A power equipment three-dimensional fault location method and system based on semantic voxels and projection mapping

Abstract

The application discloses a kind of based on semantic voxel and projection mapping's electric power equipment three-dimensional fault location method and system, comprising: camera and fault detection equipment are arranged in equipment space, obtain multi-view image and two-dimensional fault point information, generate electric power equipment three-dimensional model by three-dimensional reconstruction algorithm;Utilize principal component analysis to adjust the posture of three-dimensional model, make its main direction and standard coordinate system align;Three-dimensional model is divided into regular voxel and is projected on side, establish the mapping relationship of two-dimensional and three-dimensional space by matching with original image;Extract two-dimensional fault point and after bilateral matching, it is mapped to three-dimensional voxel surface, realize three-dimensional accurate positioning by calculating the shortest distance of voxel center normal elongation line, finally generate the visualization result of fault area. Through multi-source data fusion and spatial geometric positioning, overcome the positioning fuzzy problem caused by shielding, realize the accurate three-dimensional positioning and visualization of electric power equipment fault.

Description

Technical Field

[0001] This invention belongs to the field of intelligent detection technology for power equipment, and in particular relates to a method and system for three-dimensional fault location of power equipment based on semantic voxels and projection mapping. Background Technology

[0002] In recent years, with the development of intelligent and digital power equipment, power inspection technology has gradually moved towards automation and intelligence. Inspection tasks mainly cover the status monitoring of key equipment such as transformers, switchgear, and transmission lines, which has a significant impact on equipment operation safety. 3D reconstruction, point cloud acquisition, and multi-sensor fusion technologies have shown broad application prospects in power inspection, providing complete spatial information on equipment structures and offering data support for fault identification, maintenance decisions, and safety assurance. At the same time, with the expansion of power system scale and the increase in equipment complexity, inspection tasks place higher demands on accuracy, efficiency, and reliability. Traditional inspection methods relying on manual labor or two-dimensional signal analysis are no longer sufficient to meet the needs of modern power operation and maintenance.

[0003] Currently, most fault location methods are still based on two-dimensional signals or image information, such as fault surface identification using single-view images or acoustic sensors. These methods have significant limitations: firstly, two-dimensional data cannot fully reflect the internal structure of the equipment, and internal or obscured faults cannot be detected; secondly, two-dimensional methods struggle to accurately locate the spatial position of the fault point, leading to large location errors, especially in complex environments or scenarios with multiple overlapping devices, easily resulting in missed or false detections; furthermore, traditional methods cannot quickly provide visualized spatial information in emergency repair scenarios, making repair decisions reliant on experience-based judgment, affecting fault handling efficiency and safe equipment operation.

[0004] Therefore, designing a three-dimensional fault location scheme for power equipment based on semantic voxels and projection mapping can not only achieve accurate spatial location of fault points inside and outside the equipment, improving positioning accuracy and inspection reliability, but also intuitively display the fault location through point cloud filling and visualization rendering, improving inspection efficiency and providing intuitive basis for operation and maintenance decisions. Summary of the Invention

[0005] To address the above technical problems, this invention provides a method for three-dimensional fault location of power equipment based on semantic voxels and projection mapping.

[0006] The technical solution adopted by this invention to solve its technical problem is:

[0007] A method for three-dimensional fault localization of power equipment based on semantic voxels and projection mapping includes the following steps:

[0008] S100: Deploy cameras and fault detection equipment in the limited space of power equipment to acquire multi-view images and two-dimensional fault point information, and generate a three-dimensional model of power equipment from the multi-view images through a three-dimensional reconstruction algorithm.

[0009] S200: Principal component analysis is used to adjust the attitude of the reconstructed 3D model so that the principal direction of the 3D model is aligned with the standard coordinate system;

[0010] S300: Divide the 3D model after attitude adjustment into regular voxels, perform 2D projection on the side of the voxels to generate a projected image, match the projected image with the original multi-view image, and establish the spatial correspondence between the side of the 3D voxel and the original image.

[0011] S400: Extract two-dimensional fault coordinate points from the fault surface collected by the fault detection equipment, and match the fault points on both sides to form a matching pair;

[0012] S500: Based on spatial correspondence, the points matched by the two-dimensional fault points are mapped to the corresponding three-dimensional voxel surfaces to obtain the coordinates of the fault points on the three-dimensional surfaces;

[0013] S600: For a fault point on a three-dimensional surface, calculate the extension line of the corresponding voxel center, and determine the location of the fault point in three-dimensional space by solving the shortest distance between the extension lines.

[0014] S700: Based on the determined three-dimensional fault location, the surrounding area is visualized to generate a three-dimensional visualization result of the fault area.

[0015] Preferably, S200 includes:

[0016] S210: Calculate the centroid of the 3D point cloud and center the point cloud. Calculate the average position of all points in the 3D point cloud to obtain the centroid of the point cloud.

[0017] ;

[0018] in Let N represent the coordinates of the i-th point, and let N represent the total number of points. The centroid of the point cloud;

[0019] S220: The point cloud is translated to the vicinity of the centroid to eliminate the global translation component, resulting in a decentralized point cloud matrix.

[0020] ;

[0021] in The original point cloud matrix, The decentralized point cloud matrix, It is a column vector whose elements are all 1s. To transpose the centroid of the point cloud;

[0022] S230: Calculate the decentralized point cloud Covariance matrix:

[0023] ;

[0024] in This is the transposed decentralized point cloud matrix. The covariance matrix is ​​a symmetric matrix. Its diagonal elements describe the variance of the point cloud in the x, y, and z directions, i.e., the distribution width of the points in each direction. The off-diagonal elements describe the covariance between different directions, i.e., the tilt relationship or coupling degree of the point cloud in each direction in space.

[0025] S240: Perform eigenvalue decomposition on the covariance matrix:

[0026] ;

[0027] in column vectors This represents the three mutually perpendicular principal directions of the point cloud in space. For a diagonal eigenvalue matrix, each Sorted by the magnitude of the variance along the corresponding eigenvector direction. ;

[0028] S250: Main direction based on point cloud For decentralized point clouds Rotate the points to obtain a rotated point cloud matrix aligned with the standard coordinate system.

[0029] ;

[0030] in This is the rotated point cloud matrix, whose principal direction is now aligned with the standard coordinate system.

[0031] Preferably, S300 includes:

[0032] S310: The 3D model of the power equipment is adjusted according to the preset voxel dimensions. Mesh generation is performed on a point in three-dimensional space. The corresponding voxel index is:

[0033] ;

[0034] in The minimum coordinates of the model boundary are i, j, and k, which are voxel indices and are all rounded down.

[0035] voxel center coordinates for:

[0036] ;

[0037] S320: Project each voxel center onto the front, back, left, and right sides of the power equipment to generate a corresponding set of two-dimensional projected images. The projection process involves calculating the side normal vector and in-plane direction, rotating and centering the voxel point cloud to align the side with the two-dimensional plane. Then, the camera position and viewpoint are set, and the rotated voxel point cloud is rendered to obtain side screenshots.

[0038] S330: Matching the set of 2D projected images with the original multi-view images by combining color histograms and edge histograms: First, statistically analyze the color distribution histograms of the 2D projected images and the original multi-view images, and calculate the color similarity between the two; then, extract the edges of the 2D projected images and the original multi-view images, statistically analyze the histograms of edge intensity or direction, and calculate the structural similarity; finally, weighted summation of color similarity and structural similarity yields the final matching score; each 2D projected image is compared with all original multi-view images, and the image with the highest comprehensive similarity is selected as the matching result, thereby achieving the best correspondence between the 3D voxel side view and the original multi-view images.

[0039] Preferably, S400 includes:

[0040] S410: From the fault surfaces collected by the fault detection equipment on both sides of the power equipment, extract the vertices and two-dimensional fault coordinate points of each fault surface. , where L and R represent the adjacent two sides respectively;

[0041] S420: To ensure that the points on both sides originate from the same fault source, the fault points on each side are arranged in ascending order according to their vertical coordinates, resulting in separate sequences. , ,…and , Then, matching pairs are formed one by one according to the serial number. .

[0042] Preferably, S500 includes:

[0043] S510: Based on spatial correspondence, determine the three-dimensional voxel surface corresponding to the two-dimensional fault surface;

[0044] S520: Establish a mapping relationship between the four vertices of each faulty surface and the four vertices of the corresponding three-dimensional voxel surface;

[0045] S530: Based on the mapping relationship, the two-dimensional fault point is mapped to the three-dimensional voxel surface through the mapping function to obtain the corresponding three-dimensional coordinates: assuming the vertex of the two-dimensional fault surface is... The corresponding vertices of the three-dimensional voxel face are Then it is obtained through affine mapping:

[0046] ;

[0047] in The coordinates of the two-dimensional fault point For the corresponding three-dimensional surface coordinates, The function is for mapping.

[0048] Preferably, S600 includes:

[0049] S610: Assume the coordinates of the fault point after mapping from a two-dimensional fault point to a three-dimensional space are... Calculate the corresponding voxel grid index. :

[0050] ;

[0051] via voxel grid indexing Retrieve the previously saved voxel center coordinates;

[0052] S620: The coordinates of the centers of the fault voxels on the two faces are denoted as follows: and Along the normal direction of each voxel surface and Extend the line, and it can be represented as:

[0053] , ;

[0054] in and These represent the extended distances along the direction of the central normal vector;

[0055] S630: Find the parameters of the closest point on the two lines by solving an optimization problem that minimizes the distance between them. :

[0056] ;

[0057] Then the coordinates of the fault point in three-dimensional space for:

[0058] .

[0059] Preferably, S700 includes:

[0060] After completing the precise location of the three-dimensional fault point, the located three-dimensional fault point The core is to first determine the distribution of point clouds within a preset range. If the point cloud density within the preset range is higher than a preset threshold, then there are valid point clouds, and the area is directly rendered to highlight the fault clustering effect and achieve intuitive spatial display. If the point cloud density within the preset range is lower than the preset threshold or there are missing points, then an interpolation method based on nearest neighbor search is used to fill the missing areas to generate continuous and complete 3D point cloud visualization results.

[0061] A three-dimensional fault location system for power equipment based on semantic voxels and projection mapping includes:

[0062] The data acquisition and 3D reconstruction module is used to deploy cameras and fault detection equipment in the limited space of power equipment to acquire multi-view images and two-dimensional fault point information, and generate a 3D model of the power equipment from the multi-view images through a 3D reconstruction algorithm.

[0063] The attitude adjustment module is used to adjust the attitude of the reconstructed 3D model using principal component analysis, so that the principal direction of the 3D model is aligned with the standard coordinate system.

[0064] The voxel projection module is used to divide the 3D model after posture adjustment into regular voxels, and to perform 2D projection on the side of the voxels to generate a projected image. The projected image is then matched with the original multi-view image to establish a spatial correspondence between the side of the 3D voxel and the original image.

[0065] The fault point matching module is used to extract two-dimensional fault coordinate points from the fault surface collected by the fault detection equipment, and match the fault points on both sides to form a matching pair.

[0066] The fault point mapping module is used to map the points matched by the two-dimensional fault points to the corresponding three-dimensional voxel surfaces based on spatial correspondence, so as to obtain the coordinates of the fault points on the three-dimensional surfaces.

[0067] The fault location module is used to calculate the extension line of the corresponding voxel center for a fault point on a three-dimensional surface, and determine the location of the fault point in three-dimensional space by solving the shortest distance between the extension lines.

[0068] The visualization module is used to visualize the surrounding area based on the determined 3D fault location, generating a 3D visualization result of the fault area.

[0069] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of a three-dimensional fault location method for power equipment based on semantic voxel and projection mapping.

[0070] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for three-dimensional fault location of power equipment based on semantic voxel and projection mapping.

[0071] This invention achieves precise 3D localization and visualization of power equipment faults by fusing multi-view images with fault detection data. The scheme utilizes 3D reconstruction and unified attitude processing to construct an accurate 3D model of the equipment; establishes a reliable mapping between 2D fault information and 3D space through semantic voxel segmentation and projection matching; and effectively solves the problem of positioning ambiguity caused by occlusion or data loss based on spatial intersection positioning using normal extension lines. This significantly improves the accuracy and intuitiveness of fault localization, providing reliable support for intelligent operation and maintenance of power equipment. Attached Figure Description

[0072] Figure 1 This is a flowchart of a three-dimensional fault location method for power equipment based on semantic voxel and projection mapping in one embodiment of the present invention;

[0073] Figure 2 This is a flowchart of PCA attitude adjustment for a 3D model in one embodiment of the present invention;

[0074] Figure 3 This is a flowchart of voxel division and two-dimensional projection in one embodiment of the present invention;

[0075] Figure 4 This is a flowchart of a three-dimensional fault location method in one embodiment of the present invention. Detailed Implementation

[0076] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0077] In one embodiment, such as Figure 1 As shown, a method for three-dimensional fault localization of power equipment based on semantic voxels and projection mapping includes the following steps:

[0078] S100: Deploy cameras and fault detection equipment in the limited space of power equipment to acquire multi-view images and two-dimensional fault point information, and generate a three-dimensional model of power equipment from the multi-view images through a three-dimensional reconstruction algorithm.

[0079] S200: Principal component analysis is used to adjust the attitude of the reconstructed 3D model so that the principal direction of the 3D model is aligned with the standard coordinate system;

[0080] S300: Divide the 3D model after attitude adjustment into regular voxels, perform 2D projection on the side of the voxels to generate a projected image, match the projected image with the original multi-view image, and establish the spatial correspondence between the side of the 3D voxel and the original image.

[0081] S400: Extract two-dimensional fault coordinate points from the fault surface collected by the fault detection equipment, and match the fault points on both sides to form a matching pair;

[0082] S500: Based on spatial correspondence, the points matched by the two-dimensional fault points are mapped to the corresponding three-dimensional voxel surfaces to obtain the coordinates of the fault points on the three-dimensional surfaces;

[0083] S600: For a fault point on a three-dimensional surface, calculate the extension line of the corresponding voxel center, and determine the location of the fault point in three-dimensional space by solving the shortest distance between the extension lines.

[0084] S700: Based on the determined three-dimensional fault location, the surrounding area is visualized to generate a three-dimensional visualization result of the fault area.

[0085] The aforementioned method for 3D fault localization of power equipment based on semantic voxels and projection mapping achieves accurate localization and visualization of power equipment faults in 3D space by fusing multi-view images and fault detection data. First, by utilizing 3D reconstruction and unified attitude processing, it overcomes the limitations of single-view or single-sensor data, providing an accurate and complete 3D model of the equipment for fault analysis. Second, through semantic voxel segmentation and multimodal projection matching, a reliable correspondence between 2D fault information and 3D voxel space is established, achieving accurate mapping from 2D fault surfaces to 3D space. Finally, based on the principle of spatial intersection positioning using normal extension lines, it effectively solves the problem of positioning ambiguity caused by equipment structure occlusion or data loss, significantly improving positioning accuracy and robustness. This method effectively enhances the intuitiveness, accuracy, and automation level of fault diagnosis, providing reliable technical support for intelligent operation and maintenance of power equipment.

[0086] In one embodiment, S100 includes: Firstly, in a limited space within a power scenario, deploying cameras and fault detection equipment to acquire images of the power equipment's appearance and fault characteristics, respectively. The cameras comprehensively cover the power equipment using multi-view shooting, collecting image information from different angles; the fault detection equipment is positioned on both sides adjacent to the power equipment, acquiring the location information of potential fault areas through the excitation and echo feature extraction of acoustic signals. Subsequently, the multi-view image data is input into a 3D reconstruction algorithm (3D Gaussian) to generate a 3D model containing the equipment's geometry and surface details.

[0087] In one embodiment, principal component analysis (PCA) is used to adjust the attitude of the reconstructed 3D model of the power equipment, as shown in the flowchart below. Figure 2 As shown, S200 includes:

[0088] S210: Calculate the centroid of the 3D point cloud and center the point cloud. Calculate the average position of all points in the 3D point cloud to obtain the centroid of the point cloud.

[0089] ;

[0090] in Let N represent the coordinates of the i-th point, and let N represent the total number of points. The centroid of the point cloud;

[0091] S220: The point cloud is translated to the vicinity of the centroid to eliminate the global translation component, resulting in a decentralized point cloud matrix.

[0092] ;

[0093] in The original point cloud matrix, The decentralized point cloud matrix, It is a column vector whose elements are all 1s. To transpose the centroid of the point cloud;

[0094] S230: Calculate the decentralized point cloud Covariance matrix:

[0095] ;

[0096] in This is the transposed decentralized point cloud matrix. The covariance matrix has diagonal elements that describe the variance of the point cloud in the x, y, and z directions, i.e., the distribution width of points in each direction, while off-diagonal elements describe the covariance between different directions, i.e., the tilt relationship or coupling degree of the point cloud in each direction in space.

[0097] S240: Perform eigenvalue decomposition on the covariance matrix:

[0098] ;

[0099] in column vectors This represents the three mutually perpendicular principal directions of the point cloud in space. For a diagonal eigenvalue matrix, each Sorted by the magnitude of the variance along the corresponding eigenvector direction. ;

[0100] S250: Main direction based on point cloud For decentralized point clouds Rotate the points to obtain a rotated point cloud matrix aligned with the standard coordinate system.

[0101] ;

[0102] in This is the rotated point cloud matrix, whose principal direction is now aligned with the standard coordinate system.

[0103] In one embodiment, the reconstructed 3D model of the power equipment is divided into regular voxels, and 2D projections are performed on the four sides of the model to generate corresponding 2D images. Subsequently, these projected images are matched with the original multi-view images to establish the spatial correspondence between the sides of the 3D voxels and the actual equipment image. The flowchart is as follows: Figure 3 As shown, S300 includes:

[0104] S310: The 3D model of the power equipment is adjusted according to the preset voxel dimensions. Mesh generation is performed on a point in three-dimensional space. The corresponding voxel index is:

[0105] ;

[0106] in The minimum coordinates of the model boundary are i, j, and k, which are voxel indices and are all rounded down.

[0107] voxel center coordinates for:

[0108] ;

[0109] S320: Project each voxel center onto the front, back, left, and right sides of the power equipment to generate a corresponding set of two-dimensional projected images. The projection process involves calculating the side normal vector and in-plane direction, rotating and centering the voxel point cloud to align the side with the two-dimensional plane. Then, the camera position and viewpoint are set, and the rotated voxel point cloud is rendered to obtain side screenshots.

[0110] S330: Matching the set of 2D projected images with the original multi-view images by combining color histograms and edge histograms: First, statistically analyze the color distribution histograms of the 2D projected images and the original multi-view images, and calculate the color similarity between the two; then, extract the edges of the 2D projected images and the original multi-view images, statistically analyze the histograms of edge intensity or direction, and calculate the structural similarity; finally, weighted summation of color similarity and structural similarity yields the final matching score; each 2D projected image is compared with all original multi-view images, and the image with the highest comprehensive similarity is selected as the matching result, thereby achieving the best correspondence between the 3D voxel side view and the original multi-view images.

[0111] In one embodiment, S400 includes:

[0112] S410: From the fault surfaces collected by the fault detection equipment on both sides of the power equipment, extract the vertices and two-dimensional fault coordinate points of each fault surface. , where L and R represent the adjacent two sides respectively;

[0113] S420: To ensure that the points on both sides originate from the same fault source, the fault points on each side are arranged in ascending order according to their vertical coordinates, resulting in separate sequences. , ,…and , Then, matching pairs are formed one by one according to the serial number. .

[0114] Specifically, this method can complete the matching relationship of fault points on both sides through simple vertical sorting, ensuring that the points of each fault source on the adjacent detection surfaces are correctly associated, providing a reliable foundation for subsequently mapping the two-dimensional fault surface to three-dimensional voxels and achieving accurate three-dimensional fault location.

[0115] In one embodiment, S500 includes:

[0116] S510: Based on spatial correspondence, determine the three-dimensional voxel surface corresponding to the two-dimensional fault surface;

[0117] S520: Establish a mapping relationship between the four vertices of each faulty surface and the four vertices of the corresponding three-dimensional voxel surface;

[0118] S530: Based on the mapping relationship, the two-dimensional fault point is mapped to the three-dimensional voxel surface through the mapping function to obtain the corresponding three-dimensional coordinates: assuming the vertex of the two-dimensional fault surface is... The corresponding vertices of the three-dimensional voxel face are Then it is obtained through affine mapping:

[0119] ;

[0120] in The coordinates of the two-dimensional fault point For the corresponding three-dimensional surface coordinates, This is the mapping function. In one embodiment, after mapping the fault points on the 3D voxel surface, the center of the corresponding voxel is obtained. Extension lines are drawn along the normal direction of each voxel surface, and the location of the nearest point is determined by solving for the shortest distance between the two extension lines. Finally, the midpoint between the two nearest points is taken as the center fault point, achieving precise positioning in 3D space. The flowchart is as follows. Figure 4 As shown. S600 includes:

[0121] S610: Assume the coordinates of the fault point after mapping from a two-dimensional fault point to a three-dimensional space are... Calculate the corresponding voxel grid index. :

[0122] ;

[0123] via voxel grid indexing Retrieve the previously saved voxel center coordinates;

[0124] S620: The coordinates of the centers of the fault voxels on the two faces are denoted as follows: and Along the normal direction of each voxel surface and Extend the line, and it can be represented as:

[0125] , ;

[0126] in and These represent the extended distances along the direction of the central normal vector;

[0127] S630: Find the parameters of the closest point on the two lines by solving an optimization problem that minimizes the distance between them. :

[0128] ;

[0129] Then the coordinates of the fault point in three-dimensional space for:

[0130] .

[0131] In one embodiment, S700 includes:

[0132] After completing the precise location of the three-dimensional fault point, the located three-dimensional fault point The core is to first determine the distribution of point clouds within a preset range. If the point cloud density within the preset range is higher than a preset threshold, then there are valid point clouds, and the area is directly rendered to highlight the fault clustering effect and achieve intuitive spatial display. If the point cloud density within the preset range is lower than the preset threshold or there are missing points, then an interpolation method based on nearest neighbor search is used to fill the missing areas to generate continuous and complete 3D point cloud visualization results.

[0133] Specifically, this method ensures that the fault point and its surrounding area are clearly visible in three-dimensional space, providing a reliable visualization basis for subsequent fault analysis, maintenance decisions, or equipment monitoring.

[0134] In summary, this invention fully combines the advantages of multi-view 3D reconstruction in terms of spatial accuracy with the efficiency of voxel projection matching in fault location, effectively solving the problem of accurate fault location and visualization of power equipment in limited spaces. Specifically, the method acquires images and 2D fault information through a camera and fault detection equipment, and generates a standardized equipment model using 3D reconstruction. Subsequently, through voxel segmentation and four-view projection, the 2D fault point is mapped to a 3D voxel surface, and image matching is performed using color and edge histograms to achieve point-to-point correspondence. Furthermore, the nearest points on both sides are found by extending the voxel centroid along the normal line to determine the central fault point in 3D space. Finally, cluster rendering or interpolation filling is performed based on the surrounding point cloud distribution to generate a complete and intuitive 3D visualization effect. This method effectively improves fault location accuracy and visualization completeness, providing reliable support for the maintenance and monitoring of power equipment.

[0135] In one embodiment, a three-dimensional fault location system for power equipment based on semantic voxels and projection mapping is also provided, comprising:

[0136] The data acquisition and 3D reconstruction module is used to deploy cameras and fault detection equipment in the limited space of power equipment to acquire multi-view images and two-dimensional fault point information, and generate a 3D model of the power equipment from the multi-view images through a 3D reconstruction algorithm.

[0137] The attitude adjustment module is used to adjust the attitude of the reconstructed 3D model using principal component analysis, so that the principal direction of the 3D model is aligned with the standard coordinate system.

[0138] The voxel projection module is used to divide the 3D model after posture adjustment into regular voxels, and to perform 2D projection on the side of the voxels to generate a projected image. The projected image is then matched with the original multi-view image to establish a spatial correspondence between the side of the 3D voxel and the original image.

[0139] The fault point matching module is used to extract two-dimensional fault coordinate points from the fault surface collected by the fault detection equipment, and match the fault points on both sides to form a matching pair.

[0140] The fault point mapping module is used to map the points matched by the two-dimensional fault points to the corresponding three-dimensional voxel surfaces based on spatial correspondence, so as to obtain the coordinates of the fault points on the three-dimensional surfaces.

[0141] The fault location module is used to calculate the extension line of the corresponding voxel center for a fault point on a three-dimensional surface, and determine the location of the fault point in three-dimensional space by solving the shortest distance between the extension lines.

[0142] The visualization module is used to visualize the surrounding area based on the determined 3D fault location, generating a 3D visualization result of the fault area.

[0143] For specific limitations regarding the three-dimensional fault location system for power equipment based on semantic voxels and projection mapping, please refer to the limitations of the three-dimensional fault location method for power equipment based on semantic voxels and projection mapping mentioned above, which will not be repeated here. Each module in the aforementioned three-dimensional fault location system for power equipment based on semantic voxels and projection mapping can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0144] In one embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a three-dimensional fault location method for power equipment based on semantic voxel and projection mapping.

[0145] In one embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps of a method for three-dimensional fault location of power equipment based on semantic voxel and projection mapping.

[0146] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0147] The foregoing has provided a detailed description of a three-dimensional fault location method and system for power equipment based on semantic voxel and projection mapping, as provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention, and the descriptions of the embodiments above are merely for the purpose of helping to understand the core ideas of this invention. It should be noted that those skilled in the art can make various improvements and modifications to this invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this invention.

Claims

1. A method for three-dimensional fault localization of power equipment based on semantic voxels and projection mapping, characterized in that, Includes the following steps: S100: Deploy cameras and fault detection equipment in the limited space of power equipment to acquire multi-view images and two-dimensional fault point information, and generate a three-dimensional model of power equipment from the multi-view images through a three-dimensional reconstruction algorithm. S200: Principal component analysis is used to adjust the attitude of the reconstructed 3D model so that the principal direction of the 3D model is aligned with the standard coordinate system; S300: Divide the 3D model after attitude adjustment into regular voxels, perform 2D projection on the side of the voxels to generate a projected image, match the projected image with the original multi-view image, and establish the spatial correspondence between the side of the 3D voxel and the original image. S400: Extract two-dimensional fault coordinate points from the fault surface collected by the fault detection equipment, and match the fault points on both sides to form a matching pair; S500: Based on spatial correspondence, the points matched by the two-dimensional fault points are mapped to the corresponding three-dimensional voxel surfaces to obtain the coordinates of the fault points on the three-dimensional surfaces; S600: For a fault point on a three-dimensional surface, calculate the extension line of the corresponding voxel center, and determine the location of the fault point in three-dimensional space by solving the shortest distance between the extension lines. S700: Based on the determined three-dimensional fault location, the surrounding area is visualized to generate a three-dimensional visualization result of the fault area.

2. The method according to claim 1, characterized in that, S200 includes: S210: Calculate the centroid of the 3D point cloud and center the point cloud. Calculate the average position of all points in the 3D point cloud to obtain the centroid of the point cloud. ； in Let N represent the coordinates of the i-th point, and let N represent the total number of points. The centroid of the point cloud; S220: The point cloud is translated to the vicinity of the centroid to eliminate the global translation component, resulting in a decentralized point cloud matrix. ； in The original point cloud matrix, The decentralized point cloud matrix, It is a column vector whose elements are all 1s. To transpose the centroid of the point cloud; S230: Calculate the decentralized point cloud Covariance matrix: ； in This is the transposed decentralized point cloud matrix. The covariance matrix is ​​a symmetric matrix. Its diagonal elements describe the variance of the point cloud in the x, y, and z directions, i.e., the distribution width of the points in each direction. The off-diagonal elements describe the covariance between different directions, i.e., the tilt relationship or coupling degree of the point cloud in each direction in space. S240: Perform eigenvalue decomposition on the covariance matrix: ； in column vectors This represents the three mutually perpendicular principal directions of the point cloud in space. For a diagonal eigenvalue matrix, each Sorted by the magnitude of the variance along the corresponding eigenvector direction. ; S250: Main direction based on point cloud For decentralized point clouds Rotate the points to obtain a rotated point cloud matrix aligned with the standard coordinate system. ； in This is the rotated point cloud matrix, whose principal direction is now aligned with the standard coordinate system.

3. The method according to claim 2, characterized in that, The S300 includes: S310: The 3D model of the power equipment is adjusted according to the preset voxel dimensions. Mesh generation is performed on a point in three-dimensional space. The corresponding voxel index is: ； in The minimum coordinates of the model boundary are i, j, and k, which are voxel indices and are all rounded down. voxel center coordinates for: ； S320: Project each voxel center onto the front, back, left, and right sides of the power equipment to generate a corresponding set of two-dimensional projected images. The projection process involves calculating the side normal vector and in-plane direction, rotating and centering the voxel point cloud to align the side with the two-dimensional plane. Then, the camera position and viewpoint are set, and the rotated voxel point cloud is rendered to obtain side screenshots. S330: Matching the set of 2D projected images with the original multi-view images by combining color histograms and edge histograms: First, statistically analyze the color distribution histograms of the 2D projected images and the original multi-view images, and calculate the color similarity between the two; then, extract the edges of the 2D projected images and the original multi-view images, statistically analyze the histograms of edge intensity or direction, and calculate the structural similarity; finally, weighted summation of color similarity and structural similarity yields the final matching score; each 2D projected image is compared with all original multi-view images, and the image with the highest comprehensive similarity is selected as the matching result, thereby achieving the best correspondence between the 3D voxel side view and the original multi-view images.

4. The method according to claim 3, characterized in that, The S400 includes: S410: From the fault surfaces collected by the fault detection equipment on both sides of the power equipment, extract the vertices and two-dimensional fault coordinate points of each fault surface. , where L and R represent the adjacent two sides respectively; S420: To ensure that the points on both sides originate from the same fault source, the fault points on each side are arranged in ascending order according to their vertical coordinates, resulting in separate sequences. , ,…and , Then, matching pairs are formed one by one according to the serial number. .

5. The method according to claim 4, characterized in that, The S500 includes: S510: Based on spatial correspondence, determine the three-dimensional voxel surface corresponding to the two-dimensional fault surface; S520: Establish a mapping relationship between the four vertices of each faulty surface and the four vertices of the corresponding three-dimensional voxel surface; S530: Based on the mapping relationship, the two-dimensional fault point is mapped to the three-dimensional voxel surface through the mapping function to obtain the corresponding three-dimensional coordinates: assuming the vertex of the two-dimensional fault surface is... The corresponding vertices of the three-dimensional voxel face are Then it is obtained through affine mapping: ； in The coordinates of the two-dimensional fault point For the corresponding three-dimensional surface coordinates, The function is for mapping.

6. The method according to claim 5, characterized in that, The S600 includes: S610: Assume the coordinates of the fault point after mapping from a two-dimensional fault point to a three-dimensional space are... Calculate the corresponding voxel grid index. : ； via voxel grid indexing Retrieve the previously saved voxel center coordinates; S620: The coordinates of the centers of the fault voxels on the two faces are denoted as follows: and Along the normal direction of each voxel surface and Extend the line, and it can be represented as: , ； in and These represent the extended distances along the direction of the central normal vector; S630: Find the parameters of the closest point on the two lines by solving an optimization problem that minimizes the distance between them. : ； Then the coordinates of the fault point in three-dimensional space for: 。 7. The method according to claim 6, characterized in that, The S700 includes: After completing the precise location of the 3D fault point, the located 3D fault point The core is to first determine the distribution of point clouds within a preset range. If the point cloud density within the preset range is higher than a preset threshold, then there are valid point clouds, and the area is directly rendered to highlight the fault clustering effect and achieve intuitive spatial display. If the point cloud density within the preset range is lower than the preset threshold or there are missing points, then an interpolation method based on nearest neighbor search is used to fill the missing areas to generate continuous and complete 3D point cloud visualization results.

8. A three-dimensional fault location system for power equipment based on semantic voxels and projection mapping, characterized in that, include: The data acquisition and 3D reconstruction module is used to deploy cameras and fault detection equipment in the limited space of power equipment to acquire multi-view images and two-dimensional fault point information, and generate a 3D model of the power equipment from the multi-view images through a 3D reconstruction algorithm. The attitude adjustment module is used to adjust the attitude of the reconstructed 3D model using principal component analysis, so that the principal direction of the 3D model is aligned with the standard coordinate system. The voxel projection module is used to divide the 3D model after posture adjustment into regular voxels, and to perform 2D projection on the side of the voxels to generate a projected image. The projected image is then matched with the original multi-view image to establish a spatial correspondence between the side of the 3D voxel and the original image. The fault point matching module is used to extract two-dimensional fault coordinate points from the fault surface collected by the fault detection equipment, and match the fault points on both sides to form a matching pair. The fault point mapping module is used to map the points matched by the two-dimensional fault points to the corresponding three-dimensional voxel surfaces based on spatial correspondence, so as to obtain the coordinates of the fault points on the three-dimensional surfaces. The fault location module is used to calculate the extension line of the corresponding voxel center for a fault point on a three-dimensional surface, and determine the location of the fault point in three-dimensional space by solving the shortest distance between the extension lines. The visualization module is used to visualize the surrounding area based on the determined 3D fault location, generating a 3D visualization result of the fault area.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images