Method, system and equipment for realizing intelligent measurement of sag of power transmission line by using binocular depth camera, and medium

Abstract

The invention discloses a method for realizing intelligent sag measurement of a power transmission line by using a binocular depth camera, which comprises the following steps of: firstly, fixing the binocular camera on a rotary holder at the central position of two towers to complete camera calibration and epipolar correction; then, controlling the holder to rotate according to a fixed stepping angle, collecting a binocular image sequence covering the whole section of lead, and preprocessing the binocular image sequence; performing feature matching and image splicing by using an SIFT algorithm based on the left eye image sequence to generate a panorama and establish a panorama-local mapping relationship; thirdly, detecting an insulator and a spacer target in the panorama based on a Mask-RCNN model, accurately positioning key point coordinates of the insulator and the spacer target, and mapping the key point coordinates to an original local image; in the corresponding left and right local images, a semi-global matching (SGM) algorithm is adopted to calculate parallax and generate depth information, and then three-dimensional coordinates of a target point are obtained through conversion; and finally, performing catenary fitting according to the ordered three-dimensional point set, introducing a temperature and windage yaw dynamic correction model, and accurately calculating a sag value. According to the invention, non-contact, automatic and high-precision sag measurement is realized, the method is suitable for routing inspection of complex terrains and large-span power transmission lines, and the operation safety and efficiency are remarkably improved. The invention also provides a system, equipment and a medium for realizing the method.

Description

Technical Field

[0001] This invention belongs to the field of power construction technology and relates to a method, system, equipment and medium for intelligent measurement of transmission line sag using a binocular depth camera. Background Technology

[0002] The sag of transmission lines is a crucial aspect of power line inspection. Measuring the sag accurately reflects the condition of overhead transmission lines and helps determine potential hazards. Excessive sag can reduce the safe distance between the sag and the ground, trees, or buildings. In extreme weather conditions (such as strong winds or heavy rainfall), direct contact may occur, leading to short circuits, grounding faults, and serious accidents. Furthermore, the degree of sag reflects the overall tension on the line and the stress on the towers. If the sag is too low, the conductors experience excessive tension, increasing conductor resistance, line losses, and transmission efficiency. In severe cases, this can lead to wire breaks or insulator ruptures. Increased conductor tension on the towers, exceeding their design capacity, can cause tower tilting and severe large-scale power outages. Conversely, excessively high sag results in insufficient conductor tension, making them susceptible to wind-induced flashover in strong winds. Wind-induced conductor deflection can break down air gaps between the conductors and other objects, causing short circuits. In addition, as time goes by, line aging, load changes and environmental factors will all have a certain impact on the sag. Therefore, regular monitoring of the sag of transmission lines is of great significance for timely detection of potential line hazards and prevention of more serious accidents.

[0003] Currently, common sag measurement techniques mainly include the following: 1. Manual inspection Early sag measurements relied primarily on manual inspections, using a total station to measure the elevation angle between the conductor suspension point and the lowest point, and then calculating the sag using trigonometric formulas. This method was inefficient due to its manual operation and was significantly limited by terrain and environmental factors during long-distance power transmission.

[0004] 2. Wireless Sensor Networks (WSN) Tilt sensors are installed on conductors or spacers to measure the conductor's tilt angle in real time via sensor modules. This method provides direct measurement data, but the installation process is complex. After installation, conductor vibration and wind can cause the sensor to detach. Furthermore, maintenance costs are high when the sensor itself malfunctions.

[0005] 3. LiDAR drones Drones equipped with LiDAR acquire point cloud data of power lines, collecting 3D spatial point cloud data. This point cloud data is then used to reconstruct the catenary fitting equation of the transmission line and calculate sag. However, LiDAR equipment is expensive, and the amount of point cloud data is large, making this method costly and unsuitable for routine inspections of large-scale transmission lines.

[0006] 4. Monocular vision drones Using drone aerial images to measure the sag of power transmission lines, the distance is estimated based on the pixel positions in the camera footage and prior knowledge of the conductors. However, this method also suffers from problems such as susceptibility to interference from complex backgrounds and excessive human intervention.

[0007] Compared to applications that use laser sensors for distance measurement or require the installation of contact-based auxiliary equipment on towers to acquire data related to transmission lines, such as wireless sensor networks (WSNs), these methods also suffer from problems such as high susceptibility to interference from complex backgrounds, reliance on prior knowledge, and low efficiency due to excessive human intervention. This invention achieves completely non-contact long-distance measurement, eliminating the need for personnel to climb towers or touch conductors. This significantly reduces the complexity and risk of operation and maintenance while ensuring personnel safety.

[0008] Compared to the measurement scheme of "A UAV-mounted ranging device and its measurement method" (CN116047530A), which uses a UAV-mounted ranging sensor, the measurement process heavily relies on the flight stability and precise attitude control of the UAV platform. This results in large measurement errors under complex weather conditions and only acquires local point-to-point distance data each time, failing to quickly construct the three-dimensional shape of the entire conductor segment. This invention, by deploying a binocular camera and a rotating gimbal at a fixed ground point, achieves stable and continuous full-range image acquisition. Combined with panoramic stitching technology to construct a global view, it avoids accuracy loss caused by UAV flight jitter and can simultaneously acquire the three-dimensional coordinates of multiple key points on the conductor (insulators, spacers), providing a complete data foundation for high-precision catenary fitting.

[0009] Compared to the application scheme of "Sagging Measurement Method, Device and Computer Equipment Based on Image Parallax Depth" (CN116718122A), which utilizes dual-camera image inspection and the principle of set similarity for sag measurement, this method relies on strict camera baseline settings and precise image registration. The algorithm is heavily constrained by scene geometry, and in long-distance, large-span transmission line scenarios, the sensitivity of depth calculation drops sharply due to the small parallax, making it difficult to guarantee accuracy. This invention innovatively adopts a "panoramic first, local later" measurement strategy. First, robust panoramic image stitching is achieved through SIFT feature matching and the RANSAC algorithm. Then, the Mask-RCNN deep learning model is used to accurately locate the target in the panoramic image. Finally, the original high-resolution local image pairs are backtracked for stereo matching and depth calculation. This strategy decouples global recognition from precise local ranging, expanding the coverage of long-span lines while ensuring high-precision extraction of key point depth information. Summary of the Invention

[0010] To overcome the shortcomings of the existing technology, the present invention aims to provide a method, system, device, and medium for intelligent measurement of sag of transmission lines using a binocular depth camera. The method involves capturing local images of the transmission line using a binocular camera mounted on a rotating platform, and then stitching and fusing these local images to obtain a panoramic image. A Mask-RCNN model is used to segment insulators and spacers, mapping the segments back to the original local images. Depth information is acquired using the local binocular images to obtain global three-dimensional coordinate information of the insulators and spacers. Based on the obtained three-dimensional coordinate information, dynamic fitting of the catenary is performed to obtain the final sag measurement result. This invention utilizes a binocular depth camera to achieve sag measurement of long-distance transmission lines, establishing a measurement model with global recognition and detection and precise local matching, thus solving the problems of high technical cost, low efficiency, and insufficient accuracy in existing sag measurement methods.

[0011] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for intelligent measurement of transmission line sag using a binocular depth camera includes the following steps: Step S1: Fix the binocular depth camera on the rotating gimbal, ensuring that the gimbal is fixed in the center position of the two towers, and complete the initialization and calibration of the binocular depth camera. Step S2: Control the rotating gimbal to rotate in fixed-angle steps. Each time the camera rotates for one frame, it takes a set of images until it covers the entire span of the guide wire. Perform preprocessing on the captured images to obtain a preprocessed binocular image sequence. Step S3: Using the left camera of the binocular depth camera as the main view, import the preprocessed left camera image sequence, use the SIFT algorithm to align the images, complete the panoramic image stitching, and establish the mapping relationship between the local image and the panoramic image. Step S4: For the stitched panoramic image, use the Mask-RCNN model to perform target detection on the insulators and spacers, obtain the image coordinates, and map them back to the original panoramic image. Step S5: Based on the original image mapped back from Step 4, the target area is cropped out from the left and right images of the binocular depth camera, and a disparity map is generated by stereo matching. The depth information of the spacer rod and insulator target points is calculated, and a three-dimensional coordinate transformation is performed to generate an ordered three-dimensional point set. Step S6: Perform dynamic fitting of the catenary based on the obtained ordered three-dimensional point set to obtain the final sag calculation result.

[0012] The specific steps of step S1 are as follows: S1.1 Camera Deployment and Baseline Setting Check that the left and right binocular depth cameras and the binocular depth cameras are in normal working connection with the computer; install the bracket, binocular cameras, and rotating head according to the center position between the two towers, and adjust according to the guide wire height. Select the baseline according to the empirical formula. , , The empirical coefficient is used; the binocular depth camera bracket is adjusted to make the optical axes of the two cameras parallel and horizontal, and the left and right binocular depth cameras are rigidly connected to avoid vibration; Calibration and calibration of S1.2 binocular depth camera Customize the image resolution and acquisition frame rate of the stereo depth camera, set the image exposure time, and adjust the color balance; calibrate and standardize the intrinsic parameters of the left and right stereo depth cameras using a calibration board to obtain the corresponding intrinsic parameter parameters and focal length of each stereo depth camera. Like the main point distortion coefficient ,in, , The radial distortion coefficient is... , Tangential distortion coefficients; for binocular extrinsic parameter calibration, the extrinsic parameter matrix is ​​solved using the PnP (Perspective-n-Point) algorithm. ,in, For rotation matrix, The translation vector is used; after completing the binocular positioning, epipolar correction is performed using the Bouguet algorithm; based on the intrinsic, extrinsic, and distortion parameters of the left and right binocular depth cameras, the correction rotation matrices of the left and right binocular depth cameras are obtained. , and the corrected common intrinsic parameter matrix Generate left and right image remapping matrices , This eliminates geometric distortion and parallax angle shift between images, ensuring that the content of the left and right images is strictly aligned in the scan line direction, while retaining only horizontal parallax.

[0013] The specific steps of step S2 are as follows: S2.1 Image Pair Acquisition and Sequence Generation Calculate the step angle of the gimbal rotation based on the field of view (FOV) of the binocular depth camera and the guide wire spacing: The horizontal field of view (HFOV) represents the imaging range of a binocular depth camera, and is calculated using the following formula:

[0014]

[0015] in, The width of the built-in sensor for the binocular depth camera. For camera focal length, For the gimbal step angle, The overlap rate; According to the conductor span length And the horizontal distance between the wire and the camera Calculate the coverage length of a single image and the number of frames required to complete the panorama, where the coverage width of a single image is:

[0016] Number of image pairs required to complete panoramic shooting for:

[0017] When the calculation result is a decimal, then Round up; Left and right binocular depth cameras simultaneously acquire images from the same angle, and image pairs are generated with incrementally increasing sequence numbers based on the shooting angle and time. Save the original image size information Information such as the gimbal angle and focal length during shooting is used for subsequent operations; S2.2 Image Preprocessing The images acquired in step S2.1 are preprocessed. Distortion correction is performed using the parameters calibrated in step S1.2 to eliminate radial and tangential distortion. Based on the correction mapping relationship generated in step S1, epipolar correction is performed on the left and right images to ensure strict alignment of their contents. The images are then denoised and enhanced. Using the color temperature of the first local power line image containing insulators as a base, and using the insulator region as a reference region, the core insulator region without background interference is determined by manual selection within the first local power line image containing insulators. The total number of pixels in this region is denoted as M. All pixels within the core insulator region are iterated through, and the pixel values ​​of the red, green, and blue channels are accumulated to obtain the total brightness of each channel, Sum_R, Sum_G, and Sum_B. The average brightness of the red, green, and blue channels is obtained by dividing the total brightness of each channel by the total number of pixels in the region, M. The formulas are as follows: , , This establishes a color temperature baseline parameter, allowing subsequent images to adjust their color temperature pixel-by-pixel based on the calculated gain coefficient. The average brightness of the red, green, and blue channels in the insulator region is used as a baseline. Simultaneously, the gain coefficients of the red and blue channels relative to the green channel are obtained, and subsequent images are pixel-by-pixel adjusted by multiplying by the corresponding gain coefficient. This ensures that the color temperature of subsequent local images matches the baseline image, avoiding stitching gaps and resulting in a pre-processed binocular image sequence.

[0018] The specific steps of step S3 are as follows: S3.1 Panoramic Image Stitching and Fusion Using the left camera of a binocular depth camera as the main view, import the preprocessed left camera image sequence. Feature point extraction using the SIFT algorithm: This involves searching for image locations across all scale spaces, using an initial Gaussian blur coefficient. The number of generated layers is A Gaussian pyramid with multiple layers generates a difference Gaussian image using a Gaussian differential function, identifying candidate feature points in the image that are scale- and rotation-invariant. ,in, For pixel coordinates, For scale space, The main direction of the feature point, For each pair of adjacent images, the corresponding descriptor vector is used. , Preliminary feature point matching is performed using Fast Nearest Neighbor Search (FLANN). descriptor vector set , descriptor vector set Calculate the Euclidean distance between two vectors, and select the vectors with the smallest distance as candidate matching pairs to obtain a preliminary matching set. ; False matches are eliminated using the RANSAC algorithm: the initial homography matrix of the feature points of the candidate matching pairs is estimated, and the projection error is calculated. ,in, for The actual coordinates of the feature points are as follows: for The corresponding feature points are in The projected coordinates on the screen, the filtering error threshold is within Points within the matrix are removed to eliminate incorrect matches, and the final homography matrix is ​​obtained using the least squares method. ; Following a left-to-right splicing strategy, with Using the reference image as the base image, perform a cumulative transformation to the right; each transformation completes one stitch. The final panoramic image is obtained through cumulative transformation. ; S3.2 Establishment of Panoramic-Local Mapping Relationship Record the spatial relationship between each original image and the panoramic image. For a local image, record the homography transformation matrix of that local image with respect to the reference image. , The correspondence between a point in a local image and a point in the panoramic image is as follows:

[0019] Save the homography matrix corresponding to each local image. , and its image sequence number.

[0020] The specific steps of step S4 are as follows: S4.1 Panoramic Image Target Detection The panoramic image obtained by stitching in step S3.1 is input into the Mask-RCNN model for feature extraction to obtain a feature map containing rich semantic information. Anchor boxes of corresponding proportions are selected based on the shape of the insulator and spacer, and regions containing insulators or spacers are generated on the feature map using the Region Proposal Network (RPN). The candidate regions containing target insulators or spacers output by the RPN are subjected to ROI pooling processing to map candidate regions of different sizes to feature maps of fixed size. The feature maps after ROI pooling are then used for classification and regression tasks. The classification task is used to predict whether an object within a candidate region belongs to an insulator or a spacer. The regression task is used to accurately predict the specific location of the target object and output more accurate bounding box coordinates. In the classification task, candidate regions containing insulators or spacers are segmented at the pixel level using the Mask branch. A binary mask is generated for each candidate region, and CBAM (Spatial Attention Module) is introduced to enhance the features of the target region, suppress background interference, and generate a mask for the target object, accurately outlining the contours of the insulator and spacer. Based on the obtained target mask and bounding box coordinates, the coordinates of the bottom suspension point of the insulator and the clamping end of the spacer are extracted. The specific operations are as follows: A morphological erosion operation is performed on the binary mask of the insulator (a pixel value of 1 represents the insulator region, and a pixel value of 0 represents the background region) to preserve the core target region of the insulator; the bottommost pixel of the bottom connected region is extracted. Coordinates; Perform the same morphological erosion operation on the spacer mask to obtain the core region that can characterize the state of the spacer assembly. Extract the endpoints of the spacer clamping ends from the eroded spacer mask image, denoted as... , ; S4.2 Extraction of Original Image Pairs For the target points of spacers and insulators extracted from the panoramic image in step S4.1 The homography transformation matrix obtained in step S3.2 is then applied sequentially to the extracted target points. inverse matrix , obtain the target point Let the original target point be denoted as the point at the th position. The pixel coordinates of the original image are used to determine whether it is a local image of the original image by checking if these pixel coordinates fall within the boundary range of the original image. Image to determine target point coordinate Whether it holds true or not, if true, it means that the original local image has been found. Extract the local image corresponding to the right eye camera from the stereo cameras based on the image sequence number. .

[0021] The specific steps of step S5 are as follows: S5.1 Disparity Map and Depth Calculation Based on the local image extracted according to the image sequence number in step S4.2 , The disparity map is generated using the semi-global matching (SGM) algorithm. First, the left image... And the right picture Convert to grayscale and normalize. For each pixel in the left image, select [the pixel as the center]. A Census transform is performed on a local window, and binary codes are generated by comparing the sizes of neighboring pixels with the center pixel. This serves as the local feature descriptor for the current pixel; candidate disparities are determined based on the scene depth range. In the same row of the right figure, for candidate disparity Perform a Census transform on the corresponding pixel to generate a binary code. The Hamming distance between the left and right pixels is calculated as the initial matching cost, as shown in the following formula:

[0022] Traverse the image horizontally (from left to right), for each pixel and candidate parallax Calculate its aggregation cost : in, As a penalty for adjacent parallax, Penalize large parallax jumps; select the candidate parallax with the lowest cost as the final parallax. Generate the final disparity map. ; According to the disparity map The specific formula for calculating depth information is as follows:

[0023] in, Focal length Using the baseline distance as the baseline distance, the depth information of the target point can be calculated by substituting the disparity information of the corresponding target point into the baseline distance. S5.2 3D Coordinate Transformation Based on the disparity map in step S5.1, the target point is obtained. depth information Then, its three-dimensional coordinates in the left camera coordinate system are obtained. The specific formula is as follows:

[0024] in, , The coordinates of the camera's optical center are used as the reference system for subsequent catenary fitting; for all acquired target points, all acquired target points are arranged according to... The coordinates are sorted from smallest to largest to generate an ordered three-dimensional point set.

[0025] The specific steps of step S6 are as follows: Step S6.1: Using the ordered 3D point set from step S5.2, perform preliminary catenary fitting and select... The plane is the fitting plane, and the catenary fitting equation is:

[0026]

[0027] in, For catenary tension parameters, For horizontal tension, The weight per unit length of the conductor. This is the lowest point of the catenary; Solving for the optimal parameters in the catenary fitting equation , , ; to obtain the three-dimensional coordinates of the target point in the ordered three-dimensional point set Projected to A two-dimensional target point set is obtained on the plane. Initial values ​​are set according to the characteristics of the conductor and the measurement scenario, with the coordinates of the midpoint of the span as... Initial values, based on a three-dimensional point set The minimum value is the initial value. The catenary parameters are solved by minimizing the sum of squared residuals between the actual height of the projected point and the fitted height. Step S6.2 Environmental parameters and dynamic correction of wind deflection The fitted catenary curve is dynamically corrected to take into account temperature changes and wind effects; real-time temperature and wind speed data are used as inputs to the correction model. Because the conductor's natural length changes due to temperature variations, the catenary tension parameters change, thus affecting the catenary's morphology. Therefore, a dynamic correction model is needed to adjust the catenary fitting parameters in real time. Temperature correction model:

[0028] in, Reference temperature The length of the conductor below, The coefficient of thermal expansion of the conductor material. This is the current measured temperature; according to and The relative transformation results for the catenary parameters Fine-tuning was performed to correct the catenary shape; the dynamic correction formula for horizontal tension is:

[0029] in, Current reference temperature The initial horizontal tension is obtained based on the dynamic correction formula for this horizontal tension. The catenary tension parameters in the corrected catenary model were calculated. ; Wind deflection correction addresses the impact of wind on the conductor. It involves overlaying a lateral offset onto the original catenary model, resulting in a corrected conductor trajectory that more closely reflects actual wind deformation. The lateral displacement of the conductor... The estimation formula is:

[0030] in, For wind load, This refers to the distance between the gears. The horizontal tension of the conductor; Step S6.3 Sag Calculation Based on the catenary model after temperature and wind correction in step S6.2, the coordinates of the lowest point of the curve are extracted and substituted into the catenary equation to obtain the height of the lowest point. The coordinates of the highest point are determined by the coordinates of the suspension points of the insulators at both ends. The average value of the bottom suspension points of the insulators at both ends is taken as the final insulator suspension point coordinates. ,Right now The calculated sag results Transmitted to the terminal.

[0031] A system for intelligent measurement of transmission line sag using a binocular depth camera, for implementing the intelligent measurement method for transmission line sag described in any one of steps 1 to 6, includes: The camera deployment and calibration module is used in step S1 to fix the binocular depth camera on the rotating gimbal, ensuring that the gimbal is fixed in the center position of the two towers, and to complete the initialization and calibration of the binocular depth camera, which can provide qualified hardware deployment and accurate parameter support for subsequent image acquisition and processing. The image acquisition and preprocessing module is used in step S2 to control the rotating gimbal to rotate in fixed angle steps. Each time the gimbal rotates, the binocular camera captures a set of images until it covers the entire span of the wire. The captured images are preprocessed to obtain a preprocessed binocular image sequence. The panoramic stitching and mapping module is used in step S3. It takes the left camera of the binocular depth camera as the main view, imports the preprocessed left eye image sequence, uses the SIFT algorithm to achieve image alignment, completes panoramic image stitching, and establishes the mapping relationship between the local image and the panoramic image. The target detection and original image mapping module is used in step S4 to perform target detection of insulators and spacers in the stitched panoramic image using the Mask-RCNN model, obtain image coordinates, and map them back to the original image of the panoramic image. The depth calculation and 3D modeling module is used in step S5. Based on the original image mapped back in step S4, the target area is cropped from the left and right images of the binocular depth camera, and a disparity map is generated by stereo matching. The depth information of the spacer rod and insulator target points is calculated, and a 3D coordinate transformation is performed to generate an ordered 3D point set. The catenary fitting and sag calculation module is used in step S6 to perform dynamic catenary fitting based on the obtained ordered three-dimensional point set, and obtain the final sag calculation result.

[0032] A device for intelligent measurement of power transmission line sag using a binocular depth camera includes: Memory, used to store computer programs; A processor is used to implement the intelligent measurement method for transmission line sag described in any one of steps 1 to 6 when executing the computer program.

[0033] A computer-readable storage medium storing a computer program, which, when executed by a processor, enables intelligent measurement of transmission line sag using a binocular depth camera based on the intelligent measurement method for transmission line sag described in any one of steps 1 to 6.

[0034] Compared with the prior art, the technical solution of the present invention has the following beneficial effects: 1. Significantly improves operational efficiency. Local images are captured by rotating the gimbal, and panoramic stitching is achieved using the SIFT matching algorithm. Object detection is performed using the Mask-RCNN algorithm, and depth information is obtained by constructing a disparity map from the local original image through mapping relationships, ultimately completing the catenary fitting. The gimbal automatically acquires images and performs subsequent image processing, significantly improving operational efficiency compared to traditional manual measurement methods while reducing reliance on manual labor.

[0035] 2. No contact measurement is required, avoiding potential safety issues associated with manual sensor installation and reducing the risk of accidents due to conductor tension. Image data acquisition via a long-distance binocular module allows for operation of overhead transmission lines with large spans and complex terrain.

[0036] 3. Compared to traditional devices such as LiDAR, drones, and wireless sensor networks, this method significantly reduces overall costs while avoiding complex processing of point cloud data. Compared to monocular vision's reliance on prior knowledge to estimate depth information, this method uses a binocular module to construct a disparity map to acquire depth information, thus improving measurement accuracy while avoiding background interference.

[0037] Therefore, this invention constructs an intelligent system for sag measurement that integrates panoramic stitching and local depth information acquisition. A rotating gimbal acquires images to complete panoramic image stitching, and a binocular module acquires depth information of insulator suspension points and spacer clamping ends in the transmission line, fitting the catenary to complete the sag measurement. This automates the entire process from task planning, target recognition, data acquisition to result output, significantly improving the safety, efficiency, and accuracy of transmission line inspection. It is particularly suitable for power network operation and maintenance scenarios with complex terrain and large spans, greatly improving the problems of high cost, low efficiency, and insufficient accuracy of existing sag measurement technologies. Attached Figure Description

[0038] Figure 1 This is a flowchart of the process of the present invention.

[0039] Figure 2 This is a hardware layout diagram of the present invention.

[0040] Figure 3 This is a schematic diagram illustrating the principle of catenary fitting and sag calculation in this invention. Detailed Implementation

[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0042] The steps of the method embodiment are explained below with reference to the accompanying drawings.

[0043] in accordance with Figure 1 The specific workflow provided is as follows: A method for intelligent measurement of power line sag using a binocular depth camera, with the following steps: Step 1: Fix the binocular camera on the rotating gimbal, ensuring that the gimbal is fixed in the center of the two towers, and complete the initialization and calibration of the binocular camera.

[0044] Based on the operational scenario, select the center position between the two towers, install a rigid bracket and a rotating gimbal, and fix the binocular depth camera on the rotating gimbal. After deployment, determine the baseline of the two cameras according to the empirical formula given in step S1.1. After adjusting the binocular camera to be level using a bubble level, the binocular camera calibration and standardization operations are performed. For the left and right cameras, their intrinsic parameters are first calibrated, and the intrinsic parameters of each camera are calculated using a calibration board: focal length. Like the main point ; and distortion coefficient ,in , The radial distortion coefficient is... , Tangential distortion coefficients. After the intrinsic parameters are calibrated, binocular joint extrinsic parameter calibration and epipolar correction are performed, and the extrinsic parameter matrix is ​​solved using the PnP (Perspective-n-Point) algorithm. The Bouguet algorithm is used to perform stereo correction on the left and right camera images to ensure that the epipolar lines of the left and right images are horizontally aligned, and only horizontal parallax is retained.

[0045] Step 2: Control the rotating gimbal to rotate in fixed-angle steps. The binocular camera takes a set of images for each frame of rotation until the entire span of the guide wire is covered.

[0046] Control the rotating gimbal to fix the step angle Rotation is performed to ensure image continuity and assist in subsequent image stitching, and the overlap rate between adjacent frames is adjusted. The overlap should be no less than 30%, and an appropriate overlap rate should be selected. The step angle is determined according to the formula in step S2.1. For each step angle rotation, the left and right cameras simultaneously acquire a set of images. Based on the shooting angle and time, an incremental sequence number is generated to mark the image pairs. Save the original image size information Information such as the gimbal angle and focal length during shooting, and the required number of image pairs calculated in step S2.1. Determine whether image acquisition of the entire span area is complete. After image acquisition is complete, preprocess the acquired local images, perform distortion correction and epipolar correction to complete image alignment; use median filtering to remove noise, use histogram equalization (CLAHE) to enhance contrast, and perform color temperature unification processing based on the first image to improve the robustness of subsequent stitching and matching.

[0047] Step 3: Using the left camera of the binocular camera as the main view, the SIFT algorithm is used to align the images, complete the panoramic image stitching, and establish the mapping relationship between the local image and the panoramic image.

[0048] S3.1 Panoramic Image Stitching and Fusion Using the left camera of the binocular camera as the main view, import its preprocessed left-eye image sequence. Feature point extraction using the SIFT algorithm: This involves searching for image locations across all scale spaces, using an initial Gaussian blur coefficient. The number of generated layers is A Gaussian pyramid with multiple layers generates a difference Gaussian image using a Gaussian differential function, identifying candidate feature points in the image that are scale- and rotation-invariant. ,in, For pixel coordinates, For scale space, The main direction of the feature point, For each pair of adjacent images, the corresponding descriptor vector is used. , Preliminary feature point matching is performed using Fast Nearest Neighbor Search (FLANN). descriptor vector set , descriptor vector set Calculate the Euclidean distance between two vectors, and select the vectors with the smallest distance as candidate matching pairs to obtain a preliminary matching set. ; False matches are eliminated using the RANSAC algorithm: the initial homography matrix of the feature points of the candidate matching pairs is estimated, and the projection error is calculated. ,in, for The actual coordinates of the feature points are as follows: for The corresponding feature points are in The projected coordinates on the screen, the filtering error threshold is within Points within the matrix are removed to eliminate incorrect matches, and the final homography matrix is ​​obtained using the least squares method. ; Following a left-to-right splicing strategy, with Using the reference image as the base image, perform a cumulative transformation to the right; each transformation completes one stitch. The final panoramic image is obtained through cumulative transformation. ; S3.2 Establishment of Panoramic-Local Mapping Relationship Record the spatial relationship between each original image and the panoramic image. For a local image, record the homography transformation matrix of that local image with respect to the reference image. , The correspondence between a point in a local image and a point in the panoramic image is as follows:

[0049] Save the homography matrix corresponding to each local image. , and its image sequence number.

[0050] Step 4: For the stitched panoramic image, use the Mask-RCNN model to perform target detection on the insulators and spacers, obtain their image coordinates, and map them back to their original images.

[0051] S4.1 The panoramic image obtained by stitching in step S3.1 is input into the Mask-RCNN model for feature extraction to obtain a feature map containing rich semantic information; anchor boxes of corresponding proportions are selected based on the shape of the insulator and the spacer (based on the cylindrical contour features of the insulator and the cross or straight contour features of the spacer), and regions that may contain insulators or spacers are generated on the feature map using the Region Proposal Network (RPN); ROI pooling is performed on the candidate regions that may contain target insulators or spacers output by the RPN to map candidate regions of different sizes to feature maps of fixed size; The feature maps after ROI pooling are then used for classification and regression tasks. The classification task is used to predict whether an object in the candidate region belongs to an insulator or a spacer. The regression task is used to accurately predict the specific location of the target object and output more accurate bounding box coordinates.

[0052] In the classification task, candidate regions that may contain insulators or spacers are segmented at the pixel level using the Mask branch. A binary mask is generated for each candidate region, and CBAM (Spatial Attention Module) is introduced to enhance the features of the target region, suppress background interference, and generate a mask for the target object, accurately outlining the contours of the insulator and spacer. Based on the obtained target mask and bounding box coordinates, the coordinates of the bottom suspension point of the insulator and the clamping end of the spacer are extracted. The specific operations are as follows: A morphological erosion operation is performed on the binary mask of the insulator (a pixel value of 1 represents the insulator region, and a pixel value of 0 represents the background region) to preserve the core target region of the insulator; the bottommost pixel of the bottom connected region is extracted. Coordinates; Perform the same morphological erosion operation on the spacer mask to obtain the core region that can characterize the state of the spacer assembly. Extract the endpoints of the spacer clamping ends from the eroded spacer mask image, denoted as... , ; S4.2 Extraction of Original Image Pairs For the target points of spacers and insulators extracted from the panoramic image in step S4.1 The homography transformation matrix obtained in step S3.2 is then applied sequentially to the extracted target points. inverse matrix , obtain the target point Let the original target point be denoted as the point at the th position. The pixel coordinates of the original image are used to determine whether it is a local image of the original image by checking if these pixel coordinates fall within the boundary range of the original image. Image to determine target point coordinate Whether it holds true or not, if true, it means that the original local image has been found. Extract the local image corresponding to the right eye camera from the stereo cameras based on the image sequence number. .

[0053] Step 5: Based on the original image, crop the target area from the left and right images of the binocular camera, perform stereo matching to generate a disparity map, and calculate the depth information.

[0054] S5.1 Disparity Map and Depth Calculation Based on the local image extracted according to the image sequence number in step S4.2 , The disparity map is generated using the semi-global matching (SGM) algorithm. First, the left image... And the right picture Convert to grayscale and normalize. For each pixel in the left image, select [the pixel as the center]. A Census transform is performed on a local window, and binary codes are generated by comparing the sizes of neighboring pixels with the center pixel. This serves as the local feature descriptor for the current pixel; candidate disparities are determined based on the scene depth range. In the same row of the right figure, for candidate disparity Perform a Census transform on the corresponding pixel to generate a binary code. The Hamming distance between the left and right pixels is calculated as the initial matching cost, as shown in the following formula:

[0055] Traverse the image horizontally (from left to right), for each pixel and candidate parallax Calculate its aggregation cost : in, As a penalty for adjacent parallax, Penalize large parallax jumps; select the candidate parallax with the lowest cost as the final parallax. Generate the final disparity map. ; According to the disparity map The specific formula for calculating depth information is as follows:

[0056] in, Focal length Using the baseline distance as the baseline distance, the depth information of the target point can be calculated by substituting the disparity information of the corresponding target point into the baseline distance. S5.2 3D Coordinate Transformation Based on the disparity map in step S5.1, the target point is obtained. depth information Then, its three-dimensional coordinates in the left camera coordinate system are obtained. The specific formula is as follows:

[0057] in, , The coordinates of the camera's optical center are used as the reference system for subsequent catenary fitting; for all acquired target points, all acquired target points are arranged according to... The coordinates are sorted from smallest to largest to generate an ordered three-dimensional point set.

[0058] Step 6: After completing the local module work for all spacers and insulators, perform dynamic fitting of the catenary based on the obtained three-dimensional coordinate information to obtain the final sag calculation result.

[0059] Yiru Figure 3 As shown. Step S6.1: Using the ordered 3D point set from step S5.2, perform preliminary catenary fitting and select... The plane is the fitting plane, and the catenary fitting equation is:

[0060]

[0061] in, For catenary tension parameters, For horizontal tension, The weight per unit length of the conductor. This is the lowest point of the catenary; Solving for the optimal parameters in the catenary fitting equation , , ; to obtain the three-dimensional coordinates of the target point in the ordered three-dimensional point set Projected to A two-dimensional target point set is obtained on the plane. Initial values ​​are set according to the characteristics of the conductor and the measurement scenario, with the coordinates of the midpoint of the span as... Initial values, based on a three-dimensional point set The minimum value is the initial value. The catenary parameters are solved by minimizing the sum of squared residuals between the actual height of the projected point and the fitted height. Step S6.2 Environmental parameters and dynamic correction of wind deflection The fitted catenary curve is dynamically corrected to take into account temperature changes and wind effects; real-time temperature and wind speed data are provided by ground base stations and used as inputs to the correction model. Because the conductor's natural length changes due to temperature variations, the catenary tension parameters change, thus affecting the catenary's morphology. Therefore, a dynamic correction model is needed to adjust the catenary fitting parameters in real time. Temperature correction model:

[0062] in, Reference temperature The length of the conductor below, The coefficient of thermal expansion of the conductor material. This is the current measured temperature; according to and The relative transformation results for the catenary parameters Fine-tuning was performed to correct the catenary shape; the dynamic correction formula for horizontal tension is:

[0063] in, Current reference temperature The initial horizontal tension is obtained based on the dynamic correction formula for this horizontal tension. The catenary tension parameters in the corrected catenary model were calculated. ; Wind deflection correction addresses the impact of wind on the conductor. It involves overlaying a lateral offset onto the original catenary model, resulting in a corrected conductor trajectory that more closely reflects actual wind deformation. The lateral displacement of the conductor... The estimation formula is:

[0064] in For wind load, This refers to the distance between the gears. The horizontal tension of the conductor; Step S6.3 Sag Calculation Based on the catenary model after temperature and wind correction in step S6.2, the coordinates of the lowest point of the curve are extracted and substituted into the catenary equation to obtain the height of the lowest point. The coordinates of the highest point are determined by the coordinates of the suspension points of the insulators at both ends, and the average value of the bottom suspension points of the insulators at both ends is taken as the final value. ,Right now The calculated sag results Transmitted to the terminal; The calculated sag results can be used to further determine the operating status of power transmission and can also be used to assist in optimizing the design and planning of power transmission lines.

[0065] The following is combined with Figure 2 and Figure 3A specific embodiment of the method of the present invention is provided.

[0066] See Figure 2 In the simulated sag measurement scenario of this embodiment, a typical high-voltage transmission line is used as the measurement object. The measurement system is deployed at a ground measuring point near the center line connecting the towers on both sides of the target span, preferably in an open area near the midpoint of the two towers to ensure an unobstructed view. The ground measuring point is natural soil or a simple, leveled construction access road, requiring no special foundation for equipment placement. A rotating gimbal is erected at this location, and the binocular depth camera is fixed to the gimbal with a rigid bracket to ensure stability during shooting. The optical axis of the binocular depth camera is adjusted to be parallel and horizontal. Subsequently, the binocular camera is calibrated and polarimetrically corrected according to the method described in step S1 of this invention, completing the initialization process before sag measurement.

[0067] After completing the hardware deployment, proceed with image acquisition and preprocessing operations according to step S2. Control the rotating gimbal to rotate at a calculated step angle, and simultaneously acquire images with the binocular camera until an image sequence covering the entire span of the conductor is obtained. Perform image preprocessing operations such as distortion correction and color temperature adjustment.

[0068] The subsequent processing procedure is as follows: Panoramic stitching and mapping: Using the preprocessed left-eye image sequence as the main view, the SIFT algorithm is used to perform feature matching and stitching according to the method described in step S3 to generate a panoramic image, and a mapping relationship between the local image and the panoramic image is established.

[0069] Target detection and original image mapping: For the obtained panoramic image, according to step S4, the Mask-RCNN model is used to detect the insulators and spacers, extract their key point coordinates, and apply the inverse mapping relationship to backtrack to the original local binocular image pair.

[0070] 3D coordinate acquisition and depth calculation: In the original local binocular image pair that has been located, according to step S5, the SGM algorithm is applied to calculate the disparity map, thereby obtaining the depth information of the key points and converting it to generate an ordered 3D point set.

[0071] Catenary sag calculation: The ordered 3D point set is projected onto the fitting plane, and the catenary is initially fitted according to step S6.1. Then, according to step S6.2, real-time environmental parameters are introduced to dynamically correct the catenary for temperature and wind deflection. Finally, according to steps S6.3 and... Figure 3 Based on the principle shown, the height difference between the lowest point of the catenary and the suspension points of the insulators at both ends is calculated to obtain the sag measurement results.

[0072] The above description is merely a more detailed explanation of the technical solution of this embodiment, and is not intended to limit it. Those skilled in the art should understand that modifications can be made to the technical solution, and deficiencies can be improved or replaced, while the modifications should still be within the technical scope of this invention.

Claims

1. A method for intelligent measurement of transmission line sag using a binocular depth camera, characterized in that, Includes the following steps: Step S1: Fix the binocular depth camera on the rotating gimbal, ensuring that the gimbal is fixed in the center position of the two towers, and complete the initialization and calibration of the binocular depth camera. Step S2: Control the rotating gimbal to rotate in fixed-angle steps. Each time the camera rotates for one frame, it takes a set of images until it covers the entire span of the guide wire. Perform preprocessing on the captured images to obtain a preprocessed binocular image sequence. Step S3: Using the left camera of the binocular depth camera as the main view, import the preprocessed left camera image sequence, use the SIFT algorithm to align the images, complete the panoramic image stitching, and establish the mapping relationship between the local image and the panoramic image. Step S4: For the stitched panoramic image, use the Mask-RCNN model to perform target detection on the insulators and spacers, obtain the image coordinates, and map them back to the original panoramic image. Step S5: Based on the original image mapped back from Step S4, the target area is cropped out from the left and right images of the binocular depth camera, and a disparity map is generated by stereo matching. The depth information of the spacer rod and insulator target points is calculated, and a three-dimensional coordinate transformation is performed to generate an ordered three-dimensional point set. Step S6: Perform dynamic fitting of the catenary based on the obtained ordered three-dimensional point set to obtain the final sag calculation result.

2. The method for intelligent measurement of transmission line sag using a binocular depth camera according to claim 1, characterized in that, The specific steps of step S1 are as follows: S1.1 Camera Deployment and Baseline Setting Check that the left and right binocular depth cameras and the binocular depth cameras are in normal working connection with the computer; install the bracket, binocular cameras, and rotating head according to the center position between the two towers, and adjust according to the guide wire height. Select the baseline according to the empirical formula. , , The empirical coefficient is used; the binocular depth camera bracket is adjusted to make the optical axes of the two cameras parallel and horizontal, and the left and right binocular depth cameras are rigidly connected to avoid vibration; Calibration and calibration of S1.2 binocular depth camera Customize the image resolution and acquisition frame rate of the stereo depth camera, set the image exposure time, and adjust the color balance; calibrate and standardize the intrinsic parameters of the left and right stereo depth cameras using a calibration board to obtain the corresponding intrinsic parameter parameters and focal length of each stereo depth camera. Like the main point distortion coefficient ,in, , The radial distortion coefficient is... , Tangential distortion coefficients; for binocular extrinsic parameter calibration, the extrinsic parameter matrix is ​​solved using the PnP (Perspective-n-Point) algorithm. ,in, For rotation matrix, The translation vector is used; after completing the binocular positioning, epipolar correction is performed using the Bouguet algorithm; based on the intrinsic, extrinsic, and distortion parameters of the left and right binocular depth cameras, the correction rotation matrices of the left and right binocular depth cameras are obtained. , and the corrected common intrinsic parameter matrix Generate left and right image remapping matrices , This eliminates geometric distortion and parallax angle shift between images, ensuring that the content of the left and right images is strictly aligned in the scan line direction, while retaining only horizontal parallax.

3. The method for intelligent measurement of transmission line sag using a binocular depth camera according to claim 1, characterized in that, The specific steps of step S2 are as follows: S2.1 Image Pair Acquisition and Sequence Generation Calculate the step angle of the gimbal rotation based on the field of view (FOV) of the binocular depth camera and the guide wire spacing: The horizontal field of view (HFOV) represents the imaging range of a binocular depth camera, and is calculated using the following formula: in, The width of the built-in sensor for the binocular depth camera. For camera focal length, For the gimbal step angle, The overlap rate; According to the conductor span length And the horizontal distance between the wire and the camera Calculate the coverage length of a single image and the number of frames required to complete the panorama, where the coverage width of a single image is: Number of image pairs required to complete panoramic shooting for: When the calculation result is a decimal, then Round up; Left and right binocular depth cameras simultaneously acquire images from the same angle, and image pairs are generated with incrementally increasing sequence numbers based on the shooting angle and time. Save the original image size information Information such as the gimbal angle and focal length during shooting is used for subsequent operations; S2.2 Image Preprocessing The images acquired in step S2.1 are preprocessed. Distortion correction is performed using the parameters calibrated in step S1.2 to eliminate radial and tangential distortion. Based on the correction mapping relationship generated in step S1, epipolar correction is performed on the left and right images to ensure strict alignment of the content. The images are then denoised and enhanced. Using the color temperature of the first local power line image containing insulators as a base, and the insulator region as a reference region, the color temperature reference parameters are determined based on the average brightness of the red, green, and blue channels of the insulator region. For example, using the green channel color temperature as a reference, the gain coefficients of the red and blue channels relative to the green channel are obtained simultaneously. Subsequent images are then multiplied pixel by pixel by the corresponding gain coefficient to adjust the color temperature accordingly, ensuring that the color temperature of subsequent local images is consistent with the reference image and avoiding stitching defects. This results in a preprocessed binocular image sequence.

4. The method for intelligent measurement of transmission line sag using a binocular depth camera according to claim 1, characterized in that, The specific steps of step S3 are as follows: S3.1 Panoramic Image Stitching and Fusion Using the left camera of a binocular depth camera as the main view, import the preprocessed left camera image sequence. Feature point extraction using the SIFT algorithm: This involves searching for image locations across all scale spaces, using an initial Gaussian blur coefficient. The number of generated layers is A Gaussian pyramid with multiple layers generates a difference Gaussian image using a Gaussian differential function, identifying candidate feature points in the image that are scale- and rotation-invariant. ,in, For pixel coordinates, For scale space, The main direction of the feature point, For each pair of adjacent images, the corresponding descriptor vector is used. , Preliminary feature point matching is performed using Fast Nearest Neighbor Search (FLANN). descriptor vector set , descriptor vector set Calculate the Euclidean distance between two vectors, and select the vectors with the smallest distance as candidate matching pairs to obtain a preliminary matching set. ; False matches are eliminated using the RANSAC algorithm: the initial homography matrix of the feature points of the candidate matching pairs is estimated, and the projection error is calculated. ,in, for The actual coordinates of the feature points are as follows: for The corresponding feature points are in The projected coordinates on the screen, the filtering error threshold is within Points within the matrix are removed to eliminate incorrect matches, and the final homography matrix is ​​obtained using the least squares method. ; Following a left-to-right splicing strategy, with Using the reference image as the base image, perform a cumulative transformation to the right; each transformation completes one stitch. The final panoramic image is obtained through cumulative transformation. ; S3.2 Establishment of Panoramic-Local Mapping Relationship Record the spatial relationship between each original image and the panoramic image. For a local image, record the homography transformation matrix of that local image with respect to the reference image. , The correspondence between a point in a local image and a point in the panoramic image is as follows: Save the homography matrix corresponding to each local image. , and its image sequence number.

5. The method for intelligent measurement of transmission line sag using a binocular depth camera according to claim 1, characterized in that, The specific steps of step S4 are as follows: S4.1 Panoramic Image Target Detection The panoramic image obtained by stitching in step S3.1 is input into the Mask-RCNN model for feature extraction to obtain a feature map containing rich semantic information. Anchor boxes of corresponding proportions are selected based on the shape of the insulator and spacer, and regions containing insulators or spacers are generated on the feature map using the Region Proposal Network (RPN). The candidate regions containing target insulators or spacers output by the RPN are subjected to ROI pooling processing to map candidate regions of different sizes to feature maps of fixed size. The feature maps after ROI pooling are then used for classification and regression tasks. The classification task is used to predict whether an object within a candidate region belongs to an insulator or a spacer. The regression task is used to accurately predict the specific location of the target object and output more accurate bounding box coordinates. In the classification task, candidate regions containing insulators or spacers are segmented at the pixel level using the Mask branch. A binary mask is generated for each candidate region, and CBAM (Spatial Attention Module) is introduced to enhance the features of the target region, suppress background interference, and generate a mask for the target object, accurately outlining the contours of the insulator and spacer. Based on the obtained target mask and bounding box coordinates, the coordinates of the bottom suspension point of the insulator and the clamping end of the spacer are extracted. The specific operations are as follows: A morphological erosion operation is performed on the binary mask of the insulator (a pixel value of 1 represents the insulator region, and a pixel value of 0 represents the background region) to preserve the core target region of the insulator; the bottommost pixel of the bottom connected region is extracted. Coordinates; Perform the same morphological erosion operation on the spacer mask to obtain the core region that can characterize the geometric state of the spacer. Extract the endpoints of the spacer clamping ends from the eroded spacer mask image, denoted as... , ; S4.2 Extraction of Original Image Pairs For the target points of spacers and insulators extracted from the panoramic image in step S4.1 The homography transformation matrix obtained in step S3.2 is then applied sequentially to the extracted target points. inverse matrix , obtain the target point Let the original target point be denoted as the point at the th position. The pixel coordinates of the original image are used to determine whether it is a local image of the original image by checking if these pixel coordinates fall within the boundary range of the original image. Image to determine target point coordinate Whether it holds true or not, if true, it means that the original local image has been found. Extract the local image corresponding to the right eye camera from the stereo cameras based on the image sequence number. .

6. The method for intelligent measurement of transmission line sag using a binocular depth camera according to claim 1, characterized in that, The specific steps of step S5 are as follows: S5.1 Disparity Map and Depth Calculation Based on the local image extracted according to the image sequence number in step S4.2 , The disparity map is generated using the semi-global matching (SGM) algorithm. First, the left image... And the right picture Convert to grayscale and normalize. For each pixel in the left image, select [the pixel as the center]. A Census transform is performed on a local window, and binary codes are generated by comparing the sizes of neighboring pixels with the center pixel. This serves as the local feature descriptor for the current pixel; candidate disparities are determined based on the scene depth range. In the same row of the right figure, for candidate disparity Perform a Census transform on the corresponding pixel to generate a binary code. The Hamming distance between the left and right pixels is calculated as the initial matching cost, as shown in the following formula: Traverse the image horizontally (from left to right), for each pixel and candidate parallax Calculate its aggregation cost : in, As a penalty for adjacent parallax, Penalize large parallax jumps; select the candidate parallax with the lowest cost as the final parallax. Generate the final disparity map. ; According to the disparity map The specific formula for calculating depth information is as follows: in, Focal length Using the baseline distance as the baseline distance, the depth information of the target point can be calculated by substituting the disparity information of the corresponding target point into the baseline distance. S5.2 3D Coordinate Transformation Based on the disparity map in step S5.1, the target point is obtained. depth information Then, its three-dimensional coordinates in the left camera coordinate system are obtained. The specific formula is as follows: in, , The coordinates of the camera's optical center are used as the reference system for subsequent catenary fitting; for all acquired target points, all acquired target points are arranged according to... The coordinates are sorted from smallest to largest to generate an ordered three-dimensional point set.

7. The method for intelligent measurement of transmission line sag using a binocular depth camera according to claim 1, characterized in that, The specific steps of step S6 are as follows: Step S6.1: Using the ordered 3D point set from step S5.2, perform preliminary catenary fitting and select... The plane is the fitting plane, and the catenary fitting equation is: in, For catenary tension parameters, For horizontal tension, Weight per unit length of conductor. This is the lowest point of the catenary; Solving for the optimal parameters in the catenary fitting equation , , ; to obtain the three-dimensional coordinates of the target point in the ordered three-dimensional point set Projected to A two-dimensional target point set is obtained on the plane. Initial values ​​are set according to the characteristics of the conductor and the measurement scenario, with the coordinates of the midpoint of the span as... Initial values, based on a three-dimensional point set The minimum value is the initial value. The catenary parameters are solved by minimizing the sum of squared residuals between the actual height of the projected point and the fitted height. Step S6.2 Environmental parameters and dynamic correction of wind deflection The fitted catenary curve is dynamically corrected to take into account temperature changes and wind effects; real-time temperature and wind speed data are used as inputs to the correction model. Because the conductor's natural length changes due to temperature variations, the catenary tension parameters change, thus affecting the catenary's morphology. Therefore, a dynamic correction model is needed to adjust the catenary fitting parameters in real time. Temperature correction model: in, Reference temperature The length of the conductor below, The coefficient of thermal expansion of the conductor material. This is the current measured temperature; according to and The relative transformation results for the catenary parameters Fine-tuning was performed to correct the catenary shape; the dynamic correction formula for horizontal tension is: in, Current reference temperature The initial horizontal tension is obtained based on the dynamic correction formula for this horizontal tension. The catenary tension parameters in the corrected catenary model were calculated. ; Wind deflection correction addresses the impact of wind on the conductor. It involves overlaying a lateral offset onto the original catenary model, resulting in a corrected conductor trajectory that more closely reflects actual wind deformation. The lateral displacement of the conductor... The estimation formula is: in, For wind load, This refers to the distance between the gears. The horizontal tension of the conductor; Step S6.3 Sag Calculation Based on the catenary model after temperature and wind correction in step S6.2, the coordinates of the lowest point of the curve are extracted and substituted into the catenary equation to obtain the height of the lowest point. The coordinates of the highest point are determined by the coordinates of the suspension points of the insulators at both ends. The average value of the bottom suspension points of the insulators at both ends is taken as the final insulator suspension point coordinates. ,Right now The calculated sag results Transmitted to the terminal.

8. A system for intelligent measurement of power transmission line sag using a binocular depth camera, characterized in that, The method for intelligent measurement of transmission line sag according to any one of claims 1 to 7 includes: The camera deployment and calibration module is used in step S1 to fix the binocular depth camera on the rotating gimbal, ensuring that the gimbal is fixed in the center position of the two towers, and to complete the initialization and calibration of the binocular depth camera, which can provide qualified hardware deployment and accurate parameter support for subsequent image acquisition and processing. The image acquisition and preprocessing module is used in step S2 to control the rotating gimbal to rotate in fixed angle steps. Each time the gimbal rotates, the binocular camera captures a set of images until it covers the entire span of the wire. The captured images are preprocessed to obtain a preprocessed binocular image sequence. The panoramic stitching and mapping module is used in step S3. It takes the left camera of the binocular depth camera as the main view, imports the preprocessed left eye image sequence, uses the SIFT algorithm to achieve image alignment, completes panoramic image stitching, and establishes the mapping relationship between the local image and the panoramic image. The target detection and original image mapping module is used in step S4 to perform target detection of insulators and spacers in the stitched panoramic image using the Mask-RCNN model, obtain image coordinates, and map them back to the original image of the panoramic image. The depth calculation and 3D modeling module is used in step S5. Based on the original image mapped back in step S4, the target area is cropped from the left and right images of the binocular depth camera, and a disparity map is generated by stereo matching. The depth information of the spacer rod and insulator target points is calculated, and a 3D coordinate transformation is performed to generate an ordered 3D point set. The catenary fitting and sag calculation module is used in step S6 to perform dynamic catenary fitting based on the obtained ordered three-dimensional point set, and obtain the final sag calculation result.

9. A device for intelligent measurement of power transmission line sag using a binocular depth camera, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the intelligent measurement method for transmission line sag according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables intelligent measurement of transmission line sag using a binocular depth camera based on the intelligent measurement method for transmission line sag according to any one of claims 1 to 7.

Images