Mobile target recognition method for substation based on artificial intelligence

Abstract

The present invention discloses a method for identifying mobile targets in a substation based on artificial intelligence, and belongs to the technical field of substation automation control systems. The present invention solves the problem of incomplete monitoring in existing substations, obtains images based on a calibrated high-definition camera and performs preprocessing, thereby extracting and matching features of the processed images, and reconstructing a three-dimensional information model based on the matching result and the internal and external parameters of the high-definition camera calibration, and identifies the personnel and construction vehicle mobile targets in the substation according to the reconstructed three-dimensional information model. By calibrating the high-definition camera, and then obtaining images based on the calibrated high-definition camera and performing processing, feature extraction and matching, a three-dimensional information model can be obtained, and the three-dimensional information model can monitor the substation in all aspects, so that the personnel and construction vehicles in the substation can be accurately identified. Mobile targets, such as, avoid blind spots when monitoring, and achieve comprehensive monitoring and inspection.

Description

Technical Field

[0001] The present invention relates to the technical field of substation automation control systems, and in particular to a substation mobile target recognition method based on artificial intelligence. Background Art

[0002] Substation refers to the place in the power system where voltage and current are transformed, electric energy is received and distributed. In order to ensure the safe operation of substations, many substations have gradually replaced manual inspections with intelligent robots, which has greatly facilitated intelligent inspections of substations and improved the effectiveness and quality of inspections.

[0003] However, the inspection method relying on intelligent robots still has the following technical problems. When relying solely on inspection robots, there are generally blind spots in the inspection, and full coverage of the inspection cannot be achieved.

[0004] Therefore, the existing needs are not met, and we propose an artificial intelligence-based substation mobile target recognition method. Summary of the invention

[0005] The purpose of the present invention is to provide a substation mobile target recognition method based on artificial intelligence, which obtains a two-dimensional image of an object through a high-definition camera installed in the substation and preprocesses it, establishes a geometric model through high-definition camera calibration, solves the internal and external parameters of the high-definition camera, extracts features from the acquired image, and matches the extracted features. According to the matching results, combined with the internal and external parameters of the high-definition camera calibration, a three-dimensional information model is reconstructed, and objects are recognized based on the reconstructed three-dimensional information model, and mobile targets of personnel and construction vehicles in the substation are identified, thereby solving the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a substation mobile target identification method based on artificial intelligence, comprising the following steps:

[0007] Step 1: Image acquisition

[0008] Use a high-definition camera installed in the substation to obtain a two-dimensional image of the object and perform preprocessing;

[0009] Step 2: Camera calibration

[0010] Establish a geometric model through high-definition camera calibration and solve the internal and external parameters of the high-definition camera;

[0011] Step 3: Feature extraction

[0012] Extract features from the acquired image and match the extracted features;

[0013] Step 4: 3D reconstruction

[0014] According to the matching results, combined with the internal and external parameters calibrated by the high-definition camera, a three-dimensional information model is reconstructed;

[0015] Step 5: Target Identification

[0016] Object recognition is performed based on the reconstructed three-dimensional information model to identify moving targets such as personnel and construction vehicles in the substation.

[0017] Furthermore, in the step one, a plurality of high-definition cameras are installed in the substation, and different multi-eye camera models are constructed by high-definition cameras at multiple positions. The plurality of high-definition cameras installed in the substation are used to obtain multiple two-dimensional images of the same object from multiple perspectives, and the obtained image data are preprocessed, and the preprocessing includes filtering processing and grayscale stretching processing, and the image effect is enhanced through preprocessing.

[0018] Furthermore, the preprocessing is specifically as follows:

[0019] Filtering: The mean filtering method is used to remove noise from the image data by calculating the average value of the pixel values ​​in the area around the pixel;

[0020] Grayscale stretching processing: calculate the grayscale histogram of the original image; normalize the grayscale histogram, that is, divide the frequency of each grayscale level by the total number of image pixels to obtain the probability density function; calculate the cumulative probability density function of the normalized histogram; for each pixel, map its grayscale value to the value range of the cumulative distribution function to obtain a new grayscale level; linearly stretch the new grayscale level to map the grayscale value to the original grayscale range to obtain the final enhanced image.

[0021] Furthermore, the step 2 specifically performs the following operations:

[0022] Use a calibration reference object with known size and shape, and use a high-definition camera to take a series of images of the calibration reference object at different angles;

[0023] Through the image processing algorithm, the coordinates of the corner points in the calibration reference image are extracted;

[0024] Based on the known size and corner coordinates of the calibration reference object, the camera calibration algorithm is used to calculate the internal parameters of the HD camera, including focal length, principal point coordinates, pixel center point position and distortion parameters;

[0025] According to the actual position and posture of the calibration reference object and the captured image of the calibration reference object, the camera calibration algorithm is used to calculate the external parameters of the high-definition camera, including the position and orientation of the high-definition camera;

[0026] Based on the intrinsic and extrinsic parameters of the high-definition camera, the conversion between the pixel space coordinate system and the world coordinate system is realized.

[0027] Furthermore, in step 2, in the pixel space coordinate system, the normalized coordinates of a point are calculated by the pixel coordinates of the point and the internal parameters of the camera and the high-definition camera. After the calculation is completed, the normalized coordinates of the point are converted into the coordinates of the world coordinate system by the external parameters of the high-definition camera, thereby completing the conversion between the pixel space coordinate system and the world coordinate system.

[0028] Furthermore, the step three specifically performs the following operations:

[0029] Extract feature points of objects or scenes from the preprocessed images;

[0030] Extract feature points using feature extraction algorithm;

[0031] Use matching algorithms to match feature points from different perspectives;

[0032] According to the matching results, reliable matching pairs are screened out.

[0033] Furthermore, the feature extraction algorithm, including the SIFT algorithm and the SURF algorithm, extracts feature points from the image through feature extraction, and the feature points are the edges, corners and textures of the object; the matching algorithm adopts Brute-Force matching, which aligns the feature points in multiple images through matching so that they are in the same coordinate system.

[0034] Furthermore, the step 4 specifically performs the following operations:

[0035] Based on the matching results of step 3 and the internal and external parameters of the high-definition camera calibration in step 2, the disparity between each matching pair is calculated using the triangulation method;

[0036] Using the parallax information, the three-dimensional coordinates of the objects in the scene are calculated, and based on the three-dimensional coordinates, a three-dimensional information model of the scene is constructed;

[0037] Optimize and render the constructed 3D information model;

[0038] Exporting the optimized and rendered 3D information model, the exported 3D information model is used to perform the operation in step 5;

[0039] Among them, model rendering is to present the three-dimensional information model in a visual way. Model rendering adopts computer graphics technology. Computer graphics technology includes lighting model and texture mapping. The lighting effect is simulated by the lighting model, and the texture map is mapped to the three-dimensional information model through texture mapping.

[0040] Furthermore, the optimization process of the three-dimensional information model is specifically as follows:

[0041] Smooth the 3D information model to reduce the bumps and depressions on the model surface;

[0042] Perform detail enhancement on the 3D information model, including adding texture and improving resolution;

[0043] Optimize the 3D information model to improve the lighting performance of the 3D information model.

[0044] Furthermore, the three-dimensional information model is smoothed, including:

[0045] Extracting the initial smoothing radius corresponding to the smoothing process;

[0046] Extracting the number of data nodes of the model data corresponding to the three-dimensional information model;

[0047] Extracting model rendering sensitivity range parameters required by the three-dimensional information model during the rendering process;

[0048] The smoothing radius adjustment coefficient is obtained by using the number of data nodes of the model data and the model rendering sensitivity range parameter; wherein the smoothing radius adjustment coefficient is obtained by the following formula:

[0049]

[0050] Where s represents the smoothing radius adjustment coefficient; M represents the number of data nodes of the model data corresponding to the three-dimensional information model; M c Indicates the preset reference value of the number of nodes; L c Indicates the model rendering sensitivity reference value corresponding to the preset 3D information model; L min Indicates the lower limit of the model rendering sensitivity range parameter;

[0051] The smoothing radius after adjustment is obtained by combining the smoothing radius adjustment coefficient with the initial smoothing radius; wherein the adjusted smoothing radius is obtained by the following formula:

[0052]

[0053] Among them, R t Represents the smoothing radius after adjustment; R c Indicates the initial smoothing radius corresponding to the smoothing process; L s Indicates the actual rendering sensitivity required for subsequent model rendering corresponding to the current smoothing process; L max Indicates the upper limit of the model rendering sensitivity range parameter.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] The present invention obtains a two-dimensional image of an object through a high-definition camera installed in a substation and performs preprocessing, establishes a geometric model through high-definition camera calibration, solves the internal and external parameters of the high-definition camera, and can perform feature extraction on the obtained image based on the calibrated high-definition camera, thereby extracting feature points, and then matching the extracted feature points, and reconstructing a three-dimensional information model based on the matching results and the internal and external parameters of the high-definition camera calibration. Finally, the object is identified according to the reconstructed three-dimensional information model, and the personnel and construction vehicles in the substation are identified. By calibrating the high-definition camera, and then obtaining the image based on the calibrated high-definition camera and performing processing, feature extraction and matching, a three-dimensional information model can be obtained. The three-dimensional information model can monitor the substation in all aspects, thereby accurately identifying mobile targets such as personnel and construction vehicles in the substation, avoiding blind spots during monitoring, and realizing comprehensive monitoring and inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 The present invention is a flow chart of a method for identifying mobile targets in a substation based on artificial intelligence. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] In order to solve the following technical problems in the existing inspection method of intelligent robots in substations, there are generally blind spots in the inspection when relying solely on inspection robots, and the technical problem of being unable to achieve full coverage of the inspection is mentioned. Figure 1 , this embodiment provides the following technical solutions:

[0059] The substation mobile target recognition method based on artificial intelligence includes the following steps:

[0060] Step 1: Image acquisition

[0061] Use a high-definition camera installed in the substation to obtain a two-dimensional image of the object and perform preprocessing;

[0062] Step 2: Camera calibration

[0063] Establish a geometric model through high-definition camera calibration and solve the internal and external parameters of the high-definition camera;

[0064] Step 3: Feature extraction

[0065] Extract features from the acquired image and match the extracted features;

[0066] Step 4: 3D reconstruction

[0067] According to the matching results, combined with the internal and external parameters calibrated by the high-definition camera, a three-dimensional information model is reconstructed;

[0068] Step 5: Target Identification

[0069] Object recognition is performed based on the reconstructed three-dimensional information model to identify moving targets such as personnel and construction vehicles in the substation.

[0070] Furthermore, in the step one, a plurality of high-definition cameras are installed in the substation, and different multi-eye camera models are constructed by high-definition cameras at multiple positions. The plurality of high-definition cameras installed in the substation are used to obtain multiple two-dimensional images of the same object from multiple perspectives, and the obtained image data are preprocessed, and the preprocessing includes filtering processing and grayscale stretching processing, and the image effect is enhanced through preprocessing.

[0071] Among them, preprocessing is specifically as follows:

[0072] Filtering: The mean filtering method is used to remove noise from the image data by calculating the average value of the pixel values ​​in the area around the pixel;

[0073] Grayscale stretching processing: calculate the grayscale histogram of the original image; normalize the grayscale histogram, that is, divide the frequency of each grayscale level by the total number of image pixels to obtain the probability density function; calculate the cumulative probability density function of the normalized histogram; for each pixel, map its grayscale value to the value range of the cumulative distribution function to obtain a new grayscale level; linearly stretch the new grayscale level to map the grayscale value to the original grayscale range to obtain the final enhanced image.

[0074] Technical effects of the above content: Multiple high-definition cameras are installed in the substation, and multiple high-definition cameras can be used to form multi-camera shooting, so that multiple two-dimensional images of the same object can be obtained from multiple perspectives, providing data support for subsequent three-dimensional reconstruction. The acquired images need to be preprocessed before three-dimensional reconstruction to enhance the image effect. The preprocessing is specifically to filter the image and perform grayscale stretching. The filtering process is to remove noise in the image data by using the mean filtering method to calculate the average value of the pixel values ​​in the area around the pixel. Grayscale stretching is a commonly used enhancement method. Due to the influence of factors such as lighting conditions, photographic equipment, and shooting distance, the grayscale range in the image may be compressed to a smaller range, resulting in reduced image contrast and loss of details. Through grayscale stretching processing, the grayscale range in the original image can be mapped to a wider range, thereby enhancing the contrast and details of the image. Grayscale stretching processing maps the grayscale range of the original image to a wider range, making the difference between grayscale levels more obvious, thereby enhancing the contrast of the image. For overexposed images, grayscale stretching processing can restore the details of the bright part. For underexposed images, stretching can restore the details of the dark part. After grayscale stretching processing, pixels with similar grayscale values ​​are mapped to different grayscale values, thereby enhancing the detail information in the image and making subtle textures and edges clearer. Grayscale stretching processing can significantly improve the contrast and details of the image, and make the dark and bright details in the image richer, which is more conducive to subsequent three-dimensional reconstruction.

[0075] Step 2: Perform the following operations:

[0076] Use a calibration reference object with known size and shape, and use a high-definition camera to capture a series of images of the calibration reference object at different angles;

[0077] Through the image processing algorithm, the coordinates of the corner points in the calibration reference image are extracted;

[0078] Based on the known size and corner coordinates of the calibration reference object, the camera calibration algorithm is used to calculate the internal parameters of the HD camera, including focal length, principal point coordinates, pixel center point position and distortion parameters;

[0079] According to the actual position and posture of the calibration reference object and the captured image of the calibration reference object, the camera calibration algorithm is used to calculate the external parameters of the high-definition camera, including the position and orientation of the high-definition camera;

[0080] Based on the intrinsic and extrinsic parameters of the high-definition camera, the conversion between the pixel space coordinate system and the world coordinate system is realized.

[0081] Among them, in the pixel space coordinate system, the normalized coordinates of a point are calculated by the pixel coordinates of the point and the internal parameters of the camera and the high-definition camera. After the calculation is completed, the normalized coordinates of the point are converted into the coordinates of the world coordinate system through the external parameters of the high-definition camera, thereby completing the conversion between the pixel space coordinate system and the world coordinate system.

[0082] Technical effects of the above content: Through camera calibration, the internal and external parameters of the HD camera can be obtained. The internal parameters of the HD camera include focal length, principal point coordinates, pixel center point position and distortion parameters, which can be used to convert image coordinates into camera coordinates. The distortion coefficient is used to eliminate the radial distortion and tangential distortion of the image. The external parameters of the HD camera include the position and orientation of the HD camera, which can be used to convert the camera coordinates into world coordinates. Camera calibration can eliminate the distortion of the HD camera and improve the precision and accuracy of the image. When we need to perform three-dimensional reconstruction, camera calibration is a very important basic step. It provides basic data for subsequent image processing and analysis. Multiple cameras are required to shoot the same scene at the same time, and each camera needs to be calibrated to finally obtain the three-dimensional information model of the scene.

[0083] It should be noted that the corresponding relationship between the coordinates of the known points in space on the calibration reference and the image coordinate points, the camera geometry model and the constraints for obtaining the internal and external parameters of the camera are always established, and then these parameters are accurately determined using the optimization algorithm. The calibration reference can be either three-dimensional or two-dimensional. If it is a three-dimensional calibration object, only one image is needed to determine all the camera parameters, but the accuracy of the three-dimensional calibration object is required to be very high. If it is a two-dimensional reference object, different images need to be obtained from different angles, and the required parameters are obtained through the mutual constraints between them. Commonly used camera calibration methods include Zhang Zhengyou calibration method, etc. The calibrated high-definition camera parameters will be used for subsequent image processing and three-dimensional reconstruction processes.

[0084] Step 3: Perform the following operations:

[0085] Extract feature points of objects or scenes from the preprocessed images;

[0086] Extract feature points using feature extraction algorithm;

[0087] Use matching algorithms to match feature points from different perspectives;

[0088] According to the matching results, reliable matching pairs are screened out.

[0089] Feature extraction algorithms, including SIFT algorithm and SURF algorithm, extract feature points from images through feature extraction. Feature points are edges, corners and textures of objects. The matching algorithm uses Brute-Force matching to align feature points in multiple images through matching so that they are in the same coordinate system.

[0090] The technical effect of the above content is as follows: feature extraction and matching are performed on the preprocessed image, and feature points are extracted using a feature extraction algorithm. The feature extraction algorithm may adopt a SIFT algorithm or a SURF algorithm, so as to extract feature points from the image. Then, a matching algorithm is used to match feature points from different perspectives. The matching algorithm adopts Brute-Force matching, and feature points in multiple images are aligned through matching so that they are in the same coordinate system. Finally, reliable matching pairs are screened out based on the matching results. The feature points under different perspectives are matched through the matching algorithm, which can provide a basis for subsequent three-dimensional reconstruction.

[0091] Step 4: Perform the following operations:

[0092] Based on the matching results of step 3 and the internal and external parameters of the high-definition camera calibration in step 2, the disparity between each matching pair is calculated using the triangulation method;

[0093] Using the parallax information, the three-dimensional coordinates of the objects in the scene are calculated, and based on the three-dimensional coordinates, a three-dimensional information model of the scene is constructed;

[0094] Optimize and render the constructed 3D information model;

[0095] The optimized and rendered three-dimensional information model is exported, and the exported three-dimensional information model is used to perform the operation in step five.

[0096] The optimization processing of the three-dimensional information model is specifically as follows:

[0097] Smooth the 3D information model to reduce the bumps and depressions on the model surface;

[0098] Perform detail enhancement on the 3D information model, including adding texture and improving resolution;

[0099] Optimize the 3D information model to improve the lighting performance of the 3D information model.

[0100] Among them, model rendering is to present the three-dimensional information model in a visual way. Model rendering adopts computer graphics technology. Computer graphics technology includes lighting model and texture mapping. The lighting effect is simulated by the lighting model, and the texture map is mapped to the three-dimensional information model through texture mapping.

[0101] Technical effect of the above content: Based on the above matching results and the internal and external parameters calibrated by the high-definition camera, the triangulation method can be used to calculate the disparity between each matching pair, and the obtained disparity information can be used to calculate the three-dimensional coordinates of the objects in the scene, so that the three-dimensional information model of the scene can be constructed according to the three-dimensional coordinates, so as to achieve the purpose of three-dimensional reconstruction. The construction of the three-dimensional information model requires optimization processing and model rendering, so as to improve the accuracy and usability of the three-dimensional information model. The optimization processing is to smooth the three-dimensional information model, enhance the details and optimize the three-dimensional information model. The smoothing processing can reduce the protrusions and depressions on the model surface, making the surface of the three-dimensional information model smoother, and the texture can be increased and the resolution can be improved. The optimization can improve the lighting performance of the three-dimensional information model. According to the above optimization processing, the three-dimensional information The quality of the information model is improved, and the optimized 3D information model is then rendered. Model rendering is to present the 3D information model in a visual way, using computer graphics technology. Computer graphics technology includes illumination model and texture mapping. The illumination model simulates the illumination effect, and texture mapping is used to map the texture map to the 3D information model. Texture mapping can restore the 3D scene and objects. Finally, the optimized and rendered 3D information model is exported and saved in OBJ or FBX format. The exported 3D information model can be used to identify moving targets of personnel and construction vehicles in the substation. The 3D information model can monitor the substation in all aspects, so as to accurately identify moving targets such as personnel and construction vehicles in the substation, avoid blind spots during monitoring, and realize comprehensive monitoring and inspection.

[0102] Working principle: Through high-definition camera calibration, a geometric model is established to solve the internal and external parameters of the high-definition camera. Based on the calibrated high-definition camera, multiple high-definition cameras installed in the substation are used to obtain multiple two-dimensional images of the same object from multiple perspectives. Then, the obtained images are preprocessed to enhance the image effect, and then feature extraction is performed on the preprocessed images. After the feature points are extracted, matching is performed. Based on the matching results and the internal and external parameters of the high-definition camera calibration, a three-dimensional information model can be reconstructed. Finally, according to the reconstructed three-dimensional information model, objects are identified, and moving targets such as personnel and construction vehicles in the substation are identified. By calibrating the high-definition camera, images are obtained based on the calibrated high-definition camera and processed, feature extracted and matched to obtain a three-dimensional information model. The three-dimensional information model can monitor the substation in all aspects, so that it can accurately identify moving targets such as personnel and construction vehicles in the substation, avoid blind spots during monitoring, and achieve comprehensive monitoring and inspection.

[0103] Specifically, the three-dimensional information model is smoothed, including:

[0104] Extract the initial smoothing radius corresponding to the smoothing process;

[0105] Extracting the number of data nodes of the model data corresponding to the three-dimensional information model;

[0106] Extracting model rendering sensitivity range parameters required by the three-dimensional information model during the rendering process;

[0107] The smoothing radius adjustment coefficient is obtained by using the number of data nodes of the model data and the model rendering sensitivity range parameter; wherein the smoothing radius adjustment coefficient is obtained by the following formula:

[0108]

[0109] Where s represents the smoothing radius adjustment coefficient; M represents the number of data nodes of the model data corresponding to the three-dimensional information model; M c Indicates the preset reference value of the number of nodes; L c Indicates the model rendering sensitivity reference value corresponding to the preset 3D information model; L min Indicates the lower limit of the model rendering sensitivity range parameter;

[0110] The smoothing radius after adjustment is obtained by combining the smoothing radius adjustment coefficient with the initial smoothing radius; wherein the adjusted smoothing radius is obtained by the following formula:

[0111]

[0112] Among them, R t Represents the smoothing radius after adjustment; R c Indicates the initial smoothing radius corresponding to the smoothing process; L s Indicates the actual rendering sensitivity required for subsequent model rendering corresponding to the current smoothing process; L max Indicates the upper limit of the model rendering sensitivity range parameter.

[0113] The technical effect of the above technical solution is: by extracting the number of data nodes and rendering sensitivity range parameters of the three-dimensional information model and calculating the smoothing radius adjustment coefficient, the smoothing radius can be automatically adjusted to adapt to different model complexities and rendering requirements. This adaptability ensures that the smoothing process will neither oversimplify the model details nor retain too much noise or jaggedness due to insufficient smoothing. The calculation formula of the smoothing radius adjustment coefficient comprehensively considers the two factors of the number of data nodes and rendering sensitivity, so that the smoothing process can flexibly respond to different situations. For example, when the number of model data nodes is large, the smoothing radius may be increased to reduce details; when the rendering sensitivity requirements are high, the smoothing radius may be reduced to maintain details. Through the adjusted smoothing radius, the rendering effect of the model can be optimized, so that the model has a smoother surface and better visual effects while maintaining certain details.

[0114] On the other hand, the calculation formula of the smoothing radius adjustment coefficient takes into account two key factors: the number of data nodes and rendering sensitivity. Therefore, it can more accurately reflect the impact of model complexity and rendering requirements on the smoothing radius. This helps to improve the accuracy of the smoothing process. Since the maximum value function is used to avoid the situation where the denominator is zero, and reasonable scaling and offset are adopted, the calculation formula can remain stable in many cases. This helps to ensure the stability of the smoothing process under different models and rendering conditions. It allows technicians to clearly understand how the smoothing radius is adjusted according to model complexity and rendering requirements. This helps to enhance the interpretability and maintainability of the technology.

[0115] At the same time, the smoothing radius obtained by the above method can effectively improve the rationality of the smoothing radius and its matching with the actual needs of the smoothing processing of the three-dimensional model. Reasonable smoothing radius setting can remove noise and jaggedness on the model surface, make the model surface smoother, and improve the visual effect. By adaptively adjusting the smoothing radius, smoothing can be performed while maintaining the details of the model to avoid excessive smoothing and loss of details. Appropriate smoothing can reduce the amount of calculation during rendering and improve rendering efficiency, especially when dealing with large-scale or complex models. Smoothing can improve the surface quality of the model, make the model more realistic and delicate, and improve the overall quality of the model.

[0116] In summary, the above technical scheme and its calculation formula show good technical effects in terms of adaptability and flexibility, and have good technical effects on performance indicators in terms of accuracy, stability, interpretability and scalability.

[0117] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0118] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A substation mobile target recognition method based on artificial intelligence, characterized in that: The following steps are involved: Step 1: Image acquisition Use a high-definition camera installed in the substation to obtain a two-dimensional image of the moving target and perform preprocessing; Step 2: Camera calibration Establish a geometric model through high-definition camera calibration and solve the internal and external parameters of the high-definition camera; Step 3: Feature extraction Extract features from the acquired image and match the extracted features; Step 4: 3D reconstruction According to the matching results, the three-dimensional information model is reconstructed by combining the internal and external parameters of the high-definition camera; Step 5: Target Identification Identifying mobile targets based on the three-dimensional information model, wherein the mobile targets include personnel and construction vehicles in the substation; The step 2 specifically performs the following operations: Use a calibration reference object with known size and shape, and use a high-definition camera to take a series of images of the calibration reference object at different angles; Through the image processing algorithm, the coordinates of the corner points in the calibration reference image are extracted; Based on the known size and corner coordinates of the calibration reference object, the camera calibration algorithm is used to calculate the internal parameters of the HD camera, including focal length, principal point coordinates, pixel center point position and distortion parameters; According to the actual position and posture of the calibration reference object and the captured image of the calibration reference object, the camera calibration algorithm is used to calculate the external parameters of the high-definition camera, including the position and orientation of the high-definition camera; Based on the internal and external parameters of the high-definition camera, the conversion between the pixel space coordinate system and the world coordinate system is realized; The step 4 specifically performs the following operations: Based on the matching results of step 3 and the internal and external parameters of the high-definition camera calibration in step 2, the disparity between each matching pair is calculated using the triangulation method; Using the parallax information, the three-dimensional coordinates of the objects in the scene are calculated, and based on the three-dimensional coordinates, a three-dimensional information model of the scene is constructed; Optimize and render the constructed 3D information model; Exporting the optimized and rendered 3D information model, the exported 3D information model is used to perform the operation in step 5; Model rendering is to present the three-dimensional information model in a visual way. Model rendering adopts computer graphics technology. Computer graphics technology includes illumination model and texture mapping. The illumination model is used to simulate the illumination effect. Texture mapping is used to map the texture map onto the three-dimensional information model. The optimization process of the three-dimensional information model is specifically as follows: Smooth the 3D information model to reduce the bumps and depressions on the model surface; Enhance the details of the 3D information model, including adding texture and improving resolution; Optimize the 3D information model to improve the lighting performance of the 3D information model.

2. The method for identifying mobile targets in a substation based on artificial intelligence according to claim 1 is characterized in that: In the step 1, a plurality of high-definition cameras are installed in the substation, and different multi-camera models are constructed by using the high-definition cameras at multiple positions. The plurality of high-definition cameras installed in the substation are used to obtain multiple two-dimensional images of the same object from multiple perspectives, and the obtained image data are preprocessed, and the preprocessing includes filtering processing and grayscale stretching processing, and the image effect is enhanced through the preprocessing.

3. The method for identifying mobile targets in substations based on artificial intelligence according to claim 2 is characterized in that: The preprocessing is specifically as follows: Filtering: The mean filtering method is used to remove noise from the image data by calculating the average value of the pixel values ​​in the area around the pixel; Grayscale stretching processing: calculate the grayscale histogram of the original image; normalize the grayscale histogram, that is, divide the frequency of each grayscale level by the total number of image pixels to obtain the probability density function; calculate the cumulative probability density function of the normalized histogram; for each pixel, map its grayscale value to the value range of the cumulative distribution function to obtain a new grayscale level; linearly stretch the new grayscale level to map the grayscale value to the original grayscale range to obtain the final enhanced image.

4. The method for identifying mobile targets in a substation based on artificial intelligence according to claim 1 is characterized in that: In the step 2, in the pixel space coordinate system, the normalized coordinates of a point are calculated by the pixel coordinates of the point and the internal parameters of the high-definition camera. After the calculation is completed, the normalized coordinates of the point are converted into the coordinates of the world coordinate system by the external parameters of the high-definition camera, thereby completing the conversion between the pixel space coordinate system and the world coordinate system.

5. The method for identifying mobile targets in a substation based on artificial intelligence according to claim 1 is characterized in that: The step three specifically performs the following operations: Extract feature points of objects or scenes from the preprocessed images; Extract feature points using feature extraction algorithm; Use matching algorithms to match feature points from different perspectives; According to the matching results, reliable matching pairs are screened out.

6. The method for identifying mobile targets in a substation based on artificial intelligence according to claim 5 is characterized in that: The feature extraction algorithm includes SIFT algorithm and SURF algorithm, which extracts feature points from the image through feature extraction, and the feature points are edges, corners and textures of objects; the matching algorithm adopts Brute-Force matching, which aligns the feature points in multiple images through matching.

7. The method for identifying mobile targets in substations based on artificial intelligence according to claim 6 is characterized in that: Smoothing of 3D information models, including: Extract the initial smoothing radius corresponding to the smoothing process; Extracting the number of data nodes of the model data corresponding to the three-dimensional information model; Extracting model rendering sensitivity range parameters required by the three-dimensional information model during the rendering process; The smoothing radius adjustment coefficient is obtained by using the number of data nodes of the model data and the model rendering sensitivity range parameter; wherein the smoothing radius adjustment coefficient is obtained by the following formula: Where s represents the smoothing radius adjustment coefficient; M represents the number of data nodes of the model data corresponding to the three-dimensional information model; M c Indicates the preset reference value of the number of nodes; L c Indicates the model rendering sensitivity reference value corresponding to the preset 3D information model; L min Indicates the lower limit of the model rendering sensitivity range parameter; The smoothing radius after adjustment is obtained by combining the smoothing radius adjustment coefficient with the initial smoothing radius; The adjusted smoothing radius is obtained by the following formula: Among them, R t Represents the smoothing radius after adjustment; R c Indicates the initial smoothing radius corresponding to the smoothing process; L s Indicates the actual rendering sensitivity required for subsequent model rendering corresponding to the current smoothing process; L max Indicates the upper limit of the model rendering sensitivity range parameter.

Images