Point cloud-based device size measurement method, computer equipment and storage media

Abstract

This application relates to the field of equipment measurement technology, and more particularly to a point cloud-based method for measuring equipment dimensions, a computer device, and a storage medium. This application effectively reduces noise and redundant information and improves data quality by identifying and deleting outlier data points. Downsampling is performed on the target point cloud data after deleting outlier data points, reducing the number of data points, lowering the computational complexity of subsequent processing, and improving overall processing efficiency. Registration processing is performed on the downsampled point cloud data, unifying point cloud data collected from different perspectives or at different times into the same coordinate system, enhancing data consistency and comparability, and providing an accurate benchmark for subsequent feature extraction and dimension calculation. The registered point cloud data is segmented, and key feature points are extracted based on the segmented point cloud data. Finally, based on the extracted key feature points, a refined measurement of the equipment's geometry is achieved.

Description

Technical Field

[0001] This application relates to the field of equipment measurement technology, and in particular to a point cloud-based method for measuring equipment dimensions, a computer device, and a storage medium. Background Technology

[0002] Power equipment, such as substation equipment, is a crucial component of the power system. Accurate measurement of its geometric dimensions is essential for equipment maintenance, repair, upgrades, and the design and installation of new equipment. With the rapid development of 3D scanning technology, especially the application of lidar scanning technology, new solutions have been provided for equipment dimension measurement. LiDAR scanning equipment can quickly acquire point cloud data of the equipment surface. This point cloud data contains the geometric information of the equipment surface, enabling the measurement of the equipment's geometric dimensions.

[0003] However, during the point cloud data acquisition process, due to environmental factors (such as light, wind, and equipment surface material), the acquired point cloud data often contains a large amount of noise and redundant information, which seriously affects the accuracy and efficiency of subsequent processing. Summary of the Invention

[0004] Based on this, a point cloud-based device size measurement method, computer equipment, and storage medium are proposed, aiming to solve the technical problems of poor accuracy and low efficiency in existing laser point cloud-based device size measurement.

[0005] A first aspect of this application provides a point cloud-based method for measuring device dimensions, the method comprising:

[0006] Identify outlier data points in the laser point cloud data of the device under test;

[0007] All outlier data points are removed from the laser point cloud data to obtain the target point cloud data;

[0008] The target point cloud data is downsampled to obtain sampled point cloud data;

[0009] The sampled point cloud data is registered to obtain registered point cloud data;

[0010] The registered point cloud data is segmented;

[0011] Extract key feature points from the segmented point cloud data;

[0012] The geometric dimensions of the device under test are calculated based on the key feature points.

[0013] Optionally, before identifying outlier data points in the laser point cloud data of the device under test, the method further includes:

[0014] Acquire multi-angle laser point cloud data of the device under test;

[0015] Obtain common feature points from the multi-angle laser point cloud data;

[0016] The multi-angle laser point cloud data are fused based on the common feature points to obtain the laser point cloud data of the device under test.

[0017] Optionally, the outlier data points in the laser point cloud data of the device under test include:

[0018] For each data point in the laser point cloud data, determine the target neighborhood of the data point;

[0019] Obtain the average distance between the data point and all data points in the corresponding target neighborhood;

[0020] The average distance is compared with a preset average distance threshold to obtain the comparison result;

[0021] The average distance and a preset average distance threshold are used to identify whether the corresponding data point is an outlier.

[0022] Optionally, the downsampling process of the target point cloud data to obtain sampled point cloud data includes:

[0023] Initialize a three-dimensional voxel mesh, the size of which is determined according to a preset downsampling resolution;

[0024] Iterate through all data points in the target point cloud data and assign each data point to its corresponding voxel grid:

[0025] For each voxel grid, retain one representative data point;

[0026] All the retained representative data points are combined to form a new point cloud data, resulting in sampled point cloud data.

[0027] Optionally, the registration process for the sampled point cloud data to obtain registered point cloud data includes:

[0028] One data point is selected from the sampled point cloud data as a reference data point, and the data points in the sampled point cloud data other than the reference data point are used as data points to be registered.

[0029] The reference data points and each of the data points to be registered are combined into a data point pair;

[0030] For each pair of data points, initialize a transformation matrix;

[0031] Determine the target point pair set from the plurality of data point pairs;

[0032] An error function is calculated based on the target point pair set, and the error function represents the deviation between the data point to be registered and the reference data point after transformation by the transformation matrix.

[0033] The transformation matrix is ​​iteratively updated to minimize the error function;

[0034] When the iteration meets the preset iteration stopping condition, the target transformation matrix is ​​obtained;

[0035] The target transformation matrix is ​​used to register the data points to be registered.

[0036] Optionally, segmenting the registered point cloud data includes:

[0037] Select an unprocessed data point from the registered point cloud data as the initial seed point;

[0038] Iterate through each neighboring point of the initial seed point and calculate the spatial distance and normal direction difference between the neighboring point and the current point in the region;

[0039] When the spatial distance and normal direction difference between the adjacent points are both less than a preset threshold, the adjacent points are added to the current region and marked as processed.

[0040] Iterate through the neighboring points of the newly added point until no new neighboring points are added to the current region, then segment the current region.

[0041] Optionally, extracting the key feature points from the segmented point cloud data includes:

[0042] The segmented point cloud data is standardized to obtain standardized point cloud data.

[0043] Generate a covariance matrix based on the standardized point cloud data;

[0044] Calculate the eigenvalues ​​and corresponding eigenvectors of the covariance matrix;

[0045] Select the k largest eigenvalues ​​from the eigenvalues;

[0046] Obtain the target feature vectors corresponding to the top k largest feature values;

[0047] The key feature points are obtained based on the preset mapping matrix and the target feature vector.

[0048] Optionally, the method further includes:

[0049] Based on the geometric dimensions and the key feature points, a three-dimensional model of the device under test is generated.

[0050] A second aspect of this application provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the point cloud-based device size measurement method.

[0051] A third aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the point cloud-based device size measurement method.

[0052] This application first identifies and removes outlier data points from the laser point cloud data of the device under test, resulting in cleaner and more accurate target point cloud data. Removing outlier data points effectively reduces noise and redundant information, improving data quality and thus increasing the accuracy of device geometric dimension measurement. Downsampling of the target point cloud data reduces the number of data points, lowers the computational complexity of subsequent processing, and improves overall processing efficiency. Registration processing is then performed on the downsampling point cloud data, unifying point cloud data collected from different perspectives or at different times into the same coordinate system, enhancing data consistency and comparability, and providing an accurate benchmark for subsequent feature extraction and dimension calculation. The registered point cloud data is then segmented, and key feature points are extracted based on the segmented point cloud data. Finally, based on the extracted key feature points, a refined measurement of the device's geometric shape is achieved. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a schematic flowchart of the device size measurement method based on point cloud provided in the embodiments of this application.

[0055] Figure 2 This is a functional block diagram of the point cloud-based device size measurement device provided in the embodiments of this application.

[0056] Figure 3 This is a schematic diagram of the structure of the computer device provided in the embodiments of this application. Detailed Implementation

[0057] 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 some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0058] Figure 1 This is a flowchart illustrating a point cloud-based device size measurement method provided in an embodiment of this application. The point cloud-based device size measurement method includes the following steps.

[0059] S11, identify outlier data points in the laser point cloud data of the device under test.

[0060] In this context, the device under test (DUT) refers to a device or object whose geometric dimensions need to be measured. In the context of laser scanning measurement, the DUT can be any object whose three-dimensional shape, size, or surface feature information needs to be acquired, such as mechanical equipment, building structures, automotive parts, or works of art. In power facility applications, the DUT can be power equipment, such as substation equipment.

[0061] Laser point cloud data is a collection of three-dimensional coordinate points obtained by scanning the surface of a device under test using a lidar device. Laser point cloud data contains information about the shape, size, and features of the device's surface. Each point cloud element contains three-dimensional coordinate (XYZ) information.

[0062] Because the shape and structure of equipment are often complex, scanning from a single angle is often insufficient to fully capture its features. Therefore, high-precision and high-resolution LiDAR scanning equipment can be used to perform omnidirectional scanning of the equipment under test. Omnidirectional scanning can acquire point cloud data of the equipment under test from various angles (multi-angle LiDAR point cloud data), thereby providing a more comprehensive understanding of the shape and structure of the equipment under test.

[0063] In an optional implementation, for multi-angle laser point cloud data, a feature matching-based algorithm can be used to fuse point cloud data acquired from different angles into a complete point cloud dataset. Specifically, common feature points can be obtained from the multi-angle laser point cloud data, and the multi-angle laser point cloud data can be fused based on the common feature points to obtain the laser point cloud data of the device under test.

[0064] In one optional implementation, feature point extraction algorithms (such as SURF, SIFT, etc.) are used to extract feature points, such as corner points and edge points, from the point cloud data at each angle. These feature points are easily identifiable and matched in data from different angles. For feature points in point cloud data from different angles, the distance or similarity between feature points is calculated. Matching feature points are then found from the extracted feature points based on the distance or similarity; these matching feature points are the common feature points. Based on the common feature points, a transformation matrix is ​​calculated to convert the point cloud data from different angles to a unified coordinate system, ensuring that the data from different angles can be accurately stitched together. Applying the transformation matrix to convert the point cloud data from different angles to a unified coordinate system ensures that the point cloud data from all angles are in the same coordinate system, providing a foundation for subsequent data fusion. The point cloud data from different angles are then fused within the unified coordinate system.

[0065] After obtaining the laser point cloud data of the device under test (DUT), it is necessary to identify outlier data points. Outlier data points are those that differ significantly from the majority of the point cloud data in terms of spatial location, reflection intensity, or distribution characteristics. These points may be caused by noise, equipment errors, environmental interference, or abnormal characteristics of the DUT itself. Identifying and removing outlier data points can improve the quality of the point cloud data. Furthermore, the presence of outlier data points can affect the visualization of the point cloud data, causing unevenness, breaks, and other artifacts on the model surface. Removing outlier data points can optimize the visualization of the point cloud data, making the model smoother and more continuous, thus improving its readability and aesthetics.

[0066] In an optional implementation, the outlier data points in the laser point cloud data of the device under test include:

[0067] For each data point in the laser point cloud data, determine the target neighborhood of the data point;

[0068] Obtain the average distance between the data point and all data points in the corresponding target neighborhood;

[0069] The average distance is compared with a preset average distance threshold to obtain the comparison result;

[0070] The average distance and a preset average distance threshold are used to identify whether the corresponding data point is an outlier.

[0071] For each data point P in the laser point cloud data, a target neighborhood needs to be determined. The target neighborhood consists of the K nearest neighbors of the data point P. The value of K can be chosen based on the density and noise level of the point cloud data. The purpose of determining the neighborhood is to analyze the relationship between the data point and its surrounding data points, thereby determining whether the data point is an outlier.

[0072] After determining the target neighborhood of each data point, the distances between data point P and all data points within its target neighborhood are calculated, thus yielding the average distance d. The average distance reflects how closely the data point is connected to its surrounding data points. A large average distance indicates that the data point is far from its neighbors and may be an outlier.

[0073] After calculating the average distance for each data point, each average distance needs to be compared with a preset average distance threshold. The average distance threshold is usually determined based on the overall distribution characteristics of the point cloud data. It can be the median of the average distances for all data points plus or minus a certain standard deviation, or it can be obtained through other methods. The purpose of the comparison is to determine whether the average distance for a given data point exceeds the normal range, thereby determining whether the data point is an outlier.

[0074] Finally, based on the comparison results, it can be determined whether each data point is an outlier. If the average distance of a data point exceeds a preset average distance threshold, then the data point is considered an outlier and needs to be removed from the point cloud data. If the average distance of a data point is within the preset average distance threshold, then the data point is considered a normal data point and needs to be retained in the point cloud data.

[0075] In one optional implementation, the median μ and standard deviation σ of the average distances corresponding to all data points can be calculated. It is then determined whether the average distance d corresponding to data point P satisfies the condition |d-μ|>λ*σ, where λ is a constant (e.g., 2 or 3) used to control the sensitivity for identifying outliers. Using the median μ and standard deviation σ as statistical benchmarks allows for automatic adaptation to the characteristics of different datasets. Whether it's the density, distribution pattern, or noise level of data points, the calculated μ and σ can be adaptively adjusted to ensure the accuracy and robustness of outlier identification. Introducing the constant λ allows for flexible control of the sensitivity to outlier identification. A larger value of λ results in a wider range of identified outliers and a lower tolerance for noise and outliers; conversely, a smaller value of λ results in a narrower range of identified outliers and a higher tolerance for noise and outliers. This method is based on statistical measures and does not require complex iterative calculations or optimization processes. Only the median μ and standard deviation σ need to be calculated once, followed by a single comparison of the average distance for each data point.

[0076] The aforementioned optional implementation method, by determining a target neighborhood for each data point, calculating its average distance to other data points within that neighborhood, and comparing this average distance to a preset average distance threshold, can accurately identify which data points are outliers. This judgment based on the spatial relationships between data points can effectively remove noise and outliers from point cloud data, improving the accuracy of data processing. By identifying and removing outliers in advance, computational load can be reduced, and processing efficiency improved. Especially for large-scale point cloud data, this can significantly shorten processing time and improve overall work efficiency.

[0077] S12, remove all outlier data points from the laser point cloud data to obtain the target point cloud data.

[0078] After removing all outlier data points from the laser point cloud data, the target point cloud data is obtained. Removing outlier data points effectively removes noise and outliers from the point cloud data, improving data quality. The target point cloud data is smoother and more continuous, better reflecting the true shape and structure of the device under test.

[0079] S13, perform downsampling processing on the target point cloud data to obtain sampled point cloud data.

[0080] Target point cloud data typically contains a large number of data points, leading to a massive computational burden and slow processing speed for subsequent processing (such as dimensional measurement, 3D reconstruction, object recognition, and path planning). Downsampling the target point cloud data can significantly reduce the number of data points, thereby reducing computational complexity and improving processing efficiency. Reducing the number of data points also effectively lowers storage requirements and saves storage space.

[0081] In an optional implementation, the downsampling process of the target point cloud data to obtain sampled point cloud data includes:

[0082] Initialize a three-dimensional voxel mesh, the size of which is determined according to a preset downsampling resolution;

[0083] Iterate through all data points in the target point cloud data and assign each data point to its corresponding voxel grid:

[0084] For each voxel grid, retain one representative data point;

[0085] All the retained representative data points are combined to form a new point cloud data, resulting in sampled point cloud data.

[0086] The size of the 3D voxel mesh, i.e., the size of each voxel, is determined based on the preset downsampling resolution. The downsampling resolution determines the degree of downsampling. Higher resolution means smaller voxels, preserving more detail in the point cloud data; lower resolution means larger voxels, resulting in sparser point cloud data.

[0087] The target point cloud data is mapped onto a 3D voxel grid. Specifically, all data points of the target point cloud data are traversed, and each data point is assigned to a corresponding 3D voxel grid based on its coordinates. This ensures that subsequent aggregation of data points and selection of representative data points can be performed based on the voxel grid.

[0088] For each voxel grid containing data points, a representative data point is selected to represent all data points within that voxel grid. The representative data point can be the centroid (average value) of all data points within the voxel grid, or the data point closest to the center of the voxel grid. The purpose of selecting a representative data point is to reduce the number of data points while preserving as much of the overall shape and key features of the point cloud data as possible.

[0089] The representative data points in all voxel grids are collected to form new point cloud data, which is the downsampled sampled point cloud data.

[0090] In the above optional implementation, by using voxel grid downsampling, a large number of data points in the target point cloud data are effectively compressed into a representative data point in each voxel grid, thereby significantly reducing the number of data points. This reduction in data points directly lowers the computational complexity of subsequent processing and improves processing speed. Although the number of data points is reduced, by rationally selecting representative data points, the overall shape and key feature information of the point cloud can be preserved.

[0091] S14, the sampled point cloud data is registered to obtain registered point cloud data.

[0092] During the fusion of multi-angle point cloud data, although the coordinates have been unified, downsampling is performed after fusion. However, during downsampling, factors such as the selection of representative data points and the size of the voxel mesh may introduce errors or distortions. These errors can cause inconsistencies in local areas of the originally unified coordinate point cloud data. By performing registration processing on the sampled point cloud data, the consistency of the sampled point cloud data can be further ensured globally, reducing the impact of errors caused by downsampling and ensuring data integrity and consistency.

[0093] In an optional implementation, the registration process of the sampled point cloud data to obtain registered point cloud data includes:

[0094] One data point is selected from the sampled point cloud data as a reference data point, and the data points in the sampled point cloud data other than the reference data point are used as data points to be registered.

[0095] The reference data points and each of the data points to be registered are combined into a data point pair;

[0096] For each pair of data points, initialize a transformation matrix;

[0097] Determine the target point pair set from the plurality of data point pairs;

[0098] An error function is calculated based on the target point pair set, and the error function represents the deviation between the data point to be registered and the reference data point after transformation by the transformation matrix.

[0099] The transformation matrix is ​​iteratively updated to minimize the error function;

[0100] When the iteration meets the preset iteration stopping condition, the target transformation matrix is ​​obtained;

[0101] The target transformation matrix is ​​used to register the data points to be registered.

[0102] One data point is selected from the sampled point cloud data as a reference data point (reference point cloud). All data points in the sampled point cloud data other than the reference point cloud are used as data points to be registered (point clouds to be registered). The reference point cloud will serve as a fixed reference in the subsequent registration process.

[0103] Each selected reference data point is paired with one of the data points to be registered, forming a data point pair. The number of data point pairs corresponds to the number of data points to be registered. For each data point pair, the transformation matrix is ​​initialized as an identity matrix, assuming that the point cloud to be registered and the reference point cloud are initially completely superimposed. In subsequent iterations, the transformation matrix is ​​updated to minimize the error.

[0104] Find the nearest point pairs between the reference point cloud and the point cloud to be registered to form a target point pair set. To accelerate the matching process, data structures such as KD-trees can be used for nearest neighbor search.

[0105] Calculate the error function based on the target point pair set. The error function is usually expressed as the sum of squares of the point pair distances between the transformed point cloud to be registered and the reference point cloud.

[0106] A least squares optimization algorithm is employed to iteratively update the transformation matrix (including rotation and translation vectors) to minimize the error function. After each iteration, the point cloud to be registered is transformed according to the new transformation matrix.

[0107] When the error function converges to below a preset threshold or reaches a preset number of iterations, the iteration stops, and the final transformation matrix, i.e., the target transformation matrix, is obtained.

[0108] The final transformation matrix is ​​applied to transform the point cloud to be registered, aligning it with the reference point cloud. This process continues until all point clouds to be registered are aligned with the reference point cloud, resulting in complete point cloud data in a unified coordinate system.

[0109] In the above optional implementation, by selecting a reference data point and using other data points as data points to be registered, forming data point pairs with the reference data point for registration, the relative positional relationship between each data point to be registered and the reference data point can be accurately calculated. By iteratively updating the transformation matrix to minimize the error function, the accuracy of the registration result can be ensured. By determining the target point pair set from multiple data point pairs and calculating the error function based on the target point pairs in the target point pair set, the impact of noise and outliers on the registration result can be reduced, thereby improving the robustness of the registration process. Furthermore, since any data point can be selected as the reference data point, and the iteration stopping condition can be adjusted as needed, it offers high flexibility and can adapt to point cloud registration tasks with different scenarios and requirements.

[0110] S15, the registered point cloud data is segmented.

[0111] Segmenting registered point cloud data refers to dividing a pre-registered point cloud dataset into multiple parts or subsets according to certain rules or algorithms. Automated point cloud segmentation algorithms based on region growing can be used to segment registered point cloud data. These algorithms determine how to segment the point cloud into different parts based on local geometric features such as normal vectors and curvature.

[0112] Different parts of a device may have different geometric features and properties. Segmentation allows for specific feature extraction for each part, thus more accurately describing and representing the various components of the device under test, providing an accurate foundation for subsequent feature extraction and dimensional calculations. Furthermore, for large or complex point cloud data, direct processing of the entire dataset can be very time-consuming and computationally intensive. Segmentation breaks down point cloud data into smaller, more manageable parts without requiring manipulation of the entire dataset, thereby improving processing efficiency.

[0113] In one specific implementation, segmenting the registered point cloud data includes:

[0114] Select an unprocessed data point from the registered point cloud data as the initial seed point;

[0115] Iterate through each neighboring point of the initial seed point and calculate the spatial distance and normal direction difference between the neighboring point and the current point in the region;

[0116] When the spatial distance and normal direction difference between the adjacent points are both less than a preset threshold, the adjacent points are added to the current region and marked as processed.

[0117] Iterate through the neighboring points of the newly added point until no new neighboring points are added to the current region, then segment the current region.

[0118] An unprocessed data point can be randomly selected from the registered point cloud data, or selected based on a certain strategy, as the initial seed point. A new empty region is initialized to store the points belonging to that region. Using the initial seed point as the center, its neighboring points are traversed (usually determined by spatial distance). For each neighboring point, its spatial distance and normal direction difference with points in the current region are calculated. If both the spatial distance and normal direction difference of a neighboring point are less than a preset threshold, the neighboring point is added to the current region to achieve region growth and marked as processed. The traversal and judgment of neighboring points continues, using the newly added point as the center. When no new neighboring point can be added to the current region (i.e., all neighboring points do not meet the addition conditions or have all been processed), the region growth process terminates. The current region is marked as an independent device part, and relevant information is stored.

[0119] Find the next unprocessed data point in the registered point cloud data and use it as a new seed point. Repeat the above region growth and termination process until all data points in the registered point cloud data have been processed, meaning each data point belongs to a specific device part.

[0120] The segmented device portions undergo further processing, such as noise removal and boundary smoothing, to improve segmentation accuracy. Depending on the needs, the segmentation results are visualized or stored in a specific file format. Through these steps, the registered point cloud data is effectively segmented into multiple regions representing different device portions, providing an accurate foundation for subsequent feature extraction, geometric dimension calculation, and further measurement and modeling work.

[0121] Existing automated point cloud segmentation algorithms often struggle to accurately segment various parts of substation equipment with complex shapes and similar structures, leading to errors in subsequent processing. This application proposes a method to randomly or strategically select an unprocessed data point as the initial seed point. This flexibility allows the algorithm to better select the starting point in complex scenarios, avoiding segmentation errors caused by improper initial seed point selection. During region growing, this application considers not only the spatial distance between adjacent points but also the difference in normal direction as a criterion. This comprehensive evaluation method more accurately reflects the similarity of local geometric features between data points, effectively distinguishing different parts in equipment with complex shapes and similar structures. To adapt to equipment with different shapes and structures, the thresholds for spatial distance and normal direction difference can be automatically adjusted based on the local features of the point cloud or the segmentation progress to further improve segmentation accuracy. This application emphasizes traversing and judging adjacent points centered on newly added points. This meticulous traversal method ensures that each point is fully considered, avoiding omissions or incorrect segmentation. By continuously searching for the next unprocessed data point as a new seed point and repeating the region growth and termination process, this application ensures that each point in the point cloud data is correctly classified into a certain device part, thereby achieving comprehensive segmentation.

[0122] S16, extract key feature points from the segmented point cloud data.

[0123] The boundary and geometric feature points of the device under test (DUT) are extracted from the segmented point cloud data. Edge detection algorithms can be used to process the segmented point cloud data. Edge detection algorithms can identify the edge contours of the DUT surface, i.e., the boundary lines of the DUT's shape. Boundary lines help in understanding the overall shape and contour of the DUT.

[0124] Based on the shape characteristics of the device under test (DUT), representative key feature points are identified. Key feature points may include vertices (such as the highest or lowest point of the DUT), inflection points (such as points where the shape of the DUT changes significantly), and intersection points (such as the junctions of different parts or components of the device). Geometric feature points can more accurately describe the shape and structure of the DUT.

[0125] By extracting boundary and geometric feature points, key shape information of the device under test can be obtained from the segmented point cloud data, providing accurate basic data for subsequent processing and analysis. These feature points not only help in understanding the overall structure and shape of the device, but also provide important reference for subsequent tasks such as dimensional measurement, 3D modeling, and motion simulation.

[0126] In an optional implementation, extracting the key feature points of the device under test from the segmented point cloud data includes:

[0127] The segmented point cloud data is standardized to obtain standardized point cloud data.

[0128] Generate a covariance matrix based on the standardized point cloud data;

[0129] Calculate the eigenvalues ​​and corresponding eigenvectors of the covariance matrix;

[0130] Select the k largest eigenvalues ​​from the eigenvalues;

[0131] Obtain the target feature vectors corresponding to the top k largest feature values;

[0132] The key feature points of the device under test are obtained based on the preset mapping matrix and the target feature vector.

[0133] The segmented point cloud data is standardized to eliminate the dimensional differences between different dimensions (such as X, Y, and Z coordinates), making the data comparable.

[0134] The covariance matrix is ​​a matrix that describes the correlation between different dimensions of data, reflecting the spatial relationship between data points in point cloud data.

[0135] Eigenvalues ​​represent the magnitude of the variance of data in various directions, while eigenvectors indicate these directions. By calculating eigenvalues ​​and eigenvectors, the most important (i.e., the directions with the largest variance) in the data can be identified; these directions correspond to the main features of the data.

[0136] The k most important features are selected based on their magnitude. Larger eigenvalues ​​indicate more dramatic data changes along the direction represented by the corresponding eigenvector, meaning they contain more information. By selecting the k largest eigenvalues, the most crucial information in the data can be retained while removing noise and redundant information, thus achieving dimensionality reduction.

[0137] Obtain the eigenvectors corresponding to the selected k largest eigenvalues. These eigenvectors form the basis vectors of the data in the new feature space. The target eigenvectors are the basis of the new coordinate system after dimensionality reduction, defining how the data is represented in the new space.

[0138] The mapping matrix W is typically composed of selected target feature vectors, used to map the original d-dimensional data to a new k-dimensional feature space. Through the mapping matrix, the representation of the device under test in the dimensionality-reduced feature space, i.e., key feature points, can be obtained. This not only achieves dimensionality reduction but also preserves key feature information, helping to reduce the computational load of subsequent processing while improving accuracy and efficiency. Key feature points, as the data representation after dimensionality reduction, provide a more compact and efficient data representation for subsequent size calculations and model construction.

[0139] The above-described optional implementation methods, by standardizing the segmented point cloud data, can eliminate noise and outliers in the data, improving data quality and consistency. The covariance matrix generated based on the standardized point cloud data can capture the relationships between different feature points in the dataset, facilitating more accurate extraction of key feature points. By calculating the eigenvalues ​​and corresponding eigenvectors of the covariance matrix and selecting the k largest eigenvalues, effective filtering and optimization of feature points can be achieved, helping to extract the most representative and discriminative feature points, thereby improving the efficiency and accuracy of feature extraction. Using a preset mapping matrix and target feature vectors to obtain the key feature points of the device under test enhances the flexibility and controllability of feature extraction.

[0140] S17, Calculate the geometric dimensions of the device under test based on the key feature points.

[0141] After obtaining multiple key feature points, the geometric dimensions of the device under test can be calculated based on the three-dimensional coordinates of these key feature points.

[0142] In an optional implementation, calculating the geometric dimensions of the device under test based on the key feature points includes:

[0143] Obtain the three-dimensional coordinates of the key feature points, including the X, Y, and Z coordinates;

[0144] Obtain the maximum X-coordinate among the multiple X-coordinates;

[0145] Obtain the maximum Y-coordinate among the multiple Y-coordinates;

[0146] Obtain the maximum Z-coordinate among the multiple Z-coordinates;

[0147] The geometric dimensions of the device under test are calculated based on the maximum X coordinate, the maximum Y coordinate, and the maximum Z coordinate.

[0148] Since point cloud data includes three-dimensional coordinates, the coordinates of each key feature point in three-dimensional space can be obtained. These coordinates consist of three components: X, Y, and Z, which represent the horizontal, vertical, and angular coordinates in three-dimensional space, respectively.

[0149] After obtaining the X-coordinates of all key feature points, it is necessary to find the maximum value among these X-coordinates. This maximum X-coordinate typically represents the maximum size of the device under test (DUT) in the X direction (usually the length direction). Similarly, it is necessary to find the maximum value among all Y-coordinates; this maximum Y-coordinate represents the maximum size of the DUT in the Y direction (usually the width direction). Finally, it is necessary to find the maximum value among all Z-coordinates; this maximum coordinate represents the maximum size of the DUT in the Z direction (usually the height direction).

[0150] For devices with regular shapes (such as cuboids, cylinders, etc.), the key feature points are usually the boundary points or corner points of the device under test. The volume of the device under test can be calculated directly from its height, width, and length.

[0151] For irregularly shaped devices under test, since the length, width, and height cannot be directly calculated using simple distance calculations, integration or fitting algorithms are needed to approximate the geometric dimensions of the device. Integration methods obtain approximate dimensions by mathematically integrating the surface or volume of the device. Fitting algorithms, on the other hand, find a set of parameters that best describe the shape of the device (such as the major and minor axes of an ellipsoid), and then calculate the dimensions based on these parameters.

[0152] In an optional implementation, the method further includes:

[0153] Based on the geometric dimensions and the key feature points, a three-dimensional model of the device under test is generated.

[0154] Modeling techniques such as triangular meshes or (Non-Uniform Rational B-Splines, NURBS) can be used to generate a 3D model of the device under test based on the calculated geometric dimensions and key feature points.

[0155] In an optional implementation, generating a 3D model of the device under test based on the geometric dimensions and the key feature points includes:

[0156] Create the basic geometry in the 3D modeling software based on the stated geometric dimensions;

[0157] The basic geometry is adjusted according to the actual dimensions of the device under test to obtain the target geometry;

[0158] The key feature points are mapped onto the target geometry, and detailed structures are added to the target geometry to obtain the basic model;

[0159] The surface of the base model is smoothed.

[0160] A three-dimensional model of the device under test is generated based on the smoothed base model.

[0161] Based on the acquired geometric dimensions (such as length, width, height, radius, etc.) of the device under test, corresponding basic geometric shapes, such as cuboids, cylinders, and cones, are created in 3D modeling software (such as AutoCAD, SolidWorks, 3ds Max, etc.). Creating these basic geometric shapes provides a framework for subsequent modeling work, giving the model construction a clear starting point and reference.

[0162] By using matrix transformation algorithms in linear algebra (such as translation, rotation, scaling, etc.), the initially created basic geometry is precisely adjusted so that its size, position, and orientation match the actual size and position of the device under test, providing an accurate reference for adding details and complex structures later.

[0163] Key feature points (such as edge points, intersections, and extreme points) on the identified and recorded surface of the equipment are mapped onto the adjusted target geometry, serving as precise reference points for adding details and complex structures. Then, based on the position, curvature, and orientation information of the key feature points, detailed structures such as flanges, protrusions, grooves, and cables are added one by one to the target geometry, making the model more refined and realistic, and accurately reflecting the actual shape and detailed features of the equipment under test.

[0164] The base model, after adding detailed structures, undergoes surface smoothing to eliminate unnecessary sharp edges and protrusions, resulting in a smoother and more natural surface. Based on this smoothed base model, appropriate textures and colors are added according to the actual materials and appearance characteristics of the device under test, further enhancing its realism. Finally, the constructed device components and details are precisely assembled according to the actual structure of the device under test, forming a complete 3D model of the device, thus completing the final model construction.

[0165] In one optional implementation, the calculated geometric dimensions and generated 3D model can be output in a standard format. Standard formats may include, but are not limited to: STL (Stereolithography File Format), OBJ (Object File Format), PLY (Polygon File Format), IGES (Initialized Graphics Exchange Specification), and STEP (Product Model Data Exchange Standard). Users can choose the appropriate format for output and saving based on their actual needs. For example, STL format may be more suitable for models requiring 3D printing; while IGES or STEP formats are more advantageous for scenarios requiring collaborative design across multiple CAD systems. In the daily operation and maintenance of the power industry, outputting the geometric dimensions of the equipment under test not only facilitates equipment maintenance, repair, and upgrades but also provides strong support for the installation design of new equipment.

[0166] In one optional implementation, after obtaining the geometric dimensions of the device under test, an automated method can be used to compare the measured geometric dimensions with the actual or design dimensions of the device under test one by one to determine whether there are any deviations or errors. If the deviation is within an acceptable range, it indicates that the measured geometric dimensions are accurate and reliable, that is, the current measurement method is accurate and reliable; if the deviation is large, the current measurement method needs to be further optimized and adjusted.

[0167] Figure 2 This is a functional block diagram of the point cloud-based device size measurement device provided in the embodiments of this application.

[0168] In some embodiments, the point cloud-based device size measurement device 20 may include multiple functional modules composed of program code segments. The program code of each program segment in the point cloud-based device size measurement device 20 may be stored in the memory of a computer device and executed by at least one processor to perform (see details). Figure 1 (Description) Functionality for measuring device dimensions based on point clouds.

[0169] In this embodiment, the point cloud-based device size measurement device 20 can be divided into multiple functional modules according to its functions. These functional modules may include: an identification module 201, a deletion module 202, a sampling module 203, a registration module 204, a segmentation module 205, an extraction module 206, a calculation module 207, and a generation module 208. The term "module" in this application refers to a series of computer-readable instruction segments that can be executed by at least one processor and perform a fixed function, stored in memory. In this embodiment, the functions of each module will be detailed in subsequent embodiments.

[0170] The identification module 201 is used to identify outlier data points in the laser point cloud data of the device under test.

[0171] The deletion module 202 is used to delete all outlier data points from the laser point cloud data to obtain the target point cloud data.

[0172] The sampling module 203 is used to downsample the target point cloud data to obtain sampled point cloud data.

[0173] The registration module 204 is used to perform registration processing on the sampled point cloud data to obtain registered point cloud data.

[0174] The segmentation module 205 is used to segment the registered point cloud data.

[0175] The extraction module 206 is used to extract key feature points from the segmented point cloud data.

[0176] The calculation module 207 is used to calculate the geometric dimensions of the device under test based on the key feature points.

[0177] The generation module 208 is used to generate a three-dimensional model of the device under test based on the geometric dimensions and the key feature points.

[0178] It should be understood that the various variations and specific embodiments of the point cloud-based device size measurement method provided in the above embodiments are also applicable to the point cloud-based device size measurement device in this embodiment. Through the detailed description of the point cloud-based device size measurement method described above, those skilled in the art can clearly understand the implementation process of the point cloud-based device size measurement device in this embodiment. For the sake of brevity, it will not be described in detail here.

[0179] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements all or part of the steps of the point cloud-based device size measurement method.

[0180] See Figure 3 The diagram shown is a structural schematic of a computer device provided in an embodiment of this application. In a preferred embodiment of this application, the computer device 3 includes a memory 301, at least one processor 302, and at least one communication bus 303.

[0181] Those skilled in the art should understand that Figure 3 The structure of the computer device shown does not constitute a limitation of the embodiments of this application. The computer device 3 may also include more or fewer other hardware or software, or different component arrangements than shown.

[0182] In some embodiments, the computer device 3 is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits, programmable gate arrays, digital processors, and embedded devices. The computer device 3 may also include client devices, which include, but are not limited to, any electronic product that can interact with a client via a keyboard, mouse, remote control, touchpad, or voice control device, such as personal computers, tablet computers, smartphones, and digital cameras.

[0183] It should be noted that the computer device 3 described is merely an example. Other existing or future electronic products that are suitable for this application should also be included within the scope of protection of this application and are incorporated herein by reference.

[0184] In some embodiments, the memory 301 stores a computer program that, when executed by the at least one processor 302, implements all or part of the steps in the point cloud-based device size measurement method described above. The memory 301 includes read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data. Further, the computer-readable storage medium may primarily include a program storage area and a data storage area, wherein the program storage area may store an operating system, at least one application program required for a function, etc.

[0185] In some embodiments, the at least one processor 302 is the control core (CrolUnit) of the computer device 3, connecting various components of the entire computer device 3 via various interfaces and lines. It executes programs or modules stored in the memory 301 and calls data stored in the memory 301 to perform various functions of the computer device 3 and process data. For example, when the at least one processor 302 executes the computer program stored in the memory, it implements all or part of the steps of the point cloud-based device size measurement method described in this application embodiment; or it implements all or part of the functions of the point cloud-based device size measurement device. The at least one processor 302 may be composed of integrated circuits, for example, it may be composed of a single packaged integrated circuit, or it may be composed of multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips.

[0186] In some embodiments, the at least one communication bus 303 is configured to enable communication between the memory 301 and the at least one processor 302, etc. Although not shown, the computer device 3 may also include a power supply (e.g., a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor 302 via a power management device, thereby enabling functions such as charging, discharging, and power consumption management. The power supply may also include one or more DC or AC power supplies, a rechargeable power fault detection circuit, a power converter or inverter, a power status indicator, and other arbitrary components. The computer device 3 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0187] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium. This software functional module, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, a computer device, or a network device, etc.) or a processor to execute portions of the methods described in the various embodiments of this application.

[0188] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0189] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. A method for measuring device dimensions based on point clouds, characterized in that, The point cloud-based device size measurement method includes: Identify outlier data points in the laser point cloud data of the device under test; All outlier data points are removed from the laser point cloud data to obtain the target point cloud data; The target point cloud data is downsampled to obtain sampled point cloud data, including: initializing a three-dimensional voxel grid, the size of which is determined according to a preset downsampling resolution; traversing all data points in the target point cloud data and assigning each traversed data point to a corresponding voxel grid; for each voxel grid, retaining one representative data point; and combining all the retained representative data points into new point cloud data to obtain sampled point cloud data. The sampling point cloud data is registered to obtain registered point cloud data, including: selecting a data point from the sampling point cloud data as a reference data point, and selecting the data points in the sampling point cloud data other than the reference data point as data points to be registered; forming data point pairs with the reference data point and each data point to be registered; initializing a transformation matrix for each data point pair; determining a target point pair set from the multiple data point pairs; calculating an error function based on the target point pair set, the error function representing the deviation between the data point to be registered and the reference data point after transformation by the transformation matrix; iteratively updating the transformation matrix to minimize the error function; obtaining a target transformation matrix when the iteration meets a preset iteration stopping condition; and registering the data points to be registered using the target transformation matrix. The segmentation of the registered point cloud data includes: selecting an unprocessed data point from the registered point cloud data as an initial seed point; traversing each neighboring point of the initial seed point and calculating the spatial distance and normal direction difference between the neighboring point and the current region's point; when the spatial distance and normal direction difference between the neighboring points are both less than a preset threshold, adding the neighboring point to the current region and marking it as processed; traversing the neighboring points of the newly added point until no new neighboring points are added to the current region, and then segmenting the current region. Extract key feature points from the segmented point cloud data; The geometric dimensions of the device under test are calculated based on the key feature points.

2. The device size measurement method based on point cloud according to claim 1, characterized in that, Before identifying outlier data points in the laser point cloud data of the device under test, the method further includes: Acquire multi-angle laser point cloud data of the device under test; Obtain common feature points from the multi-angle laser point cloud data; The multi-angle laser point cloud data are fused based on the common feature points to obtain the laser point cloud data of the device under test.

3. The device size measurement method based on point cloud according to claim 1, characterized in that, The outlier data points in the laser point cloud data of the device under test include: For each data point in the laser point cloud data, determine the target neighborhood of the data point; Obtain the average distance between the data point and all data points in the corresponding target neighborhood; The average distance and a preset average distance threshold are used to identify whether the corresponding data point is an outlier.

4. The device size measurement method based on point cloud according to claim 1, characterized in that, The extraction of the key feature points from the segmented point cloud data includes: The segmented point cloud data is standardized to obtain standardized point cloud data. Generate a covariance matrix based on the standardized point cloud data; Calculate the eigenvalues ​​and corresponding eigenvectors of the covariance matrix; Select the k largest eigenvalues ​​from the eigenvalues; Obtain the target feature vectors corresponding to the top k largest feature values; The key feature points are obtained based on the preset mapping matrix and the target feature vector.

5. The device size measurement method based on point cloud according to claim 1, characterized in that, The method further includes: Based on the geometric dimensions and the key feature points, a three-dimensional model of the device under test is generated.

6. A computer device, characterized in that, The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the point cloud-based device size measurement method as described in any one of claims 1 to 5.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the point cloud-based device size measurement method as described in any one of claims 1 to 5.

Images