A point cloud scanning processing method

Abstract

The present application discloses a point cloud scanning processing method, which relates to the field of computer graphics and three-dimensional reconstruction technology. This method not only realizes the precise registration and seamless fusion of multi-layer point cloud data, but also effectively removes noise, smoothes point cloud data, and significantly reduces the amount of data through a series of preprocessing steps, including statistical filtering, Gaussian filtering, average filtering, and random sampling or voxel grid filtering technology. In terms of sub-cluster processing, principal component analysis is used to perform deep data dimensionality reduction and feature extraction, recursively segment sub-clusters, set and adjust plug-in sub-clusters, and ensure accurate segmentation and detailed adjustment of sub-clusters. Finally, relying on the advanced Alpha shape algorithm, a three-dimensional surface model is constructed that is faithful to the main structure of the point cloud and delicately presents local details, and then a high-quality three-dimensional model is accurately extracted, laying a solid foundation for subsequent graphics processing and analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of computer graphics and three-dimensional reconstruction, and in particular to a point cloud scanning processing method. Background Art

[0002] With the rapid development of 3D scanning technology, point cloud data, as its core output form, has been widely used in 3D reconstruction, virtual reality, augmented reality, 3D printing, and geographic information systems (GIS). However, the processing and analysis of point cloud data still face many challenges, especially in terms of the accuracy and efficiency of model reconstruction. Traditional point cloud processing methods often focus on overall data simplification or noise reduction, but when processing point cloud data with complex boundaries and curvature changes, it is often difficult to achieve ideal reconstruction effects. This is because point cloud data in areas with complex boundaries and curvature changes have higher information density and more complex geometric features, requiring more refined processing strategies. The original point cloud data usually contains a lot of redundant information and noise, and may have problems such as missing or overlapping data due to the limitations of the scanning equipment or the influence of the scanning environment. Therefore, effective preprocessing and simplification of point cloud data is a key step in the 3D reconstruction process. However, traditional point cloud processing methods mainly focus on data denoising, smoothing, and sampling, aiming to improve data quality and reduce data volume, but often ignore the geometric features and topological structures in point cloud data, resulting in difficulty in accurately expressing the details and shapes of objects during reconstruction. Especially when point cloud data contain complex boundaries and curvature changes, traditional processing methods often cannot obtain satisfactory results. Summary of the invention

[0003] This application solves the problems of inaccurate point cloud data alignment, low processing efficiency, and low model construction quality in the prior art by providing a multi-layer point cloud data processing and three-dimensional model construction method, and realizes efficient and accurate alignment, preprocessing, segmentation, adjustment and simplification of point cloud data, as well as the construction of high-quality three-dimensional surface models and three-dimensional models, ensuring the accuracy and detail integrity of the model.

[0004] The present application provides a point cloud scanning processing method, comprising:

[0005] S1, obtain point cloud data and determine whether it is a single layer: use laser scanner, camera, 3D scanner to capture the surface point information of the object and convert it into point cloud data; if the point cloud data has obvious features at different heights, it may indicate that it is multi-layered;

[0006] S2, fusion of multi-layer point cloud data: using the iterative closest point algorithm, through iterative search for the closest point pair and calculation of the optimal transformation matrix, to achieve accurate registration of point cloud data and unified coordinate system, the point cloud space is divided into multiple small cubic grids based on the voxel method for deduplication and fusion;

[0007] S3, preprocessing the fused point cloud data: using statistical filtering algorithm to calculate the distance between each data point and its neighboring points and remove noise points, using Gaussian filtering for local smoothing, and using voxel grid filtering to reduce the amount of point cloud data;

[0008] S4, sub-cluster segmentation: perform PCA analysis to determine the principal component direction of the point cloud data, thereby dividing two sub-clusters with different characteristics, and continuously refine them through recursive segmentation until the preset conditions are met;

[0009] S5, plug-in sub-cluster processing, adjustment and inference of subsequent changes of objects: S501, calculate the covariance matrix of all data points in the sub-cluster to reflect the distribution characteristics of the data points in space, decompose the eigenvalues ​​of the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors; S502, check the boundary areas of adjacent sub-clusters after sub-cluster segmentation through the boundary detection algorithm, and extract point cloud data along the boundary line to define as plug-in sub-clusters; S503, adjust the center point position and curvature radius of the plug-in sub-cluster;

[0010] S6, simplifying the point cloud data within the sub-cluster: by calculating the curvature difference or distance between the data point and its adjacent points, analyzing the contribution, and giving priority to retaining the data points with large contribution; using filtering methods to remove data points with small contribution, retaining the key geometric shape and curvature features of the sub-cluster;

[0011] S7, constructing a three-dimensional surface model: calculating the average point distance of the simplified point cloud data, multiplying the average point distance by a coefficient determined according to the density and distribution characteristics of the point cloud to obtain an Alpha value, where the coefficient value range is generally between 0.5 and 2, running the Alpha shape algorithm, checking the distance between point pairs, and constructing an Alpha shape around the point cloud data;

[0012] S8, extracting the 3D model: extracting the OBJ general 3D model format 3D model from the Alpha shape, ensuring that the connection relationship between the vertices, edges and faces of the model is consistent with the Alpha shape, while maintaining the normal direction of the face, and processing boundaries, holes and overlapping parts by clipping and merging.

[0013] Preferably, the iterative closest point algorithm of S2 includes:

[0014] Iteratively find the closest point pairs between the two sets of point clouds, and calculate the optimal transformation matrix between these closest points in each iteration. The matrix contains information about rotation and translation, so that the two sets of point clouds are aligned in space; by repeating this process continuously, all data are unified into the same coordinate system.

[0015] Preferably, the voxel-based method of S2 comprises:

[0016] The point cloud space is divided into multiple small cube grids, each cube represents a voxel. For each voxel, all the points contained in it are traversed, and the average coordinates of the traversed points in the X, Y, and Z directions are calculated respectively. The average value in the X direction is obtained by adding the X coordinates of all points and dividing by the number of points; the average values ​​in the Y and Z directions are obtained in the same way. The average values ​​in the three directions are combined to obtain the average position of all points in the voxel, which is expressed in three-dimensional coordinates.

[0017] Preferably, the step S501, calculating the covariance matrix, includes:

[0018] Covariance Matrix Formula , where C represents the covariance matrix. Assuming the data has d dimensions, it is a d*d matrix. The data point represents the i-th data point in the data set, which is a d-dimensional vector. The mean of the data point represents the mean vector of all data points in the data set. It is also a d-dimensional vector. The mean vector is obtained by calculating the average of all data points in each dimension. n is the number of data points. is a normalization factor used to ensure an unbiased estimate of the covariance matrix.

[0019] Preferably, the S5 comprises:

[0020] S504, setting a plug-in strategy for the low curvature sub-cluster: selecting 2 to 5 columns of point cloud data as plug-in sub-cluster candidates, extracting the selected point cloud data columns, and formally defining them as plug-in sub-clusters;

[0021] S505, analyzing and processing subclusters in the three-dimensional space by integrating boundary and time characteristics: using boundary detection algorithm to identify boundary points, using Euclidean distance formula to calculate the actual straight-line distance between two points, analyzing the relative position relationship between the boundary point and its adjacent points, and determining whether the boundary is concave inward or convex outward;

[0022] S506, determining the initial number of columns of the plug-in sub-clusters: setting a basic number of columns of the plug-in sub-clusters as a starting point for all sub-clusters, and adjusting the number of columns of the plug-in sub-clusters according to the boundary length, the concave-convex variability of the boundary shape, the consistency of the normal direction, the surface roughness, and the curvature variation factor in the time series;

[0023] S507, dynamically adjusting the number of sub-cluster columns and finally determining the number of columns: real-time monitoring of the boundary smoothness, detail retention, matching degree of time series changes, and normal direction consistency of the plug-in process to increase or decrease the number of plug-in sub-cluster columns;

[0024] S508, constructing an object change trend prediction model: using point cloud data to analyze the shape and structure of the object, dividing the surface of the object into multiple sub-clusters, each sub-cluster represents a small area on the surface of the object, calculating and analyzing the curvature change characteristics of the sub-clusters to infer the future changes in the shape of the object, and using linear regression method to fit the relationship between the curvature change, the adjustment amount of the number of plug-in columns and the state of the object to construct a prediction model.

[0025] Preferably, the S507 further includes:

[0026] For problem areas found in real-time monitoring, such as jagged edges, time series mismatch, or significant changes in normal direction, for boundary areas near time nodes with drastic changes in curvature, significant changes in time series, or large changes in normal direction, if the processing effect is not good, increase the number of plug-in sub-cluster columns; conversely, for areas with redundant columns and limited contribution to the overall effect, reduce the number of columns.

[0027] Preferably, the step S6, which preferentially retains data points with large contribution, includes:

[0028] Contribution formula = α × curvature difference + β × distance difference, where α and β are weight coefficients used to balance the proportion of curvature difference and distance difference in contribution calculation.

[0029] Preferably, the method further comprises:

[0030] S901, add time dimension information: when acquiring point cloud data, ensure that a timestamp is attached to each data point. The timestamp should record the precise acquisition time of the data point, with an accuracy of at least milliseconds and an accuracy of at least 1 millisecond;

[0031] S902, performing in-depth analysis on the time series information in the point cloud data: identifying the changing trend of data points over time and identifying mutation points in the data;

[0032] S903, sub-cluster segmentation in the time dimension: using the time window as the evaluation basis, when the timestamps of data points fall within the same time window range, these points are regarded as potential members of the same sub-cluster;

[0033] S904, setting up plug-in sub-clusters in the time dimension: giving priority to areas showing obvious changes in the time series, increasing the number of columns of plug-in sub-clusters at these key boundaries; reducing the number of columns for stable sub-cluster boundaries in the time series;

[0034] S905, adjust the plug-in sub-cluster: analyze the parameters of the initial position, size, number of columns and initial curvature radius of the time series plug-in sub-cluster, adjust the position and curvature radius of the center point of the plug-in sub-cluster to best match the changes in adjacent sub-clusters and time series; identify its significant change points and time periods. If it is found that the plug-in sub-cluster fails to fully reflect specific changes in the time series, increase or decrease the number of columns or fine-tune the boundaries to correspond to the turning points of the time series.

[0035] Preferably, the S902 includes:

[0036] A change point detection algorithm is used to accurately identify mutation points in the data, which are key time points corresponding to significant changes in the shape, position or density of objects.

[0037] Preferably, the S7, Alpha shape algorithm, further comprises:

[0038] Calculate the average value of the distance between adjacent points of the simplified point cloud as the average point distance, and multiply the average point distance by a coefficient determined according to the density and distribution characteristics of the point cloud to obtain the Alpha value, and the coefficient value range is generally between 0.5 and 2; after determining the Alpha value, run the Alpha shape algorithm to traverse the point cloud data, and set each pair of points in the traversal as point A and point B, and the algorithm will calculate the distance between the two points; draw a circle with point A as the center and the Alpha value as the radius; check whether point B is within the circle. If point B is not within the circle, the distance between point A and point B is far, and they represent a significant feature of the object surface, such as an edge or corner, and the pair of points is considered to be part of the Alpha shape boundary; if point B is within the circle, the distance between point A and point B is close, and they are considered to belong to the same part of the object surface, and the direct connection between the points can be ignored when constructing the Alpha shape.

[0039] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0040] This embodiment achieves efficient and accurate registration and fusion of multi-layer point cloud data by adopting the iterative closest point (ICP) algorithm and the voxel method, thereby improving the quality and efficiency of data processing. Through preprocessing steps, including denoising, filtering and downsampling, cleaner, smoother and easier to process point cloud data is obtained. Through PCA analysis and recursive segmentation strategy, accurate segmentation of point cloud data is achieved, and subclusters that can more accurately reflect the intrinsic structure and characteristics of point cloud data are obtained. The quality and consistency of point cloud data are further improved by calculating the curvature of subclusters, setting and adjusting plug-in subclusters, and simplifying subcluster point clouds based on curvature. Finally, a three-dimensional surface model is constructed and a three-dimensional model is extracted, realizing the complete conversion from point cloud data to a three-dimensional model. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of a point cloud scanning processing method according to an embodiment of the present invention;

[0042] Figure 2 A flowchart of setting plug-in sub-clusters in a time series in a point cloud scanning processing method according to an embodiment of the present invention;

[0043] Figure 3 The present invention is a flowchart of dynamically adjusting plug-in sub-clusters in a point cloud scanning processing method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] To facilitate the understanding of the present invention, the present application will be described more comprehensively below with reference to the relevant drawings; the drawings show preferred embodiments of the present invention, but the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to enable a more thorough and comprehensive understanding of the disclosed content of the present invention.

[0045] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which the present invention belongs; the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention; the term "and / or" used herein includes any and all combinations of one or more related listed items.

[0046] In the face of multiple challenges in point cloud data processing, including the inaccuracy of multi-layer data fusion, the impreciseness of sub-cluster segmentation, and the low processing efficiency caused by the huge amount of data, this embodiment proposes a point cloud scanning processing method. The method aims to effectively respond to and solve the above challenges through a series of advanced technical means, thereby achieving efficient and accurate processing of point cloud data and extracting high-quality three-dimensional models.

[0047] Embodiment 1: Figure 1 It is a flow chart of a point cloud scanning processing method according to an embodiment of the present invention.

[0048] like Figure 1 As shown, a point cloud scanning processing method comprises the following steps:

[0049] S1, obtain point cloud data and determine whether the point cloud data is single-layer: capture the point information on the surface of the object through a laser scanner, camera, or 3D scanner, and convert it into point cloud data; by observing the distribution of point cloud data, if there are obvious gaps between point cloud data at different heights, it may indicate that the point cloud data is multi-layered; by analyzing the density of the point cloud, if the point cloud data is dense in some areas and sparse in other areas, and this alternation of density and sparseness is related to height, it may also be a feature of multi-layer point cloud data; by calculating the height difference of the point cloud data, if there are multiple height peaks in the point cloud data and there is a 5mm height difference between these peaks, when the height difference is detected to exceed this threshold, it is considered to be multi-layer point cloud data. .

[0050] S2, fusion of multi-layer point cloud data: the iterative closest point (ICP) algorithm is used to iteratively find the closest point pairs between the two groups of point clouds. In each iteration, the optimal transformation matrix between these closest points is calculated. This matrix contains information about rotation and translation, so that the two groups of point clouds can be optimally aligned in space. The number of iterations of the ICP algorithm is set to a maximum of 100 times. After each iteration, the average distance error between the two groups of point clouds is calculated. The iteration is stopped when the error is less than 0.001mm or the number of iterations reaches the upper limit. By repeating this process continuously, the ICP algorithm can achieve accurate alignment of multi-layer point cloud data and ensure that all data are unified into the same coordinate system.

[0051] In order to further improve the quality and processing efficiency of point cloud data, a voxel-based method is used for deduplication and fusion. The point cloud space is divided into multiple small cube grids, each cube represents a voxel, and the voxel grid size is set to 5mm*5mm*5mm according to the density of the point cloud and the required accuracy. For each voxel, all the points contained in it are traversed, and the average coordinates of the traversed points in the three directions of X, Y, and Z are calculated respectively. The average value in the X direction is obtained by adding the X coordinates of all points and dividing them by the number of points; similarly, the average values ​​in the Y and Z directions can also be obtained. Combining the average values ​​in these three directions, the average position of all points in the voxel is obtained. This position can be represented by a three-dimensional coordinate, which represents the center position of all points in the voxel. It can effectively remove duplicate points and reduce the amount of point cloud data. At the same time, it can ensure that the unique features and details of each layer of point cloud are retained to avoid information loss. In practical applications, if there are noise or outliers in the point cloud data, it may have a certain impact on the calculation of the average position. It is necessary to combine other preprocessing steps such as filtering and denoising to improve the accuracy of the calculation results.

[0052] S3, preprocessing the fused point cloud data: the data preprocessing of point cloud data includes denoising, filtering and downsampling; the distance between each data point and its neighboring points is calculated by statistical filtering algorithm, and the average of these distances is calculated to remove noise points whose neighboring point distance is much larger than the average distance, so as to eliminate noise points in the point cloud data; the standard deviation of the filter is set to 0.5mm by Gaussian filtering and the average filtering of the neighborhood size is selected to perform local smoothing on each point in the point cloud to achieve filtering effect and remove outliers; in order to reduce the data volume of the point cloud and improve processing efficiency, the voxel grid is filtered to reduce the data volume of the point cloud and improve processing efficiency while maintaining the overall shape of the point cloud. The effect after filtering is to achieve a noise point removal rate of 95% and a 10% reduction in the number of point clouds; the smoothness of the point cloud is improved by 30% after Gaussian filtering, and the standard deviation is reduced from 0.05mm of the original point cloud to 0.035mm.

[0053] S4, segmentation of sub-clusters: Perform PCA analysis to standardize the point cloud data to eliminate the dimensional differences between different dimensions, calculate the covariance matrix to capture the related rows between dimensions, solve the eigenvalues ​​and eigenvectors to determine the main component direction of the point cloud data, and thus divide two sub-clusters with different characteristics. The main component direction corresponding to the eigenvector represents the main variation direction of the data. The eigenvectors corresponding to the three largest eigenvalues ​​are selected as the basis for segmentation. Recursive segmentation is used to continuously refine the sub-clusters, select a main component direction as the current segmentation axis, project all points in the sub-cluster onto the segmentation axis, and obtain a one-dimensional projection value. According to the distribution of the projection value, select a suitable threshold (the initial number of sub-clusters is about 100, which is reduced to about 50 after recursive segmentation, while maintaining more than 98% of the geometric features; the threshold can be the mean, median or a percentile of the projection value, depending on the distribution of the data and the segmentation requirements). Finally, the sub-cluster is divided into two new sub-clusters according to the threshold, one containing points with projection values ​​less than the threshold, and the other containing points with projection values ​​greater than or equal to the threshold. The recursive segmentation process will continue until the distribution of data points in all sub-clusters meets the preset conditions (the specific preset conditions need to be determined according to the actual application scenario and data processing requirements, and the means and methods involved can refer to the relevant existing technologies, which will not be elaborated in the present invention):

[0054] S401, segmenting two sub-clusters with different characteristics: after determining the direction of the principal component through the above PCA analysis, construct a plane passing through the center of the point cloud and having a normal direction consistent with the direction of the principal component, using this plane as a segmentation basis, selecting a sub-cluster containing points located on one side of the plane, these points have similar characteristics on the principal component; and the other sub-cluster contains points located on the other side of the plane, these points present different characteristics in the direction of the principal component;

[0055] S402, recursively segment two subclusters with different characteristics: introduce a recursive segmentation strategy, combine principal component analysis (PCA) and plane segmentation, and refine and optimize the segmentation results of the subclusters by continuously comparing the farthest distance from the data point to the subcluster center with the preset threshold until the conditional position is met. During PCA analysis, the eigenvectors corresponding to the first three largest eigenvalues ​​are selected as the principal component directions. Specifically, the farthest distance from the data point in each subcluster to the center of its subcluster is calculated. This distance reflects the degree of dispersion of the data points inside the subcluster; if the distance exceeds the preset farthest distance threshold, it indicates that the data points inside the subcluster are relatively dispersed and further segmentation optimization is required; at this time, the same segmentation operation is recursively performed on the two subclusters: that is, the plane passing through the subcluster center and the normal direction is consistent with the corresponding principal component direction is used again for segmentation; then, the comparison process of calculating the farthest distance and the threshold is repeated to ensure continuous optimization of the segmentation; this recursive segmentation operation will continue until the farthest distance from the data points in all subclusters to the corresponding subcluster center is less than the preset threshold. During recursive segmentation, the maximum distance threshold is set to 1 / 10 of the average size of the sub-cluster. When the distance from the point in the sub-cluster to the center exceeds this threshold, further segmentation is performed.

[0056] S5, processing and adjustment of sub-clusters: perform point cloud curvature analysis on sub-clusters, select the k points closest to each data point as points in the neighborhood, determine its neighborhood range, calculate the normal vector between the data point and its neighboring points (the direction of the surface normal at the data point), and set the plug-in sub-clusters based on the degree of curvature change and adjust the position, shape, size and other parameters of the plug-in sub-clusters to ensure the overall coherence and consistency of the point cloud data:

[0057] S501, obtain the curvature of the sub-cluster: calculate the covariance matrix of all data points in the sub-cluster. The covariance matrix formula is , where C represents the covariance matrix, which is a d*d matrix (assuming the data has d dimensions), is a data point (representing the i-th data point in the data set, which is a d-dimensional vector), is the mean of the data points (representing the mean vector of all data points in the data set, which is also a d-dimensional vector. The mean vector is obtained by calculating the average of all data points in each dimension), n is the number of data points, is a normalization factor used to ensure an unbiased estimate of the covariance matrix. The covariance matrix reflects the covariance relationship between the dimensions of the data points, that is, the distribution characteristics of the data points in space; the covariance matrix is ​​subjected to eigenvalue decomposition to obtain a set of eigenvalues ​​and corresponding eigenvectors, where the eigenvalues ​​represent the "strength" or "importance" of the matrix in the direction of the corresponding eigenvector; the curvature of the sub-clusters in different directions is evaluated by comparing the size of the eigenvalues ​​in different directions. Usually, the size of the eigenvalues ​​reflects the distribution range or discreteness of the data in each direction;

[0058] S502, setting up plug-in subclusters: In the process of setting up plug-in subclusters, firstly check the boundary areas of adjacent subclusters after subcluster segmentation, especially the boundaries with significant curvature changes exceeding the preset threshold; after subcluster segmentation, the number of subclusters is reduced from the original 100 to 50, while maintaining more than 98% of the geometric features; for the above-mentioned boundaries, the approximate position of the subcluster boundary is identified by the boundary detection algorithm, and a set of preliminary edge points is obtained, which may contain noise and unnecessary details, and the noise and pseudo-edges in the edge extraction results are removed, the real boundary information is retained, and the edge line is further refined to make it closer to the actual subcluster boundary, wherein the setting of the curvature threshold depends on experimental data and experience; in the identified boundary area, one or more columns of point cloud data are extracted along the boundary line to be defined as plug-in subclusters, and the number of columns of the plug-in subcluster is determined based on the degree of curvature change. In the area with large curvature change, a larger number of columns are selected as plug-in subclusters. After the plug-in subcluster is processed, the boundary smoothness is improved by 20%, and the model reconstruction error is reduced from 0.5mm to 0.3mm, so as to ensure that the characteristics of the boundary can be fully reflected;

[0059] S503, adjust the plug-in sub-cluster: adjust the plug-in sub-cluster to ensure that the transition between it and the adjacent sub-clusters is smoother and more natural; in the specific operation, first observe and analyze the curvature of the adjacent sub-clusters, and adjust the position of the plug-in sub-cluster based on this; the adjustment content mainly includes two aspects: one is the position of the midpoint of the plug-in sub-cluster. By adjusting the position of these points, the plug-in sub-cluster can better adapt to the shape of the adjacent sub-clusters; the second is the curvature radius of the plug-in sub-cluster. Adjusting the curvature radius can make the curvature change of the plug-in sub-cluster consistent with the adjacent sub-clusters or show a gradual trend; such an adjustment process may require multiple iterations and optimizations, and continuously adjust the parameters of the plug-in sub-cluster until the best visual effect and data quality are achieved to ensure the coherence and consistency of the entire data set.

[0060] S6, based on curvature, simplifies the point cloud data in the sub-cluster: calculate the curvature difference (the difference in curvature value between the current data point and its neighboring points, reflecting the importance of the data point in the change of the sub-cluster shape) and distance difference (the spatial distance between the current data point and its neighboring points, reflecting the influence of the data point on the spatial distribution of the sub-cluster) between each data point in the sub-cluster, so as to analyze the contribution of each data point to the sub-cluster shape and curvature. Contribution formula = α × curvature difference + β × distance difference, α and β are weight coefficients, which are used to balance the proportion of curvature difference and distance difference in contribution calculation. The α curvature difference weight is between 0.6-0.8, and the β distance difference weight is 1-α. According to the size of the contribution, the data points are sorted, and the data points with high contribution to the sub-cluster shape and curvature are retained. The filtering method is used to set the filtering threshold, traverse the data points, apply the filtering conditions, and delete the redundant data points to process the remaining data points, and remove those data points with small contribution to reduce the amount of data. During the filtering process, special attention should be paid to the setting of the filtering threshold. The threshold range should be obtained by experimental simulation based on the actual situation to ensure that the setting of these parameters can retain the key geometric shape and curvature characteristics of the sub-cluster as much as possible, avoiding data distortion caused by over-simplification.

[0061] S7, using the Alpha shape algorithm to construct a three-dimensional surface model that reflects the point cloud structure: First, calculate the average value of the distance between adjacent points in the simplified point cloud as the average point distance and set it as D, and multiply the average point distance by a coefficient determined according to the point cloud density and distribution characteristics to obtain the Alpha value (the coefficient value range is generally between 0.5 and 2). Set the Alpha value to 1.5D coefficient is selected as 1.5 according to the point cloud density. After determining the Alpha value, run the Alpha shape algorithm, traverse the point cloud data, for each pair of points in the traversal (set as point A and point B), draw a circle with point A as the center and the Alpha value as the radius, and check whether point B is in the circle. If point B is not in the circle, it means that the distance between point A and point B is far, and they represent a significant feature of the object surface (such as an edge or corner), so this pair of points is part of the Alpha shape boundary; if point B is in the circle, it means that the distance between point A and point B is close, and they belong to the same part of the object surface, so the point direct connection between the pair of points can be ignored when constructing the Alpha shape. For each pair of points A and B, if the circle with A as the center and Alpha as the radius does not contain point B, then it is considered that points A and B constitute the boundary of the Alpha shape. Through this process, the Alpha shape surrounding the point cloud data is gradually constructed, thus forming a three-dimensional surface model that reflects the point cloud structure.

[0062] S8, extract the 3D model based on the Alpha shape: extract the OBJ general 3D model format 3D model from the Alpha shape, and ensure that the points, edges and faces in the Alpha shape are converted to the representation of vertices, edges and faces in the 3D model file respectively. Specifically, each point in the Alpha shape should become a vertex of the model; each edge formed by a pair of connected points should correspond to an edge in the model; and the face surrounded by the edges needs to be converted into a face in the model. The Hausdorff distance between the extracted 3D model and the original point cloud data is 0.2mm. During the conversion process, the connection relationship between the vertices, edges and faces of the model is kept consistent with the Alpha shape, and the consistency of the normal direction of the face is maintained at 98%. For the boundary, it is necessary to ensure that it is still clearly identifiable in the converted model; for holes, the normal direction of the face must also be maintained to ensure the correct display and subsequent processing of the model. For the boundary of the Alpha shape, special attention should be paid to whether it is still clearly identifiable in the converted model; for the existing holes, they must be accurately represented in the model to avoid misunderstanding or ambiguity in subsequent processing. For the overlapping parts, it is necessary to eliminate the overlap by cutting, merging, etc. to ensure the accuracy of the model. After completing these conversion and processing steps, a 3D model corresponding to the original point cloud data can be obtained.

[0063] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0064] This application uses a series of advanced and efficient technical means such as multi-layer point cloud data fusion, preprocessing, sub-cluster segmentation, sub-cluster processing and adjustment, point cloud data simplification, and Alpha shape algorithm to build a three-dimensional surface model, which realizes efficient and accurate processing of point cloud data and accurate construction of three-dimensional models. The application of this series of technical means not only effectively improves the accuracy and efficiency of point cloud data processing, so that point cloud data can better reflect the shape and characteristics of actual objects, but also ensures that the constructed three-dimensional model has good coherence, consistency and detail expression, providing a reliable foundation for subsequent model analysis and application.

[0065] Embodiment 2: Although embodiment 1 performs plug-in processing on subclusters whose curvature exceeds the threshold, this method does not make the overall point cloud data achieve more uniform and consistent curvature characteristics after processing. Optimizing only some subclusters may lead to uneven curvature distribution, affect subsequent 3D modeling or analysis work, and may cause uneven model surface or loss of details, reducing model accuracy and quality.

[0066] Therefore, the embodiment of the present application is optimized based on the above embodiment, and specifically includes the following steps:

[0067] S504, implement the plug-in sub-cluster setting strategy for low curvature sub-clusters to enhance the overall curvature consistency: The aforementioned steps S501 and S502 effectively divide and refine the sub-clusters in the point cloud data through PCA analysis and recursive segmentation strategy, laying the foundation for structured processing. However, the processing of the boundary area of ​​the low curvature sub-clusters is equally critical and is essential to maintaining the consistency and coherence of the data set. Therefore, this embodiment implements a plug-in sub-cluster strategy for low curvature sub-clusters, supplements and improves the sub-cluster segmentation steps, and ensures the uniformity and consistency of the curvature distribution in the processing flow. By setting plug-in sub-clusters, the curvature changes in the boundary area of ​​the low curvature sub-clusters are optimized, and a smoother and more natural transition with adjacent sub-clusters is achieved.

[0068] For low curvature, suppose the average curvature of all subclusters in the point cloud data is calculated, and the median of these average curvatures is found (assuming it is 0.01, note that this is only an example value, and the actual value will vary depending on the specific data set). A low curvature threshold is set based on the median, and a certain proportion of the median (50% in this example) is taken as the threshold, that is, 0.005. Any subcluster with an average curvature lower than 0.005 is considered a low curvature subcluster, and specific processing strategies may be required, such as plugging subclusters, to enhance the curvature consistency and processing effect of the overall point cloud data.

[0069] In the specific operation, the boundary detection algorithm is first used to accurately identify the boundary points of the low curvature sub-cluster, which is the premise for ensuring the accuracy of the plug-in sub-cluster setting. Then, in the identified boundary area, one or more columns of point cloud data are selected along the boundary line as candidates for the plug-in sub-cluster. Considering that the curvature of the low curvature sub-cluster itself changes relatively slowly, 2 to 5 columns of point cloud data are selected as candidates for the plug-in sub-cluster when selecting the number of columns, so as to reduce data redundancy as much as possible while retaining the boundary features and avoid unnecessary computational burden. After that, the selected point cloud data columns are extracted and formally defined as plug-in sub-clusters. This step ensures that the boundaries of the plug-in sub-cluster and the low curvature sub-cluster are closely aligned, providing a basis for subsequent optimization and adjustment.

[0070] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0071] This application adopts a plug-in sub-cluster setting strategy for low curvature sub-clusters, and combines the boundary detection algorithm to accurately identify and select point cloud data columns, thereby realizing the optimization processing of the boundary area of ​​the low curvature sub-cluster. By setting the plug-in sub-clusters and adjusting their parameters, the transition between the low curvature sub-cluster and the adjacent sub-clusters is smoother and more natural, thereby enhancing the consistency of the overall curvature and improving the accuracy and quality of subsequent three-dimensional modeling or analysis work.

[0072] Embodiment three: Figure 2 The present invention is a flowchart of a point cloud scanning processing method for setting plug-in sub-clusters in a time series according to an embodiment of the present invention.

[0073] When processing and analyzing point cloud data, usually only spatial location information can be obtained, and time dimension information is lacking. This makes it difficult to consider the changes in data over time during fusion, segmentation, and adjustment, which may affect the accuracy and effectiveness of the processing. For example, when fusing multi-layer point cloud data, the lack of timestamps may make it impossible to accurately judge the time relationship between different layers of data, affecting the accuracy of the fusion results. Similarly, when segmenting and adjusting sub-clusters, the lack of timestamp information may also make it impossible to accurately identify and process boundary features that change over time, affecting the final 3D modeling accuracy and quality.

[0074] Therefore, the embodiment of the present application is optimized based on the above embodiment, and specifically includes the following steps:

[0075] S901, add time dimension information: When acquiring point cloud data, ensure that a timestamp is attached to each data point. The timestamp should record the precise acquisition time of the data point, with an accuracy of at least 1 millisecond to ensure that subtle time changes in the point cloud data can be captured. In particular, this step is particularly important when the point cloud data comes from time series scanning. In order to ensure the accuracy of time series scanning data, the built-in clock of the scanning device must be highly accurate and the necessary clock synchronization operations must have been performed. Clock synchronization is particularly critical for multiple devices working at the same time. It can effectively avoid the problem of inconsistent timestamps caused by clock deviation between different devices, thereby ensuring the reliability of subsequent data analysis. In addition, the start and end times of each scanning sequence should also be accurately recorded, and ensure that these timestamps are synchronized with the built-in clock of the scanning device.

[0076] S902, time series analysis: in-depth analysis of the time series information in the point cloud data is carried out, with the main purpose of identifying the changing trend of data points over time. In this process, a change point detection algorithm is used to accurately identify mutation points in the data. These mutation points are regarded as key time points, which often correspond to significant changes in the shape, position or density of objects. The above-mentioned change point detection algorithm can use any one of the CUSUM (Cumulative Sum) algorithm and the Likelihood Ratio Test (LRT). Please refer to the relevant prior art for details, and the present invention will not elaborate on this.

[0077] S903, sub-cluster segmentation in the time dimension: When performing sub-cluster segmentation, the spatial attributes and time series information of the point cloud data are comprehensively considered to ensure that the segmented sub-clusters maintain coherence and consistency in both time and space dimensions. Specifically, a time window is introduced as an evaluation basis to measure the similarity of timestamps between data points. Only when the timestamps of data points fall within the same time window range will these points be considered as potential members of the same sub-cluster. The specific time window size is set according to actual conditions.

[0078] At the same time, the main features of the point cloud data are extracted and reduced in dimension by combining PCA analysis, and the sub-clusters are refined by recursive segmentation strategy, which further ensures the rationality of the sub-cluster division. In this process, time series analysis plays an important role in helping to identify significant changes in point cloud data at specific time points or time periods. Therefore, when performing sub-cluster segmentation, special attention should be paid to ensure that the data change characteristics at these key time nodes or time periods can be fully reflected in the final sub-cluster division results. For the detailed steps of PCA analysis in this step, refer to the detailed steps of PCA analysis in step S4.

[0079] S904, plug-in sub-cluster settings under the time dimension: When setting plug-in sub-clusters, the spatial relationship between sub-clusters and time series information need to be comprehensively considered. Sub-cluster boundaries with high curvature or significant curvature changes, especially those areas that show obvious changes in the time series, should be given priority. This means that at these key boundaries, the number of columns of 1-3 plug-in sub-clusters needs to be appropriately increased to accurately capture and reflect the dynamic changes in the time series. On the contrary, for more stable sub-cluster boundaries in the time series, given that the changes in the plug-in sub-clusters may be relatively limited, it is necessary to reduce the number of columns and retain 2-3 columns, thereby maintaining the necessary accuracy while improving processing efficiency.

[0080] S905, adjustment of plug-in sub-clusters in the time dimension: When adjusting the plug-in sub-clusters, it is necessary to comprehensively consider the time series information, the spatial consistency of the sub-clusters and the curvature factors. First, the time series is deeply analyzed to accurately identify the significant change points and time periods. Then, for the plug-in sub-clusters in these key time points or time periods, it is evaluated whether they accurately reflect the dynamic changes in the time series. In order to ensure the coherence and consistency of the plug-in sub-clusters in the time dimension, it is required that the timestamps of the data points in the sub-clusters are similar, and the change trajectory is consistent with the fluctuation of the time series. Using an iterative optimization method, according to the initial spatial segmentation and time series analysis results, the initial position, size, number of columns, and initial curvature radius of the plug-in sub-cluster are defined, and the position and curvature radius of the center point of the plug-in sub-cluster are adjusted to best match the changes in adjacent sub-clusters and time series. If it is found that the plug-in sub-cluster fails to fully reflect the specific changes in the time series, it is adjusted according to the actual situation, such as increasing or decreasing the number of columns to capture the time changes more carefully, or fine-tuning the boundaries to ensure accurate correspondence to the turning points of the time series. Through meticulous operations, the parameter configuration of the plug-in sub-clusters is continuously optimized and iterated until the visual effect and data quality reach the best, laying a solid foundation for subsequent data analysis and processing.

[0081] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0082] This application adopts a method of adding time dimension information to point cloud data to achieve comprehensive capture and analysis of the spatiotemporal characteristics of point cloud data. By attaching a timestamp accurate to the millisecond level to each data point and ensuring the precise synchronization of the scanning device clock, it is possible to capture subtle time changes in the point cloud data, so that in the process of multi-layer point cloud data fusion, sub-cluster segmentation and adjustment, the changes in data over time can be fully considered. At the same time, combined with time series analysis, PCA dimensionality reduction, recursive segmentation strategy and comprehensive consideration of curvature factors, the coherence and consistency of sub-clusters in both time and space dimensions are achieved, ensuring the accuracy and quality of the final three-dimensional modeling. This series of innovative measures not only improves the accuracy and effectiveness of point cloud data processing, but also provides a more reliable foundation for subsequent data analysis and application.

[0083] Embodiment 4: Figure 3 The present invention is a flowchart of dynamically adjusting plug-in sub-clusters in a point cloud scanning processing method according to an embodiment of the present invention.

[0084] In the above embodiment, the number of columns of the plug-in sub-clusters is set based on the static degree of curvature change, lacking a dynamic adjustment mechanism, which may lead to data redundancy or loss of important boundary features. At the same time, only high-curvature sub-clusters are processed, resulting in uneven curvature distribution, which in turn affects the accuracy and quality of subsequent three-dimensional modeling. Although the plug-in strategy is also implemented for low-curvature sub-clusters to enhance the consistency and coherence of the data set, this increases the processing complexity and time cost, and the selection of the number of columns of the plug-in sub-clusters still depends on the static threshold and lacks dynamic adjustment capabilities, resulting in the curvature consistency may still not be optimal.

[0085] Therefore, the embodiment of the present application is optimized based on the above embodiment, and specifically includes the following steps:

[0086] S505, comprehensive boundary and time characteristic analysis: For each subcluster to be processed, a specific boundary detection algorithm is used to accurately identify its boundary points. Based on the three-dimensional coordinate information of the boundary points, the Euclidean distance formula is used to calculate the straight-line distance between the two points to obtain the actual length of the boundary, so as to deeply understand the scale and scope of the boundary. By analyzing the relative position relationship between the boundary point and its adjacent points, it is determined whether the boundary is concave inward or convex outward, so as to comprehensively evaluate the shape change of the boundary. Combined with the curvature change and timestamp information, further explore the change of boundary characteristics to ensure the accuracy and effectiveness of subsequent processing. In three-dimensional space, if point A and point B Then the Euclidean distance formula between them is:

[0087] d=

[0088] S506, determining the initial number of columns of the plug-in sub-clusters: Determining the initial number of columns of the plug-in sub-clusters is a multi-dimensional comprehensive consideration process involving boundary length, shape change, normal direction consistency, surface roughness, and time series analysis results.

[0089] A basic number of plug-in sub-cluster columns is set as the starting point for all sub-clusters. For the boundary length, its actual length is directly used as the basis for adjusting the number of columns. The longer the boundary, the more columns of the plug-in sub-cluster should be increased to maintain processing accuracy. For the shape change of the boundary, especially the convexity and concavity, the more complex the shape change, the more columns need to be increased accordingly. The consistency of the normal direction is quantified by calculating the standard deviation of the normal direction on the boundary point. A large standard deviation indicates a significant change in the normal direction and high surface direction inconsistency. At this time, the number of plug-in sub-cluster columns needs to be increased to better capture this change. The surface roughness is quantified using the root mean square roughness or average roughness of the statistical index. The number of columns also needs to be increased in areas with high roughness to ensure processing accuracy. And pay special attention to the curvature changes of the sub-clusters in the previous few point cloud scan data, and increase the number of plug-in sub-cluster columns according to the curvature anomalies or mutation points to achieve more refined processing. At the same time, the time series analysis results should also be considered to analyze the time distribution of each point in the point cloud. For areas with large time spans or significant changes, it may be necessary to increase the number of plug-in sub-cluster columns to capture the changing characteristics of the time series. Taking all the above factors into consideration, a weighted score is used to determine a reasonable initial number of plug-in sub-cluster columns for each sub-cluster that meets the processing accuracy requirements and takes into account efficiency. The specific weighted scoring criteria are defined by the actual situation.

[0090] S507, dynamically adjust the number of sub-cluster columns and finally determine the number of columns: through real-time monitoring of the effect of the plug-in processing, including key indicators such as edge smoothness, detail retention, matching degree of time series changes, and maintenance of normal direction consistency, and using visualization tools or data analysis methods to provide instant feedback, to ensure the accuracy and timeliness of the adjustment decision.

[0091] For problem areas found in real-time monitoring, such as jagged edges, time series mismatch, or significant changes in normal direction, dynamic adjustments are made based on curvature, timestamps, and curvature change information in the previous point cloud scan data. For boundary areas near time nodes with drastic curvature changes, significant changes in time series, or large changes in normal direction, if the processing effect is not good, the number of plug-in sub-cluster columns is appropriately increased to improve accuracy; conversely, for areas with redundant columns and limited contribution to the overall effect, the number of columns is appropriately reduced to optimize computational efficiency. When the edge smoothness is monitored to be lower than 90% or the time series matching is lower than 85%, the number of plug-in sub-cluster columns is increased by 1 column; conversely, if the edge smoothness is higher than 95% and the time series matching is higher than 95%, and increasing the number of columns does not improve the overall effect much, the number of columns is reduced by 1 column.

[0092] After each adjustment, re-execute the plug-in process and verify the adjustment effect to ensure the effectiveness of the improvement measures. This process is iterated until the processing effect and efficiency reach the preset standards and meet the actual application requirements. At the end of the dynamic adjustment process, the number of plug-in sub-cluster columns, processing effect and comparison data before and after each adjustment are recorded in detail for easy tracing and reproduction.

[0093] After completing the dynamic adjustment of the number of sub-cluster columns, the number of plug-in columns for each sub-cluster is determined based on the final processing effect and efficiency requirements. In order to achieve this goal, key indicators such as edge smoothness, detail retention, matching of time series changes, and maintenance of normal direction consistency are comprehensively considered to ensure that the determined number of plug-in sub-cluster columns can achieve the optimal processing effect. The final number of plug-in sub-cluster columns will be used for subsequent data processing and analysis.

[0094] S508, constructing an object change trend prediction model: In order to further mine and utilize the time series information in the point cloud data, a prediction method based on the change in curvature and the change in the number of plug-in sub-clusters will be adopted. The future changes in the shape of the object are predicted by calculating and analyzing the curvature change characteristics of the sub-clusters, and the future changes in the structure of the object are predicted by recording and analyzing the change pattern of the number of plug-in sub-clusters. The core of this method is to deeply analyze the previous point cloud scanning data, carefully capture the curvature change characteristics of each sub-cluster, including the size, direction and change trend of the curvature, and record the specific adjustment of the number of plug-in sub-clusters in each scan, such as the frequency and amplitude of the increase and decrease of the adjustment.

[0095] This process is similar to scanning a part with a balloon expansion, especially a spherical balloon, from an uninflated to a fully inflated state. With the continuous injection of gas, the balloon gradually changes from flat to round and full, and the change trajectory of its surface curvature coincides with the evolution of the object shape in the point cloud data, which has undergone a transition from a nearly straight low curvature to a highly curved high curvature. If a continuous point cloud scan is performed on the inflated balloon, the changes in the balloon's shape and curvature state can be accurately recorded through the point cloud data.

[0096] Specifically, the surface of the capsule is divided into many subclusters, each of which represents a small area on the surface of the capsule and changes as the capsule expands. When the capsule expands, the curvature of each small area will increase significantly, and at the same time, in the scanned point cloud data, the curvature of the corresponding subcluster will also increase significantly. In order to more accurately capture and express this change and infer the subsequent change trend of the capsule, it is necessary to conduct in-depth analysis of the previous point cloud scanning data to carefully capture the curvature change characteristics of each capsule subcluster, including the size, direction and change trend of the curvature. In each scanning process, the number of columns of the plug-in subcluster is flexibly adjusted according to the change of curvature. If the curvature of a subcluster changes greatly, the number of plug-in subcluster columns in the subcluster is increased to capture the change of curvature more carefully. At the same time, the specific situation of the number of plug-in columns of each subcluster in each scan is also recorded, so that the change of surface curvature during the expansion of the capsule can be more comprehensively understood. In this way, the change of surface curvature during the expansion of the capsule can be more accurately captured and expressed, and the subsequent development trend of the capsule can be inferred based on this change. Just as when processing point cloud data, the number of plug-in sub-cluster columns needs to be flexibly adjusted according to the curvature change, and the future shape or position change of the object needs to be predicted accordingly. If a sub-cluster shows a significant increase in curvature in the first few scans, accompanied by a corresponding increase in the number of plug-in columns to accommodate this change, then it is reasonable to infer that in subsequent point cloud scans, the part of the object mapped by the sub-cluster is likely to continue to experience large shape changes or position shifts.

[0097] In order to predict this change more scientifically, a prediction model can be constructed to more accurately predict the future state of the object. When constructing a prediction model for the change trend of an object, if there is a clear linear or nonlinear correlation between the curvature change, the adjustment of the number of plug-in columns and the subsequent development of the object, linear regression is used to collect the time series data of the curvature change, the adjustment of the number of plug-in columns and the state of the object in the point cloud data, and the influence of the curvature change and the adjustment of the number of plug-in columns on the state of the object is determined to fit the relationship between these variables, thereby constructing a prediction model. In addition, considering the two extreme states of the object change, namely the maximum and minimum curvature change and the maximum and minimum adjustment of the number of plug-in columns, the state of the object is inferred by the interpolation method. This method is based on the assumption that the change of the object is continuous and smooth, so the intermediate state can be reasonably estimated through the two extreme states. If the point cloud data carries timestamp information, the time series analysis method ARIMA model is introduced to accurately capture the dynamic evolution trend of the data over time, effectively predict key indicators such as future curvature and number of plug-in columns, and then scientifically infer the subsequent change trend of the object. The combined use of these two methods provides a powerful tool for accurate prediction. For example, if a subcluster has a large change in curvature in the first few scans and the number of plug-in subcluster columns increases accordingly, it can be inferred that the object part represented by the subcluster may continue to undergo large shape changes in subsequent scans, or its position may move.

[0098] Finally, in order to verify the accuracy of the inference, these inference results will be compared with the subsequent actual point cloud scanning data. If the inference results are consistent with the actual data, it means that the inference method is effective; if there is a difference, it is necessary to adjust the inference model or parameters according to the actual data to improve the accuracy of the inference.

[0099] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0100] This application uses comprehensive boundary and time characteristic analysis, dynamic adjustment of sub-cluster column number, construction of object change trend prediction model and other technical means to achieve accurate identification and processing of sub-cluster boundaries in point cloud data, ensure the reasonable determination and dynamic adjustment of the number of plug-in sub-cluster columns, and thus improve the accuracy and efficiency of data processing. At the same time, by constructing an object change trend prediction model based on curvature changes and plug-in sub-cluster column number changes, a scientific prediction of subsequent changes in objects is achieved, providing new methods and ideas for the deep mining and utilization of point cloud data.

[0101] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A point cloud scanning processing method, characterized in that: include: S1, obtain point cloud data and determine whether it is a single layer: use laser scanner, camera, 3D scanner to capture the surface point information of the object and convert it into point cloud data; if the point cloud data has obvious features at different heights, it may indicate that it is multi-layered; S2, fusion of multi-layer point cloud data: using the iterative closest point algorithm, through iterative search for the closest point pair and calculation of the optimal transformation matrix, to achieve accurate registration of point cloud data and unified coordinate system, the point cloud space is divided into multiple small cubic grids based on the voxel method for deduplication and fusion; S3, preprocessing the fused point cloud data: using statistical filtering algorithm to calculate the distance between each data point and its neighboring points and remove noise points, using Gaussian filtering for local smoothing, and using voxel grid filtering to reduce the amount of point cloud data; S4, segmentation of sub-clusters: perform PCA analysis to determine the principal component direction of the point cloud data, thereby dividing two sub-clusters with different characteristics, and continuously refine them through recursive segmentation until the preset conditions are met; S5, insert sub-clusters to process, adjust and infer subsequent changes of objects: S501, calculate the covariance matrix for all data points in the sub-cluster to reflect the distribution characteristics of the data points in space, decompose the eigenvalues ​​of the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors; S5 02, check the boundary area of ​​adjacent sub-clusters after the sub-cluster segmentation by using the boundary detection algorithm, and extract the point cloud data along the boundary line to define it as the plug-in sub-cluster; S503, adjust the center point position and curvature radius of the plug-in sub-cluster; S6, simplifying the point cloud data within the sub-cluster: by calculating the curvature difference or distance between the data point and its adjacent points, analyzing the contribution, and giving priority to retaining the data points with large contribution; using filtering methods to remove data points with small contribution, retaining the key geometric shape and curvature features of the sub-cluster; S7, constructing a three-dimensional surface model: calculating the average point distance of the simplified point cloud data, multiplying the average point distance by a coefficient determined according to the density and distribution characteristics of the point cloud to obtain an Alpha value, where the coefficient ranges from 0.5 to 2, running the Alpha shape algorithm, checking the distance between point pairs, and constructing an Alpha shape around the point cloud data; S8, extracting the 3D model: extracting the OBJ general 3D model format 3D model from the Alpha shape, ensuring that the connection relationship between the vertices, edges and faces of the model is consistent with the Alpha shape, while maintaining the normal direction of the face, and processing boundaries, holes and overlapping parts by clipping and merging.

2. A point cloud scanning processing method as claimed in claim 1, characterized in that: The iterative closest point algorithm of S2 includes: Iteratively find the closest point pairs between the two sets of point clouds, and calculate the optimal transformation matrix between these closest points in each iteration. The matrix contains information about rotation and translation, so that the two sets of point clouds are aligned in space; by repeating this process continuously, all data are unified into the same coordinate system.

3. A point cloud scanning processing method as claimed in claim 1, characterized in that: The S2 voxel-based approach includes: The point cloud space is divided into multiple small cube grids, each cube represents a voxel. For each voxel, all the points contained in it are traversed, and the average coordinates of the traversed points in the X, Y, and Z directions are calculated respectively. The average value in the X direction is obtained by adding the X coordinates of all points and dividing by the number of points; the average values ​​in the Y and Z directions are obtained in the same way. The average values ​​in the three directions are combined to obtain the average position of all points in the voxel, which is expressed in three-dimensional coordinates.

4. A point cloud scanning processing method as claimed in claim 1, characterized in that: The S5 comprises: S504, setting a plug-in strategy for the low curvature sub-cluster: selecting 2 to 5 columns of point cloud data as plug-in sub-cluster candidates, extracting the selected point cloud data columns, and formally defining them as plug-in sub-clusters; S505, analyzing and processing subclusters in the three-dimensional space by integrating boundary and time characteristics: using boundary detection algorithm to identify boundary points, using Euclidean distance formula to calculate the actual straight-line distance between two points, analyzing the relative position relationship between the boundary point and its adjacent points, and determining whether the boundary is concave inward or convex outward; S506, determining the initial number of columns of the plug-in sub-clusters: setting a basic number of columns of the plug-in sub-clusters as a starting point for all sub-clusters, and adjusting the number of columns of the plug-in sub-clusters according to the boundary length, the concave-convex variability of the boundary shape, the consistency of the normal direction, the surface roughness, and the curvature variation factor in the time series; S507, dynamically adjusting the number of sub-cluster columns and finally determining the number of columns: real-time monitoring of the boundary smoothness, detail retention, matching degree of time series changes, and normal direction consistency of the plug-in process to increase or decrease the number of plug-in sub-cluster columns; S508, constructing an object change trend prediction model: using point cloud data to analyze the shape and structure of the object, dividing the surface of the object into multiple sub-clusters, each sub-cluster represents a small area on the surface of the object, calculating and analyzing the curvature change characteristics of the sub-clusters to infer the future changes in the shape of the object, and using linear regression method to fit the relationship between the curvature change, the adjustment amount of the number of plug-in columns and the state of the object to construct a prediction model.

5. A point cloud scanning processing method as claimed in claim 1, characterized in that: Also includes: S901, adding time dimension information: when acquiring point cloud data, ensure that a timestamp is attached to each data point; The timestamp should record the precise time when the data point was collected, with an accuracy of at least milliseconds; S902, performing in-depth analysis on the time series information in the point cloud data: identifying the changing trend of data points over time and identifying mutation points in the data; S903, sub-cluster segmentation in the time dimension: using the time window as the evaluation basis, when the timestamps of data points fall within the same time window range, these points are regarded as potential members of the same sub-cluster; S904, setting up plug-in sub-clusters in the time dimension: giving priority to areas showing obvious changes in the time series, increasing the number of columns of plug-in sub-clusters at these key boundaries; reducing the number of columns for stable sub-cluster boundaries in the time series; S905, adjust the plug-in sub-cluster: analyze the parameters of the initial position, size, number of columns and initial curvature radius of the time series plug-in sub-cluster, adjust the position and curvature radius of the center point of the plug-in sub-cluster to best match the changes in adjacent sub-clusters and time series; identify its significant change points and time periods. If it is found that the plug-in sub-cluster fails to fully reflect specific changes in the time series, increase or decrease the number of columns or fine-tune the boundaries to correspond to the turning points of the time series.

6. A point cloud scanning processing method as claimed in claim 1, characterized in that: The step S501, calculating the covariance matrix, includes: Covariance Matrix Formula , where C represents the covariance matrix. Assuming the data has d dimensions, it is a d*d matrix. The data point represents the i-th data point in the data set, which is a d-dimensional vector. The mean of the data points represents the mean vector of all data points in the data set. It is also a d-dimensional vector. The mean vector is obtained by calculating the average value of all data points in each dimension. n is the number of data points. is a normalization factor used to ensure an unbiased estimate of the covariance matrix.

7. A point cloud scanning processing method as claimed in claim 1, characterized in that: S6, preferentially retaining data points with large contribution, including: Contribution formula = α × curvature difference + β × distance difference, where α and β are weight coefficients used to balance the proportion of curvature difference and distance difference in contribution calculation.

8. A point cloud scanning processing method as claimed in claim 4, characterized in that: The S507 further includes: For problem areas with jagged edges, time series mismatch, or significant changes in normal direction found in real-time monitoring, and for boundary areas near time nodes with drastic changes in curvature, significant changes in time series, or large changes in normal direction, if the processing effect is not good, increase the number of plug-in sub-cluster columns; conversely, for areas with redundant columns and limited contribution to the overall effect, reduce the number of columns.

9. A point cloud scanning processing method as claimed in claim 5, characterized in that: The S902 includes: A change point detection algorithm is used to accurately identify mutation points in the data, which are key time points corresponding to significant changes in the shape, position or density of objects.

10. A point cloud scanning processing method according to claim 1, characterized in that: The S7, Alpha shape algorithm, also includes: The average value of the distance between adjacent points of the simplified point cloud is calculated as the average point distance, and the average point distance is multiplied by a coefficient determined according to the density and distribution characteristics of the point cloud to obtain the Alpha value, and the coefficient ranges from 0.5 to 2; after determining the Alpha value, run the Alpha shape algorithm to traverse the point cloud data, and for each pair of points in the traversal, set them as point A and point B, and the algorithm will calculate the distance between the two points; draw a circle with point A as the center and the Alpha value as the radius; check whether point B is within the circle. If point B is not within the circle, the distance between point A and point B is far, and they represent a significant feature of the object surface, then the pair of points is determined to be part of the Alpha shape boundary; if point B is within the circle, the distance between point A and point B is close, then they are determined to belong to the same part of the object surface, and the points between the pair are ignored when constructing the Alpha shape, and they are directly connected.

Images