Special-shaped curved surface component accurate comparison and analysis method based on 3D laser scanning

Abstract

The invention relates to the technical field of data processing, in particular to a special-shaped curved surface component accurate comparison and analysis method based on 3D laser scanning. The method comprises the following steps: firstly, carrying out primary clustering on point clouds according to density features to obtain a plurality of first point cloud clusters, and realizing unsupervised classification; secondly, the fluctuation condition of the overall curvature of each first point cloud cluster can be quantified by calculating the error between the actual overall curvature and the predicted overall curvature under different numbers of neighbor points, secondary clustering is carried out on the first point cloud clusters, hub category point cloud clusters with different geometric features are recognized according to the fluctuation condition of the overall curvature, and the recognition accuracy is improved; then, the optimal neighbor point number of the hub category point cloud cluster of each category is obtained and used for curved surface fitting, the stability of geometric feature measurement is improved, and the final distance measurement is fused with spatial position and local geometric information; and finally, through optimization of an objective function and threshold judgment, the point cloud registration precision and the reliability of hub contrastive analysis are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method for precise comparative analysis of irregular curved surface components based on 3D laser scanning. Background Technology

[0002] As an indispensable means of transportation for long-distance travel, automobiles have become integrated into all aspects of people's lives. Among them, automobile wheel hubs, as irregular curved surface components, need to undergo multiple inspection processes during production. Shape inspection is one of the important processes. Due to the irregular shape of the wheel hub, 3D scanning is usually used to obtain its shape point cloud data. The scanned point cloud data is then compared and analyzed with standard point cloud data to determine whether the wheel hub is qualified.

[0003] The ICP algorithm, a classic point cloud data comparison method, is widely used in the registration of point cloud data for irregular curved surface components. Traditional ICP algorithms use only Euclidean distance as the criterion for point pair selection. This approach loses most of the details between points, failing to consider that points in space may not be their true geometric counterparts. Therefore, when registering wheel hub point cloud data, it ignores the geometric features of the wheel hub itself, thus affecting the reliability of wheel hub shape detection. Summary of the Invention

[0004] To address the technical problem that existing ICP algorithms, which use Euclidean distance for point cloud registration, neglect the geometric features of the wheel hub itself, thus affecting the reliability of wheel hub shape detection, the present invention aims to provide a precise comparative analysis method for irregular curved surface components based on 3D laser scanning. The specific technical solution adopted is as follows: This invention proposes a method for accurate comparative analysis of irregular curved surface components based on 3D laser scanning, the method comprising: Perform a clustering operation on the hub point cloud to obtain multiple first point cloud clusters; A second clustering is performed on the first point cloud cluster to obtain the hub category point cloud cluster. In the second clustering, the initial number of nearest neighbor points is preset, and the number of nearest neighbor points is increased iteratively. For each iteration, the first point cloud cluster obtains the actual overall curvature based on the number of nearest neighbor points, obtains the error between the predicted overall curvature and the actual overall curvature corresponding to the historical iteration process, and obtains the data basis of the second clustering based on all errors under all iterations. In the point cloud cluster of wheel hub category, analyze the range of local curvature under each number of nearest neighbor points, and determine the optimal number of nearest neighbor points for each wheel hub category point cloud cluster; The wheel hub category point cloud cluster is compared point by point with the preset standard wheel hub category point cloud cluster to obtain the final measurement distance; the objective function of the ICP algorithm is optimized based on the final measurement distance, and wheel hub comparison analysis is performed based on the optimized objective function; the final measurement distance is obtained from the initial distance and the second measurement distance of the ICP algorithm; the second measurement distance is obtained based on the geometric feature difference corresponding to the optimal number of nearest neighbors.

[0005] Furthermore, the hub point cloud includes: rim point cloud, bolt hole point cloud, spoke point cloud, and transition point cloud.

[0006] Furthermore, the method for obtaining the overall curvature includes: In each iteration, in each hub category point cloud cluster, with each hub point cloud as the center, the fitted surface of each hub point cloud is obtained by fitting according to the number of corresponding nearest neighbor points; The average curvature of the fitted surface of all points within the point cloud cluster of the wheel hub category is used as the overall curvature.

[0007] Furthermore, obtaining the data basis for the secondary clustering based on all errors across all iterations includes: In each iteration, the absolute value of the difference between the actual overall curvature of the first point cloud cluster and the corresponding predicted overall curvature is taken as the error; For each first point cloud cluster, the average of all errors generated by the first point cloud cluster in all iterations is used as the data basis.

[0008] Furthermore, the method for obtaining the range of the local curvature includes: For each number of nearest neighbors, the maximum difference in curvature between the fitted surfaces in the point cloud cluster of the wheel hub category is used as the range of the local curvature.

[0009] Furthermore, the method for determining the optimal number of nearest neighbors for each hub category point cloud cluster includes: Using the range of local curvature on the fitted surface of the wheel hub category point cloud cluster under different numbers of nearest neighbors as the ordinate and different numbers of nearest neighbors as the abscissa, a line graph is plotted for each wheel hub category point cloud cluster; the optimal number of nearest neighbors is selected based on the coordinate points in the line graph.

[0010] Furthermore, the method for obtaining the optimal number of nearest neighbor points for the fitted surface of each wheel hub category point cloud cluster based on the line graph includes: The number of nearest neighbor points corresponding to the point with the lowest vertical axis is selected as the optimal number of nearest neighbor points for the hub category point cloud cluster.

[0011] Furthermore, obtaining the second metric distance based on the geometric feature differences corresponding to the optimal number of nearest neighbors includes: In each hub category point cloud cluster, with each hub point cloud as the center, the fitted surface of each hub point cloud is obtained by fitting based on the number of optimal nearest neighbors; The second metric distance of the bolt hole point cloud is the cosine of the angle between the normal vectors of the fitted surfaces in the source and target point clouds; The second metric distance of the transition point cloud is the absolute value of the difference in curvature between the fitted surfaces in the source point cloud and the target point cloud; The second metric distance of the spoke point cloud is the DTW distance between the FPFH feature descriptors of the fitted surfaces in the source and target point clouds; The second metric distance of the rim point cloud is the product of the absolute value of the difference in curvature between the fitted surfaces in the source and target point clouds and the absolute value of the product of the unit normal vectors.

[0012] Furthermore, the method for obtaining the final metric distance of each wheel hub category point cloud cluster includes: The final distance measurement is the weighted sum of the initial distance and the second distance measurement; the weights of the initial distance and the second distance measurement are obtained by taking the weights using the inverse of the variance; the sum of the weights of the initial distance and the second distance measurement is a positive integer 1.

[0013] Furthermore, the step of optimizing the objective function of the ICP algorithm based on the final metric distance, and performing wheel hub comparison analysis based on the optimized objective function, includes: The initial distance of the ICP algorithm objective function is replaced with the final metric distance to obtain the optimized objective function; the optimized objective function value is compared with a preset threshold to obtain the wheel hub comparison analysis results.

[0014] The present invention has the following beneficial effects: This invention takes into account that the standard ICP algorithm uses Euclidean distance as the metric distance when performing point cloud registration, which ignores the geometric features of the hub point cloud, resulting in poor registration effect. This invention first obtains multiple first point cloud clusters based on the density characteristics of point cloud clusters of different wheel hub categories. This allows point clouds with the same or similar density characteristics to be grouped into the same cluster, achieving preliminary unsupervised classification of wheel hub point clouds and laying the foundation for subsequent secondary clustering. Then, by using the actual and predicted overall curvature under different numbers of nearest neighbors, the fluctuation of the overall curvature of each first point cloud cluster can be quantified. Secondary clustering of the first point cloud clusters is then performed, identifying wheel hub category point cloud clusters with different geometric features based on the fluctuation of the overall curvature, thus improving the accuracy of identification. Next, the optimal number of nearest neighbors for different wheel hub category point cloud clusters is obtained. Fitting a surface based on the optimal number of nearest neighbors effectively improves the stability and registration accuracy of the second metric distance, further ensuring that the final metric distance simultaneously considers the spatial distance of the point clouds and the differences in local geometric features. Finally, by comparing and analyzing the optimized objective function with a preset threshold, the accuracy of the overall point cloud registration can be significantly improved, thereby enhancing the accuracy and reliability of wheel hub comparison analysis. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 The flowchart illustrates a method for precise comparative analysis of irregular curved surface components based on 3D laser scanning, as provided in one embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for precise comparative analysis of irregular curved surface components based on 3D laser scanning proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of the precise comparative analysis method for irregular curved surface components based on 3D laser scanning provided by the present invention.

[0020] Please see Figure 1 The diagram illustrates a flowchart of a method for precise comparative analysis of irregular curved surface components based on 3D laser scanning, according to an embodiment of the present invention. The method includes: Step S101: Perform a clustering of the hub point cloud to obtain multiple first point cloud clusters.

[0021] First, a 3D laser scanner is used to scan the car wheel hub component to obtain 3D wheel hub point cloud data. Since the amount of point cloud data is enormous, a downsampling method is needed to reduce the data volume, thereby reducing subsequent computational complexity. In one specific implementation of this invention, uniform sampling is chosen as the downsampling method.

[0022] For the first clustering, the hub point cloud is used as input, and the density-based clustering (DPC) algorithm is employed. The distance between point clouds is set to Euclidean distance, and the cutoff distance is set to 1% of the total number of points whose distance to each point is less than the cutoff distance. The output yields multiple clusters, which constitute the first point cloud clusters. Notably, using density-based clustering (DPC) eliminates the need for pre-setting the number of first point cloud clusters. It groups point clouds with similar density characteristics into the same cluster based on the density features of the hub point cloud itself, enabling preliminary classification of the hub point cloud.

[0023] Step S102: Perform secondary clustering on the first point cloud cluster to obtain the hub category point cloud cluster; in the secondary clustering, a preset initial number of nearest neighbor points is set, and the number of nearest neighbor points is increased iteratively. For each iteration, the first point cloud cluster obtains the actual overall curvature based on the number of nearest neighbor points, obtains the error between the predicted overall curvature and the actual overall curvature corresponding to the historical iteration process, and obtains the data basis of the secondary clustering based on all errors under all iterations.

[0024] First, because point cloud clusters of different wheel hub categories exhibit different geometric features, and because there are cases where point cloud clusters belonging to the same wheel hub category are still clustered into multiple clusters among the multiple first point cloud clusters obtained after step S101, a secondary clustering process is needed to cluster multiple first point cloud clusters belonging to the same wheel hub category into one cluster, resulting in multiple wheel hub category point cloud clusters. After the secondary clustering, each wheel hub category point cloud cluster contains only one cluster.

[0025] Since point cloud clusters of different wheel hub categories exhibit different geometric features, the differences in these geometric features can be used to identify which wheel hub category each first point cloud cluster belongs to—this is the process of secondary clustering. Specifically, the geometric differences between point cloud clusters of different wheel hub categories manifest in the varying degrees of overall curvature fluctuation across different wheel hub category point cloud clusters with different numbers of nearest neighbors. Based on these differences in overall curvature fluctuation across different wheel hub category point cloud clusters with different numbers of nearest neighbors, the first point cloud clusters can be clustered into wheel hub category point cloud clusters.

[0026] Therefore, it is necessary to iterate and change the number of nearest neighbors to obtain the overall curvature fluctuation of each hub category point cloud cluster under each number of nearest neighbors. In a specific implementation of this invention, the initial number of nearest neighbors for iteration is set to 3, and the final number of nearest neighbors for iteration is set to 33. The overall curvature fluctuation is obtained through the error between the actual overall curvature and the predicted overall curvature. The actual overall curvature is the overall curvature actually obtained in the previous iteration and the current iteration, while the predicted overall curvature is the predicted overall curvature for the next iteration based on the actual overall curvature obtained in all previous iterations. The larger the error, the larger the error between the actual and predicted overall curvature, i.e., the greater the overall fluctuation; the smaller the error, the smaller the error between the actual and predicted overall curvature, i.e., the smaller the overall fluctuation. In a specific implementation of this invention, the prediction method uses a moving average method.

[0027] Since the geometric features of wheel hub category point cloud clusters do not change with the number of nearest neighbor points, if one first point cloud cluster belongs to one wheel hub category point cloud cluster and another first point cloud cluster belongs to another wheel hub category point cloud cluster, and the error of the first point cloud cluster in the current iteration is greater than the error of the other first point cloud cluster in the current iteration, then the error of the first point cloud cluster in all iterations is greater than the error of the other first point cloud cluster in all iterations. Therefore, this embodiment of the invention obtains the error of all first point cloud clusters in all iterations, and clusters all first point cloud clusters into wheel hub category point cloud clusters according to the magnitude of the error using a clustering algorithm. In a specific implementation of this embodiment of the invention, the clustering algorithm uses K-means clustering, and the number of clusters is set to 4.

[0028] After obtaining the wheel hub category point cloud clusters, the error of each wheel hub category point cloud cluster in each iteration is further obtained. Since the geometric features of each wheel hub category point cloud cluster are different, the error of each wheel hub category point cloud cluster in each iteration is different. By sorting the errors by size, it is possible to determine which wheel hub category point cloud cluster each wheel hub category point cloud cluster belongs to, which improves the accuracy of identifying different wheel hub category point cloud clusters and further improves the accuracy of subsequent point cloud registration.

[0029] As an example, in a specific implementation of this invention, the wheel hub point cloud includes: rim point cloud, bolt hole point cloud, spoke point cloud, and transition point cloud. The analysis of obtaining the point cloud cluster for each wheel hub category by sorting by error magnitude is as follows: Taking the wheel rim point cloud and bolt hole point cloud as examples, since the wheel rim has the largest radius and the flattest surface, as the number of neighboring points increases, the curvature difference between each neighboring point is the smallest. That is, the curvature obtained in the later iteration will show the lowest curvature fluctuation compared to the curvature obtained in the previous iteration. Therefore, the wheel hub point cloud cluster with the smallest error belongs to the wheel rim point cloud. On the other hand, for the bolt hole point cloud, since the bolt hole has the smallest radius, there is a very large difference between the surface surrounding the bolt hole and the curvature inside the bolt hole. As the number of neighboring points increases, the curvature difference between each neighboring point is the largest. That is, the curvature obtained in the later iteration will show the highest curvature fluctuation compared to the curvature obtained in the previous iteration. Therefore, the wheel hub point cloud cluster with the largest error belongs to the bolt hole point cloud.

[0030] Step S103: In the point cloud cluster of the wheel hub category, analyze the range of local curvature under each number of nearest neighbor points, and determine the optimal number of nearest neighbor points for each point cloud cluster of the wheel hub category.

[0031] Accurate curvature estimation relies on surface fitting to each point and its nearest neighbors. When the number of selected nearest neighbors is very small, noise from even a few points can significantly impact the fitting accuracy, leading to serious errors in the curvature estimate. Conversely, when the number of selected nearest neighbors is very large, the fitting range crosses the geometric features of the object's surface, such as from a plane to a curved surface or across an edge. This results in an averaged fit, dulling the geometric features of sharp surfaces and causing inaccurate or lost true curvature information. Selecting the same number of nearest neighbors for all wheel hub type point cloud clusters leads to severe edge misalignment and blurred region boundaries. Therefore, different optimal number of nearest neighbors should be used when performing surface fitting on point cloud clusters of different wheel hub types. Using the optimal number of nearest neighbors maximizes noise suppression and yields the smoothest and most stable geometric feature analysis.

[0032] First, when the number of nearest neighbor points is very small, the curvature obtained by fitting each point cloud is easily affected by individual noise points and position errors, resulting in irregular and large fluctuations in the curvature values ​​at different point clouds. Therefore, at this stage, as the number of nearest neighbor points increases iteratively from small to large, the range of local curvature will show a rapid and unstable downward trend, because adding a few nearest neighbor points can dilute the influence of individual noise points and gradually narrow the range of local curvature.

[0033] Secondly, when the number of nearest neighbor points increases to a large extent, such that the fitted surface spans different geometric features (e.g., from a plane to a surface), the fitting result will be averaged by the geometric features of different point clouds, and some features will be blunted, which is reflected in the local curvature range being pulled towards the average value. Therefore, at this stage, as the number of nearest neighbor points increases iteratively from small to large, the local curvature range will show an upward trend.

[0034] Therefore, in order to find the optimal neighborhood scale, i.e. the optimal number of nearest neighbors, between suppressing noise and preserving geometric features, the optimal number of nearest neighbors for each hub category point cloud cluster can be determined by the range variation trend of local curvature under different numbers of nearest neighbors in each iteration.

[0035] Step S104: Compare the wheel hub category point cloud cluster with the preset standard wheel hub category point cloud cluster point by point to obtain the final measurement distance; optimize the objective function of the ICP algorithm based on the final measurement distance, and perform wheel hub comparison analysis based on the optimized objective function; the final measurement distance is obtained from the initial distance and the second measurement distance of the ICP algorithm; the second measurement distance is obtained based on the geometric feature difference corresponding to the optimal number of nearest neighbors.

[0036] In the wheel hub point cloud space, the standard ICP algorithm uses an initial distance for point cloud registration, without considering the geometric features of the wheel hub itself. Since the wheel hub is a complex, irregularly shaped curved surface, simply using the standard initial distance for point cloud registration ignores its geometric characteristics. Because different wheel hub category point cloud clusters have their own unique geometric feature representations, using the optimal number of nearest neighbors can suppress noise to the greatest extent and obtain the most stable and smooth geometric features. Therefore, to effectively compare wheel hub category point cloud clusters with a preset standard wheel hub category point cloud cluster, this embodiment of the invention obtains the second metric distance for each wheel hub category point cloud cluster by using the geometric feature differences corresponding to the optimal number of nearest neighbors. The ICP algorithm is then optimized using its own initial distance and second metric distance to obtain the final metric distance. This allows the ICP algorithm to consider the geometric feature differences of different wheel hub category point cloud clusters, significantly improving the accuracy and reliability of wheel hub comparison analysis. Therefore, the optimized objective function can be used to perform comparative analysis between wheel hub category point cloud clusters and the preset standard wheel hub category point cloud clusters to determine the product quality inspection results.

[0037] In summary, this invention takes into account that the standard ICP algorithm uses Euclidean distance as the metric distance when performing point cloud registration, which ignores the geometric features of the wheel hub point cloud, resulting in poor registration effect. This invention first obtains multiple first point cloud clusters based on the density characteristics of point cloud clusters of different wheel hub categories. This allows point clouds with the same or similar density characteristics to be grouped into the same cluster, achieving preliminary unsupervised classification of wheel hub point clouds and laying the foundation for subsequent secondary clustering. Then, by using the actual and predicted overall curvature under different numbers of nearest neighbors, the fluctuation of the overall curvature of each first point cloud cluster can be quantified. Secondary clustering of the first point cloud clusters is then performed, identifying wheel hub category point cloud clusters with different geometric features based on the fluctuation of the overall curvature, thus improving the accuracy of identification. Next, the optimal number of nearest neighbors for different wheel hub category point cloud clusters is obtained. Fitting a surface based on the optimal number of nearest neighbors effectively improves the stability and registration accuracy of the second metric distance, further ensuring that the final metric distance simultaneously considers the spatial distance of the point clouds and the differences in local geometric features. Finally, by comparing and analyzing the optimized objective function with a preset threshold, the accuracy of the overall point cloud registration can be significantly improved, thereby enhancing the accuracy and reliability of wheel hub comparison analysis.

[0038] Preferably, in some possible implementations of the embodiments of the present invention, the hub point cloud includes: rim point cloud, bolt hole point cloud, spoke point cloud, and transition point cloud. The rim is the outermost area of ​​the car wheel hub, with the largest radius, fewest patterns, and simplest structure; the bolt hole is the area with the smallest radius in the car wheel hub, containing a large number of threads and having the most complex structure; the spoke is the structure connecting the rim and the center, with a relatively flat surface; the transition is the turning part where the spoke connects to the rim, and where the spoke connects to the center, with a radius second only to the bolt hole. By dividing the hub point cloud into different hub category point cloud clusters, targeted quality monitoring of different areas can be achieved, improving the accuracy and efficiency of subsequent defect identification.

[0039] Preferred methods for obtaining the overall curvature include: In each iteration, in each hub category point cloud cluster, with each hub point cloud as the center, the fitted surface of each hub point cloud is obtained by fitting according to the number of corresponding nearest neighbor points; The average curvature of the fitted surface of all points within the point cloud cluster of the wheel hub category is used as the overall curvature.

[0040] Each point in each first point cloud cluster will obtain a curvature in each iteration. The average curvature of all points in a first point cloud cluster in one iteration is taken as the overall curvature of the current first point cloud cluster in the current iteration. Using the overall curvature can reduce the computational complexity and reflect the changing trend of all points in the first point cloud cluster as the number of nearest neighbors increases.

[0041] In one specific implementation of this invention, the average value of the maximum principal curvature of each point in the first point cloud cluster is taken as the overall curvature.

[0042] Preferably, the data basis for the secondary clustering is obtained based on all errors across all iterations, including: In each iteration, the absolute value of the difference between the actual overall curvature of the first point cloud cluster and the corresponding predicted overall curvature is taken as the error; For each first point cloud cluster, the average of all errors generated by the first point cloud cluster in all iterations is used as the data basis.

[0043] The actual overall curvature of the first point cloud cluster obtained in the previous iteration is taken as the actual overall curvature. All the actual overall curvatures in the previous iterations are fitted to predict the overall curvature of the next iteration, which is taken as the predicted overall curvature of the next iteration. The predicted overall curvature reflects the changing trend of the historical overall curvature. The iteration number is incremented by one to obtain the actual overall curvature of the next iteration. The absolute value of the difference between the actual overall curvature of one iteration and the corresponding predicted overall curvature is taken as the error, and the average error of all iterations is obtained. The larger the average error, the greater the difference between the overall curvature under the increased number of nearest neighbor points and the overall curvature under the previous number of nearest neighbor points, which is more likely to be in hub point clouds with more complex geometric features.

[0044] As an example, in a specific implementation of this invention, the hub point cloud includes: rim point cloud, bolt hole point cloud, spoke point cloud, and transition point cloud. Therefore, the specific steps for obtaining the data basis for secondary clustering and performing secondary clustering are as follows: S1: The number of nearest neighbors is denoted as K, with an initial value of 3. When the iteration terminates, the number of nearest neighbors K is 33. S2: K=3, calculate the overall curvature of each cloud cluster at the first point; S3: Increment K by 1, repeat S2 until K=33, each first point cloud cluster gets 31 actual overall curvatures; S4: Based on the actual overall curvature of each first point cloud cluster under the first 3 iterations, the predicted overall curvature under the 4th iteration is obtained by the moving average method. The absolute value of the difference between the actual overall curvature under the 4th iteration and the predicted overall curvature under the 4th iteration is obtained as the error. S5: Increment K by 1, repeat S4 until K=33. Each first point cloud cluster has 28 errors. Take the average of the 28 errors of each first point cloud cluster as the average error. S6: Take the average error of each first point cloud cluster as input, use the Kmeans algorithm to perform clustering, set the number of clusters to 4, and the algorithm outputs 4 hub clusters; S7: For the four wheel hub clusters, repeat steps S1-S5 to obtain the average error of each wheel hub cluster. Sort the average error of each of the four wheel hub clusters by size. Determine which wheel hub cluster belongs to each wheel hub cluster based on the geometric characteristics of each wheel hub category point cloud cluster. The average error is sorted from largest to smallest as follows: bolt hole point cloud, transition point cloud, spoke point cloud, and rim point cloud. S8: The judgment method for step S7 is as follows: For bolt hole point cloud, since the bolt hole has the smallest radius, there is a very large difference between the curvature of the surface surrounding the bolt hole and the curvature of the inner side of the bolt hole. As the K value increases, the curvature difference between the K nearest neighbor points obtained changes the most. The curvature of the fitting plane will show the highest curvature fluctuation compared to the curvature of the plane obtained in the previous fitting. For the transition point cloud, the curvature of the transition point cloud is second only to that of the bolt hole point cloud. The transition is the area connecting the spokes and the rim, and it has the second highest curvature after the bolt hole point cloud. As the K value increases, the curvature difference between the K nearest neighbor points is smaller than that of the bolt hole. The curvature of the fitting plane will show the second highest curvature fluctuation compared to the curvature of the plane obtained in the previous fitting, which is only lower than that of the bolt hole point cloud.

[0045] For the spoke point cloud, the surface flatness of the spokes is second only to the rim and better than the transition. This is due to the smooth transition of the surface pattern. As the K value increases, the curvature difference between the K nearest neighbor points is larger than that of the rim but smaller than that of the transition. The curvature of the fitted plane will show the third lowest curvature fluctuation compared to the curvature of the previously fitted plane, only higher than that of the rim point cloud.

[0046] For the rim point cloud, the rim has the largest radius and the flattest surface. As the K value increases, the curvature difference between the K nearest neighbor points changes the slowest and smallest. The curvature of the fitted plane will show the lowest curvature fluctuation compared to the curvature of the previously fitted plane.

[0047] Preferred methods for obtaining the range of local curvature include: For each number of nearest neighbors, the maximum difference in curvature between the fitted surfaces in the wheel hub category point cloud cluster is statistically analyzed, and this difference is used as the range of the local curvature. The maximum difference in curvature reflects the shape complexity of the wheel hub category point cloud cluster; the more complex the geometry, the larger the maximum difference in curvature, and the simpler the geometry, the smaller the maximum difference in curvature. By using the maximum difference in curvature for each number of nearest neighbors, the overall shape complexity of the wheel hub category point cloud cluster can be quantified.

[0048] Preferably, the method for determining the optimal number of nearest neighbors for each hub category point cloud cluster includes: Using the range of local curvature on the fitted surface of the wheel hub category point cloud cluster under different numbers of nearest neighbors as the ordinate and different numbers of nearest neighbors as the abscissa, a line graph is plotted for each wheel hub category point cloud cluster. The optimal number of nearest neighbors is selected based on the coordinate points in the line graph. By plotting the maximum difference in curvature corresponding to the number of nearest neighbors on the line graph, the calculation of the maximum curvature difference and the change in the number of nearest neighbors can be visualized, further determining the optimal analysis scale for each wheel hub category point cloud cluster.

[0049] Preferably, the method for obtaining the optimal number of nearest neighbors for the fitted surface of the point cloud cluster for each wheel hub category based on the line graph includes: The number of nearest neighbors corresponding to the point with the lowest vertical axis is selected as the optimal number of nearest neighbors to search when performing surface fitting on the point cloud cluster of this wheel hub category.

[0050] In the line graph, as the number of nearest neighbor points increases, the number of point clouds increases when performing surface fitting. If the point clouds of the wheel hub category have significantly different curvature characteristics, the maximum difference in curvature will also increase. However, for point clouds belonging to the same region, which have the same or similar regional characteristics, the maximum difference in curvature will first increase and then decrease as the number of nearest neighbor points increases, with a minimum value. By using this minimum value as the optimal number of nearest neighbor points for the wheel hub category point cloud cluster, the local neighborhood scale for surface fitting of the wheel hub category point cloud cluster can be adaptively determined according to the geometric characteristics of each wheel hub category point cloud cluster. This allows the geometric characteristics of the wheel hub category point cloud cluster to be preserved and highlighted while smoothing out irrelevant noise.

[0051] Preferably, the second metric distance is obtained based on the geometric feature differences corresponding to the optimal number of nearest neighbors, including: In each hub category point cloud cluster, with each hub point cloud as the center, the fitted surface of each hub point cloud is obtained by fitting based on the number of optimal nearest neighbors; First, the second metric distance of the bolt hole point cloud is the cosine of the angle between the normal vectors of the fitted surfaces in the source and target point clouds. Bolt holes are the most distinctive area on a car wheel hub; the normal vectors of points on the inner wall of the bolt hole all point in the radial direction. Therefore, when registering the bolt hole point cloud with the standard wheel hub category point cloud cluster, it is necessary not only to find the closest points in space but also to find points with the same normal vector direction. The formula for the second metric distance of the bolt hole point cloud is: In the formula, The second metric distance represents the distance between the i-th point in the bolt hole point cloud and the j-th point in the standard wheel hub category point cloud cluster; This represents the cosine of the angle between the normal vector of the i-th point in the bolt hole point cloud and the normal vector of the j-th point in the standard wheel hub category point cloud cluster.

[0052] Secondly, the second metric distance of the transition point cloud is the absolute value of the difference in curvature between the fitted surfaces in the source and target point clouds; the transition region is the part connecting the spokes and rim on a car wheel hub, as well as the turning point between the spokes and the center, with a radius second only to the bolt holes. The curvature is relatively high here, and the curvature factor is considered when designing the metric distance; the formula for the second metric distance of the transition point cloud is: In the formula, The second metric distance represents the distance between the i-th point in the transition point cloud and the j-th point in the standard wheel hub category point cloud cluster; This represents the maximum principal curvature of the i-th point within the transition point cloud; This represents the maximum principal curvature of the j-th point within the point cloud cluster of the standard wheel hub category.

[0053] Then, the second metric distance of the spoke point cloud is the DTW distance between the FPFH feature descriptors of the fitted surfaces in the source and target point clouds. Since the spoke point cloud exhibits numerous patterns, and the patterns on the spokes of different vehicle models vary, including but not limited to edges, turning lines, and curves, the second metric distance of the spoke point cloud is calculated using the FPFH feature descriptor. The formula for the second metric distance of the spoke point cloud is: In the formula, This represents the second metric distance between the i-th point in the spoke point cloud and the j-th point in the standard hub category point cloud cluster; This represents the FPFH feature descriptor for the i-th point within the spoke point cloud; This represents the FPFH feature descriptor of the j-th point within the point cloud cluster of the standard wheel hub category.

[0054] Where || represents the DTW distance between two FPFH feature descriptors, using the FPFH feature descriptor within the spoke point cloud as the reference template.

[0055] Finally, the second metric distance of the rim point cloud is the product of the absolute value of the difference in curvature between the fitted surfaces in the source and target point clouds and the absolute value of the product of the unit normal vectors. For the rim region, this is a large surface with low curvature, where the normal vectors of the points point on the region point towards the center, and the surface is smooth without patterns. The second metric distance here is calculated by combining curvature and normal vectors. The formula for the second metric distance of the rim point cloud is: In the formula, This represents the second metric distance between the i-th point in the rim point cloud and the j-th point in the standard wheel hub category point cloud cluster; Represents the unit normal vector of the i-th point within the rim point cloud and the unit normal vector of the j-th point in the standard wheel hub category point cloud cluster; This represents the maximum principal curvature of the i-th point within the rim point cloud and the maximum principal curvature of the j-th point within the standard hub category point cloud cluster.

[0056] When calculating the second metric distance of each wheel hub category point cloud cluster, the fitted surface of each wheel hub point cloud is obtained by fitting the optimal number of nearest neighbors. This ensures that the extracted normal vectors, curvature, and FPFH feature descriptors match the key characteristics of the corresponding wheel hub category point cloud clusters, amplifying the geometric feature differences between different wheel hub category point cloud clusters and providing a foundation for high-accuracy registration.

[0057] Preferably, the method for obtaining the final metric distance of the point cloud cluster for each wheel hub category includes: The final metric distance is the weighted sum of the initial distance and the second metric distance. The weights of the initial distance and the second metric distance are obtained by calculating the weights using the inverse of the variance; the sum of the weights of the initial distance and the second metric distance is a positive integer 1. The initial distance is Euclidean distance, which reflects the absolute spatial differences between wheel hub point clouds. The weighted sum of the Euclidean distance and the second metric distance combines macroscopic spatial positions with local geometric features, improving the comprehensiveness and accuracy of registration for wheel hub category point cloud clusters. Obtaining the weights of the initial distance and the second metric distance by calculating the weights using the inverse of the variance adaptively balances the proportion of the two metric distances according to the distribution characteristics of the actual data, avoiding subjective bias caused by manually setting weights and improving the accuracy and reliability of registration.

[0058] It should be noted that the initial distance of the ICP algorithm is the Euclidean distance. If it is directly weighted and summed with the aforementioned four second-order distance metrics, the weights will become ineffective due to the inconsistency of dimensions. Therefore, in this embodiment of the invention, before calculating the final distance metric, the initial distance and the four second-order distance metrics are first normalized to unify the differences in spatial location and local geometric features to the same scale. In a specific implementation of this embodiment, taking the normalization process of the second-order distance of bolt hole point clouds as an example: the system obtains the second-order distance of all bolt hole point clouds and uses the maximum and minimum value normalization calculation. The method normalizes the range of the second metric distance of all bolt hole point clouds to [0, 1], obtaining the normalized second metric distance of all bolt hole point clouds. It should be noted that the normalization process of the initial distance and the other three second metric distances is the same in calculation principle as the normalization process of the second metric distance of bolt hole point clouds. It is only necessary to replace the second metric distance of all bolt hole point clouds with the initial distance or the other three second metric distances and repeat the normalization process of the second metric distance of bolt hole point clouds. Unless otherwise specified, the initial distance and the four types of second metric distances mentioned later are all distances after normalization.

[0059] The formula for calculating the weights based on the reciprocal of the variance is: In the formula, The variance representing the initial distance of the point cloud clusters for each wheel hub category; The variance of the second metric distance represents the distance between the source point cloud hub category point cloud cluster.

[0060] It should be noted that if the numerical range of a distance metric is very large (large variance), it indicates that it has high discrimination but poor stability, and should be given a lower weight. Conversely, if the numerical range of a distance metric is very small (small variance), it should be given a higher weight.

[0061] It should be further explained that when or When the value is 0, it is replaced with the preset correction parameter. or It also participates in the calculation of the formula for calculating the weight of the original variance to prevent the denominator from being 0; in this embodiment of the invention, the preset correction parameter is set to 0.01, which can be adjusted according to the specific implementation environment.

[0062] Preferably, the objective function of the ICP algorithm is optimized based on the final metric distance, and a wheel hub comparison analysis is performed based on the optimized objective function, including: The initial distance of the ICP algorithm objective function is replaced with the final metric distance to obtain the optimized objective function; the optimized objective function value is compared with a preset threshold to obtain the wheel hub comparison analysis results.

[0063] After obtaining the final metric distance of the point cloud cluster for each wheel hub category, the corresponding points in the standard wheel hub category point cloud cluster can be found based on the final metric distance of each wheel hub category point cloud cluster, forming a set of point pairs. The objective function of the standard ICP algorithm is to find a transformation that minimizes the sum of the squared Euclidean distances between all found corresponding point pairs, as shown in the formula: The standard objective function is optimized to take into account the different final metric distances of point cloud clusters for each wheel hub category during calculation. The formula is as follows: in, The objective function is represented by N; N represents the number of points in the point cloud clusters of all wheel hub categories. This indicates that the hub category belongs to the first point cloud cluster. The number of points in a class; There are four categories: bolt holes, transitions, spokes, and rims. To find the first... The final distance metric function for point cloud-like structures; This indicates that the hub category belongs to the first point cloud cluster. The m-th point of the category; This indicates that in the standard wheel hub category point cloud cluster, the transformation through R,t is... The corresponding point found; R and t are the rotation matrix and translation vector, Transform R and t to the corresponding point ; The dot product symbol is used in this invention unless otherwise specified. Both represent the dot product symbol.

[0064] After obtaining the objective function, a nonlinear optimization method (such as the Levenberg-Marquardt algorithm) is used to solve for the rotation matrix R and the translation vector t. The calculated transformation (R,t) is then applied to the entire hub category point cloud cluster. In the formula, pm represents the m-th point in the point cloud cluster of the wheel hub category; This represents the point in the standard wheel hub category point cloud cluster that corresponds to the m-th point in the wheel hub category point cloud cluster.

[0065] In this process, a point pm in the hub-type point cloud cluster is transformed into a point qm in the target point cloud after an R, t transformation. However, a single transformation cannot reach the ideal position, so the process is repeated iteratively until the iteration termination condition is met, such as reaching the preset maximum number of iterations or the change in the transformation (R, t) is sufficiently small, for example, the rate of change is less than 3% for 5 consecutive iterations. The final rotation matrix R and translation vector t are then output by the CloudCompare point cloud processing software. It should be noted that the Levenberg-Marquardt algorithm and the CloudCompare point cloud processing software are well-known techniques to those skilled in the art and will not be elaborated upon here.

[0066] It should be noted that the optimization objective function value reflects the mismatch between the wheel hub category point cloud cluster and the standard wheel hub point cloud cluster. The larger the optimization objective function value, the greater the difference between the wheel hub category point cloud cluster and the standard wheel hub category point cloud cluster, indicating a worse registration effect; the smaller the optimization objective function value, the smaller the difference between the wheel hub category point cloud cluster and the standard wheel hub category point cloud cluster, indicating a better registration effect.

[0067] The objective function value is compared with a preset threshold. If the objective function value is equal to or exceeds the preset threshold, the wheel hub does not meet the factory conditions and is a defective product. If the objective function value does not exceed the preset threshold, the wheel hub meets the factory conditions and is a qualified product.

[0068] In one specific implementation of this invention, 100 wheel hubs are sampled to obtain their respective optimization objective function values. The maximum and minimum values ​​are removed, and the average of the 98 optimization objective function values ​​is used as a preset threshold.

[0069] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0070] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for precise comparative analysis of irregular curved surface components based on 3D laser scanning, characterized in that, The method includes: Perform a clustering operation on the hub point cloud to obtain multiple first point cloud clusters; A second clustering is performed on the first point cloud cluster to obtain the hub category point cloud cluster. In the second clustering, the initial number of nearest neighbor points is preset, and the number of nearest neighbor points is increased iteratively. For each iteration, the first point cloud cluster obtains the actual overall curvature based on the number of nearest neighbor points, obtains the error between the predicted overall curvature and the actual overall curvature corresponding to the historical iteration process, and obtains the data basis of the second clustering based on all errors under all iterations. In the point cloud cluster of wheel hub category, analyze the range of local curvature under each number of nearest neighbor points, and determine the optimal number of nearest neighbor points for each wheel hub category point cloud cluster; The wheel hub category point cloud cluster is compared point by point with the preset standard wheel hub category point cloud cluster to obtain the final measurement distance; the objective function of the ICP algorithm is optimized based on the final measurement distance, and wheel hub comparison analysis is performed based on the optimized objective function; the final measurement distance is obtained from the initial distance and the second measurement distance of the ICP algorithm; the second measurement distance is obtained based on the geometric feature difference corresponding to the optimal number of nearest neighbors.

2. The method for precise comparative analysis of irregular curved surface components based on 3D laser scanning according to claim 1, characterized in that, The hub point cloud includes: rim point cloud, bolt hole point cloud, spoke point cloud, and transition point cloud.

3. The method for precise comparative analysis of irregular curved surface components based on 3D laser scanning according to claim 1, characterized in that, The method for obtaining the overall curvature includes: In each iteration, in each hub category point cloud cluster, with each hub point cloud as the center, the fitted surface of each hub point cloud is obtained by fitting according to the number of corresponding nearest neighbor points; The average curvature of the fitted surface of all points within the point cloud cluster of the wheel hub category is used as the overall curvature.

4. The method for precise comparative analysis of irregular curved surface components based on 3D laser scanning according to claim 1, characterized in that, The process of obtaining the data basis for the secondary clustering based on all errors across all iterations includes: In each iteration, the absolute value of the difference between the actual overall curvature of the first point cloud cluster and the corresponding predicted overall curvature is taken as the error; For each first point cloud cluster, the average of all errors generated by the first point cloud cluster in all iterations is used as the data basis.

5. The method for precise comparative analysis of irregular curved surface components based on 3D laser scanning according to claim 3, characterized in that, The method for obtaining the range of local curvature includes: For each number of nearest neighbors, the maximum difference in curvature between the fitted surfaces in the point cloud cluster of the wheel hub category is used as the range of the local curvature.

6. The method for precise comparative analysis of irregular curved surface components based on 3D laser scanning according to claim 1, characterized in that, The method for determining the optimal number of nearest neighbors for each hub category point cloud cluster includes: Using the range of local curvature on the fitted surface of the wheel hub category point cloud cluster under different numbers of nearest neighbors as the ordinate and different numbers of nearest neighbors as the abscissa, a line graph is plotted for each wheel hub category point cloud cluster; the optimal number of nearest neighbors is selected based on the coordinate points in the line graph.

7. The method for precise comparative analysis of irregular curved surface components based on 3D laser scanning according to claim 6, characterized in that, The method for obtaining the optimal number of nearest neighbor points for the fitted surface of each wheel hub category point cloud cluster based on the line graph includes: The number of nearest neighbor points corresponding to the point with the lowest vertical axis is selected as the optimal number of nearest neighbor points for the hub category point cloud cluster.

8. The method for precise comparative analysis of irregular curved surface components based on 3D laser scanning according to claim 2, characterized in that, The step of obtaining the second metric distance based on the geometric feature differences corresponding to the optimal number of nearest neighbors includes: In each hub category point cloud cluster, with each hub point cloud as the center, the fitted surface of each hub point cloud is obtained by fitting based on the number of optimal nearest neighbors; The second metric distance of the bolt hole point cloud is the cosine of the angle between the normal vectors of the fitted surfaces in the source and target point clouds; The second metric distance of the transition point cloud is the absolute value of the difference in curvature between the fitted surfaces in the source point cloud and the target point cloud; The second metric distance of the spoke point cloud is the DTW distance between the FPFH feature descriptors of the fitted surfaces in the source and target point clouds; The second metric distance of the rim point cloud is the product of the absolute value of the difference in curvature between the fitted surfaces in the source and target point clouds and the absolute value of the product of the unit normal vectors.

9. The method for precise comparative analysis of irregular curved surface components based on 3D laser scanning according to claim 1, characterized in that, The method for obtaining the final metric distance of the point cloud cluster for each wheel hub category includes: The final distance measurement is the weighted sum of the initial distance and the second distance measurement; the weights of the initial distance and the second distance measurement are obtained by taking the weights using the inverse of the variance; the sum of the weights of the initial distance and the second distance measurement is a positive integer 1.

10. The method for precise comparative analysis of irregular curved surface components based on 3D laser scanning according to claim 1, characterized in that, The objective function of the ICP algorithm is optimized based on the final metric distance, and the wheel hub comparison analysis is performed based on the optimized objective function, including: The initial distance of the ICP algorithm objective function is replaced with the final metric distance to obtain the optimized objective function; the optimized objective function value is compared with a preset threshold to obtain the wheel hub comparison analysis results.

Images