An RTK positioning-based unmanned aerial vehicle sag measurement method

Abstract

The application belongs to the technical field of sag optical measurement, and discloses a kind of unmanned plane sag measurement method based on RTK positioning, adopts Beidou / GNSS dual-frequency RTK positioning technology and access network RTK difference link, can realize the ultra-high-precision positioning of horizontal ±5cm, vertical ±5cm;Through Zhang's calibration method and laser plane hard constraint joint completion camera, laser range finder inside and outside parameter dynamic calibration, unify sensor timing and space coordinate reference, eliminate sensor assembly deviation and system initial error;Fusion double branch U-Net network, Canny edge detection and Zernike matrix subpixel iterative algorithm, realize 0.1 pixel level wire edge fine positioning;Through airborne edge computing unit, realize image, laser point cloud, RTK pose multi-source data field real-time fusion processing, and sag parameter instant solution output, can quickly adapt to emergency inspection, line acceptance, batch detection and other efficient operation scene.

Description

Technical Field

[0001] This invention belongs to the field of sag optical measurement technology, specifically a UAV sag measurement method based on RTK positioning. Background Technology

[0002] Sag in overhead transmission lines specifically refers to the vertical difference between the line connecting the suspension points on both sides of the conductor and the lowest point of the conductor. It is crucial to the quality of line installation and operational safety. During actual operation, excessive sag will result in insufficient distance between the conductor and the ground, and insufficient safe clearance to crossings, easily leading to power accidents such as line discharge and short-circuit tripping. Conversely, insufficient sag will cause the conductor's operating tension to exceed the design threshold, potentially leading to conductor breakage and tower overturning under prolonged high loads. Various conventional measurement technologies have been developed for sag detection of transmission lines both domestically and internationally. While UAV sag measurement solutions often utilize single visible light photogrammetry or simple laser ranging, the following technical challenges remain: The positioning accuracy is difficult to meet the standard. Conventional GPS positioning mode can only achieve meter-level positioning effect, which cannot be adapted to the high-precision measurement standard of centimeter-level power sag, and the basic positioning error is too large. Optical detection methods are limited, and most existing mainstream solutions use monocular vision imaging to collect data without integrating multi-dimensional detection methods such as laser ranging and stereo vision. This results in limited data collection dimensions and low error tolerance. Image processing algorithms are limited in accuracy, traditional image recognition technology has weak anti-interference ability, and the extraction of conductor edge features is easily affected by complex background environments such as mountains, forests, sky, and ground buildings. Sub-pixel fine positioning ability is insufficient, which further reduces the accuracy of point measurement. Lacking real-time calculation capabilities, traditional equipment needs to transmit all the image data collected on-site back to the ground terminal for post-processing and calculation. It cannot achieve real-time calculation and real-time results on the job site, resulting in poor measurement timeliness and difficulty in meeting the needs of efficient operation scenarios such as emergency inspection and batch acceptance. Summary of the Invention

[0003] The purpose of this invention is to provide a method for measuring the sag of a UAV based on RTK positioning, so as to solve one or more problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for measuring the sag of a UAV based on RTK positioning, comprising the following specific steps: Furthermore, during the system calibration and configuration phase, the BeiDou / GNSS dual-frequency RTK module is activated and connected to the network RTK differential link to provide a unified coordinate system and timing reference for optical measurements. The calibration method of Zhang's calibration combined with the laser plane hard constraint is adopted to complete the dynamic calibration of the intrinsic and extrinsic parameters of the camera and laser rangefinder, and generate a coordinate mapping matrix. The ground control unit completes the configuration of dedicated optical parameters, sets the laser sampling frequency of 10Hz, camera high-definition imaging parameters, shutter speed of no less than 1 / 1000s, sub-pixel recognition accuracy threshold, and optical parallax allowable deviation threshold, and completes the calibration of the dedicated optical measurement system; during the equipment initialization process, the sensor timing calibration is completed simultaneously, the working clock of the equipment is collected uniformly, and the sensor working temperature, power supply status and signal transmission link are self-checked.

[0005] Furthermore, in the optical positioning modeling stage, the UAV is controlled to fly to the safe optical measurement area to the side of the distance to be measured, and images of the conductor area are continuously acquired by the camera. After the images of the conductor area are preprocessed by adaptive optics contrast to suppress ambient light interference, they are input into a dual-branch U-Net network to complete semantic segmentation of the conductor area and extraction of conductor edge features, thereby achieving coarse positioning of the conductor and suspension point. The Canny edge detection and Zernike sub-pixel iterative algorithm are fused to complete 0.1-pixel-level conductor edge positioning. The adaptive optics contrast preprocessing dynamically adjusts the image contrast and brightness parameters based on the real-time light intensity and background brightness distribution of the image. It automatically adapts processing strategies for different lighting scenarios such as direct sunlight, shadow occlusion, and uneven lighting, enhances the grayscale difference between the conductor and the background environment, filters image noise caused by changes in ambient lighting, and weakens background interference while preserving the core linear features of the conductor.

[0006] The laser rangefinder is started to continuously scan and acquire optical distance data of each point on the conductor. The two-dimensional pixel optical coordinates are converted into three-dimensional spatial coordinates through the pre-calibrated intrinsic and extrinsic parameter matrix. The RANSAC algorithm is used to remove optical measurement outliers, the conductor spatial orientation equation is fitted, and optical data registration is completed by combining RTK real-time spatiotemporal data to construct the conductor optical spatial model. The system filters and acquires images frame by frame, discarding invalid images. It simultaneously completes the timestamp alignment and spatial registration of image data, laser point cloud data, and RTK pose data. Through the adaptive adjustment capability of the feature weights of the dual-branch network, it enhances the extraction accuracy of linear features in complex backgrounds and iteratively optimizes the parameters of the spatial fitting curve to eliminate positioning errors.

[0007] Furthermore, the attitude reference calibration stage calls the optical parallax closed-loop visual servo control algorithm to dynamically correct the UAV attitude through the optical spatial model of the guide wire and the optical imaging deviation; the onboard vision system calculates the vertical deviation between the UAV optical axis and the guide wire in real time, and automatically adjusts the UAV yaw angle with the goal of minimizing optical parallax; the distance between the UAV and the guide wire is monitored in real time through laser optical ranging, and the UAV moves autonomously to the optimal optical imaging range. After the UAV hovers and stabilizes, the RTK reference coordinates are recorded, and 100 sets of laser optical ranging data are continuously collected and averaged to establish a unified optical measurement benchmark. After the UAV hovers, steady-state detection is performed to lock the body attitude and observation position parameters and suppress optical parallax fluctuations caused by micro-shakes during flight.

[0008] Furthermore, in the suspension point calculation stage, the UAV is manipulated to move along the vertical optical observation direction of the conductor to perform optical measurement operations on the suspension points of the towers on both sides; multi-exposure bracketing shooting and HDR image fusion technology are used to enhance the contrast of optical features in the connection area between the insulator and the conductor; and the core feature position of the conductor suspension point is locked by improving the Harris corner detection and template matching sub-pixel positioning algorithm. Simultaneously, slant distance data from the UAV to the suspension point is collected by laser optical ranging. Combined with RTK spatiotemporal reference and fuselage attitude parameters, the three-dimensional coordinates of the suspension point are calculated using the spatial forward intersection principle. Five sets of effective optical measurement data are collected for each suspension point. The coordinate accuracy is optimized by a weighted average algorithm to provide optical boundary control points for catenary modeling. The system automatically screens the validity of each set of measurement data, removes abnormal data, dynamically allocates weighting weights based on the data signal-to-noise ratio and imaging quality, iteratively optimizes the three-dimensional coordinate solution of the suspension point, locks the core points of the connection between the conductor and the insulator, and provides boundary constraints for the construction of the conductor catenary model.

[0009] Formula for calculating the three-dimensional coordinates of the suspended point at the intersection in space: ; This represents the horizontal axis coordinate value of the conductor suspension point in a unified spatial coordinate system. The vertical axis coordinate value representing the suspension point of the conductor in a unified spatial coordinate system is the core parameter for the horizontal and vertical positioning of the suspension point. The vertical axis coordinate value representing the suspension point of the conductor in a unified spatial coordinate system is the core parameter for vertical height positioning of the suspension point. The horizontal axis coordinate value represents the real-time acquisition of the drone by the RTK module, and is the horizontal reference for the drone's spatial positioning. The vertical axis coordinate value represents the real-time acquisition of the UAV by the RTK module, and is the horizontal and vertical reference for the UAV's spatial positioning. The vertical axis coordinate value represents the real-time vertical coordinate value obtained by the RTK module on the drone, which is the vertical height reference for the drone's spatial positioning. The slant distance from the UAV to the guide wire suspension point, as measured by the laser rangefinder, is a core fundamental parameter for spatial distance measurement. The pitch angle value, representing the propagation of the laser ranging signal, is obtained in real time by the UAV's attitude sensor; The azimuth angle value representing the propagation of the laser ranging signal is obtained by fusing and solving multi-attitude angle data of the UAV.

[0010] Furthermore, during the lowest point scanning stage, the optical gradient traversal search algorithm for traverse elevation is invoked to perform the sag lowest point positioning operation; the UAV performs laser and vision fusion optical scanning along the traverse direction in the middle region of the span, and collects optical elevation data and longitudinal position information of the image at each position of the traverse in real time. Through iterative calculation of elevation gradient, it dynamically converges to the region of minimum traverse elevation, and determines the core interval of the sag lowest point; the UAV is controlled to make a small translation within ±5m of the lowest point, and a set of optical data is collected every 0.5m. The traverse deformation curve is fitted by a parabolic optical interpolation algorithm to lock the optimal optical measurement position of the lowest point; Finally, 10 sets of laser ranging data and 5 optical images were collected, and the three-dimensional coordinates of the lowest point were calculated by weighted calculation, controlling the optical measurement accuracy of the lowest point to within 5cm. The system plots the elevation gradient change curve of the traverse in real time, dynamically judges the convergence trend of elevation values, adaptively densifies the sampling points in areas with significant gradient changes, reduces the sampling frequency in flat areas, cross-verifies the validity of the points by sampling data, and refines the traverse deformation trajectory by combining parabolic optical interpolation algorithm to lock the unique sag lowest point. The coordinate parameters are optimized by weighted fusion of multi-source data.

[0011] Furthermore, the sag calculation and compensation stage first calculates the basic parameters of span and height difference based on the optical three-dimensional coordinates of the two suspension points; based on the three sets of optical feature points of the two suspension points and the lowest point, combined with the physical constraints of the catenary, the improved least squares method is used to complete the fitting of the catenary model driven by optical data. An optical error compensation model is built to specifically compensate for four types of inherent optical errors, including camera lens distortion error, atmospheric refraction optical shift error, laser ranging temperature drift error, and imaging parallax residual error. Through an airborne edge computing unit, real-time calculations are performed for multi-source optical data fusion, model fitting, error compensation, and sag calculation. Through error propagation analysis, the total error of the final sag measurement is controlled within a preset range. Formula for compensating for total sag error: ; This represents the total error value in the sag measurement. This represents the sag measurement error introduced by camera lens distortion, which is an inherent deviation of the optical imaging system; This represents the sag measurement error introduced by atmospheric refraction optical offset, and the optical deviation affected by the on-site atmospheric environment; This represents the sag measurement error introduced by temperature drift in laser ranging, and the ranging deviation caused by changes in ambient temperature. This represents the sag measurement error introduced by residual imaging parallax, and the measurement deviation caused by residual parallax in imaging observation.

[0012] The error propagation analysis systematically traces the sources and propagation paths of errors along the entire measurement process, sequentially analyzing the propagation methods and superposition patterns of calibration errors, imaging errors, ranging errors, and attitude errors during data processing. It clarifies the influence weight of each error on the final sag result, optimizes the compensation order and intensity of the error compensation model based on the analysis results, prioritizes correcting high-weight error terms, and then processes low-weight error terms in turn.

[0013] The system collects on-site environmental parameters in real time, dynamically substitutes them into the error compensation model to correct the calculation parameters, and optimizes the catenary fitting coefficients through least squares iterative optimization to reduce the impact of single-point random errors on the overall model. The airborne edge computing unit simultaneously completes data fusion and calculation processing, and synchronously generates error quantification analysis data to compensate for system errors in each link.

[0014] Furthermore, in the data verification output stage, the optical measurement results are verified for data consistency. This involves comparing the sag calculation deviations of multiple sets of optical measurement points, analyzing the standard deviation of laser optical ranging data, and verifying the fitting residuals of the catenary optical model. Once verified to be error-free, the measurement results are deemed valid. A standardized measurement report is then generated, which includes core optical measurement data such as the original optical image, laser sampling data, sub-pixel positioning parameters, error compensation parameters, and catenary optical fitting parameters. Real-time sag results and optical measurement data are transmitted to the ground via wireless link to complete standardized optical measurement operations, and the UAV automatically returns to base. The system has built-in standardized data verification thresholds, and automatically triggers retest prompts when abnormal data is detected. The system encrypts and archives the original data and calculation parameters, and organizes the report content.

[0015] The beneficial effects of this invention are as follows: 1. This invention employs BeiDou / GNSS dual-frequency RTK positioning technology and connects to the network RTK differential link, achieving ultra-high precision positioning of ±5cm horizontally and ±5cm vertically. It uses Zhang's calibration method combined with laser plane hard constraints to dynamically calibrate the intrinsic and extrinsic parameters of the camera and laser rangefinder, unifying sensor timing and spatial coordinate references, and eliminating sensor assembly deviations and initial system errors. By integrating a dual-branch U-Net network, Canny edge detection, and Zernike sub-pixel iterative algorithm, it achieves 0.1-pixel-level fine positioning of traverse edges. Combined with the RANSAC algorithm to eliminate abnormal measurements, it can stably construct a high-precision traverse 3D spatial model even against complex backgrounds such as mountains, skies, and buildings.

[0016] 2. This invention utilizes an airborne edge computing unit to achieve real-time fusion processing of multi-source data, including images, laser point clouds, and RTK pose data, as well as instant calculation and output of sag parameters. It can quickly adapt to efficient operational scenarios such as emergency inspections, line acceptance, and batch testing. The system employs an optical parallax closed-loop visual servo control algorithm to automatically calibrate the UAV's observation attitude and operational position, autonomously moving to the optimal optical imaging range. Combined with a traverse elevation optical gradient traversal and adaptive sampling mechanism, it intelligently encrypts sampling in key areas and reduces redundant acquisition in flat areas. This reduces the intensity of manual operation and improves the operational efficiency and on-site response capability of sag measurement.

[0017] 3. This invention establishes an optical error compensation model covering the entire process, which can accurately compensate for four types of core system errors: camera lens distortion, atmospheric refraction optical shift, laser ranging temperature drift, and imaging parallax residual. At the same time, it collects environmental parameters such as on-site temperature, humidity, and atmospheric transparency in real time to dynamically correct the calculation model. By improving the least squares method to iteratively optimize the catenary fitting coefficient, the total error of sag measurement is stably controlled within ±3cm. To address the challenges of backlighting and high-contrast imaging at the suspension point, it adopts multi-exposure bracketing and HDR image fusion technology to enhance feature contrast. Combined with multiple sets of data weighted averaging, outlier screening, and cross-verification mechanisms, it resists interference from changes in lighting, environmental clutter, and camera shake. Attached Figure Description

[0018] Figure 1 This is a flowchart of the UAV sag measurement method based on RTK positioning according to the present invention; Figure 2 This is a flowchart of the lowest point scanning stage of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] like Figures 1 to 2 As shown, this embodiment of the invention provides a method for UAV sag measurement based on RTK positioning, including the following specific steps: In this embodiment of the invention, during the system calibration and configuration phase, the BeiDou / GNSS dual-frequency RTK module is activated and connected to the network RTK differential link to achieve a horizontal positioning accuracy of ±5cm and a vertical positioning accuracy of ±5cm, providing a unified coordinate system and timing reference for optical measurements. The calibration method of Zhang's calibration combined with the laser plane hard constraint is used to complete the dynamic calibration of the intrinsic and extrinsic parameters of the camera and laser rangefinder, generate a coordinate mapping matrix, and eliminate sensor assembly deviations. The calibration process begins by projecting a stable spatial plane constraint benchmark using a laser rangefinder. Multiple sets of calibration reference points are then set up within the calibration area. The UAV, carrying sensors, is controlled to simultaneously collect calibration board imaging data and laser plane ranging data under different spatial orientations and attitudes. The spatial position of the laser plane is then used as a rigid constraint condition in the calibration solution process. The camera's internal parameters and lens distortion parameters are calculated sequentially, and then the relative external parameters between the laser rangefinder and the camera are calculated. A complete mapping relationship between two-dimensional image pixel coordinates, laser rangefinder values ​​and three-dimensional spatial geographic coordinates is established. The accuracy of the mapping matrix is ​​optimized through iterative fitting of multiple sets of calibration data.

[0021] During the network RTK differential link access process, the airborne positioning unit actively establishes a stable data transmission channel with the ground differential reference station, receives differential correction data in real time and completes data parsing and verification, integrates and calculates the positioning data of the Beidou and GNSS dual systems, completes coordinate transformation and reference unification with the national unified geodetic coordinate system as the reference, and simultaneously establishes a spatiotemporal reference system covering the entire measurement process.

[0022] The ground control unit completes the configuration of dedicated optical parameters, sets the laser sampling frequency of 10Hz, camera high-definition imaging parameters, shutter speed of no less than 1 / 1000s, sub-pixel recognition accuracy threshold, and optical parallax allowable deviation threshold, and completes the calibration of the dedicated optical measurement system; during the equipment initialization process, the sensor timing calibration is completed simultaneously, the working clock of the equipment is collected uniformly, and the sensor working temperature, power supply status and signal transmission link are self-checked.

[0023] The sensor timing calibration uses the system clock of the airborne main control unit as the reference source, and sends a unified clock synchronization signal to the camera, laser rangefinder, and RTK positioning module. Through the timestamp alignment mechanism, the acquisition time and data output time of each sensor are bound to the system reference clock, eliminating the timing deviation caused by hardware response delay and data transmission delay of different sensors. After calibration, a timing synchronization verification report is generated, so that the image acquisition, laser ranging, and RTK positioning data are completely synchronized in the time dimension.

[0024] The system first completes the initial configuration of the threshold parameters based on the sensor hardware performance and measurement scenario requirements. After the configuration is completed, the validity of the threshold is verified by standard calibration samples. During the verification process, the pixel positioning deviation and optical parallax fluctuation scenarios in actual measurement are simulated to verify whether the threshold parameters can effectively filter unqualified measurement data and retain valid measurement signals. After the verification is passed, the threshold parameters are fixed to the measurement system.

[0025] In this embodiment of the invention, during the optical positioning modeling stage, the UAV is controlled to fly to the safe optical measurement area to the side of the distance to be measured. The camera continuously acquires images of the conductor area. After adaptive optical contrast preprocessing to suppress ambient light interference, the images of the conductor area are input into a dual-branch U-Net network to complete semantic segmentation of the conductor area and extraction of conductor edge features, thereby achieving coarse positioning of the conductor and suspension point in complex backgrounds. The Canny edge detection and Zernike subpixel iterative algorithm are fused to achieve 0.1-pixel-level conductor edge positioning. The dual-branch U-Net network adopts an encoder-decoder parallel architecture. The semantic segmentation branch and the edge feature extraction branch share the bottom visual feature extraction layer. The encoder completes multi-scale feature downsampling of the image through continuous convolution and pooling operations, and the decoder realizes feature map size restoration through deconvolution and cross-layer feature concatenation. The network input is a single-channel conductor region image after contrast adaptive preprocessing. The image size is normalized to a uniform specification. During training, an adaptive moment estimation optimizer is used to update the network weights. The batch size is set to 8 samples, and the base learning rate is set to 1×10⁻⁶. -4 The loss function constructs a joint optimization objective with segmentation constraints and edge localization constraints. The training dataset covers various transmission line inspection background scenarios such as mountains, forests, sky, and urban buildings. The network's feature adaptation and generalization ability to complex environments is enhanced through iterative training with multi-scenario samples.

[0026] The Zernike moment subpixel iterative algorithm takes the coordinates of the coarse edge of the conductor as the initial input, constructs a numerical calculation model of edge features based on the Zernike orthogonal moment principle, sets a fixed calculation order and convergence judgment threshold during the iteration process, and approximates the real edge spatial position of the conductor through successive iterations. The algorithm is designed for the continuous linear geometric features of power transmission lines. It optimizes edge weight allocation and interference filtering rules, removes discrete noise and irrelevant texture interference from the image background, and outputs high-precision pixel coordinates of the line edge after the iteration terminates. This coordinate data is directly integrated into the coordinate transformation process from two-dimensional pixels to three-dimensional space.

[0027] The laser rangefinder is started to continuously scan and acquire optical distance data of each point on the conductor. The two-dimensional pixel optical coordinates are converted into three-dimensional spatial coordinates through the pre-calibrated intrinsic and extrinsic parameter matrix. The RANSAC algorithm is used to remove optical measurement outliers, the conductor spatial orientation equation is fitted, and optical data registration is completed by combining RTK real-time spatiotemporal data to construct the conductor optical spatial model. The RANSAC algorithm is designed with specific execution rules for the continuous linear spatial distribution characteristics of transmission lines. First, it randomly selects a subset of valid data from the optical measurement sampling data to construct an initial spatial fitting model for the transmission line. Then, it performs a fit check between all the sampled data and the initial model, determines the validity of the data based on a preset residual threshold, and filters out abnormal measurement points caused by environmental occlusion, sudden changes in illumination, and abrupt changes in distance measurement. The process of data selection, model construction, and anomaly removal is iteratively executed until the model fit reaches a stable state. Finally, all valid measurement data are retained to participate in the fitting of the transmission line spatial orientation equation.

[0028] The system filters and acquires images frame by frame, removing blurry, overexposed, and occluded invalid images. Simultaneously, it completes the timestamp alignment and spatial registration of image data, laser point cloud data, and RTK pose data. Through the adaptive adjustment capability of feature weights of the dual-branch network, it enhances the extraction accuracy of linear features in complex backgrounds, iteratively optimizes the parameters of the spatial fitting curve, and eliminates positioning deviations caused by environmental noise and imaging noise.

[0029] Multi-source data registration takes a unified temporal reference as its core. First, it matches a unique and corresponding timestamp for each effective image frame, each set of laser point clouds, and each set of RTK pose data. Then, it binds the three types of data one by one according to the timestamp. Based on the pre-calibrated sensor coordinate mapping relationship, it transforms the local coordinates of different sensors into the global geodetic coordinate system to complete the accurate matching of spatial positions.

[0030] The adaptive adjustment of feature weights is based on the background complexity and the clarity of the conductor features in real-time imaging. The system automatically identifies the current scene as belonging to types such as mountains and forests, sky, and urban buildings, and dynamically allocates the feature output weights of the semantic segmentation branch and the edge feature extraction branch. When the background interference is strong, the weight ratio of the edge feature branch is increased. When the imaging clarity is high, the weights of the two branches are evenly distributed. Through real-time weight adaptation, the network always focuses on the core linear features of the conductor, weakening the interference of complex background on feature extraction.

[0031] In this embodiment of the invention, the attitude reference calibration stage calls the optical parallax closed-loop visual servo control algorithm to dynamically correct the UAV attitude through the optical spatial model of the guide wire and the optical imaging deviation; the onboard vision system calculates the vertical deviation between the UAV optical axis and the guide wire in real time, and automatically adjusts the UAV yaw angle with the goal of minimizing optical parallax, so that the observation azimuth angle is maintained at 90°±0.1°; the distance between the UAV and the guide wire is monitored in real time through laser optical ranging, and the UAV autonomously moves to the optimal optical imaging range of 20-30m. This distance is determined based on the optimal matching principle of camera resolution and laser ranging optical accuracy. The optical parallax closed-loop visual servo control algorithm uses the difference between the theoretical position of the traverse optical spatial model and the actual position of the real-time imaging as the optical parallax feedback quantity. It establishes a complete control process including deviation calculation, command conversion, execution adjustment, and result feedback. The optical parallax deviation quantity is converted into adjustment commands for the UAV's yaw angle and fuselage horizontal displacement, which are transmitted to the airborne flight control execution unit in real time. The flight control system continuously fine-tunes the UAV's observation attitude and spatial position according to the commands, and continuously reduces the vertical deviation between the UAV's optical axis and the traverse direction until the optical parallax stabilizes within the preset allowable deviation range.

[0032] The optimal optical imaging range is determined by combining three core parameters: camera imaging resolution, effective range of laser ranging, and field of view of airborne sensor. The system reads the current sensor operating parameters and the spatial distribution information of the conductor in real time, and selects the spatial range that can ensure the imaging field of view completely covers the conductor to be measured, while also meeting the requirements of ranging accuracy and feature extraction through built-in judgment rules.

[0033] After the UAV hovers and stabilizes, the RTK reference coordinates are recorded, and 100 sets of laser optical ranging data are continuously collected and averaged to establish a unified optical measurement benchmark. After the UAV hovers, steady-state detection is performed to lock the body attitude and observation position parameters and suppress optical parallax fluctuations caused by micro-shakes during flight.

[0034] The hovering steady-state detection uses an onboard inertial measurement unit to collect real-time data on the UAV's three-axis attitude angles, position offsets, and vibration amplitudes. A fixed steady-state judgment threshold is set, and the changes in the fuselage state are continuously monitored within a preset time period. When the attitude angle fluctuations, position offsets, and vibration amplitudes are all stable within the judgment threshold range, the UAV is determined to have entered a steady-state measurement state. At this time, the fuselage attitude and spatial position parameters are locked to block the interference of flight micro-shakes on the optical measurement link.

[0035] Parameter locking is activated immediately after the UAV enters a steady state. It writes the current three-axis attitude angles, horizontal position coordinates, altitude values ​​and other core parameters into the airborne control cache. The flight control system enters a dual closed-loop position and attitude holding mode to continuously counteract the slight deviations caused by airflow disturbances and fuselage vibrations, and maintain the constant state of the parameters. Manual attitude adjustment commands are not accepted during the locking period.

[0036] In this embodiment of the invention, during the suspension point calculation stage, the UAV is manipulated to move along the vertical optical observation direction of the conductor to perform optical measurement operations on the suspension points of the towers on both sides. To address the difficulties of backlighting and high-contrast optical imaging at the suspension points, multi-exposure bracketing and HDR image fusion technology are used to enhance the contrast of optical features in the connection area between the insulator and the conductor. By improving the Harris corner detection algorithm and the template matching sub-pixel positioning algorithm, the core feature positions of the conductor suspension points are locked. The improved Harris corner detection algorithm optimizes the corner response function calculation logic based on the imaging characteristics of the connection area between the conductor and the insulator, sets a fixed corner response threshold to filter out non-critical corner information, and retains only the effective feature corners at the connection between the conductor and the insulator. The template matching subpixel localization algorithm uses a preset standard feature template of the suspending point as the matching benchmark, performs global feature search and similarity comparison within the target image region, and uses subpixel interpolation calculation to improve the localization resolution during the matching process. Finally, it outputs the subpixel coordinates of the core features of the suspending point, which serve as the core input parameter for spatial forward intersection solution.

[0037] The multi-exposure bracketing shooting method continuously captures multiple images with different exposure levels on the same hanging point area according to a preset exposure parameter sequence, which is adapted to the imaging needs of strong light areas and low light areas respectively. The HDR image fusion technology performs feature alignment and pixel-level fusion of multiple original images with different exposures, preserving clear and effective imaging details in each exposure frame and suppressing insufficient imaging contrast caused by backlighting and strong light reflection.

[0038] Simultaneously, slant distance data from the UAV to the suspension point is collected via laser optical ranging. Combined with RTK spatiotemporal reference and fuselage attitude parameters, the three-dimensional coordinates of the suspension point are calculated using the principle of spatial forward intersection. Five sets of effective optical measurement data are collected for each suspension point. The coordinate accuracy is optimized by a weighted average algorithm to eliminate random errors in single optical imaging and laser ranging, providing optical boundary control points for catenary modeling. The calculation operation uses the real-time RTK positioning coordinates of the UAV, the fuselage attitude angle parameters, the slant range data obtained by laser ranging, and the pixel coordinates of the suspension point extracted from the image as the core input parameters. Based on the spatial geometric relationship of multi-view observation, a simultaneous solution equation is constructed. By cross-solving multiple sets of data from different observation positions, the calculation deviation caused by a single observation angle is eliminated. The three-dimensional spatial coordinates of the target suspension point are obtained by simultaneous solution. During the calculation process, the data validity is verified simultaneously, and abnormal results with excessive calculation deviations are eliminated.

[0039] The weighted average algorithm assigns corresponding weight coefficients based on the imaging quality, ranging stability, and solution residual size of each set of measurement data. High-quality data with clear imaging, stable ranging, and small solution residuals are given higher weights, while low-quality data have lower weights. Multiple sets of three-dimensional coordinate data are weighted and calculated with the corresponding weight coefficients to obtain the optimized suspension point coordinate results. The influence of low-quality data is weakened by the differentiated allocation of weights.

[0040] The system automatically screens the validity of each set of measurement data, removes abnormal data such as ranging jumps, incomplete imaging, and attitude deviations, dynamically allocates weighting weights based on data signal-to-noise ratio and imaging quality, iteratively optimizes the three-dimensional coordinate solution results of the suspension point, locks the core points of the connection between the conductor and the insulator, and provides boundary constraints for the construction of the conductor catenary model.

[0041] The dynamic allocation of weights is based on three core indicators: data signal-to-noise ratio, imaging integrity, and attitude stability. The system automatically calculates the scores of the three indicators for each group of measurement data and assigns weight coefficients according to the scores. The higher the indicator score, the larger the weight coefficient, and vice versa. After the weight allocation is completed, it is substituted into the weighted calculation process, and the abnormal data with too low weights are eliminated through multiple rounds of iteration.

[0042] In this embodiment of the invention, the lowest point scanning stage calls the traverse elevation optical gradient traversal search algorithm to perform the sag lowest point positioning operation; the UAV performs laser and vision fusion optical scanning along the traverse direction in the middle region of the span, and collects optical elevation data and image longitudinal position information at each position of the traverse in real time. Through elevation gradient iterative calculation, it dynamically converges to the minimum elevation region of the traverse to determine the core interval of the sag lowest point; the UAV is controlled to move slightly within ±5m of the lowest point, and a set of optical data is collected every 0.5m. The traverse deformation curve is fitted by a parabolic optical interpolation algorithm to lock the optimal optical measurement position of the lowest point; The optical gradient traversal search algorithm for traverse elevation collects optical elevation data segment by segment along the traverse direction and calculates the elevation change gradient of adjacent points. Based on the positive and negative changes of the gradient values ​​and the trend of numerical changes, it dynamically determines the extreme value range of the traverse elevation and gradually narrows the search range of the minimum elevation value. During the traversal, it adaptively adjusts the search step size in combination with the trend characteristics of the natural deformation of the traverse. In areas with gentle elevation changes, it uses a conventional step size for traversal, while in areas with drastic elevation changes, it densifies the sampling points. Through continuous gradient comparison and interval convergence, it accurately locks the core positioning area of ​​the lowest point of sag.

[0043] The parabolic optical interpolation algorithm uses multiple sets of elevation and position data collected within the core area of ​​the lowest point as input samples. Based on the continuous and smooth characteristics of conductor deformation, a parabolic interpolation model is constructed, which transforms discrete sampling point data into a continuous conductor deformation curve. The precise position corresponding to the minimum elevation value is obtained by solving the curve. During the interpolation process, the geometric characteristics of the conductor's natural deformation are preserved, avoiding distortion of the deformation trajectory caused by discrete sampling, and accurately determining the optimal measurement position of the lowest point of sag.

[0044] Finally, 10 sets of laser ranging data and 5 optical images were collected, and the three-dimensional coordinates of the lowest point were calculated by weighted calculation, controlling the optical measurement accuracy of the lowest point to within 5cm. The system plots the elevation gradient change curve of the traverse in real time, dynamically judges the convergence trend of elevation values, adaptively densifies the sampling points in areas with significant gradient changes, reduces the sampling frequency in flat areas, cross-verifies the validity of the points by sampling data, and refines the traverse deformation trajectory by combining parabolic optical interpolation algorithm to lock the unique sag lowest point. The coordinate parameters are optimized by weighted fusion of multi-source data to improve the accuracy and stability of the spatial positioning of the lowest point.

[0045] The elevation convergence trend determination constructs a trend curve based on continuously collected elevation data. It judges whether the lowest point interval has converged based on the change in the curve slope and the numerical stability. When the curve slope approaches zero and the elevation value tends to be stable, it is determined that the convergence is complete. Adaptive densification sampling is executed in conjunction with the convergence trend. Before convergence, the sampling interval is automatically shortened and the number of samples is increased in areas with drastic gradient changes. After convergence, the sampling interval is extended and the number of samples is reduced in areas with gentle gradient changes.

[0046] In this embodiment of the invention, the sag calculation and compensation stage first calculates the basic parameters of span and height difference based on the optical three-dimensional coordinates of the two suspension points; based on the three sets of optical feature points of the two suspension points and the lowest point, combined with the physical constraints of the catenary, the improved least squares method is used to complete the fitting of the catenary model driven by optical data. The catenary model fitting uses the three-dimensional coordinates of the suspension points on both sides of the conductor and the three-dimensional coordinates of the lowest point of the sag as hard constraints to construct a catenary fitting equation that fits the physical shape of the conductor. Differentiated weighting is applied to the data quality of different measurement points to increase the weight of high-quality measurement point data in the fitting process. The core coefficients of the equation are optimized through multiple iterations to reduce the impact of random errors of discrete measurement points and environmental interference errors on the overall fitting results.

[0047] Equation of a catenary: ; The spatial elevation value representing any measuring point of the traverse is a parameter that characterizes the position of the traverse in the vertical direction. The horizontal coordinate value represents any measuring point of the traverse, and is a parameter representing the position of the traverse in the horizontal direction; It represents the horizontal stress at the lowest point of the conductor's sag, and intuitively reflects the stress and bearing state of the conductor at the lowest point. The load per unit length of conductor is the load-to-weight ratio, which reflects the load distribution characteristics generated by the conductor's own weight. The horizontal coordinates corresponding to the lowest point of the conductor sag are the horizontal positioning reference points for the catenary model. The elevation value corresponding to the lowest point of the conductor sag is the vertical positioning reference point of the catenary model; This represents the hyperbolic cosine function.

[0048] An optical error compensation model was built to specifically compensate for four types of inherent optical errors, including camera lens distortion error, atmospheric refraction optical shift error, laser ranging temperature drift error, and imaging parallax residual error. Through an airborne edge computing unit, real-time calculations of multi-source optical data fusion, model fitting, error compensation, and sag calculation were completed. Through error propagation analysis, the total error of the final sag measurement was controlled within ±3cm. The optical error compensation model adopts a multi-layer BP neural network structure. The input layer receives four types of optical error quantification values ​​and on-site environmental perception parameters. The hidden layer has two layers with a fixed number of neurons. Non-linear activation functions are used between layers to complete feature mapping and numerical transformation. In the model initialization stage, a random normal distribution is used to assign weights and bias parameters. The training process takes the error compensation correction amount as the core optimization objective, sets a fixed learning rate and number of iterations, and completes network fitting training through on-site measured error samples. The model output generates a sag calculation compensation coefficient, which is directly substituted into the catenary mathematical model to complete the real-time correction of the sag parameters.

[0049] The airborne edge computing unit uses a priority scheduling mechanism to handle multi-task operations. It sorts the four core tasks of data fusion, model fitting, error compensation, and sag calculation according to the execution logic, prioritizes the completion of multi-source data fusion processing, and then proceeds with model building, error correction and parameter calculation in sequence. During the operation, it calls the cached calibration parameters, environmental parameters and benchmark data in real time. All operation links are completed locally on the airborne end.

[0050] The system collects environmental parameters such as ambient temperature, atmospheric humidity, and atmospheric transparency in real time, dynamically substitutes them into the error compensation model to correct the calculation parameters, and optimizes the catenary fitting coefficient through least squares iterative optimization to reduce the impact of single-point random errors on the overall model. The airborne edge computing unit simultaneously completes data fusion and calculation processing, and synchronously generates error quantification analysis data to compensate for system errors in each link.

[0051] In this embodiment of the invention, the data verification output stage performs data consistency verification on the optical measurement results, compares the sag calculation deviation of multiple sets of optical measurement points, analyzes the standard deviation of laser optical ranging data, verifies the fitting residual of the catenary optical model, and determines the measurement results to be valid after verification; a standardized measurement report is generated, which includes core optical measurement data such as the original optical image, laser sampling data, sub-pixel positioning parameters, error compensation parameters, and catenary optical fitting parameters; The data consistency verification is carried out from multiple dimensions. First, the sag calculation results obtained from different observation positions and different sampling times are compared to check whether the numerical fluctuation range meets the specifications. Second, the dispersion of the laser ranging data is analyzed to determine the stability of the measurement data. Finally, the fitting residual between the catenary model and the measured points is checked to evaluate the model fit. The multi-dimensional verification results are comprehensively judged. When all indicators meet the preset requirements, the measurement results are confirmed to be valid. If any indicator is abnormal, the retest process is triggered.

[0052] Real-time sag results and optical measurement data are transmitted to the ground via wireless link to complete standardized optical measurement operations, and the UAV automatically returns to base. The system has built-in standardized data verification thresholds, and automatically triggers retest prompts when abnormal data is detected. The system encrypts and archives the original data and calculation parameters, and organizes the report content.

[0053] Anomaly screening is performed by comparing real-time measurement data with the system's built-in verification thresholds. It covers various anomaly types, such as coordinate anomalies, distance jumps, excessive model residuals, and invalid imaging. Once anomaly is determined, the system immediately stops the current calculation process and sends a remeasurement command to the ground. At the same time, it records the location, time, and data type of the anomaly. After the remeasurement is completed, the anomaly data is automatically replaced.

[0054] Data transmission uses a stable airborne wireless communication module to package and transmit verified sag parameters, coordinate data, and error compensation data in a standardized format. During transmission, core measurement data is encrypted to prevent data leakage. When archiving data, raw imaging data, laser sampling data, solution parameters, and calibration parameters are stored separately and saved in an encrypted storage format to dual-end storage units on both the airborne and ground sides, preserving complete measurement traceability data.

[0055] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0056] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for measuring the sag of a UAV based on RTK positioning, characterized in that, The specific steps include the following: During the system calibration and configuration phase, the initialization of the UAV optical measurement platform, joint calibration of sensors and configuration of optical parameters are completed, the timing and spatial coordinate references of each acquisition device are unified, the operating conditions of the equipment are checked, and sensor assembly and initial system errors are eliminated. In the optical positioning and modeling stage, UAVs are used to collect traverse imaging data by stationing at fixed points. Data registration is completed by fusing laser ranging and RTK spatiotemporal reference, and measurement outliers and environmental interference errors are eliminated to construct a 3D optical spatial model of the traverse. During the attitude reference calibration stage, based on the optical three-dimensional space model of the conductor, the observation attitude and operation position of the UAV are dynamically calibrated, the optimal optical imaging observation conditions are matched, a unified optical measurement reference is established, and optical measurement system errors are suppressed. In the suspension point calculation stage, the optical imaging effect of the suspension point is optimized, the three-dimensional coordinate calculation of the suspension points on both sides is completed, abnormal measurement data is screened and eliminated, the point accuracy is optimized, and the boundary control points of the catenary modeling are obtained. In the lowest point scanning stage, the traverse scanning acquisition is completed, the minimum elevation range is converged, the traverse deformation trajectory is refined and the lowest point coordinate parameters are optimized, the three-dimensional coordinates of the sag lowest point are optimized, and the positioning of the lowest point of the deformed traverse is realized. In the sag calculation and compensation stage, a catenary model of the conductor is constructed using the data of the suspension point and the lowest point, an optical error compensation model is built, and the measurement error is corrected by combining the field environmental parameters to achieve real-time calculation of sag parameters; During the data verification and output phase, the consistency of the sag measurement results is verified, valid measurement data is selected, and parameter archiving and report organization are completed. Data feedback and work completion are completed simultaneously.

2. The method for UAV sag measurement based on RTK positioning according to claim 1, characterized in that, During the system calibration and configuration phase, the BeiDou / GNSS dual-frequency RTK module is activated and connected to the network RTK differential link to provide a unified coordinate system and timing reference for optical measurements. The calibration method of Zhang's calibration combined with the laser plane hard constraint is adopted to complete the dynamic calibration of the intrinsic and extrinsic parameters of the camera and laser rangefinder, and generate a coordinate mapping matrix.

3. The method for UAV sag measurement based on RTK positioning according to claim 2, characterized in that, During the system calibration and configuration phase, the ground control unit completes the optical parameter configuration, sets the laser high-frequency sampling frequency, camera high-definition imaging parameters, shutter speed, sub-pixel recognition accuracy threshold, and optical parallax allowable deviation threshold, and completes the optical measurement system calibration. During the equipment initialization process, the sensor timing calibration is completed simultaneously, the working clock of the equipment is collected uniformly, and the sensor working status is checked at the same time.

4. The UAV sag measurement method based on RTK positioning according to claim 3, characterized in that, In the optical positioning modeling stage, the UAV is controlled to fly to the safe optical measurement area to the side of the span to be measured, and the camera continuously acquires images of the conductor area. After the conductor area images are preprocessed to suppress ambient light interference, they are input into a dual-branch deep learning network to complete semantic segmentation and edge feature extraction of the conductor area, thereby achieving coarse positioning of the conductor and suspension point; the Zernike subpixel iterative algorithm is combined to complete the conductor edge positioning. The laser rangefinder is started synchronously to scan and acquire optical distance data of each point on the conductor. The two-dimensional pixel optical coordinates are converted into three-dimensional spatial coordinates through the pre-calibrated intrinsic and extrinsic parameter matrix. The RANSAC outlier removal algorithm is used to remove optical measurement outliers, the conductor spatial orientation equation is fitted, and optical data registration is completed by combining RTK real-time spatiotemporal data to construct the conductor optical spatial model. The system filters valid images frame by frame, completes timestamp alignment and spatial registration of multi-source data, enhances the accuracy of conductor feature extraction, iteratively optimizes spatial fitting curve parameters, and eliminates positioning deviations.

5. The UAV sag measurement method based on RTK positioning according to claim 4, characterized in that, During the attitude reference calibration stage, the optical parallax closed-loop visual servo control algorithm is invoked to dynamically correct the UAV attitude through the optical spatial model of the guide wire and the optical imaging deviation; the vertical deviation between the UAV optical axis and the guide wire direction is calculated in real time through the airborne vision system, and the UAV yaw angle is automatically adjusted with the goal of minimizing optical parallax; the distance between the UAV and the guide wire is monitored in real time through laser optical ranging, and the UAV moves autonomously to the optimal optical imaging range. After the UAV hovers and stabilizes, the RTK reference coordinates are recorded, and multiple sets of laser optical ranging data are continuously collected and averaged to establish a unified optical measurement reference. After the UAV hovers, steady-state detection is performed to lock the body attitude and observation position parameters and suppress optical parallax fluctuations caused by micro-shakes during flight.

6. The method for UAV sag measurement based on RTK positioning according to claim 5, characterized in that, In the suspension point calculation stage, the UAV is manipulated to move along the vertical optical observation direction of the conductor to perform optical measurement operations on the suspension points of the towers on both sides; multi-exposure bracketing and image fusion technology is used to enhance the contrast of optical features in the connection area between the insulator and the conductor; and the core feature position of the conductor suspension point is locked by improving the Harris corner detection and template matching sub-pixel positioning algorithm. Simultaneously, slant distance data from the UAV to the suspension point is collected by laser optical ranging. Combined with RTK spatiotemporal reference and fuselage attitude parameters, the three-dimensional coordinates of the suspension point are calculated using the spatial forward intersection principle. Multiple sets of effective optical measurement data are collected from each suspension point. The coordinate accuracy is optimized by a weighted average algorithm to provide optical boundary control points for catenary modeling. The system automatically screens the validity of measurement data, removes abnormal data, dynamically allocates weights based on data quality, and iteratively optimizes the calculation results of the suspension point coordinates.

7. The method for UAV sag measurement based on RTK positioning according to claim 6, characterized in that, During the lowest point scanning phase, the optical gradient traversal search algorithm for traverse elevation is invoked to perform the sag lowest point positioning operation. The UAV performs laser and vision fusion optical scanning along the traverse direction in the middle region of the span, and collects optical elevation data and longitudinal position information of the image at each position of the traverse in real time. Through iterative calculation of elevation gradient, it dynamically converges to the region of minimum traverse elevation to determine the core interval of the sag lowest point. The UAV is controlled to make slight translations within the lowest point region to collect multiple sets of optical data. The traverse deformation curve is fitted by a parabolic optical interpolation algorithm to lock the optimal optical measurement position of the lowest point. The system collects multiple sets of laser ranging data and optical images, and calculates the three-dimensional coordinates of the lowest point using weighted average. The system also identifies the elevation convergence trend in real time, adaptively densifies the sampling points in key areas, cross-verifies the validity of the points, refines the traverse deformation trajectory, and optimizes the coordinate parameters of the lowest point.

8. The method for UAV sag measurement based on RTK positioning according to claim 7, characterized in that, In the sag calculation and compensation stage, the basic parameters of span and height difference are first calculated based on the optical three-dimensional coordinates of the two suspension points. Based on the three sets of optical feature points of the two suspension points and the lowest point, combined with the physical constraints of the catenary, the improved least squares method is used to complete the fitting of the catenary model driven by optical data.

9. A method for UAV sag measurement based on RTK positioning according to claim 8, characterized in that, In the sag calculation and compensation stage, an optical error compensation model is built to specifically compensate for camera lens distortion error, atmospheric refraction optical shift error, laser ranging temperature drift error, and imaging parallax residual error. Through the airborne edge computing unit, the real-time calculation of multi-source optical data fusion, model fitting, error compensation, and sag calculation is completed, and the final total sag measurement error is controlled within a preset threshold. The system collects on-site environmental parameters in real time, dynamically corrects the calculation parameters of the error compensation model, iteratively optimizes the catenary fitting coefficient, reduces the impact of single-point random errors on model fitting, and synchronously generates error quantification data to compensate for system errors in each stage.

10. A method for UAV sag measurement based on RTK positioning according to claim 9, characterized in that, In the data verification and output stage, the optical measurement results are verified for data consistency. The sag calculation deviation of multiple sets of optical measurement points is compared, the standard deviation of laser optical ranging data is analyzed, and the fitting residual of the catenary optical model is checked. After verification, the measurement results are deemed valid. A standardized measurement report containing the core data of optical measurement is generated, which includes sag value, coordinate parameters, and error compensation data. The measurement data is transmitted to the ground via a wireless link, and the drone automatically returns to its home location after completing the operation. The system has built-in data verification thresholds, automatically triggers retest prompts for abnormal data, encrypts and archives original data and parameters, and standardizes report content.

Images