Transformer rack installation process detection method based on key point detection

Abstract

The invention discloses a transformer rack installation process detection method based on key point detection, and relates to the technical field of installation process detection, and the method comprises the steps: preparing a standard checkerboard target and an image collection device for scale calibration and image collection; collecting and processing the image; key points are automatically marked on the collected image, and the positions of the marked key points are detected by using a YOLOv8 model; a checkerboard target is used for determining the proportion of pixels to the actual distance, and the actual distance between the key points is calculated; recording actual distance data of the key points, performing trend analysis on a distance detection result, and displaying the distance detection result to a user through a report or an application program. Scale calibration is carried out through the standard checkerboard target with the fixed size, the accuracy of the measurement result is ensured, efficient and accurate key point detection is carried out through the YOLOv8 model, automation of key point detection and distance calculation is achieved, and the detection efficiency is improved.

Description

Technical Field

[0001] The invention relates to the technical field of installation process detection, and in particular to a transformer stand installation process detection method based on key point detection. Background Art

[0002] The transformer rack is an important part of the power system, used to support and install transformers and related electrical equipment. The installation quality of the transformer rack directly affects the safe and stable operation of the power system. Therefore, the installation process inspection of the transformer rack is very important, and it is necessary to ensure that the relative position and distance between the components meet the design requirements.

[0003] When inspecting the installation process of transformers on a bench, it is necessary to measure the distance between the transformer crossarms, the distance between the crossarms and the ground, and the distance between the transformer and the ground. The traditional distance measurement method generally uses a theodolite. However, the theodolite itself has many shortcomings:

[0004] First, the theodolite equipment is bulky and inconvenient to carry, and it is very difficult to use it on construction sites, especially in narrow spaces or high-altitude working environments;

[0005] Secondly, the operation process of the theodolite is quite complicated and requires professionals to adjust and read data. For non-professionals, it takes a long time to learn and master this skill;

[0006] In addition, only one distance data can be obtained for each measurement, which means that the instrument position and angle need to be adjusted repeatedly to complete the inspection of the entire transformer bench, resulting in low overall work efficiency;

[0007] Finally, the measurement results need to be recorded and processed manually, which is prone to human errors and slow data processing. Summary of the invention

[0008] In view of the problems existing in the prior art, such as heavy equipment, complicated operation and low efficiency, the present invention proposes a transformer stand installation process detection method based on key point detection.

[0009] Therefore, the problem to be solved by the present invention is how to use computer vision technology to automatically identify the key points of the transformer rack and to assist in measuring the actual distance through a standard chessboard target.

[0010] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0011] In the first aspect, an embodiment of the present invention provides a transformer stand installation process detection method based on key point detection, which includes: preparing a standard checkerboard target and an image acquisition device for scale calibration and image acquisition; acquiring and processing images; automatically marking key points on the acquired images, and using the YOLOv8 model to detect the positions of the marked key points; using the checkerboard target to determine the ratio of pixels to actual distances, and calculating the actual distance between key points; recording the actual distance data of the key points, performing trend analysis on the distance detection results, and displaying the distance detection results to the user through a report or application.

[0012] As a preferred solution of the transformer stand installation process detection method based on key point detection described in the present invention, the processing includes denoising the captured image; and improving the clarity of the checkerboard target and the key parts of the stand by adjusting the contrast and brightness of the image.

[0013] As a preferred solution of the transformer stand installation process detection method based on key point detection described in the present invention, the automatic marking of key points and the use of the YOLOv8 model to detect the positions of the marked key points include the following operating steps: an automated auxiliary tool preliminarily marks the key points in the image; the marked image is input into the YOLOv8 model to detect the preliminarily marked key points; based on the output result of the YOLOv8 model, each detected key point is screened according to the confidence value; if multiple images are taken, the key point detection results from different angles are integrated according to the multi-view YOLOv8 detection method, and the image reconstruction technology is used to confirm the position consistency of each key point in the three-dimensional space.

[0014] As a preferred solution of the transformer stand installation process detection method based on key point detection described in the present invention, the detection of the key points marked initially includes: converting the input image I into a tensor form T(I), using a multi-layer convolutional neural network CNN to extract image features, and identifying feature patterns related to key points; using a dense candidate box generation mechanism to generate candidate boxes at multiple positions in the image, each candidate box is assigned an initial position and confidence according to the feature extraction result of the network; using a regression function to generate the center position coordinates (x det ,y det ), and the confidence value c, and the final output is a set of key point positions and corresponding confidences.

[0015] As a preferred solution of the transformer stand installation process detection method based on key point detection according to the present invention, the key point detection results from different angles are integrated, and the image reconstruction technology is used to confirm the position consistency of each key point in three-dimensional space, which includes the following steps: calculating the projection matrix Pt of each view image through the known camera intrinsic parameters and extrinsic parameters i , the projection matrix is ​​used to map a three-dimensional point to a two-dimensional image plane, and the formula is:

[0016] Pt i =K i ×[R i |t i ]

[0017] Among them, K i is the camera’s intrinsic parameter matrix, R i and t i is the rotation matrix and translation vector of the camera; for the key point coordinates detected by multiple perspectives (x i ,y i ) using the corresponding projection matrix Pt i Perform triangulation calculations to restore the three-dimensional coordinates (X, Y, Z) of key points; use the linear least squares method to solve and reduce the inconsistency caused by image detection errors.

[0018] As a preferred solution of the transformer stand installation process detection method based on key point detection described in the present invention, the method comprises the following steps: using the checkerboard target to determine the ratio of pixels to actual distances and calculating the real distance between key points: using different scale spacings on the checkerboard target, by gradually measuring the side lengths of different adjacent checkerboard squares, obtaining the ratio r of multiple pixels to actual distances; collecting the built-in focal length sensor data and angle sensor information of the shooting device, calculating the focal length and angle offset when the image is shot according to the changes in focal length and angle, adjusting the ratio r, and correcting the error caused by distortion; in the actual measurement process, automatically selecting the most matching ratio as the basic ratio value, and dynamically adjusting it in combination with the focal length and angle corrected data, and converting the pixel distance between key points; using the ratio r′ of the corrected pixel distance to the actual distance, performing distance calculation on the detected key point positions, outputting the actual distance between the key points, and performing accuracy verification.

[0019] As a preferred solution of the transformer stand installation process detection method based on key point detection of the present invention, wherein: the pixel distance between the key points is converted to: the pixel distance D between the detected key points is converted to p,k Use the compensated ratio to convert to the actual distance D between key points a,k , the calculation formula is as follows:

[0020] Da,k = r′×D p,k .

[0021] As a preferred solution of the transformer stand installation process detection method based on key point detection of the present invention, the distance calculation of the detected key point positions includes: using the compensated ratio r' to calculate the actual distance between all detected key points:

[0022] D a,f = r′×D a,k

[0023] The formula for the accuracy verification is:

[0024]

[0025] Where n is the number of samples, ACC is the accuracy verification index, ∈ is the acceptable error range, and D a,f is the actual distance between all detected key points.

[0026] In a second aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the transformer stand installation process detection method based on key point detection as described in the first aspect of the present invention are implemented.

[0027] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the transformer stand installation process detection method based on key point detection as described in the first aspect of the present invention are implemented.

[0028] The invention has the following beneficial effects: the invention uses a fixed-size standard checkerboard target to perform scale calibration to ensure the accuracy of the measurement results, and uses the YOLOv8 model to perform efficient and accurate key point detection, thereby automating key point detection and distance calculation and improving detection efficiency. In addition, the invention combines smart phones and computer vision technology to form a portable detection system suitable for on-site detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0030] Figure 1It is a flow chart of the transformer stand installation process detection method based on key point detection in the present invention.

[0031] Figure 2 It is the calibration plate in the present invention. DETAILED DESCRIPTION

[0032] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0033] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0034] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0035] Example 1

[0036] Reference Figure 1-2 , which is the first embodiment of the present invention, and provides a transformer stand installation process detection method based on key point detection, comprising:

[0037] S1: Prepare a standard checkerboard target and a high-resolution smartphone for scale calibration and image acquisition.

[0038] In order to ensure accurate calibration of the image scale, a checkerboard target that meets standard specifications (such as the commonly used 8x8 or 10x10 checkerboard) is selected to ensure that the physical size of the checkerboard target is known so that the ratio of pixels to actual distance can be accurately converted later.

[0039] Furthermore, a smartphone equipped with a high-resolution camera (12 megapixels or more is recommended) is used to ensure that the image quality can meet the subsequent key point detection requirements.

[0040] Check the lighting conditions of the shooting environment and ensure that images are captured under natural light or uniform lighting conditions to avoid overexposure or distortion of dark areas.

[0041] If necessary, adjust the shooting angle or use fill light to ensure that both the checkerboard target and the part of the test stand to be inspected are clearly captured.

[0042] S2: Place the target near the gantry and take a clear image with a smartphone, making sure the key points and the target are clearly visible.

[0043] Place the standard checkerboard target firmly near the transformer test bench, ensure that there is an appropriate distance between the target and the test bench structure to avoid overlap or occlusion in the image, and select a suitable target position so that the checkerboard target and the key parts of the test bench can enter the field of view at the same time to facilitate subsequent key point detection.

[0044] Use a high-resolution smartphone to take pictures, keeping the shooting angle perpendicular to the checkerboard target surface to avoid image distortion caused by perspective deviation.

[0045] When shooting, make sure that the details of the checkerboard target, key parts of the gantry, and the surrounding environment are clearly visible in the image. It is recommended to take several images so that the best quality image can be selected in subsequent processing.

[0046] Furthermore, the image is denoised and contrast enhanced to improve the accuracy of key point detection.

[0047] The captured images are denoised to remove subtle noise that may affect key point detection. The clarity of the checkerboard target and key parts of the test bench are improved by adjusting the contrast and brightness of the image, ensuring that the key points in the image are easier to detect.

[0048] S3: Automatically annotate key points and use the YOLOv8 model to detect the locations of the annotated key points.

[0049] S3.1: Automated auxiliary tools perform preliminary annotation of key points in the image.

[0050] The automatic labeling function performs intelligent identification based on the characteristics of the test bench structure to mark the positions of all key points to be tested as much as possible and record the preliminary coordinates of these points.

[0051] S3.2: Input the annotated image into the YOLOv8 model to detect the key points marked preliminarily.

[0052] The YOLOv8 model identifies the key point positions in the image and outputs the specific coordinates and confidence values ​​of the key points. Specifically, the input image I is converted into a tensor form T(I) that can be processed by the neural network, and a multi-layer convolutional neural network CNN is used to extract the features of the image to identify the feature patterns related to the key points.

[0053] YOLOv8 uses a dense candidate box generation mechanism to generate candidate boxes at multiple locations in the image. These boxes contain possible key point locations. Each candidate box is given an initial position and confidence based on the feature extraction results of the network.

[0054] Use the regression function to generate the center position coordinates (x det ,y det ), and the confidence value c, and the final output is a set of key point positions and corresponding confidences.

[0055] To ensure detection accuracy, the model performs a full range of detection throughout the image to identify the best locations for key points.

[0056] S3.3: Based on the output results of the YOLOv8 model, each detected key point is filtered according to the confidence value.

[0057] Set a confidence threshold T c To filter out key points with high confidence, for confidence c <T c The key point is marked as "pending confirmation".

[0058] For key points below the confidence threshold, they are marked as "pending confirmation", which will be processed in subsequent steps to further verify the accuracy of the key point position through other methods. This process effectively excludes low-confidence points and improves the reliability of key point detection.

[0059] S3.4: If multiple images are taken, use the multi-view YOLOv8 detection method to integrate the key point detection results from different angles, and use image reconstruction technology to confirm the position consistency of each key point in three-dimensional space.

[0060] Using the known camera internal and external parameters, calculate the projection matrix Pt of each view image i , the projection matrix is ​​used to map three-dimensional points to a two-dimensional image plane. The specific formula is:

[0061] Pt i =K i ×[R i |t i ]

[0062] Among them, K i is the camera’s intrinsic parameter matrix, R i and t i is the rotation matrix and translation vector (external parameters) of the camera.

[0063] For the key point coordinates (x i ,y i ) using the corresponding projection matrix Pt i Triangulation calculations are performed to restore the three-dimensional coordinates (X, Y, Z) of the key points; and the linear least squares method is used to solve the problem to reduce the inconsistency caused by image detection errors.

[0064] It should be noted that the image reconstruction technology is used to verify the positions of key points in different perspectives to ensure the consistency of the key points in the spatial coordinates. This can supplement the detection errors that may exist under a single angle and further refine the positions of the key points.

[0065] S4: Use the checkerboard target to determine the ratio of pixels to actual distance and calculate the true distance between key points.

[0066] First, the checkerboard edge points of the checkerboard target are extracted from the captured image to obtain multiple pixel distances with known spacing. This measurement step is performed based on the edges of multiple adjacent checkerboard squares to ensure accuracy and anti-interference.

[0067] S4.1: Using the different scale spacings on the checkerboard target, by gradually measuring the side lengths of different adjacent checkerboard squares, we can obtain the ratios of multiple pixels to the actual distances. These ratios can be used as multi-scale benchmarks to adapt to different shooting angles and distances.

[0068] Specifically, edge points of the checkerboard target are extracted from the captured image through an image processing algorithm (such as edge detection), and the edge points are defined;

[0069] Measure the pixel distance D between adjacent chessboard squares p , for example, from point p i to p i+1 distance;

[0070] Assume the side length of the known chessboard square is D a (e.g. 25mm), and then calculate the ratio r of pixels to actual distance:

[0071] r=D a / D p

[0072] S4.2: Collect the built-in focal length sensor data of the shooting device, including the focal length f obtained from the device h and shooting angle θ h , collect angle sensor information, including focal length and angle information, respectively f s and θ s , according to the changes in focal length and angle, calculate the focal length and angle offset when the image is captured, and adjust the ratio r to correct the error caused by distortion:

[0073]

[0074] Among them, r' is the ratio after compensation, Δf is the change in focal length, Δf = f h -f s , Δθ is the change in angle, Δθ=θ h -θs .

[0075] Based on these data, the ratio of pixels to actual distance is adjusted to compensate for the distortion effects caused by changes in focal length or shooting angle, thereby ensuring the accuracy of ratio conversion under zoom or non-straight shooting conditions.

[0076] S4.3: In the actual measurement process, the most matching ratio is automatically selected as the basic ratio value r ma , and dynamically adjusts the pixel distance between key points in combination with the focal length and angle correction data to obtain the accurate actual distance.

[0077] Preferably, the pixel distance D between the detected key points is p,k Use the compensated ratio to convert to the actual distance D between key points a,k , the calculation formula is as follows:

[0078] D a,k = r′×D p,k

[0079] S4.4: Use the ratio of the corrected pixel distance to the actual distance to calculate the distance of the detected key point positions, output the actual distance between the key points, and perform necessary accuracy verification to ensure the accuracy and reliability of the detection results.

[0080] Specifically, the actual distance between all detected key points is calculated using the compensated ratio r′:

[0081] D a,f = r′×D a,k

[0082] Furthermore, compare the known real distance with the calculated result:

[0083]

[0084] Among them, n is the number of samples, ACC is the accuracy verification index, and ∈ is the acceptable error range.

[0085] It can be seen that the present invention, based on the traditional single ratio conversion, takes into account the use of zoom shooting and adds an automatic correction mechanism for the camera focal length and shooting angle. It can ensure the pixel distance conversion accuracy at different focal lengths and angles through multi-scale ratio conversion based on multiple scale measurement values ​​of the checkerboard target. This improvement is particularly suitable for long-distance shooting or scenes where it is difficult to adjust the angle, and improves the accuracy of distance calculation.

[0086] S5: Record the actual distance data of key points, perform trend analysis on the distance detection results, and display the detection results to the user through a report or application.

[0087] The calculated actual distance data of key points is stored in a database or local storage, and auxiliary information such as timestamp, detection location and shooting parameters are attached to each detection record for subsequent analysis and traceability.

[0088] Furthermore, trend analysis is performed on the distance data of multiple tests. Through methods such as data fitting or moving average, a trend curve of key point distance changes is generated. Combined with the test time, the results of each test cycle are integrated into a trend chart, allowing users to intuitively view the changes in key point distances, identify the improvement or deviation trend of installation accuracy, and visualize the comparison results of multiple tests in the form of bar charts, line charts, or scatter charts. The comparison chart can highlight the change range of the actual distance and mark abnormal points that exceed the standard range, helping users quickly identify the deviation of installation accuracy and improving data readability.

[0089] Finally, based on the results of trend analysis and comparison charts, a test report is automatically generated. The report should include actual distance data of key points, trend analysis charts, comparison charts, abnormal point prompts, and related text descriptions and suggestions to ensure the comprehensiveness and practicality of the report information.

[0090] This embodiment also provides a computer device, which is suitable for the transformer stand installation process detection method based on key point detection, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the transformer stand installation process detection method based on key point detection as proposed in the above embodiment.

[0091] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0092] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for detecting the transformer stand installation process based on key point detection as proposed in the above embodiment is implemented.

[0093] In summary, the present invention uses a fixed-size standard checkerboard target for scale calibration to ensure the accuracy of the measurement results, and uses the YOLOv8 model for efficient and accurate key point detection, realizing the automation of key point detection and distance calculation, and improving detection efficiency. In addition, the present invention combines smart phones and computer vision technology to form a portable detection system suitable for on-site detection.

[0094] Example 2

[0095] Referring to Table 1 and Table 2, the second embodiment of the present invention provides a transformer stand installation process detection method based on key point detection. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0096] This test is based on the transformer bench installation process detection method of key point detection. The main steps include image scale calibration, image acquisition, key point detection, distance conversion and trend analysis. First, prepare a checkerboard target that meets standard specifications and a smartphone with a resolution of 12 million pixels or more. The target adopts a standard 8x8 checkerboard pattern, and confirms in advance that the side length of each grid is 25mm to ensure that its physical size is known for subsequent image calibration. The light source of the shooting site is set to uniform illumination and supplemented with natural light to avoid overexposure or dark distortion of the image. The specific operations include:

[0097] Place the checkerboard target firmly near the transformer stand and adjust the angle to ensure that the target and key parts of the stand are in view at the same time. When taking images, keep the mobile phone camera perpendicular to the surface of the checkerboard target to avoid image distortion caused by perspective deviation. Take multiple images from multiple angles to select the best image for subsequent processing.

[0098] The collected images are preprocessed using denoising and contrast enhancement techniques to improve the accuracy of key point detection. The processed images are imported into the YOLOv8 model, which intelligently identifies the key point positions on the transformer rack and obtains the preliminary coordinates and confidence of each key point.

[0099] To ensure the accuracy of the detection results, key points with confidence values ​​below the threshold are marked as "pending confirmation", and a multi-view method is used to further verify these points. The multi-view reconstruction technology of the projection matrix is ​​used to calculate the coordinates of the key points in three-dimensional space to improve the detection accuracy.

[0100] Extract the edge points of the checkerboard target from the image and calculate the ratio of the pixel distance to the actual distance. Determine the ratio value based on the standard size of the target (for example, the pixel value corresponding to 25mm).

[0101] The built-in device focal length sensor data correction ratio is used to adjust the error caused by focal length or shooting angle offset to ensure detection accuracy under different shooting conditions; the pixel distance between key points is converted into actual distance, and the result is verified for accuracy to ensure reliability.

[0102] The actual distance between the key points detected is stored in a table with information such as timestamp, detection location, and shooting parameters. The trend curve of key point distance change is generated through trend analysis of multiple detection results. The changes in installation accuracy are displayed through bar charts and line charts, and abnormal points that exceed the standard range are marked to facilitate users to identify installation accuracy deviations and make timely adjustments. Some experimental data are as follows:

[0103] Table 1 Key point detection accuracy statistics

[0104]

[0105] Table 2 Actual distance calculation results

[0106]

[0107]

[0108] The data in Table 1 reveal the distribution of confidence in key point detection by the YOLOv8 model, showing that most key point confidences are above 80%, indicating that the method of the present invention can effectively detect high-precision key point positions under multi-view conditions. In particular, under different view reconstruction and high resolution conditions, the deviation between the initial coordinates and the final coordinates of the key points is very small, ensuring the accuracy and stability of the detection results.

[0109] The data in Table 2 compares the calculation and correction of the actual distance between each key point pair. The actual distance deviation of most key point pairs after correction in the data is kept within 0.5%, proving the effectiveness of the scale correction and multi-scale compensation technology based on the checkerboard target. This result shows that the detection method provided by the present invention can adapt to a variety of shooting angles and distance change conditions, realizes the high-precision installation detection requirements of the key parts of the test stand, and avoids the errors caused by the inaccurate shooting angle of the traditional single scale conversion method.

[0110] Through the embodiments of the method of the present invention, it can be seen that in the process of transformer stand installation detection, multi-angle key point detection and automatic scale correction are significantly better than the existing technology in terms of accuracy. Most of the existing methods rely on fixed angles or single viewing angles, which may lead to deviations in actual distances. However, this method combines the YOLOv8 model with multi-view reconstruction technology during detection to make up for the shortcomings of a single viewing angle and make the detection results more consistent. At the same time, the correction algorithm realizes dynamic scale conversion under different angles and focal lengths, significantly improving the accuracy of distance calculation between key points. Therefore, the method of the present invention not only innovatively solves the problem of multi-view key point position consistency, but also significantly optimizes the detection effect of installation accuracy.

[0111] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A transformer stand installation process detection method based on key point detection, characterized in that: include: Prepare standard checkerboard targets and image acquisition devices for scale calibration and image acquisition; Collect and process images; Automatically annotate key points of the collected images and use the YOLOv8 model to detect the locations of the annotated key points; The real distance between key points is calculated by using the checkerboard target to determine the ratio of pixels to actual distance; Record the actual distance data of key points, perform trend analysis on the distance detection results, and display the distance detection results to users through reports or applications.

2. The transformer stand installation process detection method based on key point detection according to claim 1 is characterized in that: The processing includes denoising the captured image; And by adjusting the contrast and brightness of the image, the clarity of the checkerboard target and key parts of the test stand can be improved.

3. The transformer stand installation process detection method based on key point detection according to claim 2 is characterized in that: The automatic key point labeling and the use of the YOLOv8 model to detect the positions of the key points include the following steps: Automated auxiliary tools perform preliminary annotation of key points in the image; Input the annotated image into the YOLOv8 model to detect the key points marked initially; Based on the output results of the YOLOv8 model, each detected key point is filtered according to the confidence value; If multiple images are taken, the key point detection results from different angles are integrated according to the multi-view YOLOv8 detection method, and image reconstruction technology is used to confirm the position consistency of each key point in three-dimensional space.

4. The transformer stand installation process detection method based on key point detection according to claim 3 is characterized in that: The detecting of the key points of the preliminary marking comprises: Convert the input image I into a tensor form T(I), use a multi-layer convolutional neural network CNN to extract the features of the image, and identify the characteristic patterns related to the key points; A dense candidate box generation mechanism is used to generate candidate boxes at multiple locations in the image. Each candidate box is assigned an initial position and confidence based on the feature extraction results of the network. Use the regression function to generate the center position coordinates (x det ,y det ), and the confidence value c, and the final output is a set of key point positions and corresponding confidences.

5. The transformer stand installation process detection method based on key point detection according to claim 4 is characterized in that: The step of integrating the key point detection results from different angles and using image reconstruction technology to confirm the position consistency of each key point in three-dimensional space includes the following steps: Using the known camera internal and external parameters, calculate the projection matrix Pt of each view image i , the projection matrix is ​​used to map a three-dimensional point to a two-dimensional image plane, and the formula is: Pt i =K i ×[R i |t i ] Among them, K i is the camera’s intrinsic parameter matrix, R i and t i is the camera's rotation matrix and translation vector; For the key point coordinates (x i ,y i ) using the corresponding projection matrix Pt i Perform triangulation calculations to restore the three-dimensional coordinates (X, Y, Z) of key points; The linear least square method is used to solve the problem and reduce the inconsistency caused by image detection errors.

6. The transformer stand installation process detection method based on key point detection according to claim 5 is characterized in that: Determining the ratio of pixels to actual distances by using a checkerboard target and calculating the true distances between key points comprises the following steps: Using the different scale spacings on the checkerboard target, the ratio r of multiple pixels to the actual distance is obtained by gradually measuring the side lengths of different adjacent checkerboard squares; Collect the built-in focal length sensor data and angle sensor information of the shooting device, calculate the focal length and angle offset when the image is shot according to the changes in focal length and angle, adjust the ratio r, and correct the error caused by distortion; In the actual measurement process, the most matching ratio is automatically selected as the basic ratio value, and the pixel distance between key points is converted by combining the focal length and angle corrected data for dynamic adjustment; The ratio r of the corrected pixel distance to the actual distance is used ′ , calculate the distance between the detected key points, output the actual distance between the key points, and verify the accuracy.

7. The transformer stand installation process detection method based on key point detection according to claim 6 is characterized in that: The converting of the pixel distances between key points comprises: The pixel distance D between the detected key points p,k Use the compensated ratio to convert to the actual distance D between key points a,k , the calculation formula is as follows: D a,k =r′×D p,k 。 8. The transformer stand installation process detection method based on key point detection according to claim 6 is characterized in that: The distance calculation of the detected key point position includes: Use the compensated ratio r′ to calculate the actual distance between all detected key points: D a,f =r′×D a,k The formula for the accuracy verification is: Where n is the number of samples, ACC is the accuracy verification index, ∈ is the acceptable error range, and D a,f is the actual distance between all detected key points.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the transformer stand installation process detection method based on key point detection according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the transformer stand installation process detection method based on key point detection according to any one of claims 1 to 8 are implemented.

Images