A substation knife switch three-dimensional coordinate calculation method, system, device and medium

Abstract

This invention relates to a method, system, device, and medium for calculating the three-dimensional coordinates of a substation disconnector. The method includes: acquiring visible light images and laser point cloud data of a substation disconnector under different angles and operating conditions; inputting these into the image branch and point cloud branch of a scene coordinate regression network, respectively; extracting multi-scale features, global semantic features, and local geometric features; fusing the global semantic features and local geometric features to obtain fused point cloud features; fusing the multi-scale features and fused point cloud features to obtain fused features; acquiring the shape, size, and component relationships of the disconnector in real time, and constructing real-time geometric constraints based on these; and calculating the three-dimensional coordinates of the disconnector using a three-level cascaded regression network based on the fused features and real-time geometric constraints. Compared with existing technologies, this invention avoids mismatch problems caused by neglecting spatial topological relationships and improves the accuracy of the three-dimensional coordinate calculation of the disconnector.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring of power equipment and computer vision, and in particular to a method, system, equipment and medium for calculating the three-dimensional coordinates of a substation disconnector. Background Technology

[0002] With the continuous advancement of smart grid construction, substation disconnectors, as one of the key pieces of equipment in substations, are responsible for connecting and disconnecting circuits. In the actual operation of the power system, the normal working condition of the disconnectors directly affects the reliability and security of the power system. Therefore, real-time monitoring and precise location of the status of substation disconnectors, timely detection of potential faults, and taking corresponding measures are crucial links in ensuring the safe and stable operation of the power system.

[0003] Traditional methods for calculating the three-dimensional coordinates of substation disconnectors largely rely on manual measurement, lidar measurement, or two-dimensional image analysis based on simple visual algorithms. Regarding manual measurement, traditional methods typically require technicians to mark and measure the disconnectors on-site. This approach is not only inefficient—manual marking of each disconnector usually takes two hours or more—but also carries significant safety risks. In practice, technicians must perform measurements under high-voltage conditions, making them vulnerable to electric shock, falls, and other safety threats. Furthermore, the accuracy of manual measurements is affected by the technician's skill level and experience, making it difficult to guarantee the consistency and reliability of the results. While lidar point cloud registration methods offer high-precision three-dimensional spatial information, they are costly, with high-precision lidar units typically being expensive, posing a significant economic burden for large-scale applications. Additionally, lidar installation and maintenance costs are high, requiring specialized technicians for operation and management. Two-dimensional image analysis methods based on RGB-D cameras are significantly affected by factors such as viewing angle and lighting. In practical applications, strong reflections can cause depth information to become invalid, thus affecting the accurate monitoring of disconnector status. To address the aforementioned issues, Chinese patent application CN114821042A provides an R-FCN disconnector detection method that combines local and global features. This method uses the deep learning network model R-FCN to detect the status of disconnectors in substations. By embedding network camera control software into the substation auxiliary monitoring system, it enables the acquisition of disconnector images from multiple positions and angles under different outdoor weather conditions and backgrounds, constructing diverse disconnector data. By paralleling the R-FCN output prediction network with a global feature prediction module, it compensates for the insufficient receptive field caused by the original network predicting disconnectors based solely on local features. Although this method can improve the detection accuracy of partially obscured disconnectors and reduce the false negative and false positive rates for disconnectors in complex backgrounds, it relies solely on disconnector images in different scenarios to supplement the 2D state detection of disconnectors, without considering the spatial topological relationships between disconnector components. This may lead to mismatches in scenarios with multiple disconnectors in parallel, affecting the detection accuracy and applicability of the method.

[0004] Therefore, providing a method for calculating the coordinate information of disconnectors that considers the spatial topological relationship of the components in a substation disconnector is a technical problem that needs to be solved. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a system, equipment, and medium for calculating the three-dimensional coordinates of substation disconnectors. By using a cascaded scene coordinate regression network (C-SCN) combined with multi-size feature extraction and fusion and geometric constraint optimization, it achieves end-to-end high-precision prediction of the three-dimensional coordinates of substation disconnectors, overcoming the problems of low accuracy and poor adaptability in the existing technology, and providing reliable technical support for the intelligent operation and maintenance of substation disconnectors.

[0006] The objective of this invention can be achieved through the following technical solutions: According to a first aspect of the present invention, a method for calculating the three-dimensional coordinates of a substation disconnector based on a scene coordinate regression network is provided, comprising: Visible light images and laser point cloud data of substation disconnectors were collected from different angles and under different operating conditions. These were then input into the image branch and point cloud branch of the scene coordinate regression network to extract multi-scale features, global semantic features, and local geometric features. The global semantic features and local geometric features are fused to obtain fused point cloud features, and the multi-scale features and fused point cloud features are fused to obtain fused features. The shape, size, and relationships of the disconnector components are acquired in real time, and real-time geometric constraints are constructed based on these. Based on the aforementioned fusion features and real-time geometric constraints, the scene coordinate regression network is used to perform a three-level cascaded regression to calculate the three-dimensional coordinates of the disconnector.

[0007] As a preferred technical solution, the image branch includes a parallel multi-resolution sub-network, and the method for extracting the multi-scale features is as follows: The visible light image is generated using initial subnetting. Initial scale features are generated based on the initial features using the next subnet processing. Scale-based characteristics; Based on the aforementioned fundamental features, it is generated through downsampling. The first intermediate feature of the scale and The second intermediate feature of the scale; The basic features, the first intermediate features, and the second intermediate features are assigned to parallel subnets, and the corresponding scale features are extracted respectively. All scale features are connected across scales to generate multi-scale features.

[0008] As a preferred technical solution, the point cloud branch includes a Voxel pathway network, and the method for extracting the global semantic features is as follows: The voxel resolution is adaptively adjusted according to the size of the switch, and the laser point cloud data is divided into a regular 3D voxel grid based on the voxel resolution. Based on the aforementioned 3D voxel mesh, multiple 3D sparse convolutions are performed for progressive downsampling to extract global semantic features.

[0009] As a preferred technical solution, the point cloud branch further includes a Point path network, and the method for extracting the local geometric features is as follows: Multiple representative point cloud data points were selected from the laser point cloud data as key points; Obtain the location information of each key point, obtain all neighboring points of the key point within a preset range based on the location information, fit the local plane of the key point's neighborhood using the least squares method based on all points within the neighborhood, and extract the geometric features, normals, and normal directions of the corresponding key points. The angle difference between normals is calculated based on the normals and their directions. The angle difference includes the angle between normals and the angle between the line connecting key points and the corresponding normal. A histogram is constructed based on the angle difference and geometric features, and the FPFH feature vectors of the corresponding points, i.e., local geometric features, are generated based on the histogram.

[0010] As a preferred technical solution, the method for obtaining the fused point cloud features is as follows: A 2D BEV feature map is generated based on the global semantic features, and the 2D BEV feature map is processed to generate multiple 3D candidate boxes. For each of the 3D candidate boxes, a set of grid points are uniformly sampled within it. For each grid point, the Voxel path feature and Point path feature corresponding to the grid point are extracted from the global semantic features and local geometric features. The Voxel pathway features and Point pathway features are fused to generate a primary fused feature. The initial fusion features are flattened, and the 3D candidate boxes are optimized using a multilayer perception mechanism. Based on the optimized 3D candidate boxes, fused point cloud features are generated.

[0011] As a preferred technical solution, the method for calculating the three-dimensional coordinates of the disconnector is as follows: Coarse positioning of the disconnector: Based on the fusion features, the 3D coordinates of the center of the disconnector base are calculated, and the real-time geometric constraints are introduced as regression constraints to initially lock the overall three-dimensional position of the disconnector. Fine positioning of the knife switch: Based on the coarse positioning of the knife switch and the real-time geometric constraints, the 6DoF pose of the knife switch rotation joint is calculated, and the overall three-dimensional position of the knife switch is refined based on the 6DoF pose of the knife switch rotation joint. Physical constraint optimization: Obtain the substation topology map, construct an adjacency matrix based on the topology map, model the spatial relationship between adjacent disconnectors based on the adjacency matrix, and generate a three-dimensional coordinate correction amount that satisfies the minimum safe distance constraint between disconnectors by combining the real-time geometric constraints. Optimize the overall three-dimensional position of the disconnectors after refinement based on the three-dimensional coordinate correction amount to obtain the final three-dimensional coordinates of the disconnectors.

[0012] As a preferred technical solution, the method further includes: Geometric constraints are constructed by acquiring the shape data, size data, and relative positional relationships between the components of the knife gate. A loss function is then constructed based on these geometric constraints. Finally, a scene coordinate regression network is trained based on the loss function, wherein the loss function is: , in, Let represent the basic loss function, and have . , This indicates the total number of disconnectors. This represents the calculated three-dimensional coordinates of the i-th point. This represents the true three-dimensional coordinates of the i-th point; Indicates the geometric constraint weight coefficient; Represents the weights of the distance constraint; This indicates the distance constraint between the midpoints of the disconnectors, and , This represents the calculated three-dimensional coordinates of the j-th point. Represents point pairs The preset target distance between them; This indicates the parallel constraint weighting coefficient of the disconnector component; This indicates that the knife switch components are parallelly constrained, and , and Both represent the direction vectors of the parallel knife switch components that need to be constrained; This represents the orthogonal constraint weight coefficient of the disconnector component; This indicates orthogonal constraints on the knife switch components, and , Both represent the direction vectors of the orthogonal knife switch components that need to be constrained; Indicates the weight coefficient of the coplanar constraint term; Denotes the coplanar constraint term, and , Indicates a point in the disconnect switch. Denotes the set of coplanar points. This represents the reference plane corresponding to the point set S.

[0013] According to a second aspect of the present invention, a three-dimensional coordinate calculation system for substation disconnectors based on a scene coordinate regression network is provided, the system comprising: The data acquisition module is used to acquire visible light images and laser point cloud data of the substation disconnectors under different angles and operating conditions, and to obtain the shape, size and component relationship of the disconnectors in real time. The feature extraction module receives visible light images and laser point cloud data acquired by the data acquisition module and uses the image branch and point cloud branch of the scene coordinate regression network to extract multi-scale features, global semantic features and local geometric features. The feature fusion module fuses the global semantic features and local geometric features to obtain fused point cloud features, and fuses the multi-scale features and fused point cloud features to obtain fused features. The constraint construction module constructs real-time geometric constraints based on the shape, size, and relationships between the knife switch components. The three-dimensional coordinate calculation module, based on the aforementioned fusion features and real-time geometric constraints, utilizes the aforementioned scene coordinate regression network to perform a three-level cascaded regression to calculate the three-dimensional coordinates of the disconnector.

[0014] According to a third aspect of the present invention, an electronic device for calculating the three-dimensional coordinates of a substation disconnector is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described thereon.

[0015] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1) This invention calculates the three-dimensional coordinates of the disconnector through a three-level cascaded regression mechanism. First, coarse positioning of the disconnector is performed by calculating the 3D coordinates of the disconnector base center based on fused features and introducing real-time geometric constraints as regression constraints to initially lock the overall three-dimensional position of the disconnector. Then, fine positioning of the disconnector is performed by calculating the 6DoF pose of the disconnector's rotating joints in conjunction with real-time geometric constraints to further refine the position. Finally, through physical constraint optimization, the adjacency matrix is ​​constructed using the substation topology map to model spatial relationships and generate three-dimensional coordinate correction quantities that meet the minimum safety distance. Through the progressively refined three-dimensional coordinate calculation process, the overall positioning and component positioning of the disconnector are realized, ensuring that the spatial position and topological connection of each disconnector are accurately captured in multi-disconnector parallel scenarios, avoiding mismatch problems caused by not considering spatial topological relationships, and improving the accuracy of the three-dimensional coordinate calculation of the disconnector.

[0017] 2) In this invention, the voxel resolution is adaptively adjusted according to the size of the switch. For critical areas of the switch, the voxel resolution is adjusted to a small value to ensure that enough details can be retained, while for unimportant areas such as the background where the switch is located, the voxel resolution is adjusted to a large value to improve computational efficiency. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0020] Example 1 To address the problems of existing technologies, this invention provides a method for calculating the three-dimensional coordinates of substation disconnectors based on a scene coordinate regression network. This method is suitable for accurate calculation of the three-dimensional spatial position and status monitoring of substation disconnector equipment. High-precision three-dimensional coordinate prediction is achieved through multi-modal data fusion and geometric constraint optimization. Considering the complexity of substation scenes, a lightweight HRNet-18 is used to extract multi-scale features from images, and a Voxel-Point dual-path network is designed to process point cloud data. Combining FPFH local geometric features and voxel global semantic features, end-to-end coordinate prediction is achieved through a three-level cascaded regression, including base localization, joint pose adjustment, topology constraint optimization, and geometric constraints. The process is as follows: Figure 1 As shown, it includes: S1. Collect visible light images and laser point cloud data of substation disconnectors under different angles and operating conditions, and input them into the image branch and point cloud branch of the scene coordinate regression network respectively to extract multi-scale features, global semantic features and local geometric features.

[0021] S11. Using visible light cameras, lidar and other equipment installed in the substation, collect multimodal data of substation disconnectors under different angles and operating conditions, including visible light images and laser point cloud data, wherein the size of the visible light image is 1024×1024 pixels.

[0022] S12. Extract multi-scale features.

[0023] For the image branch, which targets the slender mechanical structure (generally with an aspect ratio > 10:1) unique to substation disconnectors, traditional downsampling networks lead to the loss of effective features, with experimental results showing a feature retention rate of < 30% at the 1 / 16 scale. To address these technical issues, this invention employs a lightweight HRNet-18 as its basic architecture, including a parallel multi-resolution subnet. This achieves in-situ preservation of multi-scale features with only a 15% increase in computational cost, specifically maintaining the continuity of the disconnector linkage through a 1 / 4 scale feature map.

[0024] S121. Visible light images are generated using initial subnetting. Initial scale features are generated based on the initial features using the next subnet processing. The scale-based features are defined by a subnet consisting of a stem layer comprising 3×3 convolutional units, BN units, and ReLU.

[0025] S122, Generated by downsampling based on basic features The first intermediate feature of the scale and The second intermediate feature of the scale.

[0026] S123. Assign the basic features, the first intermediate features, and the second intermediate features to the parallel subnets, and extract the corresponding scale features respectively.

[0027] S124. Connect all scale features across scales to generate multi-scale features.

[0028] Through the above processing, the key size information of the original input image can be preserved, and the feature response to small targets such as knife switch contacts can be improved.

[0029] S13. Extract global semantic features and local geometric features.

[0030] In this invention, the laser point cloud branch adopts a Voxel-Point dual-path network. The Voxel path network is responsible for capturing the overall semantic information of the point cloud data and executing steps S131 to S132 to first gain a macroscopic understanding of the entire disconnector device and identify the approximate categories of different areas. The Point path network focuses on the local geometric details of each point and executes steps S133 to S136 to accurately capture the geometric features such as the outline and shape of the disconnector arm. These two paths are independent of each other but work together.

[0031] S131. Adaptively adjust the voxel resolution according to the size of the knife gate. Divide the laser point cloud data into a regular 3D voxel mesh based on the voxel resolution. Specifically, adaptively adjust the voxel resolution according to the size of the knife gate, and set the voxel resolution to 1cm for key areas such as the base. 3 ~5cm 3For less important areas such as the background, the voxel resolution is set to a larger value, such as 10cm. 3 The above distinction ensures that the details of the disconnect switch can be completely preserved, while also improving computational efficiency.

[0032] S132. Based on a 3D voxel grid, perform multiple 3D sparse convolutions to progressively downsample and extract global semantic features.

[0033] The processing in steps S131-S132 provides contextual information and structured feature maps, which facilitates the generation of candidate regions.

[0034] S133. Select multiple representative point cloud data from the laser point cloud data as key points. The selection method can be to sample the farthest point at FPS and use these key points as the skeleton of the scene where the knife switch is located.

[0035] To further refine the key points using a small PointNet or MLP, perform the following steps: S134. Obtain the location information of each key point, obtain all neighboring points of the key point within a preset range based on the location information, fit the local plane of the key point's neighborhood using the least squares method based on all points within the neighborhood, and extract the geometric features, normals, and normal directions of the corresponding key points to reflect the local orientation of the point cloud surface.

[0036] S135. Calculate the angle difference between normals based on the normal and the normal direction. The angle difference includes the angle between normals and the angle between the line connecting key points and the corresponding normal.

[0037] S136. Construct a histogram based on angle difference and geometric features, and generate FPFH feature vectors for corresponding points based on the histogram, i.e., local geometric features.

[0038] Specifically, in step S136, the angle difference and geometric features are statistically quantified and placed into different intervals of the histogram to form a comprehensive description.

[0039] S2. The global semantic features and local geometric features are fused to obtain the fused point cloud features, and the multi-scale features and the fused point cloud features are fused to obtain the fused features.

[0040] Methods for obtaining fused point cloud features include: S211. Generate a 2D BEV feature map based on global semantic features, and process the 2D BEV feature map to generate multiple 3D candidate boxes.

[0041] S212. For each 3D candidate box, a set of grid points are uniformly sampled within it. For each grid point, the Voxel path feature and Point path feature corresponding to that grid point are extracted from the global semantic features and local geometric features.

[0042] Specifically, for the Point pathway network, nearby key points are found and their local geometric features are aggregated; for the Voxel pathway network, the global semantic features corresponding to the grid point are found.

[0043] S213. Perform feature fusion on the Voxel pathway features and Point pathway features, such as splicing and weighted summation, to generate primary fused features.

[0044] S214. Flatten the primary fusion features, optimize the 3D candidate boxes using a multi-layer perception mechanism, and generate fusion point cloud features based on the optimized 3D candidate boxes.

[0045] S3. Real-time acquisition of the shape, size, and relationships of the disconnector components, and construction of real-time geometric constraints based on these.

[0046] To further improve the accuracy of the three-dimensional coordinate calculation of the disconnector, geometric constraints are introduced to optimize the network prediction results. Based on the structural characteristics and spatial geometric relationships of the disconnector in the substation, a corresponding geometric constraint model is established. For example, the geometric constraint equations between the three-dimensional coordinates are determined according to the shape, size and relative positional relationships between the components of the disconnector.

[0047] Furthermore, when training the scene coordinate regression network, geometric constraints were introduced to construct the loss function, enabling the network to learn features that conform to reality. The steps are as follows: obtain the shape data, size data, and relative positional relationships between the components of the knife gate to construct geometric constraints; construct a loss function based on the geometric constraints; and train the scene coordinate regression network based on the loss function. The loss function is: , in, Let represent the basic loss function, and have . , This indicates the total number of disconnectors. This represents the calculated three-dimensional coordinates of the i-th point. This represents the true three-dimensional coordinates of the i-th point; Indicates the geometric constraint weight coefficient; Represents the weights of the distance constraint; This indicates the distance constraint between the midpoints of the disconnectors, and , This represents the calculated three-dimensional coordinates of the j-th point. Represents point pairs The preset target distance between them; This indicates the parallel constraint weighting coefficient of the disconnector component; This indicates that the knife switch components are parallelly constrained, and , and Both represent the direction vectors of the parallel knife switch components that need to be constrained; This represents the orthogonal constraint weight coefficient of the disconnector component; This indicates orthogonal constraints on the knife switch components, and , Both represent the direction vectors of the orthogonal knife switch components that need to be constrained; Indicates the weight coefficient of the coplanar constraint term; Denotes the coplanar constraint term, and , Indicates a point in the disconnect switch. Denotes the set of coplanar points. This represents the reference plane corresponding to the point set S.

[0048] S4. Based on fusion features and real-time geometric constraints, a three-level cascaded regression is performed using a scene coordinate regression network to calculate the three-dimensional coordinates of the disconnector.

[0049] S41. The coarse positioning head in the scene coordinate regression network performs coarse positioning of the knife switch. The process is as follows: calculate the 3D coordinates of the knife switch base center based on the fusion features, introduce real-time geometric constraints as regression constraints, and initially lock the overall three-dimensional position of the knife switch.

[0050] S42. The coarse positioning head in the scene coordinate regression network performs fine positioning of the knife gate. The process is as follows: calculate the 6DoF pose of the knife gate rotation joint based on the result of the coarse positioning of the knife gate and the real-time geometric constraints, and refine the overall three-dimensional position of the knife gate based on the 6DoF pose of the knife gate rotation joint.

[0051] S43. Perform physical constraint optimization on the coarse positioning head in the scene coordinate regression network. The process is as follows: obtain the substation topology map, construct the adjacency matrix based on the topology map, model the spatial relationship between adjacent disconnectors based on the adjacency matrix, and generate a three-dimensional coordinate correction amount that satisfies the minimum safe distance constraint between disconnectors by combining real-time geometric constraints. Optimize the overall three-dimensional position of the disconnector after refinement based on the three-dimensional coordinate correction amount to obtain the final three-dimensional coordinates of the disconnector.

[0052] Through the above steps, end-to-end prediction from raw data input to three-dimensional coordinate output is achieved, avoiding the error accumulation problem caused by multi-step processing in traditional methods, and improving computational efficiency and accuracy.

[0053] Example 2 The above is an introduction to the method embodiments. The following is a system embodiment to further illustrate the present invention: a three-dimensional coordinate calculation system for substation disconnectors based on a scene coordinate regression network. The system includes: The data acquisition module is used to collect visible light images and laser point cloud data of substation disconnectors under different angles and operating conditions, and to obtain the shape, size and component relationships of the disconnectors in real time.

[0054] The feature extraction module receives visible light images and laser point cloud data acquired by the data acquisition module and uses the image branch and point cloud branch of the scene coordinate regression network to extract multi-scale features, global semantic features, and local geometric features.

[0055] The feature fusion module fuses global semantic features and local geometric features to obtain fused point cloud features, and fuses multi-scale features and fused point cloud features to obtain fused features.

[0056] The constraint construction module constructs real-time geometric constraints based on the shape, size, and relationships of the disconnector components. The 3D coordinate calculation module, based on fused features and real-time geometric constraints, uses a scene coordinate regression network to perform a three-level cascaded regression to calculate the 3D coordinates of the disconnector.

[0057] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0058] This invention also provides an electronic device for calculating the three-dimensional coordinates of substation disconnectors, including a central processing unit (CPU), which can execute various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM can also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0059] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0060] The processing unit executes the various methods and processes described above, such as methods S1 to S4. For example, in some embodiments, methods S1 to S4 may be implemented as computer software programs tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of methods S1 to S4 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S4 by any other suitable means (e.g., by means of firmware).

[0061] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0062] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0063] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0064] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for calculating the three-dimensional coordinates of substation disconnectors based on scene coordinate regression networks, characterized in that, include: Visible light images and laser point cloud data of substation disconnectors were collected from different angles and under different operating conditions. These were then input into the image branch and point cloud branch of the scene coordinate regression network to extract multi-scale features, global semantic features, and local geometric features. The global semantic features and local geometric features are fused to obtain fused point cloud features, and the multi-scale features and fused point cloud features are fused to obtain fused features. The shape, size, and relationships of the disconnector components are acquired in real time, and real-time geometric constraints are constructed based on these. Based on the aforementioned fusion features and real-time geometric constraints, the scene coordinate regression network is used to perform a three-level cascaded regression to calculate the three-dimensional coordinates of the disconnector.

2. The method for calculating the three-dimensional coordinates of substation disconnectors based on a scene coordinate regression network according to claim 1, characterized in that, The image branch includes parallel multi-resolution subnetworks, and the method for extracting the multi-scale features is as follows: The visible light image is generated using initial subnetting. Initial scale features are generated based on the initial features using the next subnet processing. Scale-based characteristics; Based on the aforementioned fundamental features, it is generated through downsampling. The first intermediate feature of the scale and The second intermediate feature of the scale; The basic features, the first intermediate features, and the second intermediate features are assigned to parallel subnets, and the corresponding scale features are extracted respectively. All scale features are connected across scales to generate multi-scale features.

3. The method for calculating the three-dimensional coordinates of substation disconnectors based on a scene coordinate regression network according to claim 1, characterized in that, The point cloud branch includes the Voxel pathway network, and the method for extracting the global semantic features is as follows: The voxel resolution is adaptively adjusted according to the size of the switch, and the laser point cloud data is divided into a regular 3D voxel grid based on the voxel resolution. Based on the aforementioned 3D voxel mesh, multiple 3D sparse convolutions are performed for progressive downsampling to extract global semantic features.

4. The method for calculating the three-dimensional coordinates of substation disconnectors based on a scene coordinate regression network according to claim 3, characterized in that, The point cloud branch also includes a Point Path Network, and the method for extracting the local geometric features is as follows: Multiple representative point cloud data points were selected from the laser point cloud data as key points; Obtain the location information of each key point, obtain all neighboring points of the key point within a preset range based on the location information, fit the local plane of the key point's neighborhood using the least squares method based on all points within the neighborhood, and extract the geometric features, normals, and normal directions of the corresponding key points. The angle difference between normals is calculated based on the normals and their directions. The angle difference includes the angle between normals and the angle between the line connecting key points and the corresponding normal. A histogram is constructed based on the angle difference and geometric features, and the FPFH feature vectors of the corresponding points, i.e., local geometric features, are generated based on the histogram.

5. The method for calculating the three-dimensional coordinates of substation disconnectors based on a scene coordinate regression network according to claim 4, characterized in that, The method for obtaining the fused point cloud features is as follows: A 2D BEV feature map is generated based on the global semantic features, and the 2D BEV feature map is processed to generate multiple 3D candidate boxes. For each of the 3D candidate boxes, a set of grid points are uniformly sampled within it. For each grid point, the Voxel path feature and Point path feature corresponding to the grid point are extracted from the global semantic features and local geometric features. The Voxel pathway features and Point pathway features are fused to generate a primary fused feature. The initial fusion features are flattened, and the 3D candidate boxes are optimized using a multilayer perception mechanism. Based on the optimized 3D candidate boxes, fused point cloud features are generated.

6. The method for calculating the three-dimensional coordinates of substation disconnectors based on a scene coordinate regression network according to claim 1, characterized in that, The method for calculating the three-dimensional coordinates of the disconnector is as follows: Coarse positioning of the disconnector: Based on the fusion features, the 3D coordinates of the center of the disconnector base are calculated, and the real-time geometric constraints are introduced as regression constraints to initially lock the overall three-dimensional position of the disconnector. Fine positioning of the knife switch: Based on the coarse positioning of the knife switch and the real-time geometric constraints, the 6DoF pose of the knife switch rotation joint is calculated, and the overall three-dimensional position of the knife switch is refined based on the 6DoF pose of the knife switch rotation joint. Physical constraint optimization: Obtain the substation topology map, construct an adjacency matrix based on the topology map, model the spatial relationship between adjacent disconnectors based on the adjacency matrix, and generate a three-dimensional coordinate correction amount that satisfies the minimum safe distance constraint between disconnectors by combining the real-time geometric constraints. Optimize the overall three-dimensional position of the disconnectors after refinement based on the three-dimensional coordinate correction amount to obtain the final three-dimensional coordinates of the disconnectors.

7. The method for calculating the three-dimensional coordinates of substation disconnectors based on a scene coordinate regression network according to claim 1, characterized in that, The method further includes: Geometric constraints are constructed by acquiring the shape data, size data, and relative positional relationships between the components of the knife gate. A loss function is then constructed based on these geometric constraints. Finally, a scene coordinate regression network is trained based on the loss function, wherein the loss function is: ， in, Let represent the basic loss function, and have . , This indicates the total number of disconnectors. This represents the calculated three-dimensional coordinates of the i-th point. This represents the true three-dimensional coordinates of the i-th point; Indicates the geometric constraint weight coefficient; Represents the weights of the distance constraint; This indicates the distance constraint between the midpoints of the disconnectors, and , This represents the calculated three-dimensional coordinates of the j-th point. Represents point pairs The preset target distance between them; This indicates the parallel constraint weighting coefficient of the disconnector component; This indicates that the knife switch components are parallelly constrained, and , and Both represent the direction vectors of the parallel knife switch components that need to be constrained; This represents the orthogonal constraint weight coefficient of the disconnector component; This indicates orthogonal constraints on the knife switch components, and , Both represent the direction vectors of the orthogonal knife switch components that need to be constrained; Indicates the weight coefficient of the coplanar constraint term; Denotes the coplanar constraint term, and , Indicates a point in the disconnect switch. Denotes the set of coplanar points. This represents the reference plane corresponding to the point set S.

8. A three-dimensional coordinate calculation system for substation disconnectors based on a scene coordinate regression network, characterized in that, The system includes: The data acquisition module is used to acquire visible light images and laser point cloud data of the substation disconnectors under different angles and operating conditions, and to obtain the shape, size and component relationship of the disconnectors in real time. The feature extraction module receives visible light images and laser point cloud data acquired by the data acquisition module and uses the image branch and point cloud branch of the scene coordinate regression network to extract multi-scale features, global semantic features and local geometric features. The feature fusion module fuses the global semantic features and local geometric features to obtain fused point cloud features, and fuses the multi-scale features and fused point cloud features to obtain fused features. The constraint construction module constructs real-time geometric constraints based on the shape, size, and relationships between the knife switch components. The three-dimensional coordinate calculation module, based on the aforementioned fusion features and real-time geometric constraints, utilizes the aforementioned scene coordinate regression network to perform a three-level cascaded regression to calculate the three-dimensional coordinates of the disconnector.

9. An electronic device for calculating the three-dimensional coordinates of a substation disconnector, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

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

Images