A power equipment position dynamic correction method based on image feature matching

Abstract

The application belongs to the technical field of transformer controller, and specifically discloses a power equipment position dynamic deviation correction method based on image feature matching, which comprises the following steps: collecting a power equipment image sequence through a camera, extracting an initial feature point set from the sequence to obtain original data representation containing perspective effect; according to the extracted initial feature point set, analyzing the scale difference caused by distance change by using a scale invariant feature transform algorithm to determine the spatial offset vector of the feature points at different distances; if the modulus of the determined spatial offset vector exceeds a preset threshold, correcting the perspective effect in the image sequence through a perspective transform matrix to obtain a corrected feature point set; obtaining edge contour information based on the corrected feature point set from the corrected image, judging whether the contour definition meets the recognition requirement to obtain an optimized equipment boundary description; and the application aims to solve the problem of poor recognition accuracy of the state of power equipment in a complex environment in the prior art.

Description

Technical Field

[0001] This invention relates to the field of transformer control technology, specifically to a method for dynamic position correction of power equipment based on image feature matching. Background Technology

[0002] In the field of power equipment inspection and maintenance, image processing technology is crucial due to its ability to achieve non-contact and efficient equipment condition monitoring. Especially in the positioning and correction of power equipment, image technology provides key support for ensuring equipment operation safety and power grid stability. However, research and application in this field still face many challenges, urgently requiring innovative methods to improve technical accuracy and adaptability. Currently, most methods for power equipment location detection rely on image analysis techniques with fixed parameters, which often prove inadequate in complex environments. Particularly during equipment image acquisition, changes in the distance between the camera and the equipment can lead to different perspective effects in the image, directly interfering with the accuracy of location determination. Existing solutions often struggle to dynamically adapt to these changes, especially when distance differences are significant, making it easy to misinterpret equipment features in the image, leading to location correction failures.

[0003] A deeper technical challenge lies in the fact that variations in image acquisition distance not only affect the aspect ratio but also interfere with the spatial representation of device features. Distance changes cause varying degrees of clarity in the details of device edges within the image, leading to deviations in feature point recognition and spatial offset calculations. This deviation may manifest as over-magnification of details in close-up shots, while in long-distance shots, it may result in loss of detail, making it impossible to accurately capture device boundaries. This difference in feature representation caused by distance becomes the core obstacle affecting the accuracy of position correction.

[0004] Therefore, how to dynamically adjust the parameters of image analysis to accurately identify equipment features and calculate spatial offset when the distance between the camera and power equipment is constantly changing has become a key issue in improving the position correction effect of power equipment. Solving this problem not only involves technological breakthroughs but also directly affects the reliability of real-time monitoring and adjustment of the positions of equipment such as transformers and circuit breakers during power line inspections. Summary of the Invention

[0005] This invention provides a dynamic position correction method for power equipment based on image feature matching, aiming to solve the problem of poor recognition accuracy of power equipment status in complex environments in the prior art.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0007] A dynamic position correction method for power equipment based on image feature matching includes: acquiring a sequence of images of power equipment via a camera; extracting an initial feature point set from the sequence to obtain a raw data representation including perspective effects; analyzing the proportional differences caused by distance changes using a scale-invariant feature transformation algorithm based on the extracted initial feature point set to determine the spatial offset vectors of feature points at different distances; if the magnitude of the determined spatial offset vector exceeds a preset threshold, correcting the perspective effect in the image sequence using a perspective transformation matrix to obtain a corrected feature point set; obtaining edge contour information from the corrected image based on the corrected feature point set, determining whether the contour clarity meets the recognition requirements, and obtaining an optimized equipment boundary description; for the optimized equipment boundary description, using an optical flow algorithm to track the dynamic motion trajectory under distance changes to determine the real-time position offset; if the real-time position offset deviates from the standard range, fitting the correspondence between the offset and distance changes using the least squares method to obtain an adaptive parameter adjustment model; updating the analysis parameters in the image processing flow based on the adaptive parameter adjustment model, determining whether the updated parameters improve the feature recognition accuracy, and obtaining a final correction command sequence for power grid stability monitoring.

[0008] In one aspect of this disclosure, the step of acquiring an image sequence of power equipment via a camera, extracting an initial feature point set from the sequence, and obtaining a raw data representation including a perspective effect includes:

[0009] By continuously acquiring image sequences of power equipment through cameras, a raw dataset containing multiple frames of images is obtained.

[0010] Image preprocessing techniques were used to denoise and adjust the contrast of the acquired raw dataset to obtain a clear image set after processing.

[0011] For the processed set of clear images, the initial feature points in each frame are extracted to obtain a feature data set containing multiple feature point sets.

[0012] If the number of feature points in the feature data set is lower than the preset threshold, then perform secondary feature extraction on the corresponding frame image to determine the supplemented feature point set.

[0013] Based on the supplemented feature point set, analyze the impact of perspective effect on the feature point position, and obtain the adjustment data set with perspective correction;

[0014] By adjusting the data group, the original data representation of the power equipment is constructed, and it is determined whether the data representation meets the preset integrity standard.

[0015] If the data representation does not meet the completeness standard, the adjusted data group is iteratively optimized to obtain the final power equipment data representation result.

[0016] In one aspect of this disclosure, the step of analyzing the proportional differences caused by distance changes using a scale-invariant feature transformation algorithm based on the extracted initial feature point set, and determining the spatial offset vectors of feature points at different distances, includes:

[0017] For the feature point set, the scale-invariant feature transformation algorithm is used to handle the proportional differences caused by distance changes and obtain preliminary spatial offset data of feature points under different distance conditions.

[0018] Based on the preliminary spatial offset data, the vector calculation method is used to analyze the correspondence between spatial offset and offset vector, and to determine the specific offset vector of the feature point under positional changes.

[0019] By using offset vectors, a model of the positional variation of feature points under different distance conditions is constructed to obtain the spatial distribution adjustment results of the feature point set;

[0020] If the positional changes of feature points exceed a preset threshold, the point set analysis is recalibrated to obtain the calibrated positional distribution data.

[0021] Based on the calibrated location distribution data and the distance conditions, the influence of the scale difference on the feature point set is analyzed, and the final spatial offset adjustment scheme is determined.

[0022] By using the final spatial offset adjustment scheme, the position of the initially extracted feature point set is updated to obtain the feature point distribution result that meets the distance change condition.

[0023] In one aspect of this disclosure, if the magnitude of the determined spatial offset vector exceeds a preset threshold, then the perspective effect in the image sequence is corrected by a perspective transformation matrix to obtain a corrected feature point set, including:

[0024] If the spatial offset exceeds the preset threshold by vector detection, the affine transformation process is initiated to perform a preliminary analysis of the perspective deformation in the image sequence and obtain deformation distribution data.

[0025] Based on the deformation distribution data, the image sequence is corrected using a perspective transformation matrix to obtain the corrected image data;

[0026] Using the corrected image data, the location of the feature point set is reconstructed to determine the distribution of the reconstructed point set.

[0027] If local offsets still exist in the reconstructed point set distribution, then perform a quadratic matrix correction on the local area to obtain the adjusted point set data;

[0028] Based on the adjusted point set data, the impact of perspective distortion on overall image processing is analyzed, and distortion correction parameters are obtained.

[0029] By using deformation correction parameters, the image sequence is subjected to final correction processing to determine the processed image result.

[0030] In one aspect of this disclosure, the step of obtaining edge contour information from the corrected image based on the corrected feature point set, determining whether the contour sharpness meets the recognition requirements, and obtaining an optimized device boundary description includes:

[0031] Starting with the feature point set, the correction data is initially screened, and outliers are filtered using a preset threshold to obtain the cleaned point set data;

[0032] Based on the cleaned point set data, perform edge contour extraction operation, divide the contour region by geometric constraints, and determine the preliminary set of contour lines;

[0033] For the initial set of contour lines, the continuity of the lines is analyzed. If a broken area is detected, the lines are supplemented by interpolation to obtain a complete contour description.

[0034] Based on the complete contour description, consistency verification is performed in conjunction with the recognition criteria. If the contour lines deviate from the standard deviation from the preset range, local smoothing is performed to obtain the adjusted contour information.

[0035] The adjusted contour information is mapped to the device boundary, and boundary analysis technology is used to segment the boundary region to determine the accuracy distribution of the boundary data.

[0036] Based on the accuracy distribution of boundary data, information is integrated and optimized to generate the final device boundary description and determine the boundary results that meet the identification criteria.

[0037] In one aspect of this disclosure, the method of using an optical flow algorithm to track dynamic motion trajectories under distance changes and determine real-time position offsets for the optimized device boundary description includes:

[0038] For the device boundary description, an optical flow algorithm is used to capture dynamic motion, obtain motion trajectory data under distance changes, and obtain a preliminary trajectory path;

[0039] Based on the initial trajectory path, real-time position monitoring is performed, and position changes are analyzed by comparing frame by frame to determine the position offset value for each frame.

[0040] By combining the position offset value with the monitoring results of distance changes, the trajectory capture is refined to obtain continuous dynamic motion trends.

[0041] Based on the continuous dynamic movement trend, the real-time position is further analyzed. If the position offset value exceeds the preset threshold, the offset judgment mechanism is triggered to obtain an abnormal offset indicator.

[0042] For abnormal offsets, a position analysis correction operation is performed. By comparing historical trajectory data, it is determined whether the corrected position matches the expected trajectory.

[0043] Based on the corrected position data, the dynamic description of the device boundary is updated, and the final offset value is obtained by real-time calculation to determine the adjusted motion trajectory.

[0044] In one aspect of this disclosure, if the real-time position offset deviates from the standard range, an adaptive parameter adjustment model is obtained by fitting the correspondence between the offset and the distance change using the least squares method, including:

[0045] For the comparative analysis of position offset and standard range, if the position offset exceeds the standard range, the real-time offset value and distance change data are obtained through the data acquisition module to determine the preliminary indication of the abnormal state.

[0046] Based on the initially identified abnormal states, the least squares method is used to fit the offset value and distance changes, construct a calculation framework for parameter adjustment, and obtain the corresponding parameter relationship data.

[0047] By combining parameter relationship data with real-time dynamic monitoring results, the position offset is continuously tracked to determine whether it continues to deviate from the threshold of the standard range.

[0048] If the threshold continues to deviate from the standard range, an adaptive parameter adjustment operation will be performed based on the dynamically monitored data stream to obtain the adjusted parameter configuration.

[0049] By adjusting the parameter configuration, the parameter relationships of the model construction framework are updated to determine the new position offset control range;

[0050] Based on the new position offset control range, the real-time monitored data is logically verified to determine the range, and the final offset status identifier is obtained.

[0051] By using the final offset status identifier, combined with the historical records of distance changes, the storage and updating of execution parameter relationships are performed to determine the benchmark data for subsequent monitoring.

[0052] In one aspect of this disclosure, the step of adjusting the analysis parameters in the image processing flow according to adaptive parameters, determining whether the updated parameters improve the accuracy of feature recognition, and obtaining the final correction instruction sequence for power grid stability monitoring includes:

[0053] The raw data stream in the image processing flow is acquired, and a preliminary comparison is made between the adaptive parameters and the analysis parameters. If the comparison results show that the parameters deviate from the preset threshold range, the parameter update mechanism is triggered to obtain the parameter set after preliminary adjustment.

[0054] Using the initially adjusted parameter set, parameters are loaded for the feature recognition module in the image processing flow. The support vector machine algorithm is used to classify the output of feature recognition and determine the classified feature data group.

[0055] Based on the classified feature data group, the impact of the parameters updated in real time on the accuracy value is analyzed. If the accuracy value does not reach the preset threshold, the analysis parameters are fine-tuned a second time to obtain the optimized parameter configuration.

[0056] By optimizing the parameter configuration, corresponding correction instructions are generated. Combined with the results of feature recognition, a complete instruction sequence is constructed to determine the control instruction set suitable for power grid stability.

[0057] Based on the control command set, the data stream for power grid stability monitoring is matched in real time. If the matching result shows a deviation, the command sequence is locally adjusted to obtain the final correction command sequence.

[0058] The final correction command sequence is used to update the stability monitoring execution logic, obtain the matching status between real-time monitoring data and command sequence, and determine the stability status of power grid operation.

[0059] Based on the stability of the power grid operation, the execution log of the command sequence is continuously recorded, and the relevant data is stored in a preset database to obtain a complete monitoring record.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] This invention constructs a complete solution by integrating feature extraction, perspective correction, and dynamic tracking technologies. To address the scale differences caused by perspective effects, this invention employs a scale-invariant feature transformation algorithm to analyze the spatial offset of feature points and corrects the image sequence using a perspective transformation matrix to optimize the accuracy of the feature point set. For dynamic offsets under varying distances, it utilizes an optical flow algorithm to track the motion trajectory and combines it with the least squares method to fit the relationship between offset and distance, constructing an adaptive parameter adjustment model to dynamically update the analysis parameters and improve recognition accuracy. Finally, this invention generates a correction command sequence to achieve real-time monitoring and stable control of power grid equipment. Its core lies in adaptive parameter adjustment and multi-algorithm collaborative optimization, ensuring high-precision identification and correction of equipment status in complex environments and providing technical assurance for the safe operation of the power grid. Attached Figure Description

[0062] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0063] Figure 1 This is a flowchart of the dynamic position correction method for power equipment based on image feature matching according to the present invention. Detailed Implementation

[0064] The present invention will be further described below with reference to embodiments. These embodiments are merely some, not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the protection scope of the present invention.

[0065] Please see Figure 1 As shown in the figure, this embodiment discloses a method for dynamic position correction of power equipment based on image feature matching, which may specifically include:

[0066] Step 101: Acquire a sequence of images of power equipment using a camera, extract an initial set of feature points from the sequence, and obtain a raw data representation that includes perspective effects.

[0067] A continuous sequence of images of power equipment is captured via a camera to obtain a raw dataset containing multiple frames. Image preprocessing techniques are used to denoise and adjust the contrast of the raw dataset, resulting in a clear image set. For this clear image set, initial feature points are extracted from each frame, yielding a feature data set containing multiple feature point sets. If the number of feature points in the feature data set is below a preset threshold, secondary feature extraction is performed on the corresponding frame to determine a supplementary feature point set. Based on the supplementary feature point set, the impact of perspective on the feature point positions is analyzed, resulting in an adjusted data set with perspective correction. Using this adjusted data set, the original data representation of the power equipment is constructed, and its compliance with preset completeness standards is assessed. If the data representation does not meet the completeness standards, the adjusted data set is iteratively optimized to obtain the final power equipment data representation result.

[0068] Specifically, an image sequence of power equipment is acquired via a camera. A high-resolution industrial camera can be used to continuously capture images of the power equipment at a rate of 30 frames per second, forming an image sequence of 1000 frames with a resolution of 1920x1080 pixels. This ensures that details and dynamic changes of the equipment are captured, and the data is stored in lossless PNG format to preserve the original information. Next, an initial feature point set is extracted from this sequence. The SIFT algorithm is used to detect key points in each frame of the image. The scale space layer is set to 3, and the threshold is 0.04, extracting approximately 500 feature points. The coordinates, scale, and orientation of each feature point are recorded to form a feature point set matrix of size 500x4 for subsequent processing. At the same time, the salience of the equipment area is analyzed by the feature point distribution density, and areas with a density higher than 0.1 are marked as key areas of interest. Subsequently, the raw data representation including perspective effects is obtained. Based on the feature point set, the RANSAC algorithm is used to estimate the perspective transformation matrix of each frame of the image. The number of iterations is set to 1000, and the inlier threshold is 5 pixels. A 4x4 transformation matrix is ​​calculated, achieving an accuracy of over 95%. The transformation matrix is ​​applied to the image sequence to generate corrected image data, ensuring that the perspective distortion error is less than 2%. The corrected data is then associated with the original feature point set to form a structured representation including perspective effects, stored in JSON format, containing feature point coordinates and transformation parameters, for subsequent 3D reconstruction or defect detection tasks. Through the above process, from image acquisition to feature extraction to perspective correction, a complete data processing chain is formed, ensuring data accuracy and consistency, and providing a reliable foundation for power equipment condition analysis.

[0069] Step 102: Based on the extracted initial feature point set, the scale-invariant feature transformation algorithm is used to analyze the proportional differences caused by distance changes, and to determine the spatial offset vector of the feature points at different distances.

[0070] For the feature point set, a scale-invariant feature transformation algorithm is used to handle the proportional differences caused by distance variations, obtaining preliminary spatial offset data of feature points under different distance conditions. Based on the preliminary spatial offset data, a vector calculation method is used to analyze the correspondence between spatial offset and offset vector, determining the specific offset vector of the feature points under positional changes. Using the offset vector, a positional change model of the feature points under different distance conditions is constructed, obtaining the spatial distribution adjustment result of the feature point set. For the distribution adjustment result, if the positional change of the feature points exceeds a preset threshold, a secondary calibration is performed on the point set analysis to obtain calibrated positional distribution data. Based on the calibrated positional distribution data, combined with distance conditions, the impact of proportional differences on the feature point set is analyzed, determining the final spatial offset adjustment scheme. Using the final spatial offset adjustment scheme, the initially extracted feature point set is updated to obtain a feature point distribution result that meets the distance change conditions.

[0071] Specifically, in the field of power equipment image analysis, based on the extracted initial feature point set, a series of information technology methods can be used to analyze the scale difference caused by distance changes and determine the spatial offset vector of feature points at different distances. First, for the feature point set, a scale-invariant feature transform algorithm is used to perform scale analysis on the feature points of each frame in the image sequence. A Gaussian kernel standard deviation of 1.6 is set to construct an 8-layer scale space. The scale response value of each feature point is calculated, and feature points with a scale change rate greater than 1.2 are selected, forming a data matrix of approximately 300 core feature points, with a size of 300x3, containing coordinate and scale information. Subsequently, using these core feature points, the impact of distance changes on scale is analyzed. By calculating the Euclidean distance difference between feature points in adjacent frames, with a distance range of 2.5 to 10.0 meters, the scale coefficient change rate is found to be approximately 0.15 per meter. Combined with a distance change model, a 3x3 scale scaling matrix of feature points at different distances is derived for subsequent spatial calibration. Next, to determine the spatial offset vector, an optical flow algorithm was introduced. With a window size of 15x15 pixels, the displacement of feature points between consecutive frames was calculated, yielding an average offset of 3.2 pixels per frame. Combined with equipment installation height data (assuming a height of 5.0 meters), the actual spatial offset vector was calculated to be 0.08 meters per frame, forming an offset vector table of size 300x2, recording the lateral and longitudinal offsets of each feature point. To form a complete logical chain, the offset vector was correlated with the operating status of the power equipment. By analyzing the periodic changes in the offset vector, a period threshold of 10 frames was set to determine if there were any abnormal vibrations in the equipment. If the periodic change in the offset vector exceeded 0.05 meters, it was marked as a potential risk point, generating a risk assessment report stored in XML format, containing feature point numbers and offset data. Through this process, from feature point scale analysis to proportional difference calculation, and then to spatial offset vector determination and risk assessment, a closed-loop analysis process was formed, providing data support for power equipment monitoring.

[0072] Step 103: If the magnitude of the determined spatial offset vector exceeds a preset threshold, the perspective effect in the image sequence is corrected by the perspective transformation matrix to obtain the corrected feature point set.

[0073] If vector detection reveals a spatial offset exceeding a preset threshold, an affine transformation process is initiated to perform a preliminary analysis of perspective distortion in the image sequence, obtaining distortion distribution data. Based on this data, a perspective transformation matrix is ​​used to correct the image sequence, yielding corrected image data. Using this corrected data, the feature point set is reconstructed to determine its distribution. If local offsets still exist in the reconstructed distribution, a second matrix correction is performed on the local area to obtain adjusted point data. Based on this adjusted data, the impact of perspective distortion on overall image processing is analyzed, resulting in distortion correction parameters. Finally, these parameters are used to perform the final correction process on the image sequence, determining the final image result.

[0074] Specifically, in the field of power equipment image analysis, when the spatial offset vector exceeds a preset threshold, a series of information technology methods can be used to correct the perspective effect of the image sequence and obtain the corrected feature point set. First, assuming the preset spatial offset vector threshold is 0.1 meters, when the detected offset vector reaches 0.12 meters, the system automatically triggers the correction process, using a perspective transformation matrix to perform geometric correction on the image sequence. The matrix size is 3x3, containing rotation, translation, and scaling parameters. The rotation angle is calculated based on the feature point offset direction, assumed to be 5.2 degrees. The translation parameter is based on the offset vector decomposed into a horizontal distance of 0.07 meters and a vertical distance of 0.05 meters. The scaling factor is set to 1.05 to compensate for perspective distortion. Subsequently, the matrix is ​​used to transform the feature point coordinates of each frame in the image sequence. The initial feature point set contains 500 points, and the data matrix size is 500x2. The horizontal and vertical coordinates of each point are recorded. After transformation, the corrected coordinates of each point are recalculated to form a new feature point set matrix, still 500x2 in size. The correction accuracy is analyzed by comparing the coordinate differences before and after correction. The average error is assumed to be 0.03 meters, which meets the system requirements. To ensure logical integrity, the corrected feature point set is combined with the structural stability analysis of power equipment. The system automatically extracts the spatial distribution features of the feature point set and calculates the distribution uniformity. The uniformity index is assumed to be 0.85 (range 0 to 1). If it is lower than 0.8, it is marked as a structural anomaly risk area, and a corresponding stability assessment log is generated and stored in JSON format, containing feature point distribution data and risk level information. Through the above process, from offset vector detection to affine transformation correction, and then to feature point set update and stability analysis, a complete closed-loop processing mechanism is formed, providing a reliable data foundation for power equipment image monitoring.

[0075] Step 104: Obtain edge contour information from the corrected image based on the corrected feature point set, determine whether the contour clarity meets the recognition requirements, and obtain the optimized device boundary description.

[0076] Starting with the feature point set, the calibration data undergoes initial screening, using a preset threshold to filter outomas and obtain cleaned point set data. Based on this cleaned point set data, edge contour extraction is performed, dividing the contour region using geometric constraints to determine a preliminary set of contour lines. For this preliminary set of contour lines, line continuity is analyzed; if broken areas are detected, interpolation methods are used to supplement the lines, resulting in a complete contour description. Based on this complete contour description, consistency verification is performed using recognition standards. If the contour lines deviate from the standard deviation by a preset range, local smoothing is performed to obtain adjusted contour information. This adjusted contour information is mapped to the device boundary, and boundary analysis techniques are used to segment the boundary region and determine the accuracy distribution of the boundary data. Based on the accuracy distribution of the boundary data, optimization information is integrated to generate the final device boundary description, confirming the boundary results that meet the recognition standards.

[0077] Specifically, in the field of power equipment image analysis, for the corrected feature point set, the system extracts edge contour information and evaluates contour sharpness through a series of information technology methods, ultimately generating an optimized equipment boundary description. First, the system extracts edge information from the corrected feature point set. Assuming the feature point set contains 600 points and the data matrix size is 600x2, the horizontal and vertical coordinates of each point are recorded. An improved Canny edge detection algorithm is used, with a low threshold of 50 and a high threshold of 150, to generate an initial contour line, calculating 320 contour points. Subsequently, the system evaluates contour sharpness by calculating the average gradient intensity of the contour line, assuming an average of 12.5 (range 0 to 255), and setting a sharpness threshold of 10.0. If the average is higher than the threshold, the sharpness is considered acceptable; otherwise, the optimization process is triggered. Next, the system optimizes the contour lines using a curve fitting algorithm based on the least squares method. The smoothness of the fitted contour lines is improved, with the smoothness index increasing from 0.6 to 0.82 (range 0 to 1). The contour point coordinates are then recalculated to form an optimized boundary description dataset containing 310 points in a 310x2 matrix. To form a complete logical chain, the system combines the optimized boundary description with the appearance integrity analysis of the power equipment, automatically calculating the boundary closure degree. The closure degree is assumed to be 0.93 (range 0 to 1). Areas below 0.9 are marked as potential appearance defects, and an analysis report is generated and stored in XML format, containing boundary point data and defect risk level information. Through this process, from feature point set to edge contour extraction, and then to clarity assessment and boundary optimization, a closed-loop processing mechanism is formed, providing data support for the appearance monitoring of power equipment.

[0078] Step 105: Considering that the actual physical displacement of the device is usually very small, the observed position offset is mainly due to the perspective effect and scaling caused by the change in distance between the camera and the device. Therefore, it is necessary to establish a mathematical model of position offset and distance change. Based on this, for the optimized device boundary description, the optical flow algorithm is used to track the dynamic motion trajectory under distance change and determine the real-time position offset.

[0079] For the device boundary description, an optical flow algorithm is used to capture dynamic motion and acquire motion trajectory data under distance changes to obtain a preliminary trajectory path. Based on the preliminary trajectory path, real-time position monitoring is performed, and position changes are analyzed by comparing frames to determine the position offset value for each frame. Using the position offset value and the distance change monitoring results, the trajectory capture is refined to obtain a continuous dynamic motion trend. Based on the continuous dynamic motion trend, the real-time position is further analyzed. If the position offset value exceeds a preset threshold, an offset judgment mechanism is triggered, resulting in an abnormal offset flag. For the abnormal offset flag, a position analysis correction operation is performed, comparing historical trajectory data to determine whether the corrected position matches the expected trajectory. Based on the corrected position data, the dynamic description of the device boundary is updated, and the final offset value is obtained using real-time calculation to determine the adjusted motion trajectory.

[0080] Specifically, in the field of power equipment image analysis, for the optimized equipment boundary description, the system uses information technology to track dynamic motion trajectories and calculate position offsets to support real-time monitoring of equipment status. The system first obtains boundary point information from the optimized boundary description dataset. Assuming the dataset contains 300 points and a matrix size of 300x2, the system records the horizontal and vertical coordinates of each point. Then, it uses the Lucas-Kanade method from the optical flow algorithm for dynamic tracking, setting the search window size to 15x15 pixels. It calculates the displacement vectors of boundary points in two adjacent frames to obtain the velocity vector of each point, assuming an average velocity of 2.3 pixels per frame (range 0 to 10). Next, the system analyzes the displacement vector, calculating the total offset by accumulating displacement data from 10 consecutive frames. Assuming a total offset of 23.5 pixels, this is corrected by combining wind speed data from the equipment installation environment (assuming a wind speed of 5.2 m / s, ranging from 0 to 20). If the offset exceeds a preset threshold of 20.0 pixels, an abnormal displacement warning is triggered, automatically recording the abnormal time point and generating a log file in JSON format, containing the offset and environmental parameter information. To further refine the logic chain, the system combines the offset data with equipment vibration frequency analysis, assuming a vibration frequency of 3.8 Hz (ranging from 0 to 10). Frequency features are extracted using a Fourier transform algorithm. If the correlation coefficient between the frequency and the offset change trend exceeds 0.75, it is marked as a potential mechanical loosening risk, and the equipment status database is automatically updated, forming a complete processing flow from dynamic trajectory tracking to risk assessment, providing data support for monitoring the operational stability of power equipment.

[0081] Step 106: If the real-time position offset deviates from the standard range, the least squares method is used to fit the correspondence between the offset and the distance change to obtain an adaptive parameter adjustment model.

[0082] For the comparative analysis of positional offset and standard range, if the positional offset exceeds the standard range, the data acquisition module obtains real-time offset values ​​and distance change data to determine the initial indication of the abnormal state. Based on the initial indication of the abnormal state, the least squares method is used to fit the offset values ​​and distance changes, constructing a calculation framework for parameter adjustment and obtaining the corresponding parameter relationship data. Using the parameter relationship data, combined with the real-time dynamic monitoring results, the positional offset is continuously tracked to determine whether it continuously deviates from the standard range threshold. If it continuously deviates from the standard range threshold, an adaptive parameter adjustment operation is performed based on the dynamic monitoring data stream to obtain the adjusted parameter configuration. Using the adjusted parameter configuration, the parameter relationships of the model-built framework are updated to determine a new positional offset control range. Based on the new positional offset control range, the real-time monitoring data undergoes logical verification for range determination to obtain the final offset status indicator. Using the final offset status indicator, combined with the historical distance change records, the parameter relationships are stored and updated to determine the benchmark data for subsequent monitoring.

[0083] Specifically, in the field of power equipment condition monitoring, the system automatically initiates data analysis and parameter adjustment processes when real-time location offset deviates from the standard range, in order to construct a model of the correspondence between offset and distance change. First, the system extracts offset data from the equipment monitoring database. Assuming the collected offset dataset contains 500 sample points, each recording the offset and corresponding distance change value, the offset range is -10.5 to 15.7 pixels, and the distance change range is 0.2 to 5.8 meters. Then, the system uses the least squares method to fit these data, calculating the linear relationship parameters between offset and distance change. Assuming the slope of the fitted data is 2.4 and the intercept is 1.1, the system generates the corresponding regression equation. Next, the system analyzes the residual distribution based on the fitting results. Assuming the standard deviation of the residuals is 0.9 pixels, if the proportion of sample points with residuals exceeding 2.0 pixels exceeds 10%, they are automatically marked as outliers, removed, and the model is refitted to ensure model accuracy.

[0084] Furthermore, the system calculates adaptive adjustment parameters based on the fitted model. Assuming a current distance change of 3.5 meters, the predicted offset is 9.5 pixels, and this parameter is updated in the equipment monitoring algorithm for subsequent real-time calibration. Simultaneously, to form a complete logical chain, the system correlates the adjusted parameters with the equipment's operating temperature data. Assuming a temperature range of 20.3 to 45.6 degrees Celsius, correlation analysis reveals a correlation coefficient of 0.62 between temperature and the offset prediction error. If this coefficient exceeds 0.5, the temperature factor is automatically incorporated into the model correction logic, generating a comprehensive adjustment strategy, which is stored as an XML file to ensure the adaptability of equipment monitoring.

[0085] Step 107: Adjust the analysis parameters in the model update image processing flow according to the adaptive parameters, determine whether the updated parameters improve the accuracy of feature recognition, and obtain the final correction instruction sequence for power grid stability monitoring.

[0086] The process begins by acquiring the raw data stream from the image processing flow. A preliminary comparison is performed between the adaptive parameters and the analysis parameters. If the comparison shows that the parameters deviate from a preset threshold range, a parameter update mechanism is triggered, resulting in a pre-adjusted parameter set. Using this pre-adjusted parameter set, parameters are loaded into the feature recognition module within the image processing flow. A support vector machine algorithm is used to classify the output of the feature recognition, determining the classified feature data set. Based on the classified feature data set, the impact of the real-time updated parameters on the accuracy value is analyzed. If the accuracy value does not reach the preset threshold, the analysis parameters are fine-tuned a second time to obtain an optimized parameter configuration. Using the optimized parameter configuration, corresponding correction commands are generated. Combined with the feature recognition results, a complete command sequence is constructed to determine a control command set suitable for power grid stability. Based on the control command set, the data stream for power grid stability monitoring is matched in real time. If the matching result shows a deviation, the command sequence is locally adjusted to obtain the final correction command sequence. Using the final correction command sequence, the execution logic of stability monitoring is updated, the matching status between the real-time monitoring data and the command sequence is obtained, and the stability status of the power grid operation is determined. Based on the stability of the power grid operation, the execution log of the command sequence is continuously recorded, and the relevant data is stored in a preset database to obtain a complete monitoring record.

[0087] Specifically, in a transformer inspection scenario at a substation, the distance between the camera and the equipment dynamically changes within a range of 10 to 15 meters. The system first uses an adaptive parameter adjustment model to calculate that when the distance increases from 10 meters to 15 meters, the scale threshold for SIFT feature extraction needs to be adjusted to 0.85 times the original value, and the gradient threshold for edge detection needs to be adjusted to 1.2 times the original value. Based on this, the system updates the corresponding parameters in the image processing flow and uses a support vector machine to classify and evaluate the feature recognition results. If the recognition accuracy improves from the initial 90% to 96%, reaching the preset threshold, the system generates specific correction commands such as "adjust the gimbal tilt angle +0.3°" and "horizontal translation -1.5cm," forming a command sequence and issuing it to the actuator. The system monitors the equipment position data in real time after the commands are executed. If it finds that the actual position still deviates from the expected position by ±0.5cm, it makes local fine adjustments to the command sequence, such as correcting the translation command to "-2.0cm". Ultimately, the system records all command execution logs and corresponding power grid stability status data, storing them in the inspection database to form a closed-loop feedback. Through this step, the system can maintain an equipment location identification accuracy of over 95% even under dynamically changing distance conditions, effectively supporting the stable operation of the power grid.

[0088] 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 dynamic position correction of power equipment based on image feature matching, characterized in that, include: By acquiring a sequence of images of power equipment using a camera, an initial set of feature points is extracted from the sequence to obtain a raw data representation that includes perspective effects. Based on the extracted initial feature point set, the scale-invariant feature transformation algorithm is used to analyze the proportional differences caused by distance changes, and to determine the spatial offset vector of the feature points at different distances; If the magnitude of the determined spatial offset vector exceeds a preset threshold, the perspective effect in the image sequence is corrected by the perspective transformation matrix to obtain the corrected feature point set. From the corrected image, edge contour information is obtained based on the corrected feature point set. It is then determined whether the contour clarity meets the recognition requirements, and an optimized device boundary description is obtained. For the optimized device boundary description, an optical flow algorithm is used to track the dynamic motion trajectory under distance changes and determine the real-time position offset. If the real-time position offset deviates from the standard range, the least squares method is used to fit the correspondence between the offset and the distance change to obtain an adaptive parameter adjustment model. The analysis parameters in the image processing flow are updated based on the adaptive parameters. It is then determined whether the updated parameters improve the accuracy of feature recognition, and the final correction instruction sequence is obtained for power grid stability monitoring.

2. The method for dynamic position correction of power equipment based on image feature matching according to claim 1, characterized in that: The process of acquiring a sequence of images of power equipment via a camera, extracting an initial set of feature points from the sequence, and obtaining a raw data representation including perspective effects includes: By continuously acquiring image sequences of power equipment through cameras, a raw dataset containing multiple frames of images is obtained. Image preprocessing techniques were used to denoise and adjust the contrast of the acquired raw dataset to obtain a clear image set after processing. For the processed set of clear images, the initial feature points in each frame are extracted to obtain a feature data set containing multiple feature point sets. If the number of feature points in the feature data set is lower than the preset threshold, then perform secondary feature extraction on the corresponding frame image to determine the supplemented feature point set. Based on the supplemented feature point set, analyze the impact of perspective effect on the feature point position, and obtain the adjustment data set with perspective correction; By adjusting the data group, the original data representation of the power equipment is constructed, and it is determined whether the data representation meets the preset integrity standard. If the data representation does not meet the completeness standard, the adjusted data group is iteratively optimized to obtain the final power equipment data representation result.

3. The method for dynamic position correction of power equipment based on image feature matching according to claim 1, characterized in that: The step of analyzing the proportional differences caused by distance changes using a scale-invariant feature transformation algorithm based on the extracted initial feature point set, and determining the spatial offset vectors of feature points at different distances, includes: For the feature point set, the scale-invariant feature transformation algorithm is used to handle the proportional differences caused by distance changes and obtain preliminary spatial offset data of feature points under different distance conditions. Based on the preliminary spatial offset data, the vector calculation method is used to analyze the correspondence between spatial offset and offset vector, and to determine the specific offset vector of the feature point under positional changes. By using offset vectors, a model of the positional variation of feature points under different distance conditions is constructed to obtain the spatial distribution adjustment results of the feature point set; If the positional changes of feature points exceed a preset threshold, the point set analysis is recalibrated to obtain the calibrated positional distribution data. Based on the calibrated location distribution data and the distance conditions, the influence of the scale difference on the feature point set is analyzed, and the final spatial offset adjustment scheme is determined. By using the final spatial offset adjustment scheme, the position of the initially extracted feature point set is updated to obtain the feature point distribution result that meets the distance change condition.

4. The method for dynamic position correction of power equipment based on image feature matching according to claim 1, characterized in that: If the magnitude of the determined spatial offset vector exceeds a preset threshold, the perspective effect in the image sequence is corrected using a perspective transformation matrix to obtain a corrected feature point set, including: If the spatial offset exceeds the preset threshold by vector detection, the affine transformation process is initiated to perform a preliminary analysis of the perspective deformation in the image sequence and obtain deformation distribution data. Based on the deformation distribution data, the image sequence is corrected using a perspective transformation matrix to obtain the corrected image data; Using the corrected image data, the location of the feature point set is reconstructed to determine the distribution of the reconstructed point set. If local offsets still exist in the reconstructed point set distribution, then perform a quadratic matrix correction on the local area to obtain the adjusted point set data; Based on the adjusted point set data, the impact of perspective distortion on overall image processing is analyzed, and distortion correction parameters are obtained. By using deformation correction parameters, the image sequence is subjected to final correction processing to determine the processed image result.

5. The method for dynamic position correction of power equipment based on image feature matching according to claim 1, characterized in that: The step of obtaining edge contour information from the corrected image based on the corrected feature point set, determining whether the contour sharpness meets the recognition requirements, and obtaining the optimized device boundary description includes: Starting with the feature point set, the correction data is initially screened, and outliers are filtered using a preset threshold to obtain the cleaned point set data; Based on the cleaned point set data, perform edge contour extraction operation, divide the contour region by geometric constraints, and determine the preliminary set of contour lines; For the initial set of contour lines, the continuity of the lines is analyzed. If a broken area is detected, the lines are supplemented by interpolation to obtain a complete contour description. Based on the complete contour description, consistency verification is performed in conjunction with the recognition criteria. If the contour lines deviate from the standard deviation from the preset range, local smoothing is performed to obtain the adjusted contour information. The adjusted contour information is mapped to the device boundary, and boundary analysis technology is used to segment the boundary region to determine the accuracy distribution of the boundary data. Based on the accuracy distribution of boundary data, information is integrated and optimized to generate the final device boundary description and determine the boundary results that meet the identification criteria.

6. The method for dynamic position correction of power equipment based on image feature matching according to claim 1, characterized in that: The optimized device boundary description employs an optical flow algorithm to track dynamic motion trajectories under varying distances and determine real-time position offsets, including: For the device boundary description, an optical flow algorithm is used to capture dynamic motion, obtain motion trajectory data under distance changes, and obtain a preliminary trajectory path; Based on the initial trajectory path, real-time position monitoring is performed, and position changes are analyzed by comparing frame by frame to determine the position offset value for each frame. By combining the position offset value with the monitoring results of distance changes, the trajectory capture is refined to obtain continuous dynamic motion trends. Based on the continuous dynamic movement trend, the real-time position is further analyzed. If the position offset value exceeds the preset threshold, the offset judgment mechanism is triggered to obtain an abnormal offset indicator. For abnormal offsets, a position analysis correction operation is performed. By comparing historical trajectory data, it is determined whether the corrected position matches the expected trajectory. Based on the corrected position data, the dynamic description of the device boundary is updated, and the final offset value is obtained by real-time calculation to determine the adjusted motion trajectory.

7. The method for dynamic position correction of power equipment based on image feature matching according to claim 1, characterized in that: If the real-time position offset deviates from the standard range, an adaptive parameter adjustment model is obtained by fitting the correspondence between the offset and the distance change using the least squares method, including: For the comparative analysis of position offset and standard range, if the position offset exceeds the standard range, the real-time offset value and distance change data are obtained through the data acquisition module to determine the preliminary indication of the abnormal state. Based on the initially identified abnormal states, the least squares method is used to fit the offset value and distance changes, construct a calculation framework for parameter adjustment, and obtain the corresponding parameter relationship data. By combining parameter relationship data with real-time dynamic monitoring results, the position offset is continuously tracked to determine whether it continues to deviate from the threshold of the standard range. If the threshold continues to deviate from the standard range, an adaptive parameter adjustment operation will be performed based on the dynamically monitored data stream to obtain the adjusted parameter configuration. By adjusting the parameter configuration, the parameter relationships of the model construction framework are updated to determine the new position offset control range; Based on the new position offset control range, the real-time monitored data is logically verified to determine the range, and the final offset status identifier is obtained. By using the final offset status identifier, combined with the historical records of distance changes, the storage and updating of parameter relationships are performed to determine the benchmark data for subsequent monitoring.

8. The method for dynamic position correction of power equipment based on image feature matching according to claim 1, characterized in that: The step of adjusting the analysis parameters in the image processing flow according to adaptive parameters, determining whether the updated parameters improve the accuracy of feature recognition, and obtaining the final correction command sequence for power grid stability monitoring includes: The raw data stream in the image processing flow is acquired, and a preliminary comparison is made between the adaptive parameters and the analysis parameters. If the comparison results show that the parameters deviate from the preset threshold range, the parameter update mechanism is triggered to obtain the preliminary adjusted parameter set. Using the initially adjusted parameter set, parameters are loaded for the feature recognition module in the image processing flow. The support vector machine algorithm is used to classify the output of feature recognition and determine the classified feature data group. Based on the classified feature data group, the impact of the parameters updated in real time on the accuracy value is analyzed. If the accuracy value does not reach the preset threshold, the analysis parameters are fine-tuned a second time to obtain the optimized parameter configuration. By optimizing the parameter configuration, corresponding correction instructions are generated. Combined with the results of feature recognition, a complete instruction sequence is constructed to determine the control instruction set suitable for power grid stability. Based on the control command set, the data stream for power grid stability monitoring is matched in real time. If the matching result shows a deviation, the command sequence is locally adjusted to obtain the final correction command sequence. The final correction command sequence is used to update the stability monitoring execution logic, obtain the matching status between real-time monitoring data and command sequence, and determine the stability status of power grid operation. Based on the stability of the power grid operation, the execution log of the command sequence is continuously recorded, and the relevant data is stored in a preset database to obtain a complete monitoring record.

Images