Point cloud data processing method, system and device for industrial three-dimensional model construction

Abstract

The application relates to the technical field of point cloud processing, and discloses a point cloud data processing method, system and device for industrial three-dimensional model construction. The method comprises the following steps: carrying out denoising and downsampling processing on original point cloud data of an industrial part to obtain pretreated point cloud data, calculating a local geometric descriptor and converting the local geometric descriptor into an industrial feature recurrent graph through a recursive graph algorithm, inputting the recurrent graph into a convolutional neural network to extract a high-dimensional geometric feature vector, mapping the feature vector into B-spline surface control parameters through a deep belief network, and performing point cloud reconstruction quality assessment according to the control parameters to obtain geometric precision error data and surface reconstruction integrity data. The application solves the problems that the point cloud data processing method in the existing industrial three-dimensional model construction lacks intelligent feature recognition, reconstruction precision is insufficient, and manufacturing constraints are not fully considered, and improves the recognition accuracy of complex industrial part geometric features and the precision of three-dimensional model reconstruction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of point cloud processing, in particular to a point cloud data processing method, system and device for industrial three-dimensional model construction. BACKGROUND

[0002] The point cloud data processing technology for industrial three-dimensional model construction is an important technical means in modern manufacturing industry. The three-dimensional point cloud data of industrial parts is obtained mainly through laser scanning, structured light scanning and other methods, and then algorithms such as point cloud registration, filtering, segmentation and surface reconstruction are used to convert discrete point cloud data into complete three-dimensional digital models. In the prior art, point cloud data processing usually adopts ICP (Iterative Closest Point) algorithm for registration, Poisson reconstruction or moving cube algorithm for surface reconstruction, and traditional geometric algorithms are used to identify and extract the geometric features of industrial parts. These methods can achieve good results when dealing with simple geometric shapes.

[0003] However, the prior art has obvious deficiencies when dealing with complex industrial parts: traditional algorithms are prone to distortion or loss of details when identifying and reconstructing complex geometric features such as thin-walled structures, sharp edges and internal cavities, especially for industrial parts with multi-scale geometric features, existing methods are difficult to adaptively identify and reconstruct different types of geometric features; a large amount of manual parameter adjustment is required in the point cloud processing process, the degree of automation is not high, and the processing efficiency is low; the existing algorithms lack consideration of industrial manufacturing constraints, and the reconstruction results often do not meet the actual processing requirements.

[0004] Based on the above analysis, the core problem of the prior art is the lack of intelligent feature identification and reconstruction methods. Since traditional geometric algorithms cannot automatically learn and adapt to different feature patterns of industrial parts, feature identification is inaccurate when dealing with complex geometric structures; the inaccuracy of feature identification further affects the quality of subsequent surface reconstruction, making the reconstructed model unable to accurately maintain the geometric features of the original part; at the same time, the lack of consideration of industrial manufacturing constraints makes the reconstruction results have manufacturability problems in actual application, so a point cloud data processing method that can intelligently identify industrial features and combine manufacturing constraints for high-precision reconstruction is needed. SUMMARY

[0005] The present application provides a point cloud data processing method, system and device for industrial three-dimensional model construction, which solves the problems of lack of intelligent feature identification, insufficient reconstruction accuracy and insufficient consideration of manufacturing constraints in the prior art point cloud data processing method for industrial three-dimensional model construction, and improves the identification accuracy of complex industrial part geometric features and the accuracy of three-dimensional model reconstruction.

[0006] In a first aspect, the application provides a point cloud data processing method for industrial three-dimensional model construction, which comprises: performing denoising and downsampling processing on original point cloud data of an industrial part to obtain preprocessed point cloud data; calculating local geometric descriptors according to the preprocessed point cloud data, performing two-dimensional image conversion processing on the local geometric descriptors through a recursive graph algorithm to obtain an industrial feature recursive graph; inputting the industrial feature recursive graph into a convolutional neural network for feature extraction processing to obtain a high-dimensional geometric feature vector; performing non-linear mapping processing on the high-dimensional geometric feature vector through a deep belief network to obtain B-spline surface control parameters; and performing point cloud reconstruction quality evaluation processing according to the B-spline surface control parameters to obtain geometric accuracy error data and surface reconstruction integrity data.

[0007] In a second aspect, the application provides a point cloud data processing system for industrial three-dimensional model construction, which comprises:

[0008] a denoising module configured to perform denoising and downsampling processing on original point cloud data of an industrial part to obtain preprocessed point cloud data;

[0009] a calculation module configured to calculate local geometric descriptors according to the preprocessed point cloud data, perform two-dimensional image conversion processing on the local geometric descriptors through a recursive graph algorithm to obtain an industrial feature recursive graph;

[0010] an extraction module configured to input the industrial feature recursive graph into a convolutional neural network for feature extraction processing to obtain a high-dimensional geometric feature vector;

[0011] a mapping module configured to perform non-linear mapping processing on the high-dimensional geometric feature vector through a deep belief network to obtain B-spline surface control parameters;

[0012] an evaluation module configured to perform point cloud reconstruction quality evaluation processing according to the B-spline surface control parameters to obtain geometric accuracy error data and surface reconstruction integrity data.

[0013] In a third aspect, a point cloud data processing device for industrial three-dimensional model construction is provided, which comprises: a memory and at least one processor, the memory having instructions stored therein; and the at least one processor invoking the instructions in the memory to cause the point cloud data processing device for industrial three-dimensional model construction to perform the point cloud data processing method for industrial three-dimensional model construction described above.

[0014] In a fourth aspect, a computer readable storage medium is provided, which has instructions stored therein, and when the instructions are run on a computer, the computer is caused to perform the point cloud data processing method for industrial three-dimensional model construction described above.

[0015] In the technical solutions provided in the application, the original point cloud data of industrial parts is subjected to denoising and downsampling processing to obtain preprocessed point cloud data, effectively solving the problems of noise interference and data redundancy in traditional point cloud processing, and the statistical filtering based on manufacturing tolerance constraints ensures that the retained point cloud data meets the industrial standards, and the voxel grid downsampling of the industrial feature scale is self-adaptive, dynamically adjusting the processing precision according to different geometric features, thereby maintaining key details and controlling the computational complexity. The local geometric descriptors are calculated according to the preprocessed point cloud data, and the two-dimensional image conversion processing is performed through the recursive graph algorithm to obtain the industrial feature recursive graph, which innovatively converts the three-dimensional geometric features into two-dimensional image representation, breaking through the limitations of traditional geometric algorithms in complex feature recognition, and the recursive graph algorithm can effectively capture the local geometric relationship and spatial distribution pattern of the point cloud. The industrial feature recursive graph is input into the convolutional neural network for feature extraction processing to obtain a high-dimensional geometric feature vector, which fully utilizes the advantages of deep learning in image feature extraction, and the multi-layer convolutional structure can automatically learn industrial geometric features of different scales, forming hierarchical feature representation from local details to overall structure, and has stronger adaptability and robustness than traditional geometric feature extraction methods.

[0016] The high-dimensional geometric feature vector is subjected to nonlinear mapping processing through the deep belief network to obtain the B-spline surface control parameters, realizing intelligent conversion from abstract features to specific geometric parameters, the multi-layer structure of the deep belief network can learn complex nonlinear mapping relationships, and the generated parameters have actual manufacturability in combination with the industrial manufacturing constraints, and the unsupervised learning characteristics of the restricted Boltzmann machine enable the network to automatically discover the implicit patterns and rules in the data. The point cloud reconstruction quality evaluation processing is performed according to the B-spline surface control parameters to obtain geometric precision error data and surface reconstruction completeness data, and a complete quality evaluation system is established, the reconstruction effect is comprehensively evaluated through multi-dimensional indexes such as geometric precision, shape integrity, and feature retention rate, and it is ensured that the reconstructed model meets the precision requirements and quality standards of industrial applications. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor based on these drawings.

[0018] Figure 1 An embodiment schematic diagram of the point cloud data processing method for industrial three-dimensional model construction in the embodiments of the application;

[0019] Figure 2Fig. 1 is a schematic diagram of an embodiment of a point cloud data processing system for industrial three-dimensional model construction in the present application;

[0020] Figure 3 Fig. 1 is a schematic diagram of an embodiment of a point cloud data processing system for industrial three-dimensional model construction in the present application; DETAILED DESCRIPTION

[0021] The present application provides a point cloud data processing method, system and device for industrial three-dimensional model construction. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0022] For ease of understanding, the specific process of the embodiments of the present application is described below. Please refer to Figure 1 An embodiment of a point cloud data processing method for industrial three-dimensional model construction in the present application includes the following steps:

[0023] Step S101, denoising and downsampling processing is performed on the original point cloud data of the industrial part to obtain preprocessed point cloud data;

[0024] Step S102, local geometric descriptors are calculated according to the preprocessed point cloud data, and two-dimensional image conversion processing is performed on the local geometric descriptors by a recursive graph algorithm to obtain an industrial feature recursive graph;

[0025] Step S103, the industrial feature recursive graph is input into a convolutional neural network for feature extraction processing to obtain a high-dimensional geometric feature vector;

[0026] Step S104, non-linear mapping processing is performed on the high-dimensional geometric feature vector by a deep belief network to obtain B-spline surface control parameters;

[0027] Step S105, point cloud reconstruction quality evaluation processing is performed according to the B-spline surface control parameters to obtain geometric accuracy error data and surface reconstruction completeness data.

[0028] It can be understood that the execution subject of the present application can be a point cloud data processing system for industrial three-dimensional model construction, and can also be a terminal or a server, and the specific implementation is not limited herein. The embodiments of the present application take the server as an example for illustration.

[0029] Specifically, the original point cloud data of the industrial part is denoised and down-sampled. The manufacturing tolerance constraint statistical filtering processing identifies and removes noise points that exceed the manufacturing precision range by calculating the statistical distribution relationship of each point with its neighborhood points, wherein the manufacturing tolerance constraint refers to the allowable deviation range set according to the processing precision standard of the industrial part. The industrial feature scale adaptive voxel grid down-sampling processing dynamically adjusts the voxel size according to the geometric size of different industrial features. The voxel grid is a data structure that divides the three-dimensional space into regular cubic grids, and reduces the data amount by retaining representative points in each voxel. The industrial processing feature covariance matrix calculation involves statistical analysis of the spatial distribution of point cloud data. The covariance matrix can describe the degree of change of point cloud data in different directions, and the eigenvalue ratio is used to identify different geometric feature types such as edges, planes and corners. The industrial semantic block processing merges point cloud data with the same manufacturing attribute into data blocks. The manufacturing semantics refer to the geometric feature attributes related to the industrial processing technology. The k-neighbor index construction processing establishes the connection relationship between each point and its nearest neighbor points. The value of k represents the number of neighborhood points, which is used for subsequent topological analysis.

[0030] The multi-dimensional local geometric descriptor contains information of four dimensions of position coordinates, normal vector, curvature value and local density. The normal vector represents the vertical direction of the point cloud surface, the curvature value describes the bending degree of the surface, and the local density reflects the distribution density of the point cloud around the point. The industrial processing feature coordinate system is a spatial reference system established according to the main processing direction of the industrial part. The Euclidean distance calculation is used to measure the similarity between the descriptor vectors. The manufacturing tolerance recursive threshold judgment processing converts the distance matrix into a binary matrix by setting a distance threshold related to the manufacturing precision. When the distance is less than the threshold, the value is 1, otherwise it is 0. The special coordinate system transformation processing adopts corresponding coordinate transformation methods for different industrial features such as fillets, threads and chamfers. The fillet feature adopts polar coordinate system transformation, the thread feature adopts helical coordinate system transformation, and the chamfer feature adopts oblique coordinate system transformation. The industrial geometric invariance constraint ensures that the recursive graph can maintain the geometric properties of the industrial features after rotation, scaling and translation operations.

[0031] The industrial feature recursive graph is input into a convolutional neural network for feature extraction. Multiple convolutional layers perform convolution operations on the recursive graph using convolution kernels of different sizes to extract geometric features at different levels. The convolution operation is an operation of element-wise multiplication and summation of the convolution kernel and the local region of the image. The industrial geometric constraint verification process filters effective features by checking whether the extracted features meet the industrial geometric rules. Manufacturing constraints refer to geometric limitations that must be met during the industrial processing. The batch normalization layer normalizes the feature map to eliminate differences between different batches of data. Standardization is a process of adjusting data to a distribution with a mean of 0 and a variance of 1. The multi-scale feature fusion process combines features extracted by different convolution layers to form a comprehensive feature representation containing multiple levels of information. The fully connected layer performs high-dimensional mapping to convert the two-dimensional feature map into a one-dimensional feature vector.

[0032] The high-dimensional geometric feature vector is mapped non-linearly by a deep belief network. The restricted Boltzmann machine is a two-layer neural network structure that calculates the activation probability of the hidden layer units through an energy function. The energy function describes the stability of the network state. The manufacturing constraint function performs bias testing on geometric parameters such as radius, thickness, and angle to ensure that the parameters meet industrial standards. Contrastive divergence learning is an unsupervised learning algorithm that iteratively updates network weights until convergence. The feature encoder decodes semantic features into control point coordinates, weight coefficients, and node vectors of B-spline surfaces. B-spline surfaces are a mathematical representation method used to describe complex surfaces. The CAD geometry verification process checks whether the generated control parameters meet the requirements of geometric continuity and smoothness.

[0033] Quality assessment is performed based on B-spline surface control parameters. Surface reconstruction processes generate three-dimensional surface models based on control points, weight coefficients, and node vectors. When the weight coefficients meet the convergence conditions, it indicates that the reconstruction process is stable. The distance calculation algorithm evaluates the geometric accuracy by calculating the point-to-surface distance between the reconstructed surface and the original point cloud. The manufacturing precision threshold is set according to the industrial standards. The curvature continuity detection algorithm calculates the first and second derivative continuity of the surface to ensure smoothness. The key feature point positioning algorithm identifies the positions of important geometric features such as hole centers, edges, and corner points. The tolerance standard is set according to the requirements of the industrial drawing. The comprehensive evaluation algorithm calculates the geometric accuracy error data and surface reconstruction completeness data by weighting the test results.

[0034] In a specific embodiment, the process of step S101 can specifically include the following steps:

[0035] Perform statistical filtering processing on the original point cloud data of the industrial part based on the manufacturing tolerance constraints to obtain denoised point cloud data that meets the industrial standards;

[0036] The point cloud data is subjected to density gradient down-sampling processing based on an industrial feature scale adaptive voxel grid, to obtain multi-level density distribution point cloud data.

[0037] An industrial processing feature covariance matrix is calculated according to the multi-level density distribution point cloud data, to obtain geometric feature classification labels containing edges, planes and corner points.

[0038] The geometric feature classification labels are subjected to industrial semantic block processing, to obtain a point cloud data block set with manufacturing semantics.

[0039] The point cloud data block set is subjected to k-neighbor index construction processing based on an industrial topological continuity constraint, to obtain preprocessed point cloud data.

[0040] Specifically, the statistical filtering processing based on the manufacturing tolerance constraint first calculates the distance statistical distribution of each point in the original point cloud and its neighborhood points. The manufacturing tolerance constraint refers to the geometric deviation allowed range set according to the industrial part processing precision standard. The statistical filtering algorithm identifies outliers by calculating the average distance and standard deviation of each point in the point cloud to its neighborhood points. These outliers are usually noise generated by reflection errors or environmental interference in the laser scanning process. The algorithm traverses each point in the point cloud, finds the nearest several points in its neighborhood, and calculates the average distance of the point to the neighborhood points. When the average distance of a point exceeds the threshold of the global average distance plus the standard deviation multiple, the point is marked as a noise point and removed from the point cloud. The manufacturing tolerance constraint ensures that the retained point cloud data meets the actual manufacturing geometric precision standard by associating the statistical threshold with the industrial processing precision requirement. The denoised point cloud data retains the true geometric features of the industrial part.

[0041] The industrial feature scale adaptive voxel grid down-sampling processing dynamically adjusts the voxel size according to the geometric size of different industrial features. The voxel grid is a data structure that divides the three-dimensional space into regular cubic cells. The density gradient down-sampling first analyzes the density distribution of the point cloud in space, calculates the point cloud density of each local region, and then adjusts the resolution of the voxel grid according to the density gradient change. In areas with high point cloud density, smaller voxel sizes are used to retain detailed features, while in areas with low density, larger voxel sizes are used to reduce data redundancy. The algorithm identifies boundary regions with sharp density changes and plane regions with relatively uniform density by calculating the local density function of the point cloud, and samples different regions with corresponding voxel sizes. The industrial feature scale adaptive mechanism automatically selects the appropriate voxel resolution according to the typical size range of different industrial features such as round holes, threads and chamfers, ensuring that the down-sampled point cloud retains key geometric features while controlling data size.

[0042] The industrial processing feature covariance matrix calculation identifies different types of geometric features by analyzing the spatial distribution characteristics of the multi-level density distribution point cloud data. The covariance matrix is a statistical quantity that describes the degree of change of data in different directions. For each point and its neighborhood, the covariance values in three directions are calculated and a 3x3 matrix is formed. The algorithm calculates three eigenvalues of the covariance matrix, and determines the geometric feature type by the ratio relationship between the eigenvalues: when the first eigenvalue is 0.8, the second eigenvalue is 0.08, and the third eigenvalue is 0.008, the eigenvalue ratio is 100:10:1, indicating that the point cloud changes dramatically in a certain direction, corresponding to the edge feature; when the first eigenvalue is 0.5, the second eigenvalue is 0.48, and the third eigenvalue is 0.05, the eigenvalue ratio is 10:9.6:1, the first two eigenvalues are close and much larger than the third eigenvalue, indicating that the point cloud is distributed in a plane, corresponding to the plane feature; when the first eigenvalue is 0.35, the second eigenvalue is 0.32, and the third eigenvalue is 0.3, the eigenvalue ratio is 1.17:1.07:1, and when the three eigenvalues are close, it indicates that the point cloud changes uniformly in each direction, corresponding to the corner feature. The geometric feature classification label is generated by quantifying the eigenvalue ratio, each point is assigned a corresponding feature type label, forming a classification data structure containing edge, plane, and corner information.

[0043] The industrial semantic segmentation processing merges point cloud data with the same geometric feature classification label into data blocks, and the manufacturing semantics refer to attribute information related to industrial processing technology and geometric function. The algorithm aggregates point cloud data with the same feature type and adjacent to each other into connected regions according to the geometric feature classification label, and each connected region represents a geometric feature with specific manufacturing semantics. The segmentation processing uses a region growing algorithm, which starts from a seed point and expands to the surrounding, gradually adds points with the same feature type and spatial adjacency to the same data block, until it cannot continue to expand. The manufacturing semantic information includes the geometric type, size parameter, position information, and process attribute of the feature, and each data block carries a corresponding semantic label, forming a set of point cloud data blocks with industrial significance.

[0044] The k-neighbor index construction process of the industrial topology continuity constraint establishes the connection relationship of each point in the point cloud data block set with the adjacent points. The k-neighbor index is a kind of spatial data structure, which records the connection information of each point and its nearest k neighbor points, and the value of k is determined according to the geometric complexity and density distribution of the industrial features. The topology continuity constraint ensures that the connection relationship between adjacent points meets the geometric continuity requirements of industrial parts, avoiding unreasonable jump connections. The algorithm first calculates the Euclidean distance of each point to all other points, then selects the k points with the smallest distance as neighbors, and establishes the adjacency relationship table. The industrial topology continuity is verified by checking the geometric rationality of the adjacency relationship, including distance constraint, angle constraint and curvature continuity constraint, only the adjacency relationship that meets these constraint conditions is retained in the index structure.

[0045] In a specific embodiment, the process of performing step S102 can specifically include the following steps:

[0046] According to the pre-processed point cloud data, a multi-dimensional local geometric descriptor containing position coordinates, normal vectors, curvature values and local density is constructed;

[0047] Based on the industrial feature coordinate system, the Euclidean distance calculation process of the local geometric descriptor is performed to obtain an industrial feature distance matrix;

[0048] The industrial feature distance matrix is subjected to manufacturing tolerance recursive threshold judgment processing to obtain an industrial standard recursive binary matrix;

[0049] According to different types of industrial features such as fillets, threads and chamfers, the recursive binary matrix is subjected to special coordinate system transformation processing to obtain a multi-type industrial feature recursive image;

[0050] Based on the industrial geometric invariance constraint, the industrial feature recursive image is subjected to rotation, scaling and translation processing to obtain an industrial feature recursive graph.

[0051] Specifically, the multi-dimensional local geometric descriptor construction process extracts four key geometric attributes of each point from the pre-processed point cloud data. The position coordinates contain the x, y, z coordinate values of the point in three-dimensional space, the normal vector is obtained by calculating the perpendicular direction of the point cloud fitting plane of the neighborhood of the point, which represents the orientation information of the surface at that point. The curvature value calculation involves analyzing the neighborhood geometry of the point, by fitting a local surface and calculating its second derivative to obtain a quantitative value of the degree of surface curvature, a high curvature value indicates that the surface is sharply curved, and a low curvature value indicates that the surface is relatively flat. The local density is calculated by counting the number of other points within the neighborhood range of the point, reflecting the density of the point cloud distribution in that area. The algorithm iterates through each point in the pre-processed point cloud, searches for neighboring points within a specified radius, fits a local plane using the least squares method to calculate the normal vector, simultaneously fits a quadratic surface to calculate the principal curvature and average curvature, and counts the number of neighborhood points to calculate the local density. The four attribute values are combined to form a 12-dimensional feature vector as the local geometric descriptor of the point.

[0052] The industrial processing feature coordinate system is a special space reference system established according to the main processing direction and geometric features of industrial parts, which is different from the general Cartesian coordinate system. The Euclidean distance calculation process calculates the straight-line distance between two points in a 12-dimensional feature vector space by comparing the similarity between the local geometric descriptors of different points. The algorithm iterates through all point pairs, calculates the square sum of the difference between the geometric descriptor vectors of each pair of points, and then takes the square root to obtain the Euclidean distance value. The industrial feature distance matrix is a symmetric matrix, with rows and columns corresponding to each point in the point cloud, and the matrix element value represents the geometric similarity between the corresponding point pairs. The smaller the distance value, the more similar the geometric features of the two points, and the larger the distance value, the greater the difference in geometric features. The algorithm fills the Euclidean distance values of all point pairs into the corresponding positions of the matrix to form a distance relationship description.

[0053] The manufacturing tolerance recursive threshold judgment process converts continuous distance values into discrete binary data, and the manufacturing tolerance is a geometric deviation tolerance range set according to the industrial processing precision requirements. The recursive threshold is determined by analyzing the processing precision level of industrial parts, and when the element value in the distance matrix is less than or equal to the threshold value, the corresponding position is assigned a value of 1, indicating that the geometric features of the two points are similar; when the element value is greater than the threshold value, it is assigned a value of 0, indicating that the geometric features are significantly different. The industrial standard recursive binary matrix is generated through this thresholding process, and the matrix only contains two values of 0 and 1, forming a digital representation of a black and white pixel image. The algorithm iterates through each element of the distance matrix one by one, and performs binary conversion according to the threshold condition to generate a recursive binary matrix of the same size.

[0054] The special coordinate system transformation processing adopts corresponding mathematical transformation methods for different industrial feature types such as fillets, threads, and chamfers. The fillet feature adopts polar coordinate system transformation to convert the rectangular coordinates into radius and angle representation, highlighting the geometric characteristics of the circular arc; the thread feature adopts helical coordinate system transformation, introducing the helix angle and pitch parameters to strengthen the geometric laws of the helix; and the chamfer feature adopts bevel coordinate system transformation, adjusting the coordinate axis direction according to the chamfer angle to optimize the edge feature representation. The algorithm selects the corresponding transformation matrix according to the geometric feature classification identifier to perform coordinate transformation operation on the recursive binary matrix. The recursive images of multiple types of industrial features are converted into gray images by converting the transformed binary matrix, and each type of industrial feature corresponds to different image patterns and texture features.

[0055] The industrial geometric invariance constraint ensures that the recursive image can maintain the essential geometric properties of the industrial feature after spatial transformation. The rotation processing realizes the angle transformation of the image around the center point through matrix multiplication, the scaling processing adjusts the size ratio of the image, and the translation processing changes the position of the image in the coordinate system. The algorithm performs multiple transformation operations on each industrial feature recursive image to generate a set of transformed images. The invariance constraint is verified by checking the consistency of the geometric features before and after transformation, including the position relationship of the feature points, the geometric shape ratio, and the maintenance degree of the topological connection relationship. The industrial feature recursive graph generates a standardized image representation by integrating multiple transformation results.

[0056] In a specific embodiment, the process of performing step S103 can specifically include the following steps:

[0057] The industrial feature recursive graph is input into the multi-layer convolution layer for industrial geometric feature extraction processing to obtain a set of multi-scale industrial feature graphs;

[0058] The set of multi-scale industrial feature graphs is subjected to geometric consistency verification processing based on the industrial geometric constraint module to obtain geometric feature graphs that meet the manufacturing constraints;

[0059] The geometric feature graphs are subjected to industrial standardization processing according to the batch normalization layer to obtain standardized industrial feature graphs;

[0060] The standardized industrial feature graphs are input into the multi-scale feature fusion module for industrial semantic fusion processing to obtain fused industrial feature graphs;

[0061] The fused industrial feature graphs are subjected to high-dimensional mapping processing based on the fully connected layer to obtain high-dimensional geometric feature vectors.

[0062] Specifically, the multi-layer convolutional layer processing performs layer-by-layer feature extraction with the industrial feature recurrent graph as input. The convolutional layer is a basic unit in neural networks for image feature detection, which extracts local features through sliding window operations between the convolution kernel and the input image. In the process of industrial geometric feature extraction, different size convolution kernels detect geometric patterns of different scales. Small size convolution kernels identify detailed features such as thread texture and surface roughness, medium size convolution kernels capture local geometric shapes such as fillets and chamfers, and large size convolution kernels extract overall structural features such as plane and curved surface distribution. Each convolutional layer contains multiple feature channels, each corresponding to a different type of industrial feature detector, and generates a feature map through convolution operation. The multi-scale industrial feature map set is composed of the output of different levels of convolutional layers. Shallow convolutional layers generate high-resolution detail feature maps, and deep convolutional layers generate low-resolution semantic feature maps, forming a feature hierarchy from detail to whole. The algorithm sequentially passes the recurrent graph through each convolutional layer, applies an activation function after each convolutional operation to enhance feature expression capability, and generates a feature map set containing multiple scale information.

[0063] The industrial geometric constraint verification process filters valid features by checking whether the multi-scale feature map set conforms to the geometric rules of industrial manufacturing. Geometric consistency verification includes size continuity check, angle reasonableness check, and surface smoothness check. Size continuity ensures that the size change of adjacent features meets the processing requirements, angle reasonableness verifies that the angle relationship between features meets the industrial design standards, and surface smoothness checks the smooth transition of feature boundaries. Manufacturing constraints are verified by preset industrial standard parameters, including minimum machining radius, maximum inclination angle, surface roughness range, and other limit conditions. The algorithm traverses each pixel region in the feature map, calculates its geometric attribute parameters, and compares them with the manufacturing constraint conditions. Feature regions that do not meet the constraints are marked as invalid and removed from the feature map. The geometric feature map that meets the manufacturing constraints only retains the feature regions that pass the verification, ensuring that the feature data in subsequent processing meets the feasibility requirements of actual industrial manufacturing.

[0064] The batch normalization layer performs industrial standardization processing on the geometric feature map. Batch normalization is a technique used in deep learning to stabilize the training process by adjusting the distribution of feature data to eliminate differences between different batches. Industrial standardization processing sets normalization parameters, including target values for mean and variance, based on the precision levels and quality standards of industrial manufacturing. The algorithm calculates the statistical properties of the current batch of feature maps, including the mean and standard deviation of each channel, and then adjusts the feature values to a standard distribution. The standardization process achieves a zero-mean unit-variance distribution by subtracting the mean and dividing by the standard deviation, and then applies learnable scaling and bias parameters to adapt to specific industrial application requirements. The standardized industrial feature map has a uniform numerical range and distribution characteristics, eliminating the scale differences between different industrial parts, making the feature representation more stable and consistent.

[0065] The multi-scale feature fusion process integrates different scale information in the standardized industrial feature map. Feature fusion is the process of combining multiple feature representations into a unified representation. Industrial semantic fusion combines detailed features with overall features by analyzing the semantic relevance of different scale feature maps. The algorithm uses two fusion strategies: feature stitching and weighted summation. Feature stitching connects feature maps of different scales in the channel dimension to form a comprehensive feature representation containing multi-scale information. Weighted summation linearly combines different scale features by assigning weight coefficients according to their importance. During the fusion process, the resolution difference between different scale feature maps needs to be handled. Through upsampling or downsampling operations, the feature maps are adjusted to a uniform size. The fused industrial feature map contains complete geometric information from local details to global structures, forming a hierarchical industrial feature representation.

[0066] The fully connected layer performs high-dimensional mapping processing to convert the two-dimensional fused industrial feature map into a one-dimensional high-dimensional feature vector. The fully connected layer is a layer structure in which each input node is connected to each output node in the neural network, and linear transformation from input to output is achieved through matrix multiplication. The high-dimensional mapping process flattens the fused feature map, expands the two-dimensional matrix into a one-dimensional vector, and then performs linear transformation through the weight matrix of the fully connected layer. The algorithm calculates the product of the input vector and the weight matrix, adds the bias vector, and then applies the activation function to obtain the output vector. The dimension of the high-dimensional geometric feature vector is determined according to the complexity of the industrial application and contains the abstract representation of all geometric information extracted and processed from the original recursive graph.

[0067] In a specific embodiment, the step of performing high-dimensional mapping processing on the fused industrial feature map based on the fully connected layer can specifically include the following steps:

[0068] Flattening and expanding the fused industrial feature map to obtain a one-dimensional feature sequence containing industrial processing feature information;

[0069] Performing industrial part geometric semantic encoding processing on the one-dimensional feature sequence based on the first fully connected layer to obtain an industrial semantic encoding vector containing cylindrical surface, plane, and thread features;

[0070] Performing manufacturing feature nonlinear mapping processing on the industrial semantic encoding vector according to the ReLU activation function to obtain an activated industrial geometric feature vector;

[0071] Inputting the activated industrial geometric feature vector into the second fully connected layer to perform industrial CAD feature dimension compression processing and obtain a compressed feature vector suitable for B-spline reconstruction;

[0072] Performing industrial standard normalization processing on the compressed feature vector based on L2 norm constraint to obtain a high-dimensional geometric feature vector.

[0073] Specifically, the flattening processing converts the two-dimensional fusion industrial feature map into a one-dimensional data structure. Flattening is the operation of rearranging a multi-dimensional array into a one-dimensional vector. The fusion industrial feature map usually has three dimensions of height, width, and channel number. The flattening operation arranges all pixel values in a continuous one-dimensional array in row-major order. The algorithm iterates through each pixel position of the feature map, starting from the first row and first column, and scans each row in turn. It adds all channel values of each pixel to the one-dimensional sequence in turn until the entire feature map is processed. The industrial processing feature information maintains its original numerical relationship during the flattening process, only the organization of the data is changed from two-dimensional space representation to linear sequence representation. The one-dimensional feature sequence contains all the geometric information of the original feature map, including surface texture, edge contour, local curvature, and other geometric properties of industrial parts. Each element in the sequence corresponds to a feature value of a specific position and channel in the feature map.

[0074] The industrial part geometry semantic encoding processing converts the one-dimensional feature sequence into an encoded representation with industrial semantic meaning through the first fully connected layer. The fully connected layer performs linear transformation on the input through a weight matrix and a bias vector. The weight parameters learn the correlation between different industrial features during the training process. The algorithm calculates the matrix multiplication of the input feature sequence and the weight matrix, and each output neuron corresponds to the weighted sum result of all elements of the input sequence. The industrial semantic encoding process converts the original pixel-level features into semantic features with industrial meaning. Different dimensions of the encoding vector correspond to specific industrial geometric feature types such as cylindrical surface, plane, and thread. The cylindrical surface feature encoding contains radius and axial information, the plane feature encoding contains normal vector and plane equation parameters, and the thread feature encoding contains pitch, thread angle, and thread type parameters. The industrial semantic encoding vector realizes the mapping and conversion from low-level pixel features to high-level semantic features through the learned weight parameters.

[0075] The ReLU activation function performs a non-linear transformation on the industrial semantic encoding vector. The ReLU function is a rectified linear unit activation function that performs piecewise linear processing on the input value. The manufacturing feature non-linear mapping converts the linear encoding result into a non-linear representation through the ReLU function. The ReLU function keeps the positive value unchanged and sets the negative value to zero. The mathematical expression is output equals the maximum of input and zero. The algorithm iterates through each element of the encoding vector. When the element value is greater than zero, it remains unchanged. When the element value is less than or equal to zero, it is set to zero. The non-linear mapping introduces sparsity in the feature representation, only the activated feature dimensions participate in subsequent calculations, and the influence of negative features is suppressed. The activated industrial geometric feature vector has stronger expression ability, and the non-linear transformation enables the network to learn complex industrial feature patterns and geometric relationships.

[0076] The industrial CAD feature dimension compression process compresses the activated feature vector to a dimension scale suitable for B-spline reconstruction through a second fully connected layer. Dimension compression is a dimension reduction technique that maps high-dimensional features to a low-dimensional space through a linear transformation while preserving important geometric information. The algorithm calculates the product of the activated feature vector and the compression weight matrix, which has a row number equal to the input dimension and a column number equal to the target compression dimension. B-spline reconstruction requires a specific number of control parameters, including control point coordinates, weight coefficients, and node vectors, and the dimension of the compressed feature vector matches the total number of these parameters. The compression process selects the most important feature dimensions through learned weight parameters, removing redundant information and noise interference. The compressed feature vector suitable for B-spline reconstruction contains the core geometric information needed to construct a three-dimensional model, balancing the requirements of information preservation and computational efficiency.

[0077] The L2 norm constraint normalization process ensures that the compressed feature vector meets the numerical range requirements of the industry standard. The L2 norm is the square root of the sum of the squares of all elements of a vector, used to measure the size of the vector. Industrial standard normalization adjusts the feature vector to a standardized numerical range through L2 norm constraint, preventing excessively large or small values from affecting subsequent calculation accuracy. The algorithm first calculates the L2 norm value of the compressed feature vector, then divides each element of the vector by the norm value to obtain a normalized vector of unit length. Normalization eliminates the scale differences between different industrial parts, making the feature representation have a uniform numerical range and comparison benchmark. The high-dimensional geometric feature vector has stable numerical properties after L2 norm constraint, meeting the requirements of deep belief network input.

[0078] In a specific embodiment, the process of performing step S104 can specifically include the following steps:

[0079] Input the high-dimensional geometric feature vector into the restricted Boltzmann machine for energy function calculation processing, and obtain the activated hidden feature when the hidden layer unit activation probability is greater than the threshold value;

[0080] Based on the manufacturing constraint function, the activated hidden feature is judged for radius, thickness, and angle deviation, and the qualified constraint feature is obtained when the deviation value is less than the standard tolerance;

[0081] Input the qualified constraint feature into the restricted Boltzmann machine for contrastive divergence learning processing, and obtain the stable semantic feature when the weight update converges;

[0082] According to the feature encoder, the stable semantic feature is decoded for control point, weight, and node vector, and the candidate B-spline parameter is obtained when the decoding error meets the accuracy requirement;

[0083] Based on the quality evaluation module, the candidate B-spline parameter is subjected to CAD geometric verification processing, and the B-spline surface control parameter is obtained when the geometric continuity and smoothness meet the preset standard.

[0084] Specifically, the RBM energy function calculation process inputs the high-dimensional geometric feature vector as the visible layer unit, and the RBM is a two-layer neural network structure, which includes a visible layer and a hidden layer, and there is a connection weight between the two layers but no connection within the same layer. The energy function describes the stability of the current state of the network, and the system state is determined by calculating the interaction energy between the visible layer units and the hidden layer units. The algorithm assigns the high-dimensional geometric feature vector to the visible layer unit, and then calculates the activation probability of each hidden layer unit. The activation probability is calculated by the sigmoid function after the weighted sum of the visible layer units. When the activation probability of the hidden layer unit exceeds the set threshold value, the unit is activated and outputs 1, otherwise it outputs 0. The threshold value is set according to the complexity and feature density of the industrial part, and a lower threshold value is used for complex parts to activate more feature units, and a higher threshold value is used for simple parts to reduce redundant activation. The activated hidden features are composed of all activated hidden layer units, forming a sparse binary vector representation, which contains the key geometric information extracted from the input features.

[0085] The manufacturing constraint function verifies the industrial manufacturing feasibility of the activated hidden features, and the manufacturing constraints include geometric size, angle range, surface quality and other industrial standard requirements. The radius deviation judgment checks whether the radius value of the circular arc and cylindrical feature is within the allowed range, the thickness deviation judgment verifies whether the wall thickness and plate thickness meet the structural strength requirements, and the angle deviation judgment confirms whether the chamfer and conical surface angle meets the processing technology limit. The algorithm decodes the activated hidden features into corresponding geometric parameters, obtains the standard tolerance range through table lookup or calculation, and then calculates the deviation of the actual parameters from the standard values. When the radius deviation is less than the radius allowable value of the standard tolerance, the thickness deviation is less than the thickness allowable value, and the angle deviation is less than the angle allowable value, the feature is marked as qualified. Qualified constraint features only retain hidden units that pass all manufacturing constraint tests, filtering out feature representations that do not meet industrial manufacturing requirements, ensuring that the generated B-spline parameters have actual processing feasibility.

[0086] The contrastive divergence learning process optimizes network parameters by iteratively updating the connection weights of the restricted Boltzmann machine. Contrastive divergence is an unsupervised learning algorithm that adjusts weights by comparing the difference between data distribution and model distribution. The algorithm first reconstructs the visible layer units based on the eligible constraint features, and then calculates the hidden layer activations again based on the reconstruction results. This process is repeated multiple times to form a Gibbs sampling chain. Weight updates are achieved by calculating the statistical difference between the original data and the reconstructed data. The update amount is equal to the learning rate multiplied by the gradient value. When the weight changes of consecutive multiple iterations are less than the convergence threshold, the learning process stops. The convergence of weight updates indicates that the network has learned the stable statistical properties of the data, and the network parameters no longer change significantly. Stable semantic features are represented by the converged hidden layer units, which contain optimized industrial geometric feature encodings and have stronger expression ability and generalization performance.

[0087] The feature encoder decoding process converts stable semantic features into mathematical parameter representations of B-spline surfaces. The feature encoder is a trained neural network specifically designed to map abstract features to specific geometric parameters. Control point decoding extracts the three-dimensional coordinates of the control points of the B-spline surface from the semantic features, weight decoding calculates the influence coefficients of each control point on the surface shape, and node vector decoding determines the parameter sequence of the B-spline basis function. The algorithm inputs the stable semantic features into the encoder network and generates the corresponding B-spline parameters through multiple layers of nonlinear transformation. The decoding error is calculated by comparing the difference between the decoding output and the target parameters. When the root mean square value of the decoding error is less than the accuracy requirement threshold, the decoding result is considered to meet the quality standard. The candidate B-spline parameters contain all the mathematical information required to construct a three-dimensional surface, and the accuracy of the parameters directly affects the geometric accuracy of the final reconstructed model.

[0088] The quality evaluation process verifies the geometric reasonableness and mathematical correctness of the candidate B-spline parameters. Quality evaluation includes geometric continuity check and surface smoothness check. Geometric continuity verification checks whether the boundaries between adjacent surface patches are smoothly connected, and smoothness check checks whether the first and second derivatives of the surface are continuous. The algorithm reconstructs the surface based on the B-spline parameters, calculates the curvature distribution and normal vector variation of the surface, and checks for abrupt changes or discontinuities. CAD geometry verification includes topology consistency check, size accuracy check, and shape reasonableness check to ensure that the generated surface meets industrial design requirements. When the geometric continuity index meets the continuity standard and the smoothness index meets the smoothness standard, the candidate parameters pass the quality evaluation. The B-spline surface control parameters are composed of candidate parameters that have passed all verifications, containing complete control information required to reconstruct industrial three-dimensional models.

[0089] In a specific embodiment, the process of performing step S105 can specifically include the following steps:

[0090] The B-spline surface control parameters are used for surface reconstruction processing, and when the control point weight coefficient meets the convergence condition, a reconstructed surface model is obtained.

[0091] The reconstructed surface model is subjected to point cloud comparison and analysis processing based on a distance calculation algorithm, and when the average distance error is less than the manufacturing precision threshold, geometric precision error data is obtained.

[0092] The reconstructed surface model is input into a curvature continuity detection algorithm for shape accuracy verification processing, and when the curvature change rate meets the smoothness requirement, shape integrity evaluation results are obtained.

[0093] According to the key feature point positioning algorithm, the shape integrity evaluation results are subjected to position deviation calculation processing, and when the hole center, edge and corner position deviation meets the tolerance standard, feature retention rate data is obtained.

[0094] Based on the comprehensive evaluation algorithm, the feature retention rate data is subjected to model integrity quantification processing, and when the feature retention rate and model continuity both meet the standard, surface reconstruction integrity data is obtained.

[0095] Specifically, the B-spline surface reconstruction processing generates a three-dimensional geometric model according to the control parameters, and the B-spline surface is a parameterized surface representation method defined by control points, weight coefficients and node vectors. The reconstruction algorithm first reads the three-dimensional coordinate data of the control points, then applies the corresponding weight coefficients to adjust the influence degree of each control point on the surface shape, and finally determines the distribution of the basis function according to the node vector. The control point weight coefficient convergence condition is judged by iteratively calculating the weight change, and when the weight change of continuous multiple iterations is less than the preset threshold, it is considered to be converged. The algorithm recalculates the position of the sampling points on the surface in each iteration, and adjusts the weight coefficient by minimizing the distance error between the sampling points and the original point cloud. During the convergence process, the weight coefficient is gradually optimized until the surface shape is stable, at which time the generated reconstructed surface model contains the main geometric features of the original point cloud. The reconstructed surface model is represented in the form of triangular mesh or NURBS surface, and each surface point has accurate three-dimensional coordinates and normal vector information.

[0096] The distance calculation algorithm evaluates the reconstruction accuracy by comparing the spatial position difference between the reconstructed surface model and the original point cloud data. The point cloud comparison analysis process first establishes the correspondence between each point in the original point cloud and the nearest point on the reconstructed surface, and then calculates the perpendicular distance from the point to the surface. The algorithm iterates through each point in the original point cloud, searches for the nearest surface position on the reconstructed surface, and calculates the Euclidean distance between the two. The average distance error is obtained by taking the arithmetic mean of all point distance values, reflecting the overall deviation of the reconstructed model from the original data. The manufacturing precision threshold is set according to the machining precision level of industrial parts, with smaller thresholds for precision parts and larger thresholds for ordinary parts. When the average distance error is less than the manufacturing precision threshold, the reconstructed model is considered to meet the geometric accuracy requirements. The geometric accuracy error data includes statistical indicators such as average error, maximum error, and standard deviation, quantitatively describing the geometric accuracy of the reconstructed model.

[0097] The curvature continuity detection algorithm analyzes the local geometric properties of the reconstructed surface model. Curvature is a geometric quantity that describes the bending degree of the surface, and continuity detection verifies whether the surface has abrupt changes or non-smooth areas. The shape accuracy verification process evaluates the surface quality by calculating the principal curvature and mean curvature distribution of the surface. The algorithm uniformly samples detection points on the reconstructed surface, calculates the curvature value of each detection point, and then analyzes the curvature change rate between adjacent points. The curvature change rate is obtained by calculating the difference between the curvature values of adjacent sampling points. A large change rate indicates that the surface has non-smooth areas. The smoothness requirement is set according to the surface quality standards of industrial applications, with higher smoothness required for aviation parts and medium smoothness required for mechanical parts. The shape integrity evaluation result considers the uniformity, continuity, and smoothness indicators of the curvature distribution, generating evaluation data that describes the shape quality of the surface.

[0098] The key feature point positioning algorithm identifies geometric positions with important functional significance in industrial parts, including hole centers, edges, corner points, and other positions that are crucial to the function and assembly of the parts. The position deviation calculation process evaluates feature retention accuracy by comparing the position difference between the feature points in the reconstructed model and the design reference position. The algorithm automatically identifies hole center positions in the reconstructed surface model by fitting a circular boundary to determine the center coordinates. Then it detects edge positions by identifying surface boundary lines through curvature changes. Finally, it locates corner point positions by analyzing the intersection points of multiple edges to determine the corner coordinates. The hole center position deviation is obtained by calculating the distance between the reconstructed hole center and the design hole center. The edge position deviation is obtained by calculating the average offset of the edge line. The corner point position deviation is obtained by calculating the difference in corner coordinates. The tolerance standard is set according to the size tolerance and position tolerance requirements of the industrial drawing, with different feature types corresponding to different tolerance levels. The feature retention rate data is calculated by statistically analyzing the proportion of feature points that meet the tolerance requirements, reflecting the ability of the reconstructed model to retain key features.

[0099] The comprehensive evaluation algorithm weights and synthesizes feature retention rate data with model continuity indicators, comprehensively evaluating performance indicators across multiple dimensions such as geometric accuracy, shape quality, and feature integrity. Model integrity metric processing converts each indicator into a unified numerical score through an established evaluation function. The algorithm sets weight coefficients for indicators such as feature retention rate, model continuity, and geometric accuracy, adjusting the weight allocation according to specific application requirements. A satisfactory feature retention rate indicates that the reconstructed model retains the key functional features of the original part, while a satisfactory model continuity indicates that the reconstructed surface possesses good mathematical properties. The comprehensive evaluation function calculates the overall score through weighted summation; when the overall score exceeds a preset standard, the reconstruction quality is considered acceptable. The surface reconstruction integrity data includes scores for each sub-indicator and the comprehensive score, generating a quality assessment report.

[0100] The point cloud data processing method for industrial 3D model construction in the embodiments of this application has been described above. The point cloud data processing system for industrial 3D model construction in the embodiments of this application is described below. Please refer to [link / reference]. Figure 2 One embodiment of the point cloud data processing system for industrial 3D model construction in this application includes:

[0101] The noise reduction module is used to perform noise reduction and downsampling on the raw point cloud data of industrial parts to obtain preprocessed point cloud data.

[0102] The calculation module is used to calculate local geometric descriptors based on the preprocessed point cloud data, and to perform two-dimensional image conversion processing on the local geometric descriptors through a recursive graph algorithm to obtain an industrial feature recursive graph.

[0103] The extraction module is used to input the industrial feature recursive graph into a convolutional neural network for feature extraction processing to obtain a high-dimensional geometric feature vector;

[0104] The mapping module is used to perform nonlinear mapping processing on the high-dimensional geometric feature vector through a deep belief network to obtain the B-spline surface control parameters;

[0105] The evaluation module is used to perform point cloud reconstruction quality evaluation processing based on the B-spline surface control parameters to obtain geometric accuracy error data and surface reconstruction integrity data.

[0106] above Figure 2 The point cloud data processing system for industrial 3D model construction in this embodiment of the invention is described in detail from the perspective of modular functional entities. The point cloud data processing device for industrial 3D model construction in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0107] Reference Figure 3The embodiment of the present application also provides a point cloud data processing device for industrial three-dimensional model construction, which can be a server, and the internal structure of the server can be as shown in the accompanying drawings. Figure 3 The point cloud data processing device for industrial three-dimensional model construction comprises a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. The processor of the computer is used to provide computing and control capabilities. The memory of the point cloud data processing device for industrial three-dimensional model construction comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the point cloud data processing device for industrial three-dimensional model construction is used to store corresponding data in the embodiment. The network interface of the point cloud data processing device for industrial three-dimensional model construction is used to communicate with an external terminal through network connection. The computer program is executed by the processor to implement the above method.

[0108] Those skilled in the art can understand that Figure 3 The structure shown in the accompanying drawings is only a block diagram of part of the structure related to the present application, and does not constitute a limitation on the point cloud data processing device for industrial three-dimensional model construction to which the present application is applied.

[0109] The present application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium or a volatile computer readable storage medium, and the computer readable storage medium stores instructions, and when the instructions are run on a computer, the computer executes the steps of the point cloud data processing method for industrial three-dimensional model construction.

[0110] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-mentioned system, system and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0111] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application or the entire or part of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing an industrial three-dimensional model construction point cloud data processing device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0112] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A point cloud data processing method for industrial three-dimensional model construction, characterized in that, The method includes: The raw point cloud data of industrial parts is denoised and downsampled to obtain preprocessed point cloud data; Local geometric descriptors are calculated based on the preprocessed point cloud data, and two-dimensional image conversion processing is performed on the local geometric descriptors using a recursive graph algorithm to obtain an industrial feature recursive graph. The process of inputting the industrial feature recursive graph into a convolutional neural network for feature extraction to obtain a high-dimensional geometric feature vector includes: inputting the industrial feature recursive graph into a multi-layer convolutional layer for industrial geometric feature extraction to obtain a multi-scale industrial feature graph set; performing geometric consistency verification on the multi-scale industrial feature graph set based on an industrial geometric constraint module to obtain a geometric feature graph that conforms to manufacturing constraints; performing industrial standardization on the geometric feature graph according to a batch normalization layer to obtain a standardized industrial feature graph; inputting the standardized industrial feature graph into a multi-scale feature fusion module for industrial semantic fusion to obtain a fused industrial feature graph; and performing high-dimensional mapping on the fused industrial feature graph based on a fully connected layer to obtain the high-dimensional geometric feature vector. The high-dimensional geometric feature vectors are nonlinearly mapped using a deep belief network to obtain the control parameters of the B-spline surface. The point cloud reconstruction quality assessment is performed based on the B-spline surface control parameters to obtain geometric accuracy error data and surface reconstruction integrity data.

2. The point cloud data processing method for industrial three-dimensional model building according to claim 1, characterized in that, The process of denoising and downsampling the raw point cloud data of industrial parts to obtain preprocessed point cloud data includes: The original point cloud data of the industrial parts is subjected to statistical filtering based on manufacturing tolerance constraints to obtain denoised point cloud data that conforms to industrial standards. The denoised point cloud data is subjected to density gradient downsampling processing based on an industrial feature scale-adaptive voxel grid to obtain multi-level density distribution point cloud data. Calculate the industrial processing feature covariance matrix based on the multi-level density distribution point cloud data to obtain geometric feature classification labels including edges, planes, and corners; The geometric feature classification identifiers are processed into industrial semantic blocks to obtain a set of point cloud data blocks with manufacturing semantics; The point cloud data set is processed by constructing a k-nearest neighbor index based on industrial topology continuity constraints to obtain the preprocessed point cloud data.

3. The point cloud data processing method for industrial three-dimensional model building according to claim 1, characterized in that, The step of calculating local geometric descriptors based on the preprocessed point cloud data, and performing two-dimensional image transformation processing on the local geometric descriptors using a recursive graph algorithm to obtain an industrial feature recursive graph includes: A multidimensional local geometric descriptor containing position coordinates, normal vectors, curvature values, and local density is constructed based on the preprocessed point cloud data. Based on the industrial processing feature coordinate system, the local geometric descriptor is processed by Euclidean distance calculation to obtain the industrial feature distance matrix; The industrial feature distance matrix is ​​processed by manufacturing tolerance recursive threshold judgment to obtain an industrial standard recursive binary matrix; Based on the different industrial feature types of fillets, threads, and chamfers, the recursive binary matrix is ​​transformed using a dedicated coordinate system to obtain recursive images of multiple types of industrial features. The industrial feature recursive image is rotated, scaled, and translated based on the industrial geometric invariance constraint to obtain the industrial feature recursive map.

4. The point cloud data processing method for industrial three-dimensional model building according to claim 1, characterized in that, The full connection layer-based high-dimensional mapping processing of the fused industrial feature map obtains the high-dimensional geometric feature vector, including: The fused industrial feature map is flattened and expanded to obtain a one-dimensional feature sequence containing industrial processing feature information; The one-dimensional feature sequence is processed by a first full connection layer for industrial part geometric semantic coding to obtain an industrial semantic coding vector containing cylindrical surface, plane, and thread features; The industrial semantic coding vector is processed by a manufacturing feature nonlinear mapping according to a ReLU activation function to obtain an activated industrial geometric feature vector; The activated industrial geometric feature vector is input into a second full connection layer for industrial CAD feature dimension compression to obtain a compressed feature vector suitable for B-spline reconstruction; The compressed feature vector is processed by an industrial standard normalization based on L2 norm constraint to obtain the high-dimensional geometric feature vector.

5. The point cloud data processing method for industrial three-dimensional model building according to claim 1, characterized in that, The high-dimensional geometric feature vector is processed by a deep belief network for nonlinear mapping to obtain B-spline surface control parameters, including: The high-dimensional geometric feature vector is input into a restricted Boltzmann machine for energy function calculation, and when the hidden layer unit activation probability is greater than a threshold value, an activated hidden feature is obtained; The activated hidden feature is processed by a manufacturing constraint function for radius, thickness, and angle deviation judgment, and when the deviation value is less than the standard tolerance, a qualified constraint feature is obtained; The qualified constraint feature is input into a restricted Boltzmann machine for contrastive divergence learning, and when the weight update converges, a stable semantic feature is obtained; The stable semantic feature is processed by a feature encoder for control point, weight, and node vector decoding, and when the decoding error meets the accuracy requirement, a candidate B-spline parameter is obtained; The candidate B-spline parameter is processed by a quality evaluation module for CAD geometric verification, and when the geometric continuity and smoothness meet the preset standard, the B-spline surface control parameter is obtained.

6. The point cloud data processing method for industrial three-dimensional model building according to claim 1, characterized in that, The B-spline surface control parameter is processed for point cloud reconstruction quality evaluation to obtain geometric accuracy error data and surface reconstruction completeness data, including: The B-spline surface control parameter is processed for surface reconstruction, and when the control point weight coefficient meets the convergence condition, a reconstructed surface model is obtained; The reconstructed surface model is processed by a distance calculation algorithm for point cloud comparison analysis, and when the average distance error is less than the manufacturing precision threshold, the geometric accuracy error data is obtained; The reconstructed surface model is input into a curvature continuity detection algorithm for shape accuracy verification, and when the curvature change rate meets the smoothness requirement, a shape integrity evaluation result is obtained; The shape integrity evaluation result is processed by a key feature point positioning algorithm for position deviation calculation, and when the hole center, edge, and corner position deviation meets the tolerance standard, feature retention rate data is obtained; The feature retention rate data is processed by a comprehensive evaluation algorithm for model completeness quantification, and when the feature retention rate and model continuity both meet the standard, the surface reconstruction completeness data is obtained.

7. A point cloud data processing system for industrial three-dimensional model building, characterized by, The point cloud data processing system for industrial three-dimensional model construction comprises: a denoising module configured to denoise and downsample the original point cloud data of the industrial part to obtain preprocessed point cloud data; a computing module configured to calculate a local geometric descriptor based on the preprocessed point cloud data, and convert the local geometric descriptor into a two-dimensional image through a recursive graph algorithm to obtain an industrial feature recursive graph; an extracting module configured to input the industrial feature recursive graph into a convolutional neural network to perform feature extraction processing to obtain a high-dimensional geometric feature vector, including: inputting the industrial feature recursive graph into a multi-layer convolutional layer to perform industrial geometric feature extraction processing to obtain a multi-scale industrial feature map set; performing geometric consistency verification processing on the multi-scale industrial feature map set based on an industrial geometric constraint module to obtain a geometric feature map that meets manufacturing constraints; performing industrial standardization processing on the geometric feature map based on a batch normalization layer to obtain a standardized industrial feature map; inputting the standardized industrial feature map into a multi-scale feature fusion module to perform industrial semantic fusion processing to obtain a fused industrial feature map; and performing high-dimensional mapping processing on the fused industrial feature map based on a fully connected layer to obtain the high-dimensional geometric feature vector; a mapping module configured to perform non-linear mapping processing on the high-dimensional geometric feature vector through a deep belief network to obtain B-spline surface control parameters; an evaluating module configured to perform point cloud reconstruction quality evaluation processing based on the B-spline surface control parameters to obtain geometric precision error data and surface reconstruction completeness data.

8. A point cloud data processing device for industrial three-dimensional model building, characterized by, The computer program, when executed by the processor, causes the processor to implement the point cloud data processing method for industrial three-dimensional model construction according to any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, causes the processor to implement the point cloud data processing method for industrial three-dimensional model construction according to any one of claims 1 to 6.

Images