Method, medium and system for updating three-dimensional model of power equipment

Abstract

The invention discloses a power equipment three-dimensional model updating method, medium and system, and the method comprises the steps: obtaining scene information after the power equipment is updated, and the scene information comprises point cloud data and image data; key feature points are extracted from the image data, and the equipment type is identified based on the point cloud data; obtaining pose data and positioning data when the scanning device collects the scene information; determining equipment space coordinates according to the scene information, the pose data and the positioning data; obtaining an old three-dimensional model before the power equipment is updated according to the equipment type; performing alignment calculation on old point cloud data in the old three-dimensional model and the point cloud data according to the key feature points and the equipment space coordinates to obtain a displacement vector; and determining a change area of the old three-dimensional model according to the displacement vector, and updating the change area according to the target point cloud data subjected to displacement vector alignment processing to obtain an updated new three-dimensional model. According to the method, the updating efficiency of the three-dimensional model of the power equipment can be improved, and the data consistency is ensured.

Description

Technical Field

[0001] This invention relates to the field of power equipment model updating, and in particular to a method, medium, and system for updating a three-dimensional model of power equipment. Background Technology

[0002] The updating of existing power grid real-scene scanning models mainly relies on two methods: full-scale scanning reconstruction and manual intervention. Full-scale scanning reconstruction uses drones, LiDAR, or ground robots to rescan the entire area to generate a complete 3D model. While this method offers high accuracy, it is time-consuming (usually taking several hours to several days), and repeated scanning leads to data redundancy, making it difficult to meet real-time requirements. This is especially true in scenarios with frequent equipment replacement or temporary construction, where the digital twin model is prone to decoupling from the physical equipment status. Manual intervention updates rely on maintenance personnel manually marking equipment locations and adjusting the model, which is inefficient, error-prone, and highly dependent on professional personnel, with labor costs accounting for up to 40%. Furthermore, traditional image matching algorithms are sensitive to environmental interference such as changes in lighting and occlusion, resulting in equipment location recognition error rates as high as 5%-10%. This leads to spatial coordinate deviations between the model and the actual equipment, potentially causing misjudgments in maintenance and delaying emergency repairs. Full-scale scanning also requires the deployment of multiple high-precision devices, resulting in high hardware costs that are unaffordable for small and medium-sized power grid companies. Moreover, existing methods are difficult to adapt to complex terrain or temporary operation scenarios.

[0003] In summary, existing model update methods suffer from drawbacks such as low efficiency, poor data consistency, high cost, and poor flexibility. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method, medium and system for updating a three-dimensional model of power equipment, which can improve the updating efficiency of the three-dimensional model of power equipment and ensure data consistency.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A method for updating a 3D model of power equipment, comprising: Acquire scene information after the power equipment is updated, the scene information including point cloud data and image data; Key feature points are extracted from the image data, and the device type is identified based on the point cloud data; Acquire pose and positioning data of the scanning device when it collects the scene information; The device spatial coordinates are determined based on the scene information, the pose data, and the positioning data; Obtain the old 3D model of the power equipment before the update based on the equipment type; The displacement vector is calculated by aligning the old point cloud data in the old 3D model with the point cloud data based on the key feature points and the device spatial coordinates. The changed region of the old 3D model is determined based on the displacement vector, and the changed region is updated based on the target point cloud data after the displacement vector alignment process to obtain the updated new 3D model.

[0006] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows: A computer medium includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement a method for updating a three-dimensional model of a power equipment as described above.

[0007] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows: A system for updating a three-dimensional model of power equipment includes a data acquisition module, a data calculation module, and a model updating module; The data acquisition module is used to acquire scene information after the power equipment is updated, the scene information including point cloud data and image data; and to acquire pose data and positioning data when the scanning device collects the scene information. The data calculation module is used to extract key feature points from the image data, identify the device type based on the point cloud data, and determine the device spatial coordinates according to the scene information, the pose data, and the positioning data. The model update module is used to obtain the old 3D model of the power equipment before the update according to the equipment type; to align the point cloud data with the old point cloud data in the old 3D model according to the key feature points and the equipment spatial coordinates to calculate the displacement vector; and to determine the changed area of ​​the old 3D model according to the displacement vector, and update the changed area according to the target point cloud data after the displacement vector alignment processing to obtain the updated new 3D model.

[0008] The beneficial effects of this invention are as follows: By integrating multi-source sensing data and intelligent algorithms, a closed-loop process from data acquisition to model update is constructed. First, point cloud and image data are acquired simultaneously to form a complementary system, utilizing both the geometric accuracy of point cloud data and the texture information of image data to overcome the vulnerability of single data sources to interference in complex environments. By extracting key feature points from the image and combining them with point cloud classification to identify device types, accurate matching of device identities is achieved, avoiding redundant data processing caused by full-scale scanning. The pose and positioning data of the scanning device are introduced, and a joint map is constructed using SLAM algorithms, effectively eliminating sensor cumulative errors, improving the global consistency of spatial coordinates, and providing a reliable benchmark for subsequent alignment. The displacement vector is calculated through feature point matching and point cloud registration algorithms to accurately locate the areas in the old model that need updating, avoiding the waste of computational resources caused by full model reconstruction. Finally, an incremental update strategy is adopted, performing local mesh reconstruction only on changed areas, preserving the original data of unchanged parts while injecting the latest state information through target point cloud data, significantly reducing update time and hardware costs. Attached Figure Description

[0009] Figure 1 A flowchart illustrating a method for updating a three-dimensional model of power equipment, provided as an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a computer medium provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the architecture of an update system for a three-dimensional model of power equipment provided in an embodiment of the present invention. Detailed Implementation

[0010] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0011] Embodiments of the present invention provide a method for updating a three-dimensional model of power equipment, comprising: Acquire scene information after the power equipment is updated, the scene information including point cloud data and image data; Key feature points are extracted from the image data, and the device type is identified based on the point cloud data; Acquire pose and positioning data of the scanning device when it collects the scene information; The device spatial coordinates are determined based on the scene information, the pose data, and the positioning data; Obtain the old 3D model of the power equipment before the update based on the equipment type; The displacement vector is calculated by aligning the old point cloud data in the old 3D model with the point cloud data based on the key feature points and the device spatial coordinates. The changed region of the old 3D model is determined based on the displacement vector, and the changed region is updated based on the target point cloud data after the displacement vector alignment process to obtain the updated new 3D model.

[0012] As described above, the beneficial effects of this invention are as follows: By integrating multi-source sensing data and intelligent algorithms, a closed-loop process from data acquisition to model update is constructed. First, point cloud and image data are acquired simultaneously to form a complementary system, utilizing both the geometric precision of point cloud data and the texture information of image data to overcome the vulnerability of single data sources to interference in complex environments. By extracting key feature points from images and combining them with point cloud classification to identify device types, accurate matching of device identities is achieved, avoiding redundant data processing caused by full-scale scanning. The pose and positioning data of the scanning device are introduced, and a joint map is constructed using the SLAM algorithm, effectively eliminating sensor cumulative errors, improving the global consistency of spatial coordinates, and providing a reliable benchmark for subsequent alignment. The displacement vector is calculated through feature point matching and point cloud registration algorithms to accurately locate the areas in the old model that need updating, avoiding the waste of computational resources caused by full model reconstruction. Finally, an incremental update strategy is adopted, performing local mesh reconstruction only on changed areas, preserving the original data of unchanged parts while injecting the latest state information through target point cloud data, significantly reducing update time and hardware costs.

[0013] Furthermore, extracting key feature points from the image data includes: The device outline is determined in the image data based on the pre-trained YOLOv5 model, and key feature points are extracted within the device outline using image processing algorithms.

[0014] As described above, a device contour detection framework is constructed based on a pre-trained YOLOv5 model. This framework leverages the strong representational ability of deep neural networks for device morphological features, overcoming the misjudgment defects of traditional algorithms under complex lighting conditions. Image processing algorithms are executed within the defined device contour region, eliminating background noise interference through spatial constraints and focusing on key point detection within the device's feature regions. This phased processing architecture retains the advantages of deep learning models in object detection while inheriting the strengths of traditional image processing algorithms in feature point localization accuracy, forming a dual technical guarantee mechanism. The introduction of the pre-trained model significantly improves the robustness of device contour recognition, while the local feature extraction strategy within the contour effectively avoids the computational redundancy and mismatch risks associated with full-image search.

[0015] Furthermore, the types of devices for identifying points based on the point cloud data include: The point cloud data is classified and identified based on the pre-trained PoinNet model to determine the equipment type of the power equipment.

[0016] As described above, using point cloud data as input avoids the shortcomings of two-dimensional image texture information being affected by environmental interference, and enhances the robustness of device type recognition by utilizing the three-dimensional geometric characteristics of point cloud data. The pre-trained PoinNet model extracts high-order features from the point cloud through a deep learning framework, which, compared with traditional methods of manually designing features, can adaptively capture key classification criteria such as device shape and structure. Through an end-to-end classification and recognition process, automated determination of device type is achieved, avoiding subjective errors that may be introduced by manual annotation, and providing an accurate device type benchmark for subsequent model alignment.

[0017] Further, determining the device spatial coordinates based on the scene information, the pose data, and the positioning data includes: Extract the geometric features from the point cloud data and the texture features from the image data, respectively; The geometric features and the texture features are fused to obtain joint features; Based on the joint features, a joint map is constructed using the SLAM algorithm to determine the spatial coordinates of the devices; The global positioning information of the power equipment is determined based on the pose data and the positioning data; The device spatial coordinates are corrected based on the global positioning information.

[0018] As described above, firstly, by simultaneously extracting geometric features from point cloud data and texture features from image data, the limitations of a single data source in complex environments can be overcome: geometric features reflect the structural information of the device, while texture features capture surface details; their complementarity enhances the robustness of feature representation. Secondly, a joint map is constructed using the SLAM algorithm based on joint features, achieving dynamic calculation of device spatial coordinates through simultaneous localization and map building techniques. The SLAM algorithm integrates real-time sensor data and feature information in this process, effectively solving the positioning drift problem caused by single-frame data errors in traditional methods. Simultaneously, by combining the pose and positioning data of the scanning device, global positioning information is generated through multi-sensor fusion, eliminating accumulated errors in the local coordinate system. Finally, the device spatial coordinates are corrected based on the global positioning information, ensuring that the coordinate values ​​conform to both local feature matching results and global spatial reference, thereby significantly improving coordinate accuracy.

[0019] Furthermore, the displacement vector is calculated by aligning the old point cloud data in the old 3D model with the point cloud data based on the key feature points and the device spatial coordinates, including: The matching feature points of the key feature points in the old 3D model are determined by the feature point matching algorithm, and the key feature points are mapped to the point cloud data to obtain point cloud feature points; Determine the initial transformation matrix for converting the point cloud feature points into the matching feature points based on the device spatial coordinates; Based on the initial transformation matrix, the rotation matrix and translation matrix are calculated by aligning the old point cloud data with the new point cloud data using an iterative nearest-point point cloud registration algorithm. The displacement vector required to align the old point cloud data with the new point cloud data is determined based on the rotation matrix and the translation matrix.

[0020] As described above, the process first employs a feature point matching algorithm to establish the correspondence between key feature points in the old model and the new point cloud. This spatial mapping forms point cloud feature points, laying the foundation for data association. Subsequently, an initial transformation matrix is ​​generated based on the device's spatial coordinates, utilizing prior spatial location information to improve the convergence speed and accuracy of the registration algorithm. Then, an iterative nearest-point algorithm is used for fine registration based on the initial transformation. Rigid transformation of the point cloud data is achieved through the joint solution of rotation and translation matrices, ultimately outputting vector parameters representing the model's displacement. This scheme overcomes the limitations of traditional algorithms that rely on a single data source by combining feature matching with spatial constraints. The phased registration strategy ensures both computational efficiency and improved robustness in complex scenarios.

[0021] Furthermore, determining the changed region of the old 3D model based on the displacement vector includes: If the displacement vector is greater than a preset threshold, the points in the old point cloud data corresponding to the displacement vector are marked as change points, and the change area of ​​the old 3D model is determined based on the change points.

[0022] As described above, the criterion of "displacement vector greater than a preset threshold" can filter out minute displacements caused by environmental noise or measurement errors, and only perform subsequent processing on significant displacements that exceed a reasonable range, avoiding invalid calculations. Simultaneously, it establishes a correspondence between the old model and the new point cloud data through the spatial correlation of displacement vectors, and delineates the complete change region based on the spatial distribution characteristics of the change points. This forms a closed-loop logic from displacement determination to region calibration, improving the positioning accuracy of the model update region while reducing computational complexity.

[0023] Furthermore, updating the changed region based on the target point cloud data after the displacement vector alignment process to obtain the updated new 3D model includes: Determine the target point cloud data after the old point cloud data has been aligned based on the displacement vector within the changed region; Based on the target point cloud data, new mesh cells are generated in the changed area using the incremental Marching Cubes algorithm. The new mesh cells are then merged with the old mesh cells in the unchanged area of ​​the old 3D model to obtain an updated new 3D model.

[0024] As described above, by first determining the target point cloud data based on displacement vector alignment, the region to be updated can be accurately located, avoiding redundant data processing in unchanged regions. Then, an incremental Marching Cubes algorithm is used to generate new mesh cells. This algorithm reduces voxel computation by more than 80% compared to the traditional global Marching Cubes algorithm by extracting isosurfaces only in changed regions. Finally, by merging the old and new mesh cells, the structural integrity of unchanged regions in the original model is preserved, while the newly generated high-precision local mesh is seamlessly integrated, ensuring geometric consistency in model updates. The combination of incremental algorithm and local data processing overcomes the limitation of traditional 3D reconstruction requiring global computation, while displacement vector alignment provides a spatial registration basis for local updates. Together, they achieve a dual improvement in model update efficiency and accuracy.

[0025] Furthermore, after obtaining the updated 3D model, it also includes: Obtain the actual physical data of the power equipment and the physical data of the new three-dimensional model; Based on the actual physical data and the model physical data, the consistency between the new 3D model and the state of the power equipment is verified using a Bayesian network model. If the verification fails, the new 3D model is updated based on the actual physical data.

[0026] As described above, this approach firstly provides a multi-dimensional data foundation for state verification by simultaneously acquiring the actual physical data of the power equipment and the physical data of the new 3D model. Secondly, a Bayesian network model is introduced for verification, utilizing its probabilistic reasoning capabilities to handle the nonlinear relationship and uncertainty between the actual physical data and the model data. For example, conditional probability analysis is used to determine whether the deviation of equipment parameters is within the allowable threshold. Finally, an automatic update mechanism is triggered when verification fails, directly correcting the model parameters based on the actual physical data, rather than relying on manual intervention. This scheme overcomes the lack of dynamic verification after model updates in traditional methods through a closed-loop feedback mechanism. Furthermore, the probabilistic modeling characteristics of Bayesian networks address the problem of single threshold judgment being susceptible to noise interference under complex operating conditions.

[0027] Another embodiment of the present invention provides a computer medium including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the above-described method for updating a three-dimensional model of a power equipment.

[0028] Another embodiment of the present invention provides a system for updating a three-dimensional model of power equipment, including a data acquisition module, a data calculation module and a model updating module; The data acquisition module is used to acquire scene information after the power equipment is updated, the scene information including point cloud data and image data; and to acquire pose data and positioning data when the scanning device collects the scene information. The data calculation module is used to extract key feature points from the image data, identify the device type based on the point cloud data, and determine the device spatial coordinates according to the scene information, the pose data, and the positioning data. The model update module is used to obtain the old 3D model of the power equipment before the update according to the equipment type; to align the point cloud data with the old point cloud data in the old 3D model according to the key feature points and the equipment spatial coordinates to calculate the displacement vector; and to determine the changed area of ​​the old 3D model according to the displacement vector, and update the changed area according to the target point cloud data after the displacement vector alignment processing to obtain the updated new 3D model.

[0029] Please refer to Figure 1 Embodiment 1 of the present invention is as follows: A method for updating a 3D model of power equipment, specifically including: S1. Obtain scene information after the power equipment is updated, the scene information including point cloud data and image data.

[0030] Point cloud data refers to a set of three-dimensional spatial coordinates acquired by LiDAR or a depth camera, specifically a phase-detection laser scanner, used to construct the device's geometry. Image data refers to two-dimensional visual information acquired by an optical camera, specifically an RGB-D camera, used to simultaneously capture the device's color image (RGB) and depth information.

[0031] S2. Extract key feature points from the image data and identify the device type based on the point cloud data.

[0032] Key feature points refer to the essential characteristics that distinguish different power equipment, such as the equipment's nameplate and terminals. Equipment type refers to the type of power equipment currently being collected, such as surge arresters and insulators.

[0033] S3. Obtain pose data and positioning data when the scanning device collects the scene information.

[0034] The pose data includes the acceleration and angular velocity data of the scanning device, which can be implemented using an inertial measurement unit to eliminate sensor motion errors. The positioning data includes longitude and latitude, which can be implemented using a GPS (Global Positioning System) / RTK (Real-Time Kinematics) module.

[0035] S4. Determine the device spatial coordinates based on the scene information, the pose data, and the positioning data.

[0036] Among them, the device spatial coordinates are the joint localization results of the fusion of geometric features and texture features, which can be implemented by feature-level fusion algorithm to establish a globally consistent spatial reference system.

[0037] S5. Obtain the old 3D model of the power equipment before the update, based on the equipment type.

[0038] S6. Align the old point cloud data in the old 3D model with the point cloud data based on the key feature points and the device spatial coordinates to calculate the displacement vector.

[0039] The displacement vector is a mathematical expression describing the spatial positional difference between the old and new point clouds. Specifically, it can be implemented using rigid body transformation parameters to quantify the change in device position.

[0040] S7. Determine the changed region of the old 3D model based on the displacement vector, and update the changed region based on the target point cloud data after the displacement vector alignment process to obtain the updated new 3D model.

[0041] Specifically, the process begins by synchronously acquiring updated point cloud and image data from the device. The device's identity is then determined through visual feature extraction and point cloud classification. Combined with the scanning device's real-time pose information, a spatial coordinate system incorporating global positioning corrections is constructed. A correspondence between the old and new models is established, and precise displacement parameters are calculated through iterative registration. When the displacement exceeds a set threshold, the local area requiring updating is automatically identified, and only the changed portions are reconstructed. Using an incremental update strategy, the original model data for unchanged areas is retained, and newly acquired point cloud data is integrated into the designated area, completing the local model update.

[0042] In an optional implementation, extracting key feature points from the image data in step S2 includes: S21. Based on the pre-trained YOLOv5 model, determine the device outline in the image data, and extract key feature points within the device outline using an image processing algorithm.

[0043] The pre-trained YOLOv5 model refers to a target detection model trained on a large-scale power equipment image dataset. It extracts deep image features through a convolutional neural network to accurately locate equipment boundaries under complex lighting conditions. This model optimizes parameters through transfer learning, eliminating the impact of environmental interference on equipment contour recognition. The image processing algorithm refers to a local feature extraction method performed within the equipment contour region. It can employ scale-invariant feature transformation algorithms or accelerated robust feature algorithms, detecting stable feature points by constructing a Gaussian pyramid space. This algorithm operates only within the target contour region, avoiding background noise interference from full-image search.

[0044] Specifically, the YOLOv5 model first performs multi-scale feature fusion on the input image to generate a predicted device bounding box, outputting precise device contour coordinates. After confirming the device contour, the image processing algorithm is confined to the contour coordinate range, and selects key feature points characterizing the device structure by calculating local gradient extrema or corner response functions. This staged processing architecture eliminates environmental interference through the global perception capability of the object detection model, and then uses the sub-pixel-level localization accuracy of the local feature algorithm to capture detailed device features, forming a refined feature extraction mechanism under global constraints.

[0045] In one optional implementation, the device type identified based on the point cloud data in S2 includes: S22. Classify and identify the point cloud data based on the pre-trained PoinNet model to determine the equipment type of the power equipment.

[0046] The pre-trained PoinNet model refers to a deep learning network trained and optimized using a large-scale power equipment point cloud dataset. Specifically, it can be trained using a point cloud dataset containing typical equipment such as transformers, circuit breakers, and disconnectors through transfer learning. This model extracts local geometric features and global topological features from the point cloud through a multilayer perceptron architecture, constructs an abstract representation of the equipment morphology, and achieves end-to-end classification and recognition.

[0047] Specifically, after acquiring the device point cloud data, the point cloud is first denoised and its coordinate system is normalized. The preprocessed point cloud is then input into the PoinNet model, which captures the spatial distribution characteristics of the device structure through a multi-layer feature extraction network. The last fully connected layer of the model outputs the probability distribution of each device category, and the category with the highest probability value is selected as the recognition result. Since the point cloud data itself contains three-dimensional spatial information, data augmentation methods such as random rotation and scaling are used during model training to improve the model's adaptability to changes in device posture, thereby solving the misjudgment problem caused by changes in viewing angle in traditional image recognition methods.

[0048] In one alternative implementation, S4 includes: S41. Extract the geometric features from the point cloud data and the texture features from the image data, respectively.

[0049] S42. The geometric features and the texture features are fused to obtain a joint feature.

[0050] S43. Based on the joint features, a joint map is constructed using the SLAM algorithm to determine the spatial coordinates of the device.

[0051] S44. Determine the global positioning information of the power equipment based on the pose data and the positioning data.

[0052] S45. Correct the device spatial coordinates based on the global positioning information.

[0053] Geometric features refer to the attribute information in point cloud data that characterizes the three-dimensional spatial structure of the device. Specifically, this can be achieved using curvature calculation or normal vector estimation algorithms from point cloud processing libraries, used to describe the device's shape and spatial distribution characteristics. Texture features refer to the attributes in image data that characterize the visual information of the device's surface. Specifically, this can be achieved using SIFT or ORB algorithms from computer vision libraries, used to capture the color, pattern, and edge details of the device's surface. Feature fusion refers to the process of integrating different types of features into a unified representation. Specifically, this can be achieved using multimodal fusion layers in neural networks or weighted averaging methods, enhancing environmental resistance to interference through complementary features. SLAM algorithm refers to simultaneous localization and mapping technology, specifically implemented using the LeGO-LOAM algorithm, dynamically updating the device's position in the local coordinate system through a joint map. Global positioning information refers to the spatial reference of the device in the absolute coordinate system, specifically achieved by fusing latitude and longitude data from the GPS module with attitude angle data from the IMU sensor, used to eliminate local positioning errors.

[0054] Specifically, geometric features extract the 3D structure of the device contour by analyzing the spatial relationships between adjacent points in the point cloud, while texture features extract surface markings by detecting pixel regions with significant gradient changes in the image. These two types of features are fused and input into the SLAM system, where feature matching and motion estimation generate a joint map containing the device's location. Pose data provides the scanning device's own attitude parameters, and localization data provides the device's initial position in the global coordinate system. Combining these two data points generates global constraints, which are then used to nonlinearly optimize the device coordinates in the joint map, ultimately yielding accurate spatial coordinates.

[0055] In one alternative implementation, S6 includes: S61. The matching feature points of the key feature points in the old 3D model are determined by the feature point matching algorithm, and the key feature points are mapped to the point cloud data to obtain point cloud feature points.

[0056] S62. Determine the initial transformation matrix for converting the point cloud feature points into the matching feature points based on the device spatial coordinates.

[0057] S63. Based on the initial transformation matrix, the old point cloud data and the point cloud data are aligned and calculated using an iterative nearest-point point cloud registration algorithm to obtain the rotation matrix and translation matrix.

[0058] S64. Determine the displacement vector required to align the old point cloud data with the point cloud data based on the rotation matrix and the translation matrix.

[0059] Among them, the feature point matching algorithm refers to establishing the correspondence between key points in the old and new models through key feature points. It can be implemented using SIFT or ORB algorithms and is used to eliminate mismatches caused by changes in lighting or local occlusion in complex scenes. The initial transformation matrix refers to rigid transformation parameters generated based on device spatial coordinates, used to provide initial pose constraints for subsequent point cloud registration. The iterative nearest-neighbor point cloud registration algorithm refers to iteratively optimizing the transformation parameters by minimizing the distance error between corresponding points. KD-trees can be used to accelerate the nearest neighbor search, used to achieve fine alignment of the old and new point clouds under the initial pose constraints. The displacement vector refers to the spatial transformation parameters synthesized from the rotation matrix and translation matrix, which can be derived through the homogeneous coordinate transformation formula, used to quantify the spatial positional differences between the old 3D model and the newly acquired point cloud.

[0060] Specifically, in the point cloud alignment process, firstly, stable feature points from the old and new data are extracted using a feature point matching algorithm to establish a correspondence. For example, the SIFT algorithm is used to extract scale-invariant feature points. Then, an initial transformation matrix is ​​generated based on the device's spatial coordinates, for example, by calculating the initial rotation angle and translation using the device's positioning information in the global coordinate system. Based on this initial pose, an iterative nearest-neighbor algorithm is used for multiple rounds of registration optimization. For example, in each iteration, a KD-tree is used to quickly find the nearest neighbor pair and update the transformation matrix. Finally, the translation vector is obtained by decomposing the transformation matrix. For example, the 3D rotation components are converted to Euler angles, and the translation components are directly used as vector elements.

[0061] In one optional implementation, determining the changed region of the old 3D model based on the displacement vector in S7 includes: S71. If the displacement vector is greater than a preset threshold, the points in the old point cloud data corresponding to the displacement vector are marked as change points, and the change area of ​​the old 3D model is determined based on the change points.

[0062] In this context, the displacement vector refers to the transformation required to align the old point cloud data with the updated point cloud data in 3D space, used to quantify displacement differences in local areas of the model. The preset threshold is a pre-defined critical value for displacement judgment, specifically a value set based on equipment installation accuracy or measurement error range, such as 0.5 mm or 1 cm, used to distinguish between effective displacement and noise interference. Change points are discrete points in the old point cloud data whose displacement exceeds the preset threshold, used to identify local areas in the model where substantial displacement has occurred. The change region refers to the continuous 3D spatial range formed by change points, specifically achieved by merging change point regions using a density-based spatial clustering algorithm to form boundary contours, used to determine the complete topological structure that needs to be updated in the model.

[0063] Specifically, after aligning the old and new point cloud data, the magnitude of each old point cloud point is determined based on its corresponding displacement vector. When the displacement exceeds a preset threshold, the point is considered a valid change point caused by actual equipment displacement or structural changes, rather than a spurious change caused by measurement errors or environmental noise. Subsequently, a spatial clustering algorithm is used to aggregate adjacent change points into a continuous three-dimensional region. For example, the DBSCAN algorithm is used to divide densely connected point sets into independent clusters, and then the geometry of the change region is generated based on the cluster boundaries. This process filters out interference data caused by minute displacements and integrates discrete change points into a complete updated region through spatial correlation, avoiding the fragmentation problem of region division caused by local matching deviations in traditional methods.

[0064] In an optional implementation, S7, updating the changed region based on the target point cloud data after the displacement vector alignment process to obtain the updated new 3D model includes: S72. Determine the target point cloud data after the old point cloud data has been aligned based on the displacement vector within the changed region.

[0065] S73. Based on the target point cloud data, new mesh cells are generated in the changed area using the incremental Marching Cubes algorithm, and the new mesh cells are merged with the old mesh cells in the unchanged area of ​​the old 3D model to obtain an updated new 3D model.

[0066] The target point cloud data after displacement vector alignment refers to the local point cloud set after spatial registration of the old and new point cloud data through coordinate transformation. This is achieved using a rigid transformation based on the ICP (Iterative Closest Point) algorithm to eliminate spatial errors caused by device position offset. The incremental Marching Cubes algorithm is a mesh generation method that extracts isosurfaces only in locally changed regions. This can be achieved by dynamically adjusting the voxel spatial resolution to avoid redundant calculations in unchanged regions. The merging of old and new mesh cells involves seamlessly stitching the newly generated local mesh with the unchanged mesh retained in the original model. This can be achieved using a mesh fusion method based on vertex index matching to ensure consistency in geometric topology.

[0067] Specifically, after registering the old point cloud with the current point cloud using displacement vectors, regions where the displacement exceeds a preset threshold can be accurately identified as changed regions. Within these changed regions, the incremental Marching Cubes algorithm is used to calculate isosurfaces only for the 3D voxels of that region, generating new triangular meshes. This algorithm dynamically partitions the local voxel space, skipping the voxel traversal process of unchanged regions and directly extracting the surface geometry of the changed regions. The newly generated mesh cells are connected to the meshes of the unchanged regions in the old model through vertex coordinate matching, preserving the original mesh's topological relationships, ultimately forming a complete 3D model update.

[0068] In an alternative implementation, after S7, the following is also included: S8. Obtain the actual physical data of the power equipment and the model physical data of the new three-dimensional model.

[0069] S9. Based on the actual physical data and the model physical data, verify the consistency between the new three-dimensional model and the state of the power equipment using a Bayesian network model.

[0070] S10. If the verification fails, the new three-dimensional model is updated based on the actual physical data.

[0071] The actual physical data refers to the physical attribute parameters of the power equipment collected by sensors or detection devices. Specifically, this can be achieved by using infrared thermal imaging sensors to collect temperature data and vibration sensors to collect mechanical state data, providing a real physical basis for state verification. The model physical data refers to the equipment attribute parameters stored in the new 3D model. Specifically, this can be achieved by mapping the grid vertex coordinates to the entity data to store the equipment attribute parameters. The Bayesian network model refers to an inference tool based on a probabilistic graphical model. Specifically, this can be achieved by using conditional probability tables to construct the dependencies between equipment parameters, Markov chain Monte Carlo algorithms for parameter estimation, and posterior probability calculations to verify the data deviation range, thus solving the problem of traditional threshold judgments being sensitive to noise.

[0072] Specifically, during the model update phase, infrared thermal imaging data (temperature distribution) and vibration sensor data (mechanical condition) of the power equipment entity are collected simultaneously, and the corresponding data are extracted from the 3D model. The two sets of data are input into the Bayesian network model, and the causal relationship between parameters is analyzed through a preset node conditional probability table. For example, when the model temperature distribution exceeds the allowable range of the actual temperature, an abnormal probability event is triggered. If the joint probability of the model data and the actual data is lower than a preset threshold, the verification is deemed to have failed and the update process is initiated. The detection values ​​in the actual physical data are directly overwritten with the corresponding parameters in the model, thereby eliminating the cumulative error between the model and the entity.

[0073] Please refer to Figure 2 Embodiment two of the present invention is as follows: A computer medium 100 includes a memory 101, a processor 102, and a computer program stored on the memory 101 and running on the processor 102. When the processor 102 executes the computer program, it implements a method for updating a three-dimensional model of a power equipment according to Embodiment 1 above.

[0074] Please refer to Figure 3 Embodiment 3 of the present invention is as follows: A power equipment three-dimensional model updating system 200 is used to implement the power equipment three-dimensional model updating method of the above embodiment 1.

[0075] Specifically, the system 200 includes a data acquisition module 201, a data calculation module 202, and a model update module 203.

[0076] The data acquisition module 201 is used to acquire scene information after the power equipment is updated, the scene information including point cloud data and image data; and to acquire pose data and positioning data when the scanning device collects the scene information.

[0077] The data calculation module 202 is used to extract key feature points from the image data and identify the device type based on the point cloud data; and to determine the device spatial coordinates according to the scene information, the pose data and the positioning data.

[0078] The model update module 203 is used to obtain the old 3D model of the power equipment before the update according to the equipment type; to calculate the displacement vector by aligning the point cloud data with the old point cloud data in the old 3D model according to the key feature points and the equipment spatial coordinates; to determine the change area of ​​the old 3D model according to the displacement vector; and to update the change area according to the target point cloud data after the displacement vector alignment process to obtain the updated new 3D model.

[0079] In summary, this invention provides a method, medium, and system for updating 3D models of power equipment. It leverages the complementary geometric precision of point cloud data and the texture information of image data to overcome the limitations of a single data source in complex environments. Simultaneously, it accurately extracts key feature points using a pre-trained YOLOv5 model and image processing algorithms, and combines this with a pre-trained PointNet model to utilize the 3D geometric characteristics of point cloud data for equipment type identification, avoiding errors from manual annotation. The pose and positioning data of the scanned equipment are introduced, and a joint map is constructed using SLAM algorithms, effectively eliminating sensor cumulative errors and improving the global consistency of spatial coordinates. Displacement vectors are calculated through feature point matching and point cloud registration algorithms to accurately locate changed areas in the old model, avoiding the waste of computational resources caused by full model reconstruction. Finally, an incremental update strategy is adopted, performing local mesh reconstruction only on changed areas, significantly reducing update time and hardware costs. Furthermore, a Bayesian network model is used to verify the consistency between the updated 3D model and the actual state of the power equipment, and the model is automatically updated when verification fails, further ensuring the accuracy and reliability of the model. Overall, this solution improves the efficiency and accuracy of model updates while reducing computational complexity and hardware costs, providing efficient and precise technical support for the digital operation and maintenance of power equipment.

[0080] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for updating a three-dimensional model of power equipment, characterized in that, include: Acquire scene information after the power equipment is updated, the scene information including point cloud data and image data; Key feature points are extracted from the image data, and the device type is identified based on the point cloud data; Acquire pose and positioning data of the scanning device when it collects the scene information; The device spatial coordinates are determined based on the scene information, the pose data, and the positioning data; Obtain the old 3D model of the power equipment before the update based on the equipment type; The displacement vector is calculated by aligning the old point cloud data in the old 3D model with the point cloud data based on the key feature points and the device spatial coordinates. The changed region of the old 3D model is determined based on the displacement vector, and the changed region is updated based on the target point cloud data after the displacement vector alignment process to obtain the updated new 3D model.

2. The method for updating a three-dimensional model of power equipment according to claim 1, characterized in that, Extracting key feature points from the image data includes: The device outline is determined in the image data based on the pre-trained YOLOv5 model, and key feature points are extracted within the device outline using image processing algorithms.

3. The method for updating a three-dimensional model of power equipment according to claim 1, characterized in that, The types of devices that can identify based on the point cloud data include: The point cloud data is classified and identified based on the pre-trained PoinNet model to determine the equipment type of the power equipment.

4. The method for updating a three-dimensional model of power equipment according to claim 1, characterized in that, Determining the device spatial coordinates based on the scene information, the pose data, and the positioning data includes: Extract the geometric features from the point cloud data and the texture features from the image data, respectively; The geometric features and the texture features are fused to obtain joint features; Based on the joint features, a joint map is constructed using the SLAM algorithm to determine the spatial coordinates of the devices; The global positioning information of the power equipment is determined based on the pose data and the positioning data; The device spatial coordinates are corrected based on the global positioning information.

5. The method for updating a three-dimensional model of power equipment according to claim 1, characterized in that, The displacement vector is calculated by aligning the old point cloud data in the old 3D model with the point cloud data based on the key feature points and the device spatial coordinates, including: The matching feature points of the key feature points in the old 3D model are determined by the feature point matching algorithm, and the key feature points are mapped to the point cloud data to obtain point cloud feature points; Determine the initial transformation matrix for converting the point cloud feature points into the matching feature points based on the device spatial coordinates; Based on the initial transformation matrix, the rotation matrix and translation matrix are calculated by aligning the old point cloud data with the new point cloud data using an iterative nearest-point point cloud registration algorithm. The displacement vector required to align the old point cloud data with the new point cloud data is determined based on the rotation matrix and the translation matrix.

6. The method for updating a three-dimensional model of power equipment according to claim 5, characterized in that, Determining the changed region of the old 3D model based on the displacement vector includes: If the displacement vector is greater than a preset threshold, the points in the old point cloud data corresponding to the displacement vector are marked as change points, and the change area of ​​the old 3D model is determined based on the change points.

7. The method for updating a three-dimensional model of power equipment according to claim 6, characterized in that, The updated 3D model is obtained by updating the changed region based on the target point cloud data after the displacement vector alignment process, including: Determine the target point cloud data after the old point cloud data has been aligned based on the displacement vector within the changed region; Based on the target point cloud data, new mesh cells are generated in the changed area using the incremental Marching Cubes algorithm. The new mesh cells are then merged with the old mesh cells in the unchanged area of ​​the old 3D model to obtain an updated new 3D model.

8. The method for updating a three-dimensional model of power equipment according to claim 1, characterized in that, After obtaining the updated 3D model, the following is also included: Obtain the actual physical data of the power equipment and the physical data of the new three-dimensional model; Based on the actual physical data and the model physical data, the consistency between the new 3D model and the state of the power equipment is verified using a Bayesian network model. If the verification fails, the new 3D model is updated based on the actual physical data.

9. A computer medium, characterized in that, The device includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement a method for updating a three-dimensional model of a power equipment as described in any one of claims 1 to 8.

10. A system for updating a three-dimensional model of power equipment, characterized in that, It includes a data acquisition module, a data calculation module, and a model update module; The data acquisition module is used to acquire scene information after the power equipment is updated, the scene information including point cloud data and image data; and to acquire pose data and positioning data when the scanning device collects the scene information. The data calculation module is used to extract key feature points from the image data and identify the device type based on the point cloud data; And determine the device's spatial coordinates based on the scene information, the pose data, and the positioning data; The model update module is used to obtain the old 3D model of the power equipment before the update according to the equipment type; to align the point cloud data with the old point cloud data in the old 3D model according to the key feature points and the equipment spatial coordinates to calculate the displacement vector; and to determine the changed area of ​​the old 3D model according to the displacement vector, and update the changed area according to the target point cloud data after the displacement vector alignment processing to obtain the updated new 3D model.

Images