency noise while preserving edge features. Next, image distortion correction is performed by retrieving the intrinsic calibration parameters of the visible light camera, including the camera's focal length, principal point coordinates, and distortion coefficients. These parameters are used to correct radial and tangential distortion in the image, eliminating image distortion caused by camera lens defects. Finally, the images that have undergone noise removal and distortion correction are integrated to obtain a continuous frame image sequence.

[0030] Step 2.5 involves analyzing the continuous frame image sequence frame by frame, processing each frame to identify corners, edges, and textured regions, and extracting the initial feature point set for each frame. Specifically, this includes: first, retrieving the preprocessed continuous frame image sequence and importing it frame by frame into the feature extraction processing unit according to the image arrangement order; then, performing multi-dimensional feature recognition on a single frame image to accurately capture corner areas with drastic grayscale changes, identifying edge features to capture the boundary contours of different land features, and simultaneously identifying textured regions to mark areas with different texture features, such as vegetation cover areas and bare rock areas; next, extracting initial feature points based on the identified corners, edges, and textured regions, ensuring that feature points evenly cover different areas of the image, with at least 2000 initial feature points extracted for each frame; finally, numbering and storing the initial feature point set extracted from each frame image, with each feature point containing the pixel coordinates and feature description information, thus completing the initial feature point set extraction for all frames.

[0031] In this embodiment of the invention, by employing a technique that uses the spatial coordinates of grounding bolts to construct a basic spatial structure through a cloud module to determine the orientation scale, combining geographic information to plan flight paths and acquisition parameters, controlling the UAV to acquire images according to regulations and performing noise reduction and distortion preprocessing, and extracting initial feature points such as corner points and edges frame by frame, the technical problems of traditional image acquisition paths lacking precise scale references, irregular acquisition intervals and overlap rates leading to low data validity, image noise and distortion affecting feature extraction quality, and insufficient targeting of initial feature point extraction are effectively overcome. Thus, it achieves the goal of providing a precise spatial benchmark for image acquisition, ensuring the quality and continuity of sequential images, and obtaining a comprehensive and effective set of initial feature points.

[0032] In a preferred embodiment of the present invention, step 3 above may include: step 3.1, receiving an initial feature point set for each frame of image, projecting the feature points onto a three-dimensional coordinate system based on the spatial structure of the foundation formed by the spatial coordinates of the three transmission tower foot grounding bolts, to obtain the three-dimensional spatial distribution data of the feature points, specifically including: first, receiving the initial feature point set corresponding to each frame of image, which includes the image pixel coordinates and feature description information of each feature point; then, retrieving the parameters of the constructed three-dimensional coordinate system based on the absolute spatial coordinates of the three transmission tower foot grounding bolts, the coordinate system being based on the centroid of the triangular structure formed by the three grounding bolts. The XY plane is defined by a horizontal plane parallel to the Earth's coordinate system, and the Z-axis is defined by a direction perpendicular to the horizontal plane. Next, combining the intrinsic parameters of the visible light camera, the real-time pose data during the UAV's flight, and the geographic information data at the time of image acquisition, a mapping relationship is established between the image pixel coordinate system and the basic spatial structure's three-dimensional coordinate system. Through this mapping relationship, the initial feature points in each frame of the image are converted from pixel coordinates to three-dimensional spatial coordinates in the basic spatial structure's three-dimensional coordinate system. Finally, the three-dimensional spatial coordinates of the feature points in all frames of the image are summarized, categorized and stored according to the image frame number and the feature point number, resulting in the three-dimensional spatial distribution data of the feature points.

[0033] Step 3.2: Based on the 3D spatial distribution data of feature points, calculate the spatial distance between each feature point and its neighboring feature points, construct feature point density distribution data, identify clustered regions where the spatial distance is less than a preset threshold, and mark redundant feature points within the clustered regions. Specifically, this includes: First, retrieving the generated 3D spatial distribution data of feature points, defining a spherical neighborhood for each feature point with a radius of 0.3 meters; then, calculating the 3D linear distance between each feature point and all other feature points in its neighborhood; next, based on the calculated spatial distance data, counting the number of feature points in the neighborhood of each feature point, and constructing... Feature point density distribution data visually reflects the density distribution of feature points within the basic spatial structure. In this implementation, the preset spatial distance threshold is 0.1 meters. When the spatial distance between two feature points is less than 0.1 meters, these two feature points are determined to be in a clustered state. Then, the density distribution data of all feature points are traversed to identify clustered areas where the number of feature points exceeds a preset standard. The preset standard is that the number of feature points in each spherical neighborhood with a radius of 0.3 meters is greater than 10. Finally, all feature points in the clustered areas except for the feature point with the highest response value are marked, and the marked feature points are determined to be redundant feature points.

[0034] Step 3.3: Based on redundant feature points, gradually eliminate redundant feature points within clustered regions in ascending order of feature point response values ​​until the feature point density within each clustered region drops below a preset density threshold, obtaining a filtered feature point set. Specifically, this includes: First, retrieving the labeled redundant feature point data, and simultaneously retrieving the response value data corresponding to each feature point. This response value reflects the corner strength or edge gradient change amplitude of the feature point; a higher response value indicates stronger feature point recognition and stability. Then, in ascending order of feature point response values, further refine the redundant feature points within the clustered regions. The process involves sorting the feature points. The preset feature point density threshold is that the number of feature points within a spherical neighborhood with a radius of 0.3 meters should not exceed 8. Then, starting with the redundant feature point with the lowest response value, feature points within the clustered area are gradually removed. After removing one feature point at a time, the feature point density of that clustered area is recalculated. When the feature point density of a clustered area drops below the preset density threshold, the feature point removal operation for that area is stopped. Redundant feature point removal is then performed on all identified clustered areas sequentially. Finally, all feature points after removing redundant feature points are summarized to obtain the filtered feature point set.

[0035] Step 3.4: Based on the filtered feature point set, analyze the areas in the basic spatial structure not covered by feature points, calculate the number of feature points in each grid cell, and identify sparse texture regions where the number of feature points is lower than a preset sparsity threshold. Specifically, this includes: First, retrieving the obtained filtered feature point set and simultaneously retrieving the three-dimensional coordinate system parameters of the basic spatial structure. Divide the basic spatial structure into three-dimensional grid cells, setting the size of each grid cell to 0.5m × 0.5m × 0.5m to ensure that the divided grid can fully cover the entire transmission line corridor mapping area; then, project all feature points from the filtered feature point set... The image is projected into the pre-defined 3D mesh cells, and the number of feature points contained in each mesh cell is counted cell by cell. In this implementation, the preset sparsity threshold is that the number of feature points in each mesh cell is less than 3. Next, the feature point count statistics of all mesh cells are traversed, and the regions corresponding to the mesh cells with the number of feature points below the preset sparsity threshold are identified as texture sparse regions. Texture sparse regions mainly correspond to areas with scarce texture information, such as bare rock areas and sparse vegetation areas in the transmission line corridor. Finally, the coordinates of all identified texture sparse regions are marked to clarify the specific location and range of each sparse region in the basic spatial structure.

[0036] Step 3.5: For sparse texture regions, feature point extraction is performed again within these regions, with a focus on enhancing feature point detection in edge, corner, and structured texture regions, and supplementing with new feature points. Specifically, this includes: First, retrieving the coordinate range data of the marked sparse texture regions and extracting the corresponding image sub-blocks from the continuous frame image sequence; then, adjusting the feature extraction parameters for the extracted image sub-blocks to improve the sensitivity of the edge detection algorithm, reducing the edge gradient threshold to 60% of the normal value, while simultaneously enhancing the response strength of the corner detection algorithm and expanding the coverage of corner detection; next, supported by the adjusted feature extraction algorithm, feature points are extracted again from the image sub-blocks in the sparse texture regions, focusing on detecting edge contours and corner structures, even in areas with subtle texture changes; then, the re-extracted feature points are preliminarily screened to remove false feature points generated during detection, ensuring that the supplemented feature points possess true and effective feature description information; finally, the screened new feature points are converted into three-dimensional coordinates in the basic spatial structure three-dimensional coordinate system, completing the feature point supplementation extraction work for sparse texture regions.

[0037] Step 3.6 involves fusing the filtered feature point set with the newly extracted feature points, and then verifying the spatial uniformity of the fused feature point set. When the verification results show that the feature point distribution meets the preset uniformity index, a uniformly distributed optimized feature point set is obtained. Specifically, this includes: first, merging the obtained filtered feature point set with the newly extracted feature points to obtain the fused feature point set; then, performing spatial uniformity verification on the fused feature point set. In this implementation, the preset uniformity index is that the difference in the number of feature points between any two adjacent grid cells does not exceed two. Furthermore, the number of feature points in more than 90% of the grid cells is within the range of 3 to 8. Next, the fused feature point set is projected into a 0.5m × 0.5m × 0.5m three-dimensional grid cell, and the number of feature points in each grid cell is counted cell by cell. The difference in the number of feature points between adjacent grid cells is calculated, and the proportion of grid cells with the number of feature points within the preset range is also counted. Finally, it is determined whether the statistical results meet the preset uniformity index. When the statistical results fully meet the index requirements, the fused feature point set is determined to be uniformly distributed, and the feature point set is output as the optimized feature point set.

[0038] In this embodiment of the invention, the technical means of obtaining three-dimensional distribution data by projecting feature points based on the basic spatial structure composed of grounding bolts, calculating neighborhood distances to identify clustered areas and removing redundant feature points according to response values, analyzing grid cells to identify sparse texture areas and supplementing feature points, and completing spatial uniformity verification after fusing feature points, effectively overcomes the technical problems of clustered redundancy, missing feature points in sparse texture areas, and high mismatch rate in subsequent inter-frame matching caused by uneven spatial distribution of feature points in traditional feature point extraction. Thus, it achieves the goal of obtaining a uniformly distributed and effective optimized feature point set, improving the accuracy and stability of feature point matching.

[0039] In a preferred embodiment of the present invention, step 4 above may include: step 4.1, receiving a uniformly distributed set of optimized feature points, sorting the optimized feature point set according to the image acquisition time order, and constructing a data structure for the correspondence between feature points between adjacent frame images. Specifically, this includes: first, receiving the output uniformly distributed set of optimized feature points, wherein the feature point set contains the three-dimensional coordinates and feature description information of the feature points corresponding to each frame image in the continuous frame images of the transmission line corridor, and each feature point set is accompanied by a corresponding image acquisition timestamp and frame number; subsequently, reading the image acquisition timestamps of all optimized feature point sets, sorting them from earliest to latest according to the timestamps. The optimized feature point sets corresponding to all frames are reordered to form a sequence of feature point sets arranged according to the acquisition time, with the sequence order completely consistent with the order of the images acquired by the UAV during flight. Next, based on the sorted feature point set sequence, a data structure for the correspondence between feature points of adjacent frames is constructed. In specific operation, the optimized feature point set of the nth frame image is associated with the optimized feature point set of the (n+1)th frame image using the frame number as an index, and a feature point mapping index table between frame n and frame n+1 is established. The index table contains the number correspondence between the feature points of the two frames, and reserves fields for feature point similarity and matching tags.

[0040] Step 4.2: Based on the feature point correspondence data structure, feature point matching is performed between the optimized feature point sets of adjacent frame images to obtain a preliminary set of feature point matching pairs. Specifically, this includes: First, retrieving the constructed adjacent frame feature point correspondence data structure for the optimized feature point sets of the nth and (n+1)th frame images; then, matching is performed using feature descriptor similarity comparison. The descriptor of each feature point in the nth frame image is compared with the descriptors of all feature points in the (n+1)th frame image to calculate the similarity. The similarity calculation is based on the vector distance between the feature descriptors. As the basis for judgment, the smaller the vector distance, the higher the similarity. In this implementation, the preset similarity threshold is 0.8. When the descriptor similarity of two feature points is greater than 0.8, the two feature points are judged to be a preliminary matching pair. Then, all feature points of the nth frame image are traversed, and the similarity comparison with the feature points of the (n+1)th frame image is completed. All feature point pairs that meet the similarity threshold requirements are summarized and recorded. Finally, in the same way, the matching operation of the optimized feature point set of all adjacent frame images is completed in turn to obtain a preliminary feature point matching pair set covering all adjacent frames.

[0041] Step 4.3 involves performing geometric consistency verification on the preliminary feature point matching pair set, removing mismatched point pairs that do not conform to geometric constraints, and obtaining the feature point matching results. Specifically, this includes: First, retrieving the obtained preliminary feature point matching pair set and performing geometric consistency verification on the preliminary matching pairs, removing mismatched point pairs that do not conform to geometric constraints; then, setting the key parameters of the random sampling consensus algorithm, with the number of iterations set to 2000 and the interior point determination threshold set to 0.5 meters, which represents the maximum allowable reprojection error of the feature point pair; the specific verification process involves randomly selecting several matching pairs from the preliminary matching pair set and calculating adjacent... The initial geometric transformation relationship between frame images is determined, and then all preliminary matching pairs are substituted into this transformation relationship to calculate the reprojection error of each matching pair. Matching pairs with a reprojection error less than 0.5 meters are determined as inliers, and matching pairs with a reprojection error greater than 0.5 meters are determined as outliers. Then, the above sampling calculation process is repeated until 2000 iterations are completed. The geometric transformation relationship obtained from the calculation with the most inliers is selected as the optimal geometric constraint relationship. Finally, based on the optimal geometric constraint relationship, all outliers, i.e., mismatched point pairs, are eliminated, and all inliers, i.e., correct matching pairs that meet the geometric constraints, are retained to obtain the feature point matching result.

[0042] Step 4.4: Based on the feature point matching results, and using the foundational spatial structure formed by the grounding bolts of the three power transmission towers as a scale reference, calculate the relative transformation matrix between adjacent frame images. Specifically, this includes: First, retrieving the obtained feature point matching results and using the correct matching pairs to calculate the initial relative transformation matrix between adjacent frame images. This matrix reflects the spatial position and attitude changes between two adjacent frame images. Then, retrieving the constructed foundational spatial structure data composed of the grounding bolts of the three power transmission towers. The absolute spatial coordinates of the three vertices of the foundational spatial structure are known, and the actual physical distance between the three vertices is a fixed value. This is used as a scale reference benchmark to solve the scale drift problem that easily occurs in visual SLAM. Next, comparing the initial relative transformation matrix with the scale reference benchmark of the foundational spatial structure, calculating the scale factor deviation of the initial transformation matrix, and then performing scale calibration on the initial relative transformation matrix based on the deviation to ensure that the scale of the transformation matrix is ​​consistent with the real physical space scale. Finally, after completing the scale calibration, a relative transformation matrix containing accurate rotation and translation information between adjacent frame images is obtained. This matrix accurately reflects the spatial motion relationship of the UAV at the acquisition time of two adjacent frame images.

[0043] Step 4.5 converts the relative transformation matrix into inter-frame motion pose parameters for the UAV, including rotation matrices and translation vectors. The motion pose parameters are then smoothed and outlier detected to obtain the UAV's inter-frame motion pose. Specifically, this involves: first, decomposing the obtained relative transformation matrix into rotation matrices and translation vectors. The rotation matrix reflects the UAV's attitude change, and the translation vector reflects its position change; together, they constitute the UAV's inter-frame motion pose parameters. Then, the inter-frame motion pose parameters are smoothed using a sliding window filtering algorithm. The sliding window size is set to 5 frames, meaning the mean of the pose parameters over 5 consecutive frames is calculated, and the calculated mean is used to replace... The original pose parameters within the window are used to eliminate pose parameter fluctuations caused by slight jitter during UAV flight. Next, outlier detection is performed on the smoothed pose parameters. The preset outlier judgment threshold is a change in translation vector greater than 0.2 meters or a change in rotation matrix angle greater than 2 degrees. When the change in pose parameters of a certain frame compared to the adjacent frame exceeds this threshold, the pose parameters of that frame are judged as outliers, and the outlier is replaced by the interpolation result of the normal pose parameters of the preceding and following frames. Finally, after completing the smoothing and outlier detection of all inter-frame motion pose parameters, the continuity of the processed pose parameters is verified to ensure that the changes in pose parameters of adjacent frames are smooth and without abrupt changes, thus obtaining the inter-frame motion pose of the UAV.

[0044] In this embodiment of the invention, by employing techniques such as optimizing the feature point set by sorting it according to the image acquisition time sequence and constructing the correspondence between feature points of adjacent frames, verifying the geometric consistency of the preliminary matching pairs to eliminate mismatched point pairs, using the basic spatial structure formed by grounding bolts as a scale reference to calculate the relative transformation matrix, converting the transformation matrix into pose parameters and performing smoothing and outlier detection, the technical problems of high mismatch rate of feature point matching, lack of stable scale reference for pose estimation leading to scale drift, and insufficient stability due to outliers in pose parameters are effectively overcome in traditional visual SLAM. This results in improved feature point matching and pose estimation reliability, and the acquisition of stable and accurate inter-frame motion pose of the UAV.

[0045] In a preferred embodiment of the present invention, step 5 above may include: Step 5.1, based on the inter-frame motion pose of the UAV and the uniformly distributed optimized feature point set, mapping the optimized feature point set to the three-dimensional space corresponding to the motion pose, analyzing the spatial connection relationship and distribution pattern between feature points, and extracting the topological association characteristic data of the feature point set in space, specifically including: First, receiving the output inter-frame motion pose data of the UAV, the motion pose data includes the stable rotation matrix and translation vector of the UAV at the acquisition time of each two adjacent frames, along with the corresponding frame number, acquisition timestamp, and pose accuracy verification information; simultaneously receiving the uniformly distributed optimized feature point set, the optimized feature point set includes the pixel coordinates of the feature points in each frame image, feature description information, and the preliminary three-dimensional coordinates of the feature points in the basic spatial structure three-dimensional coordinate system; then, mapping the optimized feature point set to the three-dimensional space corresponding to the inter-frame motion pose of the UAV, specifically, combining the intrinsic parameters of the visible light camera, converting the pixel coordinates of the feature points into three-dimensional coordinates in the camera coordinate system, and then using the motion pose parameters of the UAV at the corresponding time, mapping the camera... The coordinates are converted from the original coordinate system to three-dimensional spatial coordinates in the geodetic coordinate system, completing the spatial mapping of all optimized feature points and obtaining accurate three-dimensional spatial location data of the feature points. Next, the spatial connectivity between feature points is analyzed, and a spherical neighborhood is defined for each feature point with a radius of 0.3 meters. The spatial straight-line distance between each feature point and all feature points in its neighborhood is calculated, and the number of neighborhood connections for each feature point is counted. The number of neighborhood connections reflects the spatial correlation strength between feature points; the more connections, the stronger the correlation of the feature point in the spatial topology. At the same time, the spatial distribution pattern of feature points is analyzed. The three-dimensional space is divided into three-dimensional grid cells with a size of 0.5 meters × 0.5 meters × 0.5 meters. All feature points are projected onto the corresponding grid cells, and the number of feature points in each grid cell is counted. Clustered and sparse regions of feature points are identified, and the spatial range of clustered regions and the size of feature point clusters are recorded. Finally, the data such as the number of neighborhood connections, neighborhood spatial distances, feature point cluster distribution range, and feature point density in the grid cells are integrated to form the topological correlation characteristic data of the feature point set in space.

[0046] Step 5.2: Based on the topological association characteristic data and using the foundation spatial structure formed by the grounding bolts of the three transmission towers as geometric constraints, a robust kernel function energy optimization function is constructed. First, the generated feature point topological association characteristic data is retrieved, along with the constructed foundation spatial structure data consisting of the grounding bolts of the three transmission towers. The foundation spatial structure is a triangular structure, with the absolute spatial coordinates of its three vertices known, and the geometric parameters such as the side length and azimuth angle of the triangle fixed. This serves as the geometric benchmark for the topological consistency constraint. Then, considering the noise resistance characteristics of the robust kernel function, the Huber kernel function is selected as the core kernel function of the energy optimization function. , , It is the threshold parameter in the same segmentation condition. This is the original input error value. The threshold of this kernel function is set to 1.0 to effectively suppress the interference of outliers in the topological association characteristic data on the optimization results, and to avoid the influence of feature point topological deviations in bare rock and sparse vegetation areas on the pose optimization accuracy. Next, an energy optimization function is constructed, in which the pose error term is calculated based on the deviation between the actual topological association characteristics of the feature points and the theoretical topological association characteristics derived from the current UAV motion pose. The deviation calculation covers dimensions such as the deviation of the number of neighborhood connections, the deviation of the feature point cluster size, and the deviation of the feature point density of the grid cell. The larger the deviation value, the lower the fit between the current motion pose and the actual spatial topology. Topology 1 The consistency constraint term is calculated based on the matching deviation between the spatial topology of feature points and the geometric constraints of the basic spatial structure. It focuses on constraining the spatial orientation deviation between the feature point clusters and the basic spatial structure, as well as the scale deviation between the feature point cluster region and the basic spatial structure, ensuring that the feature point topology conforms to the geometric laws of the basic spatial structure. Simultaneously, weighting coefficients are set for two terms in the energy optimization function: the pose error term is weighted at 0.7, and the topology consistency constraint term is weighted at 0.3. This highlights the core objective of pose error correction while ensuring topological consistency, ultimately forming a complete robust kernel function energy optimization function. ,in, These are the inter-frame motion pose parameters of the UAV that need to be optimized. It is the total number of three-dimensional mesh units after the basic spatial structure is divided. It is the first The weighting coefficients of the pose error term for each mesh cell, with a range of values. , It is the balance coefficient of the geometric consistency constraint term, and it is the total number of adjacent mesh element pairs. It is the first The weight coefficients of the geometric constraint terms for adjacent grid cells are taken as the average of the weights of the two adjacent grid cells. It is the first Geometric consistency error term for adjacent grid cells.

[0047] Step 5.3: Based on the energy optimization function, the error adjustment value of the current motion pose is obtained by analyzing the geometric deviation between the topological correlation characteristics and the basic spatial structure. Specifically, this includes: First, substituting the current inter-frame motion pose parameters of the UAV into the constructed energy optimization function to calculate the initial energy value. The initial energy value reflects the degree of deviation between the feature point topology and the theoretical topology and the geometric constraints of the basic spatial structure under the current motion pose. Then, the geometric deviation between the topological correlation characteristics and the basic spatial structure is analyzed. Specifically, the key parameters in the feature point topology are compared with the geometric parameters of the basic spatial structure, including the spatial positional deviation between the centroid of the feature point cluster and the centroid of the basic spatial structure, the azimuth angle of the feature point cluster and the azimuth angle of the basic spatial structure, and the scale of the feature point cluster region and the basic spatial structure. The scale deviation of the spatial structure was assessed by setting geometric deviation thresholds, positional deviation thresholds of 0.2 meters, azimuth deviation thresholds of 2 degrees, and scale deviation thresholds of 0.1 times the scale of the basic spatial structure. All deviation areas exceeding these thresholds were marked; these areas represent the pose deviation areas requiring focused adjustment. Next, with the goal of minimizing the energy optimization function value, the error adjustment value for the current motion pose was solved iteratively. In each iteration, the rotation and translation components of the UAV's motion pose were adjusted according to the direction and magnitude of the geometric deviation. The adjustment step size for the rotation component was set to 0.05 degrees, and the adjustment step size for the translation component was set to 0.01 meters. After adjustment, the energy value was recalculated, and the changes in energy value before and after adjustment were compared. When the difference in energy value calculated in two consecutive iterations is less than 1 × 10⁻⁶, the adjustment is considered complete. -6 When the iteration converges, the calculation is stopped, and the adjustment angle of the rotation component and the adjustment distance of the translation component in the last iteration are taken as the final error adjustment value.

[0048] Step 5.4 applies the error adjustment value to the inter-frame motion pose of the UAV, correcting the rotation and translation components of the motion pose to obtain corrected motion pose data. Specifically, this includes: First, retrieving the calculated error adjustment value and decomposing it into rotation adjustment angles and translation adjustment distances, where the rotation adjustment angle is in degrees and the translation adjustment distance is in meters. Then, retrieving the output original inter-frame motion pose parameters of the UAV, for the rotation component, fine-tuning the original rotation matrix according to the rotation angle in the error adjustment value to correct UAV attitude deviations and ensure the rotation angle matches the actual flight attitude. For the translation component, correcting the original translation vector according to the translation distance in the error adjustment value to adjust the UAV position coordinates and eliminate position offset errors. During the correction process, the motion pose parameters of all adjacent frames are processed frame by frame. After completing the pose correction for each frame, the corrected rotation matrix and translation vector are recorded, while the original pose parameters are retained as a backup. Finally, the corrected pose parameters of all frames are summarized to obtain the preliminary corrected motion pose data.

[0049] Step 5.5: Perform convergence verification on the corrected motion pose data. When the verification result meets the preset accuracy threshold requirements, the corrected motion pose is obtained. Specifically, this includes: First, setting the preset accuracy threshold for convergence verification. In this implementation, the convergence accuracy threshold for the rotation component is a corrected rotation angle deviation of less than 0.1 degrees, and the convergence accuracy threshold for the translation component is a corrected translation distance deviation of less than 0.05 meters. Then, perform convergence verification on the corrected motion pose data. Specifically, calculate the change in pose parameters between adjacent frames after correction and compare the magnitude of this change with the preset accuracy threshold. Simultaneously, continuously verify 5 frames of corrected motion pose data and observe their changing trends. If the change in rotation angle is less than 0.1 degrees and the change in translation distance is less than 0.05 meters for 5 consecutive frames, and the pose parameters show no significant fluctuations, the convergence verification is considered successful. If the above conditions are not met, recalculate the error adjustment value and execute the correction process again until the verification result meets the preset accuracy threshold. Finally, when the convergence verification is successful, output the final corrected motion pose.

[0050] In this embodiment of the invention, by employing techniques such as mapping the optimized feature point set to three-dimensional space to extract topological correlation characteristics, combining the basic spatial structure formed by grounding bolts as geometric constraints, constructing an energy optimization function containing a robust kernel function and including pose error terms and topological consistency constraint terms, and then calculating pose error adjustment values ​​by analyzing geometric deviations, the invention effectively overcomes the problems of chaotic topological correlation of feature points, lack of reliable geometric constraints in pose optimization, weak noise resistance leading to difficulty in correcting pose accumulation errors, and insufficient optimization accuracy in the prior art. This achieves accurate capture of pose deviation sources, improves the pertinence and accuracy of pose error adjustment, and enhances the stability and reliability of motion pose.

[0051] In a preferred embodiment of the present invention, step 6 above may include: Step 6.1, sorting and storing the corrected motion pose data according to the time sequence to construct a motion trajectory dataset of the UAV during its flight along the power transmission line corridor. Specifically, this includes: First, the 3D modeling processing unit receives the corrected motion pose data, which contains the corrected rotation matrix and translation vector of the UAV at the acquisition time of each two adjacent frames, along with the corresponding frame number, image acquisition timestamp, and flight status information; then, reading the acquisition timestamps of all corrected motion pose data, and sorting the pose data corresponding to all frames in order from earliest to latest timestamp. The sorting result is completely consistent with the time sequence of the images acquired by the UAV during its flight along the power transmission line corridor, ensuring the continuity of the trajectory data; next, organizing and storing the sorted pose data, associating each frame number with the corresponding corrected pose parameters, acquisition time, and index information of the corresponding image, and supplementing the pose data of the UAV's takeoff point and landing point to construct a complete UAV flight motion trajectory dataset; the dataset clearly presents the spatial position changes and attitude adjustment patterns of the UAV throughout the entire mapping process.

[0052] Step 6.2: Based on the motion trajectory dataset and combined with the optimized feature point set corresponding to each frame image, project the feature points in the optimized feature point set onto the geodetic coordinate system to obtain the initial 3D point cloud data set. Specifically, this includes: First, retrieving the constructed motion trajectory dataset, and simultaneously retrieving the optimized feature point set corresponding to each frame image, as well as the intrinsic parameters of the visible light camera, the transformation relationship data between the basic spatial structure and the geodetic coordinate system; then, for the optimized feature point set of each frame image, combined with the corrected motion pose parameters of the UAV at the corresponding time, establishing the projection relationship between the feature point pixel coordinates and the geodetic coordinate system; in specific operation, first converting the feature points from image pixel coordinates to 3D relative coordinates in the UAV coordinate system, and then converting the 3D relative coordinates to absolute 3D coordinates in the geodetic coordinate system using the UAV geodetic coordinates and attitude information in the motion trajectory data; completing the coordinate transformation of all optimized feature points frame by frame, summarizing the geodetic coordinates of the feature points in all frames, classifying and recording them according to the frame number and feature point number, with each feature point associated with its corresponding geodetic coordinates, feature description information, and the image frame information to which it belongs; finally, integrating all feature point data to obtain the initial 3D point cloud data set.

[0053] Step 6.3 involves spatial filtering the initial 3D point cloud dataset to remove outliers and noise points. Simultaneously, the point cloud data is registered and aligned based on the geometric constraints of the basic spatial structure to obtain a preliminarily registered 3D point cloud model. Specifically, this includes: First, spatial filtering is performed on the initial 3D point cloud dataset, using a statistical filtering algorithm to remove outliers and noise points. Filtering parameters are set, with the number of neighboring points for each point cloud set to 15. The average distance between each point and its 15 neighboring points is calculated, with a preset distance threshold of 2 meters. When the average distance between a point and its neighboring points exceeds 2 meters, it is identified as an outlier or noise point and removed from the point cloud dataset. After filtering and eliminating some data, valid point clouds that conform to spatial distribution patterns are retained. Then, the basic spatial structure data is retrieved, and the point cloud data is registered and aligned using the absolute spatial coordinates of the grounding bolts at the bases of the three transmission towers as geometric constraints. By calculating the relative positional deviation between the point cloud data and the basic spatial structure, the spatial position of the point cloud is adjusted to ensure that the coordinates of the points corresponding to the grounding bolt areas in the point cloud are consistent with the known absolute coordinates. Simultaneously, the coordinate deviation of overlapping areas in adjacent frame point clouds is ensured to be less than 0.3 meters, achieving global registration and alignment of the point cloud data. Finally, the filtered and registered point cloud data are integrated to obtain a preliminary registered 3D point cloud model.

[0054] Step 6.4: Based on the initially registered 3D point cloud model, the point cloud model is densified. The positional accuracy of each 3D point in the point cloud is optimized using bundle adjustment, while maintaining the geometric structural characteristics of the transmission line corridor terrain and features, resulting in an optimized dense 3D point cloud model. Specifically, this includes: First, based on the initially registered 3D point cloud model, densification processing is carried out. A dense point cloud generation method based on image feature matching is used to supplement the point cloud data in sparse texture areas. For texture-deficient areas such as bare rock and sparse vegetation in mountainous areas, the adjacent frames are enhanced. Image feature matching is used to mine spatial points corresponding to weak texture information, supplementing and generating dense point clouds to ensure the density uniformity of the point cloud model, with no less than 10 point clouds per square meter. Subsequently, the accuracy of the densed point cloud is optimized using bundle adjustment. UAV motion trajectory data, camera intrinsic parameters, and point cloud 3D coordinates are substituted into the bundle adjustment process to build a complete bundle adjustment optimization workflow. By synchronously adjusting the spatial coordinates of each 3D point and the UAV pose parameters, the point cloud reprojection error is minimized, with a preset reprojection error threshold of 0.0. The reprojection error of each point cloud is less than 0.02 meters, ensuring a distance of 2 meters. During the optimization process, the geometric characteristics of key facilities such as transmission lines, towers, and grounding bolts are preserved to avoid structural distortion caused by optimization. Finally, after optimization, a dense 3D point cloud model with uniform density and accurate positioning is obtained. The bundle adjustment optimization process focuses on minimizing the point cloud reprojection error. It first imports the densed 3D point cloud, the corrected UAV pose, camera intrinsic parameters, actual pixel coordinates of feature points, and absolute coordinates of key facilities, and unifies them to the ground plane. The coordinate system is then used to project each point cloud onto the image plane after pose transformation and distortion correction to obtain theoretical pixel coordinates. The reprojection error is calculated by comparing the theoretical pixel coordinates with the actual pixel coordinates, and point clouds with errors exceeding 0.02 meters are marked. Rigid constraints are applied to the point clouds of critical facilities such as power transmission lines, towers, and grounding bolts to limit the adjustment range, while fixing the known coordinates of the point clouds. Simultaneously, the UAV pose and the coordinates of ordinary terrain point clouds are adjusted. Based on the error distribution, high-error objects are prioritized for correction, and higher weights are assigned to the point clouds of critical facilities. The process is completed when the number of iterations reaches 200 or the average error change between two consecutive iterations is less than 1 × 10⁻⁶. -6 The convergence condition is set at 0.02 meters. Finally, all point cloud reprojection errors are re-verified to ensure they are less than 0.02 meters and that the geometric deviations of key facilities meet the standards. After verification, the optimized pose parameters and dense 3D point cloud are output.

[0055] Step 6.5 involves surface reconstruction of the dense 3D point cloud model to obtain a complete 3D mesh model of the transmission line corridor's terrain, vegetation, buildings, and transmission facilities. Specifically, this includes: first, retrieving the optimized dense 3D point cloud model and using the Poisson reconstruction algorithm for surface reconstruction, setting reconstruction accuracy parameters, and setting the sampling density to 0.05 meters to ensure the reconstructed surface accurately matches the spatial distribution of the point cloud; during reconstruction, constructing a continuous 3D surface using the point cloud's normal vector information to distinguish different types of features within the transmission line corridor, including terrain, vegetation, buildings, and transmission facilities; and then sampling for different features. Using a differentiated reconstruction strategy, the terrain areas retain their natural undulating shape, while the vegetation areas are simplified to preserve their overall outline and distribution range, avoiding excessive detailing that could lead to redundant model data. Transmission facilities, including towers, lines, and grounding bolts, undergo enhanced detail reconstruction to ensure the structural integrity and morphological accuracy of key components. After reconstruction, a complete 3D mesh model containing all land cover types is generated. This 3D mesh model consists of continuous polygonal faces, visually representing the 3D spatial morphology of the transmission line corridor. Finally, the 3D mesh model is smoothed to eliminate surface burrs and irregular protrusions, improving the model's visual appeal and usability.

[0056] Step 6.6: Based on the complete 3D mesh model, and using the spatial coordinates of the three transmission tower foot grounding bolts as control points, perform absolute coordinate correction and accuracy verification on the 3D mesh model. When the verification results meet the preset mapping accuracy requirements, obtain the 3D mapping results data of the transmission line corridor. Specifically, this includes: First, retrieving the complete 3D mesh model and simultaneously retrieving the absolute spatial coordinates of the three transmission tower foot grounding bolts. Using these three coordinates as absolute control points, perform absolute coordinate correction on the 3D mesh model. By calculating the deviation between the corresponding grounding bolt positions in the mesh model and the absolute coordinates of the control points, adjust the spatial position of the entire mesh model so that the coordinates of the grounding bolts in the model are completely consistent with the known absolute coordinates. The error of the corrected control point coordinates is less than 0.01 meters. Subsequently, perform precision... The accuracy verification process involves setting preset thresholds for mapping accuracy. The threshold for key component positioning accuracy is 0.03 meters, and the threshold for overall model spatial deviation is 0.05 meters. Ten key detection points, including transmission line insulators, tower crossarms, and grounding bolts, are selected. The deviation between their coordinates in the mesh model and their actual absolute coordinates is measured, while the overall spatial distortion of the model is also detected. If the coordinate deviation of all key detection points is less than 0.03 meters, and the overall spatial deviation of the model is less than 0.05 meters, the accuracy verification is considered successful, meeting the millimeter-level measurement requirements for transmission line operation and maintenance. If the accuracy requirements are not met, point cloud optimization and surface reconstruction are performed again until the verification is successful. Finally, after the accuracy verification is successful, the 3D mapping results of the transmission line corridor, including the 3D mesh model, point cloud data, and coordinates of key components, are output.

[0057] In this embodiment of the invention, by employing technical means such as constructing a UAV motion trajectory dataset, projecting optimized feature points onto a geodetic coordinate system to generate an initial point cloud, removing noise outliers through spatial filtering and registering the point cloud with the basic spatial structure, implementing densification and bundle adjustment to optimize point cloud accuracy, performing surface reconstruction to form a three-dimensional mesh model, and relying on grounding bolt control points to complete absolute coordinate correction and accuracy verification, the invention effectively overcomes the technical problems of traditional three-dimensional point cloud models, including interference from noise points and outliers, insufficient registration and alignment accuracy, model offset due to lack of absolute coordinate constraints, easy distortion of terrain and ground feature geometry, and difficulty in meeting the accuracy requirements of power transmission line operation and maintenance measurements. This achieves the technical effect of generating high-precision, high-completeness three-dimensional mapping results for power transmission line corridors, accurately restoring the true form of terrain, vegetation, buildings, and power transmission facilities within the corridor, and providing reliable data support for the safe operation, maintenance, and repair of power transmission lines.

[0058] As shown in Figure 2, embodiments of the present invention also provide a UAV-based 3D mapping system using visual SLAM, comprising: an acquisition module for deploying and measuring three grounding bolts at the base of power transmission towers with known spatial locations on the ground along a power transmission line corridor using a UAV equipped with a visible light camera; an extraction module for constructing a basic spatial structure based on the spatial coordinates of the three grounding bolts at the base of power transmission towers, and acquiring a continuous sequence of images along the power transmission line corridor according to a preset overlap rate; extracting an initial feature point set for each frame of the acquired sequence of images by performing frame-by-frame processing; and a filtering module for filtering the initial feature point set, removing redundant feature points with clustered distribution, and supplementing the extracted feature points in sparse texture areas to obtain an optimized feature point set with uniform distribution. The system comprises: a feature point set; a matching module, used to perform feature point matching between adjacent frames based on the optimized feature point set, and obtain matching results; the UAV's inter-frame motion pose is estimated based on the matching results; a correction module, used to extract the topological correlation characteristics of the feature point set in space based on the motion pose and the obtained uniformly distributed optimized feature point set; an energy optimization function with a robust kernel function is constructed based on the extracted spatial topological correlation characteristics of the feature point set; an error adjustment value is calculated by minimizing the energy optimization function; the motion pose is corrected based on the error adjustment value; and a processing module, used to construct and optimize a 3D point cloud model of the transmission line corridor by fusing the optimized feature point set obtained from multiple frames based on the corrected motion pose, and obtain the measured 3D mapping results.

[0059] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A UAV 3D mapping method based on visual SLAM, characterized in that, The method includes: Step 1, using a drone equipped with a visible light camera, deploying and measuring three grounding bolts of transmission towers with known spatial locations on the ground along the transmission line corridor; Step 2, constructing a basic spatial structure based on the spatial coordinates of the three grounding bolts, and acquiring a continuous sequence of images along the transmission line corridor according to a preset overlap rate; processing the acquired sequence of images frame by frame, and extracting an initial feature point set for each frame; Step 3, filtering the initial feature point set, removing redundant feature points with clustered distribution, and supplementing with feature points extracted from sparse texture areas to obtain a uniformly distributed optimized feature point set; Step 4, based on the optimized feature points... Step 5: Based on the motion pose, combined with the obtained uniformly distributed optimized feature point set, the topological correlation characteristics of the feature point set in space are extracted; based on the extracted spatial topological correlation characteristics of the feature point set, an energy optimization function of robust kernel function is constructed, and an error adjustment value is calculated by minimizing the energy optimization function. The motion pose is corrected by the error adjustment value to obtain the corrected motion pose; Step 6: Based on the corrected motion pose, the optimized feature point set obtained by fusing multiple frames of images is fused to construct and optimize the three-dimensional point cloud model of the transmission line corridor to obtain the measured three-dimensional mapping results.

2. The UAV 3D mapping method based on visual SLAM according to claim 1, characterized in that, Step 1: Using a drone equipped with a visible light camera, deploy and measure the actual grounding bolts of three transmission towers with known spatial locations on the ground along the transmission line corridor. This includes: controlling the drone to fly along a preset route, scanning the ground of the transmission line corridor with the visible light camera, identifying and locating the grounding bolt positions of the three transmission towers, and generating an initial image dataset containing the three grounding bolts; based on the initial image dataset, performing edge detection and center point extraction on the contour features of the three grounding bolts to calculate the two-dimensional pixel coordinates of the three grounding bolts in the image coordinate system; based on the two-dimensional pixel coordinates, combined with the drone's flight altitude and camera intrinsic parameters, converting the two-dimensional pixel coordinates into three-dimensional relative coordinates of the three grounding bolts in the drone's coordinate system; based on the real-time position data provided by the three-dimensional relative coordinates, converting the three-dimensional relative coordinates of the three grounding bolts into absolute spatial coordinates in the geodetic coordinate system; based on the absolute spatial coordinates, performing error analysis and consistency verification on the spatial position accuracy of the three grounding bolts, and outputting the spatial coordinate data of the actual grounding bolts of the three transmission towers with known spatial locations when the verification results meet the preset accuracy threshold.

3. The UAV 3D mapping method based on visual SLAM according to claim 2, characterized in that, Step 2: Based on the physical spatial coordinates of the grounding bolts at the base of the three transmission towers with known spatial locations, a basic spatial structure is constructed, and continuous sequence images are collected along the transmission line corridor according to the preset overlap rate. By processing the acquired sequence of images frame by frame, an initial feature point set is extracted for each frame. This includes: constructing a basic spatial structure composed of three spatial points based on the spatial coordinate data of the grounding bolts at the bases of three power transmission towers, and determining the spatial orientation and scale information of the basic spatial structure in the geodetic coordinate system; planning the flight path of the UAV along the power transmission line corridor based on the spatial orientation and scale information of the basic spatial structure, combined with the geographic information data of the power transmission line corridor, and calculating the distance parameters between adjacent image acquisition points; generating UAV flight control commands based on the flight path and distance parameters and sending them to the UAV to control the UAV to fly along the planned path, while simultaneously triggering the visible light camera to continuously capture images at the calculated acquisition interval, acquiring a sequence of image data streams along the power transmission line corridor; sorting and preprocessing the image data in chronological order according to the sequence of image data streams to eliminate image noise and distortion, resulting in a continuous frame image sequence; and analyzing the continuous frame image sequence frame by frame, processing each frame to identify corner points, edges, and texture regions in the image, and extracting the initial feature point set corresponding to each frame.

4. The UAV 3D mapping method based on visual SLAM according to claim 3, characterized in that, Step 3 above includes: receiving the initial feature point set of each frame image; projecting the feature points onto a three-dimensional coordinate system based on the spatial coordinates of the foundation space structure formed by the grounding bolts of the three transmission towers to obtain the three-dimensional spatial distribution data of the feature points; calculating the spatial distance between each feature point and its neighboring feature points based on the three-dimensional spatial distribution data of the feature points, constructing feature point density distribution data, identifying clustered areas where the spatial distance is less than a preset threshold, and marking redundant feature points within the clustered areas; and gradually eliminating redundant feature points within the clustered areas according to the order of feature point response values ​​from low to high, until the feature point density in each clustered area drops to a preset density threshold. The following steps are taken: First, a filtered feature point set is obtained. Based on this set, the regions in the basic spatial structure not covered by feature points are analyzed. The number of feature points within each grid cell is calculated, and texture sparse regions with fewer feature points than a preset sparsity threshold are identified. For these sparse regions, feature point extraction is performed again, with a focus on enhancing feature point detection at edges, corners, and structured texture regions, and new feature points are extracted. The filtered feature point set is then fused with the newly extracted feature points. Spatial uniformity is verified on the fused feature point set. When the verification results show that the feature point distribution meets the preset uniformity index, a uniformly distributed optimized feature point set is obtained.

5. The UAV 3D mapping method based on visual SLAM according to claim 4, characterized in that, Step 4: Based on the optimized feature point set, feature point matching is performed between adjacent frame images to obtain matching results. The inter-frame motion pose of the UAV is estimated using the matching results, including: receiving a uniformly distributed optimized feature point set, sorting the optimized feature point set according to the image acquisition time order, and constructing a feature point correspondence data structure between adjacent frame images; based on the feature point correspondence data structure, feature point matching is performed between the optimized feature point sets of adjacent frame images to obtain a preliminary set of feature point matching pairs; geometric consistency verification is performed on the preliminary set of feature point matching pairs, and mismatched point pairs that do not conform to geometric constraints are removed to obtain the feature point matching results; based on the feature point matching results, and combined with the foundation spatial structure composed of the grounding bolts of three power transmission towers as a scale reference, the relative transformation matrix between adjacent frame images is calculated; the relative transformation matrix is ​​converted into the inter-frame motion pose parameters of the UAV, including rotation matrix and translation vector, and the motion pose parameters are smoothed and outlier detected to obtain the inter-frame motion pose of the UAV.

6. The UAV 3D mapping method based on visual SLAM according to claim 5, characterized in that, Step 5 above includes: mapping the optimized feature point set to the three-dimensional space corresponding to the motion pose based on the inter-frame motion pose of the UAV and the uniformly distributed optimized feature point set; analyzing the spatial connection relationship and distribution pattern between feature points; and extracting the topological association characteristic data of the feature point set in space. Based on the topological association characteristic data, and combining the basic spatial structure formed by the grounding bolts of the three power transmission towers as geometric constraints, a robust kernel function energy optimization function is constructed. Based on the energy optimization function, the error adjustment value of the current motion pose is obtained by analyzing the geometric deviation between the topological association characteristics and the basic spatial structure. The error adjustment value is applied to the inter-frame motion pose of the UAV, and the rotation and translation components of the motion pose are corrected respectively to obtain the corrected motion pose data. The convergence of the corrected motion pose data is verified, and the corrected motion pose is obtained when the verification result meets the preset accuracy threshold requirement.

7. The UAV 3D mapping method based on visual SLAM according to claim 6, characterized in that, Step 6 above includes: sorting and storing the corrected motion pose data according to the time series to construct a motion trajectory dataset of the UAV during its flight along the power transmission line corridor; based on the motion trajectory dataset, and combined with the optimized feature point set corresponding to each frame image, projecting the feature points in the optimized feature point set onto the geodetic coordinate system to obtain an initial 3D point cloud dataset; performing spatial filtering on the initial 3D point cloud dataset to remove outliers and noise points, and simultaneously registering and aligning the point cloud data according to the geometric constraints of the basic spatial structure to obtain a preliminary registered 3D point cloud model; and based on the preliminary registered 3D point cloud model... The point cloud model is densified, and the positional accuracy of each 3D point in the point cloud is optimized by bundle adjustment while maintaining the geometric characteristics of the terrain and features of the transmission line corridor, resulting in an optimized dense 3D point cloud model. Surface reconstruction is then performed on the dense 3D point cloud model to obtain a complete 3D mesh model of the transmission line corridor terrain, vegetation, buildings, and transmission facilities. Based on the complete 3D mesh model, and using the spatial coordinates of the grounding bolts at the bases of three transmission towers as control points, the 3D mesh model undergoes absolute coordinate correction and accuracy verification. When the verification results meet the preset mapping accuracy requirements, the 3D mapping data of the transmission line corridor is obtained.

8. A visual SLAM-based UAV 3D mapping system, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to deploy and measure the physical grounding bolts of three power transmission towers with known spatial locations on the ground of the power transmission line corridor using a drone equipped with a visible light camera. The extraction module is used to construct a basic spatial structure based on the physical spatial coordinates of the grounding bolts at the base of three transmission towers with known spatial locations, and to collect a continuous sequence of images along the transmission line corridor according to a preset overlap rate; by processing the collected sequence of images frame by frame, an initial set of feature points is extracted for each frame. The filtering module is used to filter the initial feature point set, remove redundant feature points that are clustered and distributed, and supplement the feature points extracted from sparse texture regions to obtain a uniformly distributed optimized feature point set. The matching module performs feature point matching between adjacent frames based on an optimized feature point set to obtain matching results; it then estimates the inter-frame motion pose of the UAV based on the matching results. The correction module extracts the topological correlation characteristics of the feature point set in space based on the motion pose and the uniformly distributed optimized feature point set; it constructs an energy optimization function with a robust kernel function based on the extracted spatial topological correlation characteristics of the feature point set; it calculates an error adjustment value by minimizing the energy optimization function; and it corrects the motion pose using the error adjustment value to obtain the corrected motion pose. The processing module fuses the optimized feature point set obtained from multiple frames based on the corrected motion pose to construct and optimize a 3D point cloud model of the transmission line corridor, obtaining the measured 3D mapping results.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images