Three-dimensional point AI registration and tolerance analysis method and system for non-contact measurement

Abstract

The invention provides a non-contact measurement-oriented three-dimensional point AI registration and tolerance analysis method and system, and relates to the field of three-dimensional point AI registration and tolerance analysis, and the method comprises the steps: synchronously collecting point cloud data through multiple sensors, and carrying out the adaptive preprocessing, and obtaining a preprocessed point cloud pair; constructing a multi-scale geometric descriptor by using a local curvature and a normal vector, and performing multi-scale hierarchical feature extraction to obtain an enhanced point feature set; calculating a bidirectional matching probability matrix by using an optimal transmission theory, screening high-confidence point pairs through spatial compatibility constraint to obtain a high-confidence matching point pair set and a matching score, performing transformation matrix estimation and quantification uncertainty, and outputting optimal rigid body transformation and a covariance matrix thereof; self-adaptive optimization is executed in a layered mode, the search range is dynamically adjusted, transformation parameters are verified in a cross-scale mode, and a convergent accurate registration result is output. The method is used for overcoming the defect that in the prior art, low-overlapping-rate registration precision is low.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional point AI registration and tolerance analysis, and in particular to a three-dimensional point AI registration and tolerance analysis method and system for non-contact measurement. Background Art

[0002] 3D point cloud registration and tolerance analysis are core technologies in computer vision, intelligent manufacturing, and industrial inspection. With the development of non-contact measurement technologies, such as LiDAR, structured light scanning, and photogrammetry, the acquisition of 3D point cloud data has become more convenient and efficient, driving the evolution from traditional algorithms to AI-driven solutions to meet the higher requirements of modern industry for accuracy, efficiency, and automation.

[0003] 3D point cloud registration refers to the process of aligning point cloud data acquired from different perspectives or times into a unified coordinate system through spatial transformation. This is a key step in applications such as 3D reconstruction and quality inspection. Tolerance analysis is a key link between product design and manufacturing, and is used to evaluate and control the impact of part size, shape and position deviations on product functional quality.

[0004] In existing technologies, however, in low overlap scenarios (overlap rate 15%-30%), existing technologies face major challenges, and the accuracy and robustness of point cloud processing with low overlap rate are low. Summary of the Invention

[0005] The present invention provides a three-dimensional point AI registration and tolerance analysis method and system for non-contact measurement, which is used to solve the defect of low overlap rate and low registration accuracy in the prior art.

[0006] In one aspect, the present invention provides a three-dimensional point AI registration and tolerance analysis method for non-contact measurement, comprising: The point cloud data is collected synchronously by multiple sensors and then adaptively preprocessed to obtain preprocessed point cloud pairs; Based on the preprocessed point cloud pairs, a multi-scale geometric descriptor is constructed using local curvature and normal vectors, and multi-scale hierarchical feature extraction is performed to obtain an enhanced point feature set; Based on the enhanced point feature set, the optimal transmission theory is used to calculate the two-way matching probability matrix, and high-confidence point pairs are screened through spatial compatibility constraints to obtain a high-confidence matching point pair set and matching score. Based on the high-confidence matching point pair set and matching score, the transformation matrix is ​​estimated, the uncertainty is quantified, and the optimal rigid body transformation and its covariance matrix are output; Based on the optimal rigid body transformation and its covariance matrix, adaptive optimization is performed hierarchically, the search range is dynamically adjusted, and the transformation parameters are verified across scales to output a converged and accurate registration result. Based on the converged and precise registration results, a three-dimensional error field is constructed and the partition error characteristics are statistically analyzed. Abnormal areas are interactively displayed through heat maps and streamline maps, and an error analysis report and interactive visualization interface are output.

[0007] Furthermore, point cloud data is collected synchronously by multiple sensors and adaptively preprocessed to obtain preprocessed point cloud pairs, including: The point cloud data collected by multiple sensors are synchronized in time and space, and the coordinate system is unified to obtain aligned multimodal data; Based on the aligned multimodal data, the sampling density is dynamically adjusted using a voxelized grid, and the sampling granularity is determined according to the local curvature changes of the point cloud to obtain filtered data. Based on the filtered data, outlier removal and normal vector consistency detection are combined to filter outliers and output clean and density-balanced multimodal point cloud pairs, i.e., preprocessed point cloud pairs.

[0008] Furthermore, based on the preprocessed point cloud pairs, a multi-scale geometric descriptor is constructed using local curvature and normal vectors, and multi-scale hierarchical feature extraction is performed to obtain an enhanced point feature set, including: Based on the preprocessed point cloud pairs, the multi-scale covariance matrix of each cloud point is calculated, and the multi-scale eigenvalues ​​and normal vectors are extracted through singular value decomposition to obtain the local geometric code; Based on local geometric coding, spatial attention is used to focus on key structures, and channel attention is used to strengthen the discriminative feature dimensions, thus obtaining features enhanced by dual attention. Based on the features enhanced by dual attention, features of different scales are fused bottom-up through the feature propagation module to obtain the enhanced point feature set.

[0009] Furthermore, based on the enhanced point feature set, the optimal transmission theory is used to calculate the two-way matching probability matrix, and high-confidence point pairs are screened through spatial compatibility constraints to obtain a high-confidence matching point pair set and matching scores, including: Based on the enhanced point feature set, the inner product similarity of the source point cloud and the target point cloud features is calculated and normalized to obtain the constructed two-way matching probability matrix: Based on the bidirectional matching probability matrix, the similarity matrix is ​​expanded and the empty set is added. The optimal transmission plan matrix with the empty set constraint is solved to obtain the optimal transmission matching plan matrix. Based on the optimal transmission matching plan matrix, geometrically consistent point pairs are screened through distance invariance and triangle ratio consistency under rigid transformation, and a set of high-confidence matching point pairs and matching scores are obtained.

[0010] Further, based on the high-confidence matching point pair set and the matching score, a transformation matrix is estimated, and uncertainty is quantified, and an optimal rigid transformation and its covariance matrix are output, including: Based on the high-confidence matching point pair set and the matching score, the centroid of the matching point pair is calculated, and the initial transformation is obtained through covariance matrix decomposition; Based on the initial transformation, the matching error distribution is simulated using Monte Carlo simulation, and multiple sets of perturbed matching point pairs are generated; Based on the multiple sets of perturbed matching point pairs, the distribution characteristics are counted and the covariance matrix is established to represent the uncertainty of the transformation, and the constructed uncertainty propagation matrix is output; Based on the uncertainty propagation matrix, a candidate transformation is generated for each high-confidence matching point pair subset, and the optimal transformation is selected through a voting mechanism, and the optimal rigid transformation and its covariance matrix are output.

[0011] Further, based on the optimal rigid transformation and its covariance matrix, hierarchical adaptive optimization is performed, the search range is dynamically adjusted, and the transformation parameters are verified across scales, and the converged accurate registration result is output, including: Based on the optimal rigid transformation and its covariance matrix, hierarchical accurate registration is performed on the rigid transformation parameters from coarse to fine, and the optimized rigid transformation parameters are obtained; Based on the optimized rigid transformation parameters, the search range is dynamically adjusted, the point-to-local plane distance is used to accelerate convergence, and the transformation consistency under different resolutions is checked, and abnormal estimates are removed, and the converged accurate registration result is output.

[0012] Further, based on the converged accurate registration result, a three-dimensional error field is constructed and the partition error characteristics are counted, and the abnormal areas are interactively displayed through the heat map and streamline map, and an error analysis report and an interactive visualization interface are output, including: Based on the converged accurate registration result, the dense corresponding error vectors of the registered point cloud and the target point cloud are calculated, and a three-dimensional error vector field is established; Based on the three-dimensional error vector field, the mean, standard deviation and extreme value are calculated in each partition, the systematic error and random noise are distinguished, and the statistical characteristic analysis result is obtained; Based on the statistical characteristic analysis result, the error amplitude is displayed using the heat map, and the error direction pattern is displayed using the streamline map.

[0013] On the other hand, the present application also provides a three-dimensional point AI registration and tolerance analysis system for non-contact measurement, including: An acquisition module is configured to synchronously acquire point cloud data through multiple sensors and perform adaptive preprocessing to obtain a preprocessed point cloud pair; The processing module is configured to construct a multi-scale geometric descriptor using local curvature and normal vector based on the pre-processed point cloud pair, and perform multi-scale hierarchical feature extraction to obtain an enhanced point feature set; based on the enhanced point feature set, a bidirectional matching probability matrix is calculated using optimal transport theory, and a high-confidence point pair set and a matching score are obtained by filtering high-confidence point pairs through spatial compatibility constraints; based on the high-confidence point pair set and the matching score, a transformation matrix is estimated, and uncertainty is quantified, and an optimal rigid transformation and a covariance matrix thereof are output; based on the optimal rigid transformation and the covariance matrix thereof, adaptive optimization is performed in layers, the search range is dynamically adjusted, and the transformation parameters are verified across scales, and a converged accurate registration result is output; based on the converged accurate registration result, a three-dimensional error field is constructed, and partition error characteristics are counted, and an abnormal area is interactively displayed through a heat map and a streamline map, and an error analysis report and an interactive visualization interface are output.

[0014] In another aspect, the present application also provides an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the above-mentioned any one of the three-dimensional point AI registration and tolerance analysis methods for non-contact measurement when executing the program.

[0015] In another aspect, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the above-mentioned any one of the three-dimensional point AI registration and tolerance analysis methods for non-contact measurement.

[0016] In another aspect, the present application also provides a computer program product comprising a computer program, wherein the computer program is executed by a processor to implement the above-mentioned any one of the three-dimensional point AI registration and tolerance analysis methods for non-contact measurement.

[0017] The three-dimensional point AI registration and tolerance analysis method and system for non-contact measurement provided by the present application solve the low overlap ratio registration problem through a hierarchical processing strategy, and realize the collaborative optimization of registration and tolerance analysis, are designed for non-contact measurement scenarios, form a complete closed loop from data acquisition to final analysis, and significantly improve the accuracy and robustness of low overlap ratio point cloud processing; multi-modal feature fusion coding combines geometric, color and local context information, and enhances the representation ability of low overlap areas through hierarchical feature extraction; dynamic trusted area perception uses optimal transport theory combined with spatial compatibility constraints to adaptively identify and enhance the overlap area, and significantly reduces the interference of non-overlapping point pair registration; the registration uncertainty is transmitted to the tolerance analysis link, a probabilistic tolerance evaluation model is established, the influence of measurement error on tolerance analysis is quantified through Monte Carlo simulation, and end-to-end uncertainty management from registration to tolerance analysis is realized. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 1 is a flow chart of a method for non-contact measurement-oriented 3D point AI registration and tolerance analysis provided by an embodiment of the present invention; Figure 2 Schematic diagram of a 3D point AI registration and tolerance analysis system for non-contact measurement provided by an embodiment of the present invention; Figure 3 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0021] Figure 1 This is one of the flow charts of the three-dimensional point AI registration and tolerance analysis method for non-contact measurement provided by an embodiment of the present invention.

[0022] like Figure 1 As shown, the embodiment of the present invention provides a three-dimensional point AI registration and tolerance analysis method for non-contact measurement, which mainly includes the following steps: 11. Use multiple sensors to synchronously collect point cloud data and perform adaptive preprocessing to obtain preprocessed point cloud pairs; 12. Based on the preprocessed point cloud pairs, a multi-scale geometric descriptor is constructed using local curvature and normal vectors, and multi-scale hierarchical feature extraction is performed to obtain an enhanced point feature set; 13. Based on the enhanced point feature set, the optimal transmission theory is used to calculate the two-way matching probability matrix, and high-confidence point pairs are screened through spatial compatibility constraints to obtain a high-confidence matching point pair set and matching score; 14. Based on the high-confidence matching point pair set and matching score, the transformation matrix is ​​estimated, the uncertainty is quantified, and the optimal rigid body transformation and its covariance matrix are output; 15. Based on the optimal rigid transformation and its covariance matrix, hierarchical adaptive optimization is performed, the search range is dynamically adjusted, and the transformation parameters are verified across scales, and the converged accurate registration result is output; 16. Based on the converged accurate registration result, a three-dimensional error field is constructed and the partition error characteristics are counted, the abnormal areas are interactively displayed through the heat map and streamline map, and the error analysis report and interactive visualization interface are output.

[0023] In the embodiment of the application, multi-sensor synchronous acquisition and adaptive preprocessing are performed, multi-sensor (such as LiDAR, RGB camera) synchronous acquisition is performed, the problem of incomplete single sensor data is solved, multi-modal information (geometry, color, reflection intensity) is fused to enhance data richness, adaptive preprocessing (voxelization downsampling, statistical denoising) can reduce the data volume while retaining key features, and the subsequent calculation complexity is significantly reduced; multi-scale hierarchical feature extraction, multi-scale descriptors based on local curvature and normal vector, combined with double attention mechanism (space + channel), effectively capture the subtle geometric structure of the low overlap area, 128-dimensional feature vector encodes local details and global context, and the feature discrimination degree is maintained when the overlap rate is less than 20%, solving the failure problem of traditional manual features (such as FPFH) in sparse areas; optimal transport matching and spatial compatibility verification, the bidirectional matching probability matrix generated by the optimal transport theory (Sinkhorn algorithm) processes noise interference through entropy regularization, the matching accuracy is improved, and the spatial compatibility constraint (first-order distance invariance + second-order triangle proportion) further eliminates the geometric inconsistent mismatch, so that the registration success rate of the low overlap rate (overlap rate 15%-30%) scene is improved; transformation matrix estimation and uncertainty quantification, weighted SVD solution combined with Monte Carlo simulation, not only output the optimal rigid transformation, but also quantify the covariance matrix, realize the probabilistic expression of registration error, and provide a reliable error propagation model for tolerance analysis; hierarchical optimization and cross-scale verification, the hierarchical optimization strategy (adaptive ICP) from coarse to fine improves the calculation efficiency, and the dynamic adjustment of the search range (based on the overlap rate estimation) avoids local optimization. Cross-scale verification eliminates abnormal transformation parameters, shortens the registration time of the aviation structural parts (1.2M point cloud), and improves the accuracy; three-dimensional error field quantifies systematic errors (such as offset) and random noise, partitions statistics (mean / standard deviation) to locate manufacturing defects, and heat map and streamline map interactive visualization helps engineers quickly identify out-of-tolerance areas, and improves the accuracy of key profile tolerance analysis in industrial gear box detection.

[0024] As shown in Figure 1 Step 11 includes: 111. Temporally and spatially synchronizing the point cloud data acquired by the multi-sensor to obtain aligned multi-modal data in a unified coordinate system; 112. Based on aligned multimodal data, the sampling density is dynamically adjusted using voxelized grids, and the sampling granularity is determined according to the change in local curvature of the point cloud to obtain filtered data; 113. Based on the filtered data, outlier removal and normal vector consistency detection are combined to filter outliers and output clean and density-balanced multimodal point cloud pairs, i.e., preprocessed point cloud pairs.

[0025] In an embodiment of the present invention, through time-space synchronization and coordinate system unification, the systematic error of multi-sensor data is eliminated, the geometric consistency of subsequent processing is ensured, and the foundation is laid for low overlap registration; the point cloud space is divided into voxel grids (such as the initial voxel side length is 5mm), and each voxel retains a representative point. Local curvature calculation: the covariance matrix is ​​calculated for the neighborhood of each point p (radius r = 3 times the voxel side length), and the eigenvalue and curvature are obtained by singular value decomposition. Dynamic granularity adjustment: in high curvature areas (κ>threshold), the voxel side length is reduced (such as 2mm) to retain details, and in flat areas (κ≤threshold), the voxel side length is increased (such as 8mm) for sparseness; the k nearest neighbors of each point p are calculated (such as k= 50) Average distance μ and standard deviation σ, eliminate points satisfying d(p)>μ+ασ, normal vector consistency detection: adaptive downsampling reduces the data volume by more than 70% (for example, from 1.2M points to 350K points), while retaining 98.5% of key geometric features, significantly reducing the computational complexity of subsequent feature extraction and registration; calculate the angle θ between the normal vector of point p and the average direction of the neighborhood normal vector, eliminate outliers with θ>30°, and output a clean and density-balanced point cloud pair; outlier point filtering (such as statistical outlier removal and normal vector consistency detection) can reduce the noise level by 65% ​​(from 2.3mm to 0.8mm), effectively avoiding the interference of flying points and measurement errors on registration.

[0026] like Figure 1 As shown, step 12 includes: 121. Based on the preprocessed point cloud pairs, the multi-scale covariance matrix of each cloud point is calculated, and the multi-scale eigenvalues ​​and normal vectors are extracted through singular value decomposition to obtain the local geometric code; 122. Based on local geometric coding, spatial attention is used to focus on key structures, and channel attention is used to strengthen the discriminative feature dimensions, thus obtaining features enhanced by dual attention. 123. Based on the features enhanced by dual attention, features of different scales are fused from bottom to top through the feature propagation module to obtain the enhanced point feature set.

[0027] In the embodiment of the present invention, by calculating the multi-scale covariance matrix of each point (such as the neighborhood radius r1 <r2<r3)并进行奇异值分解(SVD),提取多尺度特征值(λ1,λ2,λ3)和法向量(n1,n2,n3),构建局部几何描述符(如曲率、法向量夹角等),能够同时捕捉微观曲率变化和宏观表面趋势,在低重叠率场景下(重叠率15%-30%)显著提升特征区分度; Spatial attention dynamically allocates weights of local geometric patterns through 3D convolution kernels to focus on high curvature areas and structural boundaries, suppressing interference from flat areas; channel attention evaluates the importance of feature dimensions through normalized mutual information (DNMI), strengthening discriminative features that are robust to rotation and occlusion; the dual attention mechanism combines local details with global context, allowing features to maintain high discriminability in low-overlap areas (such as 15% overlap) and reducing the false matching rate; Hierarchical feature fusion uses the feature propagation (FP) module to fuse multi-scale features (such as local descriptors with a radius from 0.1m to 1.0m) from the bottom up to construct a 128-dimensional feature vector that combines local subtle structure with global consistency. The hierarchical fusion strategy effectively addresses the limitations of single-scale features in complex scenes and reduces registration errors.

[0028] like Figure 1 As shown, step 13 includes: 131. Based on the enhanced point feature set, the inner product similarity of the source point cloud and the target point cloud features is calculated and normalized to obtain the constructed two-way matching probability matrix: 132. Based on the bidirectional matching probability matrix, the similarity matrix is ​​expanded and the empty set is added. The optimal transmission plan matrix with the empty set constraint is solved to obtain the optimal transmission matching plan matrix. 133. Based on the optimal transmission matching plan matrix, geometrically consistent point pairs are screened through distance invariance and triangle ratio consistency under rigid transformation to obtain a set of high-confidence matching point pairs and matching scores.

[0029] In the embodiment of the present invention, the enhanced point feature set is input (Each point is a 128-dimensional vector); Inner product similarity calculation: source point cloud Middle Point and target point cloud Middle Feature similarity of points: ;in ;

[0030] Among them, S ij is the feature similarity, is the feature covariance weight (default 0.2), which strengthens the consistency of local feature distribution. is the neighborhood feature covariance matrix, which improves sensitivity to local geometric structure; bidirectional normalization: row normalization: , column normalization: , the final similarity: ,in, is the row and column weight balance factor (default 0.6) to suppress outlier dominance; outputs the two-way matching probability matrix ,satisfy ; To solve the optimal transmission matching plan matrix, first perform empty set expansion and expand the similarity matrix , add row and column filling values , indicating a "no match" state; allowing points to be matched at a cost Mismatching avoids error propagation caused by forced pairing; entropy regularization optimizes transmission: ;

[0031] constraint: ; Where T is the optimal transmission matrix, is the Mahalanobis distance weight, (、 is a learnable projection matrix) to enforce feature space alignment; is the prior distribution weight, Initialized by point density distribution, improve the uniformity of matching space; solve Sinkhorn iteration (complexity ), 10 iterations converge; output the optimal transmission plan matrix ,element Represents a matching pair Confidence score of ; Verification of geometric compatibility constraints, first-order distance constraints (rigid invariance): ; in, is the local curvature change rate, and the threshold is adjusted adaptively. is the curvature sensitivity coefficient, and the constraints are relaxed in high curvature areas. is the base distance tolerance; Second-order proportionality constraint (triangle similarity): ; in, is the local surface flatness measure, is the noise suppression factor (default 0.05), tightening the constraints in the flat area; is the ratio tolerance threshold; Compatibility matrix construction: ;

[0032] Spectral Clustering is used to filter the largest connected subgraph, retain geometrically consistent matching pairs, and output a set of high-confidence matching point pairs. and scores , non-overlapping points are automatically excluded (e.g. ); Empty set constraints and covariance regularization solve the forced matching problem of low-overlap points in traditional optimal transmission, reducing the mismatch rate; adaptive geometric constraints dynamically adjust the threshold through curvature and flatness, maintaining 95% matching accuracy on complex structures; end-to-end probabilistic framework, from feature matching to geometric verification, probabilistic modeling of the entire process, provides an uncertainty quantification basis for tolerance analysis; through the fusion of multi-order geometric constraints and probabilistic transmission models, highly robust alignment of low-overlap point clouds is achieved, laying a reliable correspondence for subsequent tolerance analysis.

[0033] like Figure 1 As shown, step 14 includes: 141. Based on the set of high-confidence matching point pairs and the matching scores, the centroid of the matching point pairs is calculated, and the initial transformation is obtained by covariance matrix decomposition; 142. Based on the initial transformation, Monte Carlo simulation is used to simulate the matching error distribution and generate multiple sets of perturbed matching point pairs; 143. Based on multiple sets of perturbation matching point pairs, the statistical distribution characteristics are established and the covariance matrix is ​​established to represent the uncertainty of the transformation, and the constructed uncertainty propagation matrix is ​​output; 144. Based on the uncertainty propagation matrix, candidate transformations are generated for each subset of high-confidence matching point pairs, the optimal transformation is selected through a voting mechanism, and the optimal rigid body transformation and its covariance matrix are output.

[0034] In the embodiment of the present invention, the robustness problem of transformation matrix estimation in low overlap point cloud registration is solved by weighted SVD and probabilistic perturbation modeling, and the covariance propagation of registration error is realized, providing a reliability quantification basis for tolerance analysis; weighted SVD solves the initial transformation: input the set of credible point pairs and matching scores , perform curvature adaptive weight correction: introduce local curvature weight for each matching pair ,in is the point curvature, is the curvature bandwidth (default 0.1) and modifies the matching score: ,in is the feature similarity gain (default 0.3); Covariance matrix construction: ; Among them, γ is the regularization factor, which suppresses the singular matrix problem.

[0035] SVD decomposition and reflection processing are: ; Monte Carlo simulation matching error distribution, error disturbance model, first define the disturbance parameters: position noise , characteristic noise ,in, Point cloud spacing, Feature norm; Generate perturbation point pairs: , , the eigenvector ; Spatial Correlation Constraint: Add a spatial correlation matrix , the disturbance term is corrected to ( is the relevant radius, the default ); output Group perturbation matching point pairs (generally ); Transformation distribution statistics and covariance modeling, Lie algebra parameterization: the rotation matrix Convert to Lie algebra (6-dimensional vector: 3-dimensional rotation + 3-dimensional translation); Covariance matrix calculation: ;

[0036] in, is the Lie algebra parameter of the nth Monte Carlo simulation, is the parameter mean, is the conservative factor, The largest standard deviation in history; The uncertainty propagation matrix is: Introducing Mahalanobis distance weight: ,in is the historical maximum variance, is a conservative factor; Multi-hypothesis fusion and optimal transformation selection, generate subsets: divide the point pairs into Subset , each subset contains at least 3 non-coplanar points; candidate transformation generation: for each subset Perform weighted SVD to obtain candidate transformations ; Voting mechanism: Define voting scores: ; in, is the time decay factor (default 0.01), is the number of iterations; the transformation with the highest voting score is selected as the output; subset partitioning avoids local optimality and improves the registration success rate in low overlap rate scenarios (overlap rate 15%-30%); the time decay factor gives priority to new hypotheses to adapt to dynamic scanning scenarios.

[0037] like Figure 1 As shown, step 15 includes: 151. Based on the optimal rigid body transformation and its covariance matrix, the rigid body transformation parameters are hierarchically precisely aligned layer by layer from coarse to fine to obtain the optimized rigid body transformation parameters; 152. Based on the optimized rigid body transformation parameters, the search range is dynamically adjusted, the point-to-local plane distance is used to accelerate convergence, and the transformation consistency under different resolutions is checked to eliminate abnormal estimates and output the converged and accurate registration results.

[0038] In the embodiment of the present invention, the convergence speed and local optimum problems in low overlap point cloud registration are solved through the hierarchical optimization strategy and dynamic search mechanism. At the same time, the uncertainty of the transformation parameters is quantified by the covariance matrix to achieve robust registration from coarse to fine. and the covariance matrix , downsample those multi-scales: build pyramid levels , the downsampling rate of each layer is (default ), the coarsest layer ( ) The point cloud resolution is reduced to the original data ; Dynamically adjust the downsampling granularity:

[0039] in, The curvature gradient sensitivity coefficient (default 0.2) retains more details in high curvature areas. is the average curvature of the current layer; layer-by-layer optimization:

[0040] Layered optimization: each layer solves the transformation parameters through weighted SVD , weighted combined with matching score Confidence of the cooperative party:

[0041] in, is the Lie algebra parameter of the current layer, is the inverse covariance matrix; Adaptive ICP improvement: The point-to-point error of traditional ICP is changed to the distance from the point to the local plane of the target point:

[0042] in, Align weights for normal vectors to enforce normal vector consistency; Dynamic search range adjustment, estimating overlap_est and local point density based on overlap ratio Adjust search radius:

[0043] in, is the overlap rate weight, is the point density coefficient, is the standard deviation of the noise in the current layer, which is calculated by the eigenvalue of the covariance matrix. is the local point density; Adaptive step size: Adjust the iteration step size based on uncertainty: ; Among them, step_max is the maximum allowed step size (default 0.1).

[0044] like Figure 1 As shown, step 16 includes: 161. Based on the converged precise registration results, the dense corresponding error vector between the registered point cloud and the target point cloud is calculated, and a three-dimensional error vector field is established; 162. Based on the three-dimensional error vector field, the mean, standard deviation and extreme value are calculated by partition, the systematic error and random noise are distinguished, and the statistical characteristic analysis results are obtained; 163. Based on the statistical characteristics analysis results, the error amplitude is displayed using a heat map, and the error direction pattern is displayed using a streamline map.

[0045] In an embodiment of the present invention, a three-dimensional error vector field is constructed by calculating the dense corresponding error vector E(x,y,z)=Pt(x,y,z)−Ps′(x,y,z) between the registered point cloud Ps′ and the target point cloud Pt to generate a three-dimensional error field. The error field quantifies the registration deviation of each spatial point and can intuitively reflect the local registration accuracy. For example, in industrial parts inspection, the error vector field can accurately identify the micron-level deviation (such as 0.15mm) in the tooth profile area, which is significantly better than the traditional method (0.5mm); partition statistical characteristic analysis is performed to divide the error field into grid areas, and the mean (reflecting the system offset), standard deviation (characterizing random noise) and extreme value (identifying the maximum deviation) of each area are calculated respectively. By distinguishing between systematic error and random noise, the registration algorithm can be optimized or the sensor parameters can be adjusted in a targeted manner. For example, in aviation During structural component inspection, zoning statistics revealed a system error of 0.23mm in a certain area, which was traced back to sensor calibration deviation. After correction, the error was reduced by 60%. In interactive visualization, the heat map uses color gradients to map the error amplitude distribution. Red indicates high-error areas (such as deviations of more than 0.3mm), and blue indicates low-error areas. Users can interactively locate defect locations by zooming and rotating, such as the abnormally high-temperature area of ​​a gearbox tooth root crack. Streamline maps use arrows or curves to display the directional pattern of error vectors, revealing systematic deformation trends (such as overall translation or rotation). The combination of heat maps and streamline maps improves the detection rate of abnormal areas. Zoning statistics quickly distinguishes sensor errors from environmental noise, shortening debugging time. Interactive reports support engineers to adjust inspection strategies in real time, such as dynamically adjusting the laser scanning path through heat maps in aviation skin inspection.

[0046] like Figure 2 As shown, a 3D point AI registration and tolerance analysis system 20 for non-contact measurement is characterized by comprising: An acquisition module 21 is used to synchronously collect point cloud data through multiple sensors and perform adaptive preprocessing to obtain preprocessed point cloud pairs; The processing module 22 is used to construct a multi-scale geometric descriptor based on the preprocessed point cloud pairs using local curvature and normal vectors, and perform multi-scale hierarchical feature extraction to obtain an enhanced point feature set; based on the enhanced point feature set, the optimal transmission theory is used to calculate the two-way matching probability matrix, and high-confidence point pairs are screened through spatial compatibility constraints to obtain a high-confidence matching point pair set and a matching score; based on the high-confidence matching point pair set and the matching score, the transformation matrix is ​​estimated and the uncertainty is quantified to output the optimal rigid body transformation and its covariance matrix; based on the optimal rigid body transformation and its covariance matrix, adaptive optimization is performed hierarchically, the search range is dynamically adjusted, and the transformation parameters are verified across scales to output a converged and accurate registration result; based on the converged and accurate registration result, a three-dimensional error field is constructed and the partition error characteristics are statistically analyzed, the abnormal area is interactively displayed through a heat map and a streamline map, and an error analysis report and an interactive visualization interface are output.

[0047] Figure 3 It is a structural diagram of an electronic device provided by an embodiment of the present invention.

[0048] like Figure 3 As shown, the electronic device may include: a processor 610, a communications interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communications interface 620, and the memory 630 communicate with each other via the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the three-dimensional point AI registration and tolerance analysis method for non-contact measurement.

[0049] In addition, the logic instructions in the aforementioned memory 630 can be implemented in the form of a software functional unit and, when sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0050] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the three-dimensional point AI alignment and tolerance analysis methods for non-contact measurement provided by the above methods.

[0051] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the three-dimensional point AI registration and tolerance analysis method for non-contact measurement provided by the above methods.

[0052] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0053] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A 3D point AI registration and tolerance analysis method for non-contact measurement, characterized in that: include: Point cloud data is collected synchronously by multiple sensors and adaptively preprocessed to obtain preprocessed point cloud pairs; Based on the preprocessed point cloud pairs, a multi-scale geometric descriptor is constructed using local curvature and normal vectors, and multi-scale hierarchical feature extraction is performed to obtain an enhanced point feature set; Based on the enhanced point feature set, the optimal transmission theory is used to calculate the two-way matching probability matrix, and high-confidence point pairs are screened through spatial compatibility constraints to obtain a high-confidence matching point pair set and matching score. Based on the high-confidence matching point pair set and matching score, the transformation matrix is ​​estimated, the uncertainty is quantified, and the optimal rigid body transformation and its covariance matrix are output; Based on the optimal rigid body transformation and its covariance matrix, adaptive optimization is performed hierarchically, the search range is dynamically adjusted, and the transformation parameters are verified across scales to output a converged and accurate registration result. Based on the converged and precise registration results, a three-dimensional error field is constructed and the partition error characteristics are statistically analyzed. Abnormal areas are interactively displayed through heat maps and streamline maps, and an error analysis report and interactive visualization interface are output.

2. The three-dimensional point AI registration and tolerance analysis method for non-contact measurement according to claim 1 is characterized in that: The point cloud data is collected synchronously by multiple sensors and then adaptively preprocessed to obtain preprocessed point cloud pairs, including: The point cloud data collected by multiple sensors are synchronized in time and space, and the coordinate system is unified to obtain aligned multimodal data; Based on the aligned multimodal data, the sampling density is dynamically adjusted using voxelized grids, and the sampling granularity is determined according to the local curvature changes of the point cloud to obtain filtered data. Based on the filtered data, outlier removal and normal vector consistency detection are combined to filter outliers and output clean and density-balanced multimodal point cloud pairs, i.e., preprocessed point cloud pairs.

3. The three-dimensional point AI registration and tolerance analysis method for non-contact measurement according to claim 2 is characterized in that: Based on the preprocessed point cloud pairs, a multi-scale geometric descriptor is constructed using local curvature and normal vectors, and multi-scale hierarchical feature extraction is performed to obtain an enhanced point feature set, including: Based on the preprocessed point cloud pairs, the multi-scale covariance matrix of each cloud point is calculated, and the multi-scale eigenvalues ​​and normal vectors are extracted through singular value decomposition to obtain the local geometric code; Based on local geometric coding, spatial attention is used to focus on key structures, and channel attention is used to strengthen the discriminative feature dimensions, thus obtaining features enhanced by dual attention. Based on the features enhanced by dual attention, features of different scales are fused bottom-up through the feature propagation module to obtain the enhanced point feature set.

4. The three-dimensional point AI registration and tolerance analysis method for non-contact measurement according to claim 3 is characterized in that: Based on the enhanced point feature set, the optimal transmission theory is used to calculate the two-way matching probability matrix. High-confidence point pairs are screened through spatial compatibility constraints to obtain a high-confidence matching point pair set and matching scores, including: Based on the enhanced point feature set, the inner product similarity of the source point cloud and the target point cloud features is calculated and normalized to obtain the constructed two-way matching probability matrix: Based on the bidirectional matching probability matrix, the similarity matrix is ​​expanded and the empty set is added. The optimal transmission plan matrix with the empty set constraint is solved to obtain the optimal transmission matching plan matrix. Based on the optimal transmission matching plan matrix, geometrically consistent point pairs are screened through distance invariance and triangle ratio consistency under rigid transformation, and a set of high-confidence matching point pairs and matching scores are obtained.

5. The three-dimensional point AI registration and tolerance analysis method for non-contact measurement according to claim 4 is characterized in that: Based on the high-confidence matching point pair set and matching score, the transformation matrix is ​​estimated and the uncertainty is quantified. The optimal rigid body transformation and its covariance matrix are output, including: Based on the set of high-confidence matching point pairs and the matching scores, the centroid of the matching point pairs is calculated, and the initial transformation is obtained through covariance matrix decomposition; Based on the initial transformation, Monte Carlo simulation is used to simulate the matching error distribution and generate multiple sets of perturbed matching point pairs. Based on multiple sets of perturbation matching point pairs, the statistical distribution characteristics are established and a covariance matrix is ​​established to represent the uncertainty of the transformation, and the constructed uncertainty propagation matrix is ​​output; Based on the uncertainty propagation matrix, candidate transformations are generated for each subset of high-confidence matching point pairs, the optimal transformation is selected through a voting mechanism, and the optimal rigid body transformation and its covariance matrix are output.

6. The three-dimensional point AI registration and tolerance analysis method for non-contact measurement according to claim 5 is characterized in that: Based on the optimal rigid body transformation and its covariance matrix, it performs hierarchical adaptive optimization, dynamically adjusts the search range, and verifies the transformation parameters across scales, outputting converged and accurate registration results, including: Based on the optimal rigid body transformation and its covariance matrix, the rigid body transformation parameters are hierarchically precisely aligned layer by layer from coarse to fine to obtain the optimized rigid body transformation parameters; Based on the optimized rigid body transformation parameters, the search range is dynamically adjusted, the point-to-local plane distance is used to accelerate convergence, and the transformation consistency at different resolutions is checked to eliminate abnormal estimates and output the converged and accurate registration results.

7. The three-dimensional point AI registration and tolerance analysis method for non-contact measurement according to claim 6 is characterized in that: Based on the converged and precise registration results, a 3D error field is constructed and the partition error characteristics are statistically analyzed. Abnormal areas are interactively displayed through heat maps and streamline maps. An error analysis report and an interactive visualization interface are output, including: Based on the converged precise registration results, the dense corresponding error vector between the registered point cloud and the target point cloud is calculated, and a three-dimensional error vector field is established. Based on the three-dimensional error vector field, the mean, standard deviation and extreme value are calculated by partition, the systematic error and random noise are distinguished, and the statistical characteristics analysis results are obtained; Based on the statistical characteristics analysis results, a heat map is used to display the error amplitude, and a streamline map is used to display the error direction pattern.

8. A 3D point AI registration and tolerance analysis system for non-contact measurement, characterized by: include: The acquisition module is used to synchronously collect point cloud data through multiple sensors and perform adaptive preprocessing to obtain preprocessed point cloud pairs; A processing module is used to construct a multi-scale geometric descriptor based on the pre-processed point cloud pairs using local curvature and normal vectors, and perform multi-scale hierarchical feature extraction to obtain an enhanced point feature set; Based on the enhanced point feature set, the optimal transmission theory is used to calculate the two-way matching probability matrix, and high-confidence point pairs are screened by spatial compatibility constraints to obtain a high-confidence matching point pair set and matching scores. Based on the high-confidence matching point pair set and matching scores, the transformation matrix is ​​estimated, and the uncertainty is quantified to output the optimal rigid body transformation and its covariance matrix. Based on the optimal rigid body transformation and its covariance matrix, hierarchical adaptive optimization is performed, the search range is dynamically adjusted, and the transformation parameters are verified across scales to output a converged and precise registration result. Based on the converged and precise registration result, a three-dimensional error field is constructed and the partition error characteristics are statistically analyzed. The abnormal areas are interactively displayed through heat maps and streamline maps, and an error analysis report and an interactive visualization interface are output.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the three-dimensional point AI registration and tolerance analysis method for non-contact measurement as described in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the three-dimensional point AI registration and tolerance analysis method for non-contact measurement as claimed in any one of claims 1 to 7 is implemented.

Images