Automatic reverse generation method and system for GIM model based on point cloud

Abstract

The invention discloses a point cloud-based GIM model automatic reverse generation method and system, and relates to the technical field of point cloud modeling, and the method comprises the steps: carrying out the traversal inspection of a target power grid, and carrying out the virtual segmentation, and obtaining a plurality of video segments; performing inter-frame similarity analysis to obtain a first similarity index; if the preset similar threshold value is not met, performing lens identification cutting to obtain a target cutting result; establishing a first point cloud data group, and performing modeling analysis to obtain a first single model block; establishing a first single model, and matching a corresponding first device; and rendering the first parameterized information to the first single model to obtain a target GIM model of the target power grid. The technical problems that in the prior art, construction of the target power grid GIM depends on a large amount of manual operation, the process is tedious, efficiency is low, and the accuracy of the constructed model is poor are solved, and the technical effects that automatic reverse generation of the target power grid GIM is achieved, and the efficiency and accuracy of model construction are improved are achieved.

Description

Technical Field

[0001] This invention relates to the field of point cloud modeling technology, and specifically to a method and system for automatic reverse generation of GIM models based on point clouds. Background Technology

[0002] Gain-of-Information Models (GIMs), as a key technology for efficient management and operation, can comprehensively and accurately present various information about power grid equipment and their interrelationships, providing crucial support for power system planning, design, operation monitoring, and fault diagnosis. However, traditional techniques currently face numerous challenges in building GIM models of target power grids. On the one hand, they primarily rely on manual on-site measurements and manual modeling. This process requires a large number of professionals to go deep into the power grid field to measure and record the location, shape, and parameters of various equipment one by one, and then manually input the data into modeling software to build the model. This not only consumes a significant amount of manpower, resources, and time, but also has extremely low efficiency, seriously affecting the progress of power grid digitalization. On the other hand, manual operation makes it difficult to guarantee the accuracy of the model. Different personnel have varying measurement habits and technical levels, which can easily lead to errors during the measurement process. Furthermore, the understanding and reconstruction of complex power grid structures during manual modeling can also be flawed, resulting in GIM models that fail to accurately reflect the actual condition of the power grid.

[0003] Existing technologies suffer from technical problems such as relying on a large amount of manual operation to construct the target power grid GIM model, which is cumbersome, inefficient, and results in poor accuracy of the constructed model. Summary of the Invention

[0004] This application provides a method and system for automatic reverse generation of GIM models based on point clouds, which is used to address the technical problems in the prior art where the construction of target power grid GIM models relies on a large amount of manual operation, the process is cumbersome and inefficient, and the accuracy of the constructed models is poor.

[0005] In view of the above problems, this application provides a method and system for automatic reverse generation of GIM models based on point clouds.

[0006] The first aspect of this application provides a method for automatic reverse generation of GIM models based on point clouds, the method comprising:

[0007] The target power grid is traversed and inspected to obtain target inspection videos, and the target inspection videos are virtually segmented to obtain multiple video segments; inter-frame similarity analysis is performed on the first video segment among the multiple video segments to obtain a first similarity index; if the first similarity index does not meet a predetermined similarity threshold, the first video segment is subjected to shot recognition and segmentation to obtain a target segmentation result, wherein the target segmentation result includes multiple shot videos; a first point cloud data group is constructed based on the first shot videos among the multiple shot videos, and the first point cloud data group is modeled and analyzed according to a random sampling modeling mechanism to obtain a first single model block; a first single model is constructed based on the first single model block, and a first device corresponding to the first single model is matched, wherein the first device has first parameterized information; the first parameterized information is rendered onto the first single model to obtain the target GIM model of the target power grid.

[0008] A second aspect of this application provides an automatic reverse engineering system for GIM models based on point clouds, the system comprising:

[0009] The system includes a multi-video segment acquisition module for traversing and inspecting the target power grid to obtain target inspection videos, and virtually segmenting the target inspection videos to obtain multiple video segments; a first similarity index acquisition module for performing inter-frame similarity analysis on a first video segment among the multiple video segments to obtain a first similarity index; a target segmentation result acquisition module for performing shot recognition segmentation on the first video segment if the first similarity index does not meet a predetermined similarity threshold to obtain a target segmentation result, wherein the target segmentation result includes multiple shot videos; a first single model block acquisition module for constructing a first point cloud data group based on the first shot video among the multiple shot videos, and performing modeling analysis on the first point cloud data group according to a random sampling modeling mechanism to obtain a first single model block; a first single model construction module for constructing a first single model based on the first single model block, and matching the first device corresponding to the first single model, wherein the first device has first parameterized information; and a target GIM model acquisition module for rendering the first parameterized information onto the first single model to obtain a target GIM model of the target power grid.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] The target power grid is traversed and inspected to obtain target inspection videos, which are then virtually segmented into multiple video segments. Inter-frame similarity analysis is performed to obtain a first similarity index. If the first similarity index does not meet a predetermined similarity threshold, the first video segment is segmented by shot recognition to obtain a target segmentation result. A first point cloud data group is constructed based on the first shot video from the multiple shot videos, and the first point cloud data group is modeled and analyzed using a random sampling modeling mechanism to obtain a first single model block. A first single model is constructed based on the first single model block, and a first device corresponding to the first single model is matched, with the first device possessing first parameterized information. The first parameterized information is rendered onto the first single model to obtain the target GIM model. This achieves the technical effect of automatically generating a target power grid GIM model, improving the efficiency and accuracy of model construction. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A schematic diagram of the automatic reverse generation method for GIM models based on point clouds provided in this application embodiment;

[0014] Figure 2 A schematic diagram of the structure of the automatic reverse generation system for GIM models based on point clouds provided in this application embodiment.

[0015] Figure labeling: Module 10 for acquiring multiple video segments, Module 20 for acquiring the first similarity index, Module 30 for acquiring the target segmentation result, Module 40 for acquiring the first single model block, Module 50 for acquiring the first single model assembly, and Module 60 for acquiring the target GIM model. Detailed Implementation

[0016] This application provides a method and system for automatic reverse generation of GIM models based on point clouds, which addresses the technical problems in the existing technology of relying on a large amount of manual operation, having a cumbersome and inefficient process, and producing models with poor accuracy.

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] Example 1, as Figure 1 As shown, this application provides an automatic reverse engineering method for GIM models based on point clouds, the method comprising:

[0019] Step S100: Perform a traversal inspection of the target power grid to obtain the target inspection video, and perform virtual segmentation on the target inspection video to obtain multiple video segments.

[0020] Specifically, a comprehensive inspection of the target power grid is conducted using inspection equipment. This can be achieved by using drones equipped with high-definition cameras to fly over the power grid area along a pre-set route, or by using monitoring cameras installed at key locations on the power grid to continuously and stably collect real-time images of various parts of the grid. This generates a target inspection video containing rich information, including the appearance and relative positions of power grid equipment. Subsequently, to facilitate accurate analysis, the target inspection video needs to be virtually segmented. Based on factors such as the overall video duration, the frequency of image changes, and subsequent analysis requirements, a unit frame number greater than 2 and less than the total number of frames in the target inspection video is determined. According to this unit frame number, the target inspection video is divided into multiple relatively independent video segments in chronological order, making the information in each video segment more concentrated and targeted, laying a solid foundation for extracting key data from the video and constructing an accurate GIM model.

[0021] Step S200: Perform inter-frame similarity analysis on the first video segment among the multiple video segments to obtain the first similarity index.

[0022] Specifically, after virtual segmentation of the target inspection video into multiple video segments, analysis is performed on the first video segment. First, the first frame image is extracted from the first video segment, serving as the benchmark for subsequent comparisons. Then, the similarity of the first frame image with any non-first frame image within the first video segment is calculated one by one. During the calculation process, an inter-frame similarity evaluation function is introduced. This function considers the similarity of images across multiple feature dimensions, resulting in a series of arbitrary similarity indices. These feature dimensions cover color distribution, texture features, shape features, etc., to comprehensively measure the similarity between images. Finally, the average of all the obtained arbitrary similarity indices is calculated, and this average is the first similarity index. The first similarity index precisely quantifies the similarity between frames within the first video segment, providing a crucial basis for subsequent judgments on whether there are scene changes in the video segment. If the first similarity index does not meet a predetermined similarity threshold, subsequent operations such as shot recognition and segmentation will be triggered, which is of great significance to the accurate construction of the entire GIM model.

[0023] Step S300: If the first similarity index does not meet the predetermined similarity threshold, then the first video segment is subjected to shot recognition and segmentation to obtain the target segmentation result, wherein the target segmentation result includes multiple shot videos.

[0024] Specifically, after the first similarity index is calculated, it is compared with a pre-set similarity threshold. This pre-set similarity threshold is determined based on a large amount of experimental data and practical experience, and is used as a standard to judge changes in the scene within a video segment. If the first similarity index does not reach this threshold, it indicates that there may be significant changes in the scene within the first video segment, such as scene switching or equipment movement. At this time, the shot recognition and segmentation program is started. Using image recognition and processing algorithms, the first video segment is analyzed frame by frame. Based on the changing characteristics of the scene content, such as the motion trajectory of objects, changes in light, and sudden color changes, the boundaries of different shots are accurately identified, and the first video segment is segmented into multiple shot videos. These shot videos together constitute the target segmentation result. Each shot video focuses on a specific power grid equipment or scene, providing more targeted data for extracting more accurate information from the video and building a more accurate GIM model. This allows the GIM model built based on the video to more accurately reflect the actual situation of the target power grid.

[0025] Step S400: Construct a first point cloud data group based on the first shot video from the multiple shot videos, and perform modeling and analysis on the first point cloud data group according to the random sampling modeling mechanism to obtain the first single model block.

[0026] Specifically, after obtaining the target segmentation results containing multiple video shots, the first video shot is selected as the basic data source. Three-dimensional spatial information is extracted from each frame of the first video shot, and this information is integrated to form a first point cloud data set. Each point in the point cloud data set accurately records the coordinate information of the corresponding object's surface position, fully presenting the spatial morphology of the power grid equipment from that shot's perspective. Next, a random sampling modeling mechanism is used to process the first point cloud data set. A portion of points is randomly selected from the first point cloud data set to form a first sampled point cloud set, while the remaining points are used as a first verification point cloud set. A first sampling model is constructed using the first sampled point cloud set, which initially outlines the shape and structure of the power grid equipment. Subsequently, the first sampled model is verified using the first verification point cloud set. Specifically, the spatial distance from the points in the first verification point cloud set to the first sampling model is calculated. Points within a predetermined distance threshold are classified as matching point sets, and points outside the threshold are classified as non-matching point sets. The accuracy of the model is judged by the ratio of the number of points in the matching point cloud set to the number of points in the non-matching point cloud set. If the ratio meets the predetermined verification constraints, it indicates that the first sampling model is relatively accurate. At this time, it is determined as the first single model block, laying a solid foundation for the subsequent construction of a complete single model and the generation of the target GIM model.

[0027] Step S500: Construct a first single model based on the first single model block, and match the first device corresponding to the first single model, wherein the first device has first parameterized information.

[0028] Specifically, the first single model is constructed based on the first single model block using a mesh reconstruction algorithm. Using the point cloud data of the first single model block as a foundation, the Poisson reconstruction algorithm is employed to reconstruct the surface, transforming the discrete point cloud data into a continuous triangular mesh model. By constructing an implicit function to fit the point cloud data, the surface details of the model can be effectively recovered, resulting in a first single model with a complete geometric shape. For matching the first single model with the corresponding first device, a feature matching algorithm is used. First, the geometric features of the first single model are extracted, such as the model's shape description and size parameters. Simultaneously, feature vectors of various types of equipment are extracted from the power grid equipment database. A nearest neighbor search algorithm based on Euclidean distance is used to calculate the distance between the feature vector of the first single model and the feature vectors of each device in the database, selecting the device with the smallest distance as the first matching device. For obtaining the first parameterized information of the first device, different parameter extraction algorithms are used depending on the device type. For transformer equipment, a rule-based parameter extraction algorithm is used to extract parameters such as rated capacity and turns ratio from data sources such as equipment nameplate information and design documents according to predefined rules. For transmission line equipment, parameters such as conductor type, length, and resistance are obtained by analyzing the line's topology and electrical parameter calculation formulas. These extracted parameters are associated with the first single model to complete the parameterization of the first device, providing support for the subsequent generation of an accurate target GIM model.

[0029] Step S600: Render the first parameterized information onto the first single model to obtain the target GIM model of the target power grid.

[0030] Specifically, to render the first parametric information onto the first single model to obtain the target GIM model, the process begins by utilizing a 3D modeling software interface, such as the Revit API in the Revit platform, to enable interaction between the parametric information and the model. By writing program scripts, the first parametric information of the first device, such as electrical parameters (rated voltage, rated current, etc.), geometric parameters (height, diameter, etc.), and other attribute parameters, is formatted according to the model's data structure requirements. Then, based on these formatted parameters, attribute assignment operations are performed on the first single model. Internally, electrical parameters are associated with the electrical attribute nodes of the model, while geometric parameters are used to precisely adjust the model's shape and size, ensuring that the model closely matches the actual device in both appearance and performance parameters. After assignment, the model is rendered using the software's rendering engine. By adjusting rendering parameters such as materials and lighting, the model more realistically presents the device's appearance features. Simultaneously, the device's parameter information is visually integrated into the model display, for example, by adding parameter annotations at specific locations on the model or using color and texture changes to reflect different parameter states. Through this series of operations, the first model is transformed into a target GIM model containing rich equipment information. This not only gives it a precise geometric shape but also integrates comprehensive equipment parameters, enabling it to accurately reflect the actual situation of the target power grid and provide strong support for power grid planning, operation and maintenance, and management.

[0031] In one possible implementation, step S100 further includes:

[0032] Step S110: Obtain the unit frame count, wherein the unit frame count is greater than 2 and less than the total number of frames in the target inspection video.

[0033] Step S120: Perform virtual segmentation on the target inspection video according to the unit frame number to obtain the multiple video segments.

[0034] Specifically, the target inspection video undergoes preliminary analysis to obtain its total frame count. Subsequently, a unit frame count greater than 2 is required. This is because if the unit frame count is 1 or 2, the segmented video segments contain too little image information to reflect the dynamic changes of power grid equipment or complex scenes. Simultaneously, the unit frame count must be less than the total frame count of the target inspection video; otherwise, effective video segmentation cannot be achieved, defeating the purpose of segmentation. Determining the unit frame count may involve considering factors such as the video's frame rate, the complexity of the scene content, and the hardware's processing capabilities. For example, for videos with high frame rates and frequent scene changes, appropriately increasing the unit frame count can improve processing efficiency while ensuring information integrity; conversely, for systems with limited hardware performance, choosing a smaller unit frame count helps reduce the computational burden. The final determined unit frame count will provide a crucial basis for subsequent virtual segmentation of the target inspection video.

[0035] Based on the determined unit frame number, starting from the beginning frame of the target inspection video, video content is sequentially segmented at unit frame intervals. For example, if the unit frame number is 10 and the total number of frames in the target inspection video is 1000, starting from frame 1, frames 1-10 are sequentially segmented as the first video segment, frames 11-20 as the second video segment, and so on, until the entire target inspection video is segmented, resulting in multiple relatively independent video segments. During the segmentation process, the original video content is not substantially modified; virtual segmentation is achieved only by recording the index information of the beginning and end frames of each video segment. These video segments retain the operational status information of the target power grid at different times and facilitate subsequent individual and detailed analysis and processing of each video segment. This provides convenience for subsequent operations such as inter-frame similarity analysis and point cloud data extraction, playing a crucial foundational role in accurately constructing the GIM model of the target power grid.

[0036] In one possible implementation, step S200 further includes:

[0037] Step S210: Obtain the first frame image from the first video segment.

[0038] Step S220: Calculate the similarity between the first frame image and any non-first frame image in the first video segment to obtain an arbitrary similarity index.

[0039] Step S230: Take the mean of the arbitrary similarity indices and denote it as the first similarity index.

[0040] Specifically, after virtual segmentation of the target inspection video into multiple video segments, the operation focuses on the first video segment. Following the video data storage structure and retrieval rules, image data is extracted from the starting position of the first video segment. Since a video consists of a series of consecutive image frames, the first frame of the first video segment is at the forefront of the storage order. Based on the video encoding format and storage specifications, the starting point of the first video segment is precisely located, and the image data corresponding to that position is read, thereby obtaining the first frame image of the first video segment. This first frame image will serve as a key reference image for subsequent inter-frame similarity analysis, providing the basis for calculating the first similarity index.

[0041] The process iterates through every non-first frame image in the first video segment, and for each non-first frame image, an inter-frame similarity evaluation function is introduced to calculate its similarity to the first frame image. This calculation yields a numerical value that quantifies the similarity between the first frame image and the non-first frame image; this value is the arbitrary similarity index. A higher similarity index indicates greater similarity between the two images; conversely, a lower index indicates greater difference. By performing this similarity calculation on all non-first frame images in the first video segment, a series of arbitrary similarity indices can be obtained. These indices will provide important basis for subsequently determining the overall similarity of the first video segment.

[0042] All calculated arbitrary similarity indices are aggregated, and the average value of these indices is determined as the first similarity index. This first similarity index comprehensively reflects the overall similarity between the frames of images within the first video segment, providing a quantitative basis for subsequent judgments such as whether there are scene transitions in the first video segment. If the first similarity index does not meet the predetermined similarity threshold, subsequent operations such as shot recognition and segmentation will be triggered, which is of great significance for accurately constructing the GIM model of the target power grid.

[0043] In one possible implementation, step S220 further includes:

[0044] Step S221: Introduce an inter-frame similarity evaluation function, and calculate the similarity between the first frame image and any non-first frame image in the first video segment according to the inter-frame similarity evaluation function to obtain the arbitrary similarity index.

[0045] The expression for the inter-frame similarity evaluation function is as follows:

[0046]

[0047] f(p1,p x ) represents the first frame image p1 and any non-first frame image p x The arbitrary similarity index between them, p (1,i) p represents the i-th image feature of the first frame image p1. (x,i) p represents any non-first frame image x The i-th image feature, and n = 3.

[0048] Specifically, when it is necessary to calculate an arbitrary similarity index, an inter-frame similarity evaluation function is introduced. The expression for this function is: Where n = 3. During the calculation, for the first frame image p1 and any non-first frame image p... xThe image features of each frame are extracted. Specifically, the image features refer to the i-th image feature, where i takes values ​​of 1, 2, or 3, covering several important image attribute dimensions. These image features include hue, brightness, and saturation. This setting allows the inter-frame similarity evaluation function to more accurately measure the similarity between the first frame and any non-first frame image. In the specific calculation, the proportions of the three feature values—hue, brightness, and saturation—are calculated separately. Then, these three results are added together, and the final sum is the similarity index between the first frame image and any non-first frame image. This index comprehensively considers the similarity of images in terms of hue, brightness, and saturation, providing accurate data support for determining the first similarity index. It is crucial for accurately judging the stability of images within a video segment and for constructing a high-quality GIM model.

[0049] In one possible implementation, step S400 further includes:

[0050] Step S410: According to the random sampling modeling mechanism, randomly obtain the first sampling point cloud group in the first point cloud data group.

[0051] Step S420: The first point cloud data group after removing the first sampling point cloud group is denoted as the first verification point cloud group.

[0052] Step S430: Verify the first sampling model constructed based on the first sampling point cloud group according to the first verification point cloud group, and obtain the first verification result.

[0053] Step S440: If the first verification result meets the predetermined verification constraints, the first sampling model is used as the first single model block.

[0054] Specifically, before performing any operations related to building a model based on point cloud data, it is necessary to acquire and preprocess the point cloud data, including noise reduction, registration, and merging. Noise reduction: The Statistical Outlier Removal (SOR) algorithm is used to remove noise. In this algorithm, for a point x to be detected, its k nearest neighbors x' are considered. i Calculate the average distance between the point to be detected and its k nearest neighbors. Here, ‖xx i ‖ represents the relationship between the point to be detected x and its i-th nearest neighbor x. iThe average distance between the points is calculated using the Euclidean distance between them. This average distance is compared to a set threshold, calculated as μ + ασ, where μ is the average distance between all point pairs, σ is the standard deviation, α is an adjustable parameter (usually set to 1.0), and k is typically set to 50. If the average distance between the point to be detected and its k nearest neighbors is greater than μ + ασ, the point is identified as an outlier and removed, thus achieving noise reduction and improving the quality of the point cloud data. Point cloud registration: The Iterative Nearest Point (ICP) algorithm is used for multi-site point cloud registration. The goal of this algorithm is to achieve point cloud alignment by minimizing the distance between corresponding point pairs. Its objective function is... In this formula, R represents the rotation matrix, used to describe the rotational changes of the point cloud in space; t is the translation vector, used to represent the translation amount of the point cloud in space; x i and P i These are corresponding point pairs in two sets of point clouds, N p The number of corresponding point pairs. By continuously adjusting the rotation matrix R and translation vector t, the objective function E(R, t) is minimized, thus achieving point cloud registration. Point cloud merging: The registered multi-site point cloud data are merged into a unified point cloud model, forming a complete point cloud data foundation and providing reliable data support for subsequent modeling work.

[0055] After point cloud data preprocessing, sampling operations are performed on the first point cloud data group according to a pre-set random sampling modeling mechanism. The purpose of the random sampling modeling mechanism is to select representative point cloud samples from the first point cloud data group to ensure that the subsequently constructed model accurately reflects the characteristics of the target power grid. First, a random selection algorithm is activated. This algorithm randomly selects from all point clouds included in the first point cloud data group, with each point cloud having an equal probability of being selected. This avoids sampling bias caused by human selection or fixed rule selection, ensuring the randomness and diversity of the samples. During the random selection process, a certain number of point clouds are selected from the first point cloud data group according to a pre-determined sampling ratio or sampling quantity. These selected point clouds are grouped together to form the first sampled point cloud group. This first sampled point cloud group contains some key feature information of the target power grid within the video segment and serves as the foundation data for subsequently constructing the first sampled model. Through random sampling, the accuracy of the model can be ensured while reducing the amount of data processing and improving the efficiency of model construction.

[0056] After acquiring the first sampled point cloud group, the point cloud data already selected for the first sampled point cloud group is discarded based on this group. The first point cloud data group contains a large amount of point cloud information extracted from the first video footage, which depicts the spatial location, shape, and other features of the target power grid equipment. During the discarding process, points belonging to the first sampled point cloud group are accurately identified by comparing unique identifiers or coordinate information of the point clouds, and these points are removed from the first point cloud data group. After the discarding operation, the remaining point cloud data is re-integrated, and this new set of point cloud data is recorded as the first verification point cloud group. The first verification point cloud group plays an important role; it retains all information from the first point cloud data group except for the sampled portion. It will be used to verify the first sampling model constructed based on the first sampled point cloud group. By comparing the degree of matching between the first verification point cloud group and the first sampling model, the accuracy and reliability of the first sampling model are judged, providing a key basis for subsequently determining the first single model block.

[0057] The first sampling model constructed based on the first sampled point cloud group is validated by sequentially extracting each point cloud data from the first validation point cloud group. For each extracted point cloud, a spatial distance calculation algorithm is used to obtain the first spatial distance from the point cloud to the first sampled model. This distance reflects the proximity of the point cloud to the model surface. The calculated first spatial distance is compared with a pre-set predetermined distance threshold. If the first spatial distance is within the predetermined distance threshold range, it indicates that the point cloud has a high degree of matching with the first sampled model, and this point cloud is added to the matching point cloud set; conversely, if the first spatial distance is not within the predetermined distance threshold range, it indicates that the point cloud has a low degree of matching with the model, and this point cloud is added to the non-matching point cloud set. After all point clouds in the first validation point cloud group have undergone the above operations, the ratio of the number of point clouds in the matching point cloud set to the number of point clouds in the non-matching point cloud set is calculated. This ratio is the first validation result, which intuitively reflects the overall matching degree between the first sampled model and the first validation point cloud group. A higher ratio indicates that the first sampling model can fit the data in the first verification point cloud group well, and the model has high accuracy; conversely, a lower ratio indicates that the model may have biases and needs further optimization or reconstruction.

[0058] The first verification result is rigorously compared with pre-defined verification constraints. These constraints are determined based on various factors, including the accuracy requirements of the target power grid GIM model and the actual application scenario. They define a reasonable range for the matching degree between the first sampling model and the first verification point cloud group. If the first verification result meets the pre-defined verification constraints, it indicates that the first sampling model built based on the first sampling point cloud group has high accuracy and reliability, and can effectively reflect the characteristics and structural information of the target power grid equipment represented by the first point cloud data group. At this point, the first sampling model is formally designated as the first single model block. The first single model block is the basic unit for constructing the complete first single model, containing the key features of the target power grid from a specific perspective and within a specific data range. Subsequently, based on this reliable first single model block, further operations such as fusion and splicing with other single model blocks will be carried out to gradually construct a complete and accurate target power grid GIM model to meet the needs of practical applications such as power grid equipment management, analysis, and planning.

[0059] In one possible implementation, step S400 further includes:

[0060] Step S450: The PMVS algorithm is used to densify the first point cloud data group.

[0061] Specifically, after completing the determination of the first single model block and other related operations, the quality and accuracy of the first point cloud data set are improved by using the PMVS (Patch-based Multi-View Stereo) algorithm for densification. The PMVS algorithm is a powerful multi-view stereo vision algorithm that can fully utilize image information acquired from multiple perspectives to reconstruct the 3D structure of a scene. For the first point cloud data set, the PMVS algorithm first analyzes the geometric relationships and texture features between the multiple images corresponding to the point cloud data. By matching feature points of the same object in different images, the algorithm can determine the accurate location of these points in 3D space. During processing, the PMVS algorithm supplements the sparse point cloud in the first point cloud data set. Based on the existing point cloud, it searches for areas where point clouds may exist but have not yet been detected, and generates new point cloud data in these areas through calculation and inference. These newly generated point clouds complement the original point cloud data, making the entire first point cloud data set denser. After densification processing by the PMVS algorithm, the first point cloud data set can more accurately describe the surface shape and detailed features of the target power grid equipment. This not only helps improve the accuracy and quality of the subsequently constructed GIM model, but also provides richer and more accurate three-dimensional information for the analysis, monitoring and management of power grid equipment, enabling the model to better reflect the actual situation of power grid equipment.

[0062] In one possible implementation, step S430 further includes:

[0063] Step S431: Extract the first point cloud from the first verification point cloud group.

[0064] Step S432: Obtain the first spatial distance from the first point cloud to the first sampling model.

[0065] Step S433: If the first spatial distance is within a predetermined distance threshold, then add the first point cloud to the matching point cloud set; if the first spatial distance is not within the predetermined distance threshold, then add the first point cloud to the non-matching point cloud set.

[0066] Step S434: Take the ratio of the number of points in the matching point cloud set to the number of points in the non-matching point cloud set as the first verification result.

[0067] Specifically, after preparing the first verification point cloud group, detailed analysis begins. The first verification point cloud group consists of the remaining point clouds after removing the first sampling point cloud group from the first point cloud data group. It contains numerous data points used to verify the first sampling model. Following predetermined rules or order, a point cloud is selected from the first verification point cloud group as the first analysis object for this verification operation; this selected point cloud is the first point cloud. This extraction process depends on the order of the point clouds in the data group, such as selecting them sequentially according to their coordinate positions or acquisition times; it may also use random sampling to ensure the randomness and comprehensiveness of the verification. Regardless of the method used, extracting the first point cloud lays the foundation for subsequently calculating the spatial distance between this point cloud and the first sampling model, thereby determining the accuracy of the first sampling model. Through the analysis and processing of the first point cloud, the verification of the entire first verification point cloud group is gradually completed, thus obtaining a reliable first verification result.

[0068] In this step, a spatial distance calculation algorithm is used to obtain the first spatial distance between the first point cloud and the first sampled model. First, the precise coordinates of the first point cloud in 3D space are determined. These coordinates are typically represented in the form of (x, y, z), which uniquely identifies the first point cloud's location in space. Simultaneously, for the first sampled model, the geometric features of the model surface and their corresponding data representation are determined based on its construction method and data structure. Then, using a distance calculation algorithm, such as a 3D extension of the Euclidean distance algorithm, the coordinates of the first point cloud are compared with those of points on the surface of the first sampled model. During the calculation, the algorithm considers the distances from the first point cloud to all possible locations on the model surface, ultimately determining the shortest distance from the first point cloud to the first sampled model as the first spatial distance. This distance directly reflects the spatial proximity of the first point cloud and the first sampled model, serving as crucial data for subsequent assessments of their matching and providing important quantitative evidence for the entire model validation process.

[0069] After obtaining the first spatial distance between the first point cloud and the first sampling model, this first spatial distance is compared with a pre-set predetermined distance threshold. The predetermined distance threshold is determined comprehensively based on factors such as the accuracy requirements of the target power grid GIM model, the actual application scenario, and past experience data. It is an important standard for judging the degree of matching between the first point cloud and the first sampling model. If the first spatial distance is within the predetermined distance threshold range, it means that the relative positional relationship between the first point cloud and the first sampling model in spatial location meets expectations, indicating that the actual spatial information represented by the first point cloud matches the spatial structure described by the first sampling model relatively well. In this case, the first point cloud is added to the matching point cloud set. The matching point cloud set is used to collect point clouds with a high degree of matching with the first sampling model. These point clouds help prove that the first sampling model can accurately reflect the actual point cloud data characteristics to a certain extent. Conversely, if the first spatial distance is not within the predetermined distance threshold range (i.e., greater than or less than this threshold), it indicates that the matching degree between the first point cloud and the first sampling model is low. This is because the construction of the first sampling model at this location has deviated, or the first point cloud represents a special area not covered by the model. In this case, the first point cloud is added to the non-matching point cloud set. The non-matching point cloud set is used to record the point cloud that does not match the first sampling model. By analyzing the non-matching point cloud set, problems in the first sampling model can be found, providing a basis for subsequent optimization and adjustment of the model.

[0070] After classifying the first set of point clouds (adding them to either the matching or non-matching point cloud set), the number of point clouds in both sets is counted. Point clouds in the matching set represent those with a high degree of matching to the first sampling model; their quantity reflects the extent to which the first sampling model accurately fits the actual point cloud data. Point clouds in the non-matching set represent those with a poorer match to the first sampling model; their quantity reflects the size of the model's bias or uncovered areas. After accurately counting the number of point clouds in each set, the number of point clouds in the matching set is divided by the number of point clouds in the non-matching set, and the ratio is calculated. This ratio is the first verification result, which quantitatively and intuitively presents the overall matching status between the first sampling model and the first verification point cloud set. If the ratio is large, it indicates that the number of point clouds in the matching point cloud set is much larger than that in the unmatched point cloud set, meaning that the first sampling model can better reflect the actual data characteristics contained in the first verification point cloud set, and the model has high accuracy and reliability. Conversely, if the ratio is small, it indicates that the number of point clouds in the unmatched point cloud set is large, meaning that the first sampling model has many mismatches with the actual data, and the model may need to be rebuilt or optimized in order to build a target power grid GIM model that better meets the actual needs.

[0071] Example 2, based on the same inventive concept as the point cloud-based automatic reverse generation method for GIM models in the aforementioned examples, such as... Figure 2 As shown, this application provides an automatic reverse engineering system for GIM models based on point clouds. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0072] The multiple video segment acquisition module 10 is used to traverse and inspect the target power grid to obtain the target inspection video, and to virtually segment the target inspection video to obtain multiple video segments.

[0073] The first similarity index acquisition module 20 is used to perform inter-frame similarity analysis on the first video segment among the plurality of video segments to obtain the first similarity index.

[0074] The target segmentation result acquisition module 30 is used to perform shot recognition and segmentation on the first video segment if the first similarity index does not meet a predetermined similarity threshold, so as to obtain a target segmentation result, wherein the target segmentation result includes multiple shot videos.

[0075] The first single model block acquisition module 40 is used to construct a first point cloud data group based on the first shot video in the multiple shot videos, and to perform modeling and analysis on the first point cloud data group according to the random sampling modeling mechanism to obtain the first single model block.

[0076] The first single model building module 50 is used to build a first single model based on the first single model block and match the first device corresponding to the first single model, wherein the first device has first parameterized information.

[0077] The target GIM model acquisition module 60 is used to render the first parameterized information to the first single model to obtain the target GIM model of the target power grid.

[0078] Furthermore, the system is also used to implement the following functions:

[0079] Obtain a unit frame count, wherein the unit frame count is greater than 2 and less than the total number of frames in the target inspection video; perform virtual segmentation on the target inspection video based on the unit frame count to obtain the multiple video segments.

[0080] Furthermore, the system is also used to implement the following functions:

[0081] Obtain the first frame image in the first video segment; calculate the similarity between the first frame image and any non-first frame image in the first video segment to obtain an arbitrary similarity index; take the mean of the arbitrary similarity indices and denot it as the first similarity index.

[0082] Furthermore, the system is also used to implement the following functions:

[0083] An inter-frame similarity evaluation function is introduced, and the similarity between the first frame image and any non-first frame image in the first video segment is calculated based on the inter-frame similarity evaluation function to obtain the arbitrary similarity index; wherein, the expression of the inter-frame similarity evaluation function is: f(p1,p x ) represents the first frame image p1 and any non-first frame image p x The arbitrary similarity index between them, p (1,i) p represents the i-th image feature of the first frame image p1. (x,i) p represents any non-first frame image x The i-th image feature, and n = 3.

[0084] Furthermore, the system is also used to implement the following functions:

[0085] According to the random sampling modeling mechanism, a first sampling point cloud group is randomly obtained from the first point cloud data group; the first point cloud data group after removing the first sampling point cloud group is denoted as the first verification point cloud group; the first sampling model constructed based on the first sampling point cloud group is verified according to the first verification point cloud group to obtain a first verification result; if the first verification result meets the predetermined verification constraints, the first sampling model is used as the first single model block.

[0086] Furthermore, the system is also used to implement the following functions:

[0087] The PMVS algorithm is used to densify the first point cloud data group.

[0088] Furthermore, the system is also used to implement the following functions:

[0089] Extract the first point cloud from the first verification point cloud group; obtain the first spatial distance from the first point cloud to the first sampling model; if the first spatial distance is within a predetermined distance threshold, add the first point cloud to the matching point cloud set; if the first spatial distance is not within the predetermined distance threshold, add the first point cloud to the non-matching point cloud set; take the ratio of the number of points in the matching point cloud set to the number of points in the non-matching point cloud set as the first verification result.

[0090] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0091] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0092] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for automatic reverse generation of GIM models based on point clouds, characterized in that, include: The target power grid is traversed and inspected to obtain target inspection videos, and the target inspection videos are virtually segmented to obtain multiple video segments; Inter-frame similarity analysis is performed on the first video segment among the plurality of video segments to obtain a first similarity index; If the first similarity index does not meet the predetermined similarity threshold, then the first video segment is subjected to shot recognition and segmentation to obtain the target segmentation result, wherein the target segmentation result includes multiple shot videos; A first point cloud data group is constructed based on the first shot video from the multiple shot videos, and the first point cloud data group is modeled and analyzed according to the random sampling modeling mechanism to obtain the first single model block; A first single model is constructed based on the first single model block, and a first device corresponding to the first single model is matched, wherein the first device has first parameterized information; The first parameterized information is rendered into the first single model to obtain the target GIM model of the target power grid.

2. The automatic reverse generation method for GIM models based on point clouds according to claim 1, characterized in that, A target power grid is traversed and inspected to obtain target inspection video, and the target inspection video is virtually segmented to obtain multiple video segments, including: Obtain the unit frame count, wherein the unit frame count is greater than 2 and less than the total number of frames in the target inspection video; The target inspection video is virtually segmented based on the unit frame number to obtain the multiple video segments.

3. The automatic reverse generation method for GIM models based on point clouds according to claim 1, characterized in that, Inter-frame similarity analysis is performed on the first video segment among the plurality of video segments to obtain a first similarity index, including: Obtain the first frame image from the first video segment; The similarity index is obtained by calculating the similarity between the first frame image and any non-first frame image in the first video segment. The average of any of the similarity indices is taken and denoted as the first similarity index.

4. The automatic reverse generation method for GIM models based on point clouds according to claim 3, characterized in that, The similarity index is calculated by comparing the first frame image with any non-first frame image in the first video segment, including: An inter-frame similarity evaluation function is introduced, and the similarity between the first frame image and any non-first frame image in the first video segment is calculated based on the inter-frame similarity evaluation function to obtain the arbitrary similarity index; The expression for the inter-frame similarity evaluation function is as follows: f(p1,p x ) represents the first frame image p1 and any non-first frame image p x The arbitrary similarity index between them, p (1,i) p represents the i-th image feature of the first frame image p1. (x,i) p represents any non-first frame image x The i-th image feature, and n = 3.

5. The automatic reverse generation method for GIM models based on point clouds according to claim 1, characterized in that, A first point cloud data group is constructed based on the first shot video from the plurality of video shots, and the first point cloud data group is modeled and analyzed according to a random sampling modeling mechanism to obtain a first single model block, including: According to the random sampling modeling mechanism, the first sampling point cloud group in the first point cloud data group is randomly obtained; The first point cloud data group after removing the first sampling point cloud group is denoted as the first verification point cloud group; The first sampling model constructed based on the first sampling point cloud group is verified according to the first verification point cloud group to obtain the first verification result; If the first verification result meets the predetermined verification constraints, the first sampling model is used as the first single model block.

6. The automatic reverse generation method for GIM models based on point clouds according to claim 5, characterized in that, After constructing a first point cloud data group based on the first shot video from the plurality of shot videos, the method further includes: using the PMVS algorithm to densify the first point cloud data group.

7. The automatic reverse generation method for GIM models based on point clouds according to claim 5, characterized in that, The first sampling model constructed based on the first sampling point cloud group is validated according to the first validation point cloud group to obtain a first validation result, including: Extract the first point cloud from the first verification point cloud group; Obtain the first spatial distance from the first point cloud to the first sampling model; If the first spatial distance is within a predetermined distance threshold, the first point cloud is added to the matching point cloud set; if the first spatial distance is not within the predetermined distance threshold, the first point cloud is added to the non-matching point cloud set. The ratio of the number of points in the matching point cloud set to the number of points in the non-matching point cloud set is taken as the first verification result.

8. A point cloud-based automatic reverse engineering system for GIM models, characterized in that, The system is used to implement the automatic reverse generation method for GIM models based on point clouds as described in any one of claims 1-7, and the system comprises: A multiple video segment acquisition module is used to traverse and inspect the target power grid to obtain target inspection videos, and to virtually segment the target inspection videos to obtain multiple video segments; The first similarity index acquisition module is used to perform inter-frame similarity analysis on the first video segment among the plurality of video segments to obtain the first similarity index; The target segmentation result acquisition module is used to perform shot recognition and segmentation on the first video segment if the first similarity index does not meet a predetermined similarity threshold, so as to obtain a target segmentation result, wherein the target segmentation result includes multiple shot videos; The first single model block acquisition module is used to construct a first point cloud data group based on the first shot video in the multiple shot videos, and to perform modeling and analysis on the first point cloud data group according to the random sampling modeling mechanism to obtain the first single model block; The first single model building module is used to build a first single model based on the first single model block and match the first device corresponding to the first single model, and the first device has first parameterized information; The target GIM model acquisition module is used to render the first parameterized information to the first single model to obtain the target GIM model of the target power grid.

Images