Efficient 3d geometric duplicate checking method

Abstract

The invention discloses an efficient 3d geometric duplicate checking method, relates to the technical field of three-dimensional modeling, and lays a foundation for subsequent similarity calculation by performing cloud point generation processing and simplification processing on an input 3D model, obtaining a simplified grid model, extracting key geometric features and converting the key geometric features into feature vector geometry. According to the method, the local region features are extracted through the local region matching stage and the global geometric similarity matching stage, the local similarity score and the global similarity score are calculated through the similarity score obtained through fitting, and the model duplicate checking precision and efficiency are effectively improved. Compared with a traditional method, the scheme can ensure accurate matching of a local area and a global geometric structure on the basis of multi-feature extraction, and information loss or misjudgment is avoided. And by optimizing iterative evaluation and duplicate checking judgment, the accuracy and adaptability of duplicate checking are further ensured, and the complexity and detail differences of different models can be flexibly coped with.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional modeling, and in particular to an efficient 3D geometry duplicate checking method. Background Art

[0002] With the rapid development of digital technology, 3D modeling has been widely used in many fields, such as architectural design, manufacturing, virtual reality, game development, and digital cultural relics protection. In these fields, the creation and processing of 3D data has become a key technical link. Especially in the field of 3D modeling and printing, accurate geometric structure and complex 3D data processing are crucial.

[0003] Although there are already some 3D geometry duplication detection technologies on the market, mainstream methods usually rely on point cloud matching, mesh comparison or simple shape feature calculation, each of which has its own advantages and disadvantages. Although point cloud matching technology can effectively match point cloud data from different sources, its computational complexity is extremely high, especially when faced with large-scale and complex 3D data, the performance often cannot meet actual needs. Mesh comparison usually relies on the mesh structure of the model for comparison, and for models with small differences in details but similar shapes, it often cannot effectively identify their similarities. Although shape feature calculation can detect duplication simply and quickly, this method has poor adaptability to complex geometric bodies and has large calculation errors.

[0004] In addition, traditional 3D duplicate checking technologies also face a common problem - low computational efficiency. When processing large-scale 3D models or scene data, these methods often require a lot of computing resources, especially in situations where real-time requirements are high, such as virtual reality or industrial production processes, where a large number of models need to be checked for duplicates in a short period of time. The efficiency and accuracy of traditional methods are difficult to meet the requirements.

[0005] At the same time, existing methods mostly rely on global matching for comparison, which may cause them to miss some repeated or similar structures when facing models with slight differences. This limitation makes traditional methods less applicable in large-scale data processing, especially when the amount of data increases sharply, the efficiency of duplicate checking will drop sharply, and may even cause the operation of the entire system to be unstable. Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention provides an efficient 3D geometry duplication checking method, which solves the problems mentioned in the background technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: an efficient 3D geometry duplicate checking method, comprising the following steps:

[0008] S1. Obtain a simplified mesh model by performing cloud point generation and simplification processing on the input 3D model, and extract geometric features of the 3D model, including fixed-point normal vectors, surface normal vectors, and curvature, and then convert the geometric features into feature vector geometry using a data processing method, and mark them as feature vector geometry;

[0009] S2, performing local region matching, by segmenting the simplified grid model to obtain multiple local regions, then extracting features of the local regions to obtain a local feature set, and calculating similarity scores between adjacent local regions;

[0010] S3, performing a global geometric similarity matching phase, obtaining a local similarity score by fitting the similarity score of each region, and then fitting it with the feature vector geometry to obtain a global similarity score;

[0011] S4. Compare the preset similarity evaluation threshold with the global similarity score to obtain a model similarity evaluation result, and mark and classify the model according to the model similarity evaluation result to obtain a candidate duplicate model list;

[0012] S5. Optimize and iteratively evaluate the candidate duplicate model list, adjust the similarity calculation model, obtain the average similarity fluctuation score after adjustment, optimization and iteration, and compare it with the preset duplication detection threshold to obtain the input 3D model duplication detection result.

[0013] Preferably, said S1 includes S11 and S12;

[0014] S11, performing cloud point generation processing and simplification processing on the input 3D model, wherein the cloud point generation processing is performed by setting up at least three coordinate axes at the center point of the 3D model, and forming a plurality of satellite circles around each coordinate axis along a direction perpendicular to the coordinate axis and passing through the center point, and then photographing the geometric center point of the 3D model to generate a plurality of positions of the 3D model and a plurality of multi-angle 2D depth images, and the simplification processing is performed by reducing the number of faces obtained by photographing the input 3D model, while retaining the geometric features of the 3D model, including using edge folding and vertex aggregation methods for simplification processing to obtain a simplified mesh model;

[0015] Among them, edge folding is performed by selecting the edges with the farthest distance between the two ends of the 3D model for folding, and the new vertex positions after folding are obtained through weighted calculation, so as to achieve the purpose of simultaneously retaining the shape and curvature-related geometric characteristics of the 3D model.

[0016] Preferably, S12, extracting and processing geometric features of the simplified mesh model to reflect the shape, surface structure and local geometric properties of the input 3D model, and performing a quantitative extraction process on the complex structure of the simplified mesh model by extracting and converting the geometric features, including fixed-point normal vectors, surface normal vectors and curvature, and then converting the geometric features into feature vector geometry using a data processing method, including the normal vector of each vertex, the normal vector of each face and the curvature of each vertex, marked as feature vector geometry;

[0017] The quantitative extraction process includes S121, S122 and S123.

[0018] Preferably, S121, the normal vector is specifically obtained by weighted averaging the surface normal vectors adjacent to the vertex to provide a feedback of a local surface direction and reflect the local geometric shape of the position where the vertex is located;

[0019] S122, the surface normal vector is used to describe the direction of the plane where each face is located. For each triangular face, the surface normal vector can reflect the relative direction of the triangular face to other faces in space, and is specifically obtained by calculating the vertex coordinates of the triangular face;

[0020] S123. The curvature value is an indicator used to describe the curvature of the surface, specifically used to reflect the details and shape of the 3D model. It also reflects the curvature of the surface around a point, including the principal curvature and Gaussian curvature. The curvature value is obtained by estimating the local curvature of the mechanical energy based on the distance metric of the vertex neighborhood.

[0021] Preferably, S2 includes S21 and S22;

[0022] S21. Performing a local area matching stage, dividing the simplified network model into multiple independent local areas by geometrically dividing the simplified mesh model. Specifically, the division is performed according to the topological structure and geometric distribution of the simplified network model, so that the boundaries of the multiple independent local areas are continuous and non-overlapping. Multiple local areas are obtained, and the obtained multiple local areas are integrated to generate multiple local area sets, and each local area is composed of a fixed number of triangular faces.

[0023] Preferably, S22, extract geometric feature information for each local area in the multiple local area sets, including the local normal vectors of the vertices in the local area, the local curvature values ​​of the surface normal vectors in the local area, and the curvature values ​​of the vertices in the local area, to form a local feature set, and calculate the similarity scores between adjacent local areas.

[0024] Preferably, said S3 includes S31;

[0025] S31, perform global geometric similarity matching stage, specifically match the divided local feature set, obtain local similarity score by fitting the similarity score of each area in the local area set, reflect the geometric feature similarity between local areas, and then fit the feature vector geometry based on the local similarity score to obtain the global similarity score between models, reflect the global geometric structure similarity of the models, which is used to evaluate the structural similarity between different models.

[0026] Preferably, said S4 includes S41 and S42;

[0027] S41, comparing the preset similarity evaluation threshold with the global similarity score to obtain a model similarity evaluation result, reflecting the overall geometric similarity between the models;

[0028] When the global similarity score is ≥ the similarity evaluation threshold, it means that the models are similar in geometric structure, and the model similarity evaluation result is a repeated result;

[0029] When the global similarity score is less than the similarity evaluation threshold, it means that the models are not similar in geometric structure, and the model similarity evaluation result is a non-repetitive result;

[0030] S42. Mark and classify the models according to the model similarity evaluation results. When the model similarity evaluation results are duplicate results, mark the association relationship between the models as a duplicate relationship, and simultaneously classify the models into a candidate duplicate model list. After marking and classifying all input 3D models, a candidate duplicate model list is formed.

[0031] Preferably, said S5 includes S51 and S52;

[0032] S51. Optimize and iteratively evaluate the candidate repeated model list, and obtain the average similarity fluctuation score after adjustment, optimization and iteration by adjusting the similarity calculation model, wherein the adjustment optimization includes optimizing the proportional adjustment similarity score, the global similarity score and the simplified grid model, and iteratively calculate the proportional adjustment similarity score and the global similarity score at the same time, determine the iteration volatility, and obtain the average similarity fluctuation score.

[0033] Preferably, S52, comparing the similarity fluctuation average score with a preset duplication detection threshold to obtain a duplication detection result of the input 3D model;

[0034] When the average similarity fluctuation score is ≥ the duplicate checking threshold, the models are determined to be duplicate models and are re-substituted into the candidate duplicate model list;

[0035] When the similarity fluctuation average score is less than the duplicate determination threshold, the models are determined to be non-duplicate models and are removed from the candidate duplicate model list.

[0036] The present invention provides an efficient 3D geometry duplicate checking method, which has the following beneficial effects:

[0037] (1) The simplified mesh model is obtained by performing cloud point generation and simplification processing on the input 3D model, and the key geometric features are extracted and converted into feature vector geometry, which lays the foundation for the subsequent similarity calculation. This method extracts local area features through two stages: local area matching and global geometric similarity matching, and calculates the local similarity score and the global similarity score through the similarity score obtained by fitting, which effectively improves the accuracy and efficiency of model duplication checking. Compared with traditional methods, this scheme can ensure the accurate matching of local area and global geometric structure on the basis of multiple feature extraction, avoiding information loss or misjudgment. By optimizing iterative evaluation and duplication judgment, the accuracy and adaptability of duplication checking are further ensured, and it can flexibly cope with the complexity and detail differences of different models. Finally, by optimizing the candidate duplicate model list and outputting the final duplication checking result, the accuracy and calculation efficiency of large-scale 3D model duplication checking are effectively improved, meeting the needs of fast and accurate duplication checking.

[0038] (2) By geometrically segmenting the simplified mesh model M, the model is divided into multiple local region sets, and each local region is ensured to be composed of a fixed number of triangular faces, thereby providing a more uniform and controllable local region structure. The geometric features of the local regions, such as vertex normals, face normals, and curvature values, are fully extracted at this stage to form a local feature set, which provides high-dimensional geometric information for subsequent similarity calculations. By calculating the similarity scores between each local region and introducing appropriate weight coefficients to balance the influence of different features, it is ensured that the geometric similarity of each local region can be comprehensively evaluated.

[0039] (3) By fitting the similarity scores of local areas, the local similarity scores are calculated, and further combined with the feature vector geometry, the global similarity scores between models are obtained, thereby providing a quantitative indicator that comprehensively reflects the similarity of the geometric structures of the models. On the basis of the global similarity score, the models with similar geometric structures and dissimilar models are successfully distinguished by comparison with the preset similarity evaluation threshold, ensuring accurate similarity evaluation results. Through label classification, geometrically similar models are classified into a list of candidate duplicate models, further improving the efficiency of 3D model duplication detection. Compared with traditional methods, the present invention can perform similarity analysis at a more fine-grained local area level, and through a global evaluation mechanism, it ensures the high accuracy and efficiency of the duplication detection results, greatly reducing the misjudgment rate, and is particularly suitable for duplication detection and management of large-scale 3D model libraries.

[0040] (4) The similarity score and the global similarity score are adjusted and optimized multiple times by iterative optimization, and the accuracy of the similarity calculation is further improved by adjusting parameters such as weight coefficients and simplification coefficients. In this process, the similarity score and the global similarity score are adjusted proportionally, and then combined with iterative calculation and volatility evaluation, the similarity fluctuation average score is obtained, so that the final duplicate checking result has stronger robustness and credibility. Compared with traditional methods, the present invention can reduce errors in the judgment of small differences between models through multiple iterative optimizations, improve the accuracy of duplicate checking results, especially when dealing with large model libraries with complex geometric structures, showing excellent stability and efficiency, ensuring that the repeatability recognition of 3D models is more accurate, and avoiding misjudgment caused by simple calculations. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a schematic diagram of the steps of an efficient 3D geometry duplicate checking method of the present invention;

[0042] Figure 2 A schematic diagram of the logic flow of an efficient 3D geometry duplicate checking method of the present invention;

[0043] Figure 3 Schematic diagram of the quantitative extraction process steps. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0045] Example 1

[0046] The present invention provides an efficient 3D geometry duplicate checking method, please refer to Figure 1 , including the following steps:

[0047] S1. Obtain a simplified mesh model M by performing cloud point generation and simplification processing on the input 3D model, and extract geometric features of the 3D model, including fixed-point normal vectors, surface normal vectors, and curvature. Then, the geometric features are converted into feature vector geometry using a data processing method, and marked as feature vector geometry F.

[0048] S2, in the local region matching stage, the simplified grid model M is segmented to obtain multiple local regions, the features of the local regions are extracted to obtain a local feature set FA, and the similarity score S(A, B) between adjacent local regions is calculated;

[0049] S3, perform global geometric similarity matching, obtain the local similarity score Savg(A, B) by fitting the similarity score S(A, B) of each region, and then fit it with the feature vector geometry F to obtain the global similarity score Sglobal(D, F);

[0050] S4. Compare the preset similarity evaluation threshold Sthe with the global similarity score Sgloba l(D, F) to obtain the model similarity evaluation result, and mark and classify the model according to the model similarity evaluation result to obtain the candidate duplicate model list Cdup;

[0051] S5. Optimize and iteratively evaluate the candidate duplicate model list Cdup, and obtain the average similarity fluctuation score Sfina l(D, F) after adjustment, optimization and iteration by adjusting the similarity calculation model. Compare it with the preset duplication judgment threshold Cthe to obtain the input 3D model duplication result.

[0052] In the present embodiment, by performing cloud point generation and simplification processing on the input 3D model, a simplified mesh model M is obtained, and key geometric features are extracted, which are converted into feature vector geometry F, which lays the foundation for subsequent similarity calculation. The method extracts local area features through two stages of local area matching and global geometric similarity matching, and calculates local similarity score Savg (A, B) and global similarity score Sgloba l (D, F) by the similarity score S (A, B) obtained by fitting, effectively improving the precision and efficiency of model duplicate checking. Compared with traditional methods, this scheme can ensure the accurate matching of local area and global geometric structure on the basis of multiple feature extraction, avoiding information loss or misjudgment. By optimizing iterative evaluation and duplicate checking judgment, the accuracy and adaptability of duplicate checking are further ensured, and the complexity and detail differences of different models can be flexibly dealt with. Finally, by optimizing the candidate repeated model list Cdup, the final duplicate checking result is output, effectively improving the precision and calculation efficiency of large-scale 3D model duplicate checking, meeting the needs of fast and accurate duplicate checking.

[0053] Example 2

[0054] This embodiment is explained in Example 1, please refer to Figure 1 , specifically: the S1 includes S11 and S12;

[0055] S11, performing cloud point generation processing and simplification processing on the input 3D model, wherein the cloud point generation processing is performed by setting up at least three coordinate axes at the center point of the 3D model, and forming a plurality of satellite circles around each coordinate axis along a direction perpendicular to the coordinate axis and passing through the center point, and then photographing the geometric center point of the 3D model to generate a plurality of positions of the 3D model and a plurality of multi-angle 2D depth images, and the simplification processing is performed by reducing the number of faces obtained by photographing the input 3D model, while retaining the geometric features of the 3D model, including using edge folding and vertex aggregation methods for simplification processing to obtain a simplified mesh model;

[0056] The cloud point generation includes generating a point cloud map from a 3D model, by setting up at least three coordinate axes at the center point of the 3D model, and forming a number of satellite circles around each coordinate axis along a direction perpendicular to the coordinate axis and passing through the center point, wherein the radius is the radius of the circumscribed sphere of the 3D model plus a fixed value and multiplied by a coefficient greater than 1, and evenly placing cameras at a certain angle to shoot directly at the center point to generate a number of positions and a number of multi-angle 2D depth images of the 3D model, wherein the 2D depth image is used for subsequent processing of geometric features, extraction and conversion of geometric features, and a quantitative extraction process of the complex structure of the simplified mesh model M, including fixed-point normal vectors, surface normal vectors and curvature, and then using a data processing method to convert the geometric features into feature vector geometry F, and at the same time generating a feature vector geometry F for each input 3D model, and storing it in a file as a vector database; when searching, extracting the feature vector geometry F of the 3D model to be inspected, and then measuring the difference in the feature vector geometry F between the models, and sorting the extracted feature vector geometry F;

[0057] Among them, edge folding is performed by selecting the cloud points at both ends of the 3D model that are farthest from the edge, and the new vertex positions after folding are obtained through weighted calculation, so as to achieve the purpose of retaining the shape and curvature-related geometric characteristics of the 3D model at the same time;

[0058] Among them, the weighted calculation takes into account the vertex coordinates at both ends of the edge and their geometric relationship in the mesh. The purpose of weighted calculation is to ensure that the newly generated vertex position does not introduce too large deformation, so as to maintain the original geometric shape as much as possible;

[0059] The simplified grid model M is obtained by the following calculation formula:

[0060]

[0061] In the formula, Mo represents the input 3D model, α represents the simplification coefficient, which is specifically in the range of 0<α<1, and specifically represents the ratio of the number of simplified faces to the input 3D model Mo.

[0062] By simplifying the input 3D model, especially using edge folding and vertex aggregation methods, the complexity of the model can be significantly reduced while effectively retaining the geometric features of the model. The edge folding technology folds the farthest edge and weightedly calculates the position of the new vertex to ensure that the newly generated mesh structure maintains the geometric form while avoiding excessive deformation. This process not only reduces the amount of calculation and improves processing efficiency, but also ensures that the simplified model can still accurately reflect the shape and curvature characteristics of the original model.

[0063] The generation of the simplified mesh model M reduces the number of faces by controlling the simplification coefficient, so that the complexity of the model can be effectively controlled. The setting of the simplification coefficient is between 0 and 1, and the degree of simplification can be adjusted according to needs, thereby ensuring that key information is not lost during the simplification process. This processing method is particularly suitable for model preprocessing of large-scale data sets. It can greatly reduce the consumption of storage and computing resources without damaging the geometric structure of the model, thereby improving the efficiency of subsequent duplicate checking and analysis. Overall, through this simplified processing, not only the computing efficiency is improved, but also the fidelity of shape and curvature in the geometric processing process is enhanced, thereby achieving effective simplification and optimization of the model, providing a more efficient and accurate basis for subsequent geometric matching and similarity evaluation.

[0064] S12, extract and process geometric features of the simplified mesh model M to reflect the shape, surface structure and local geometric properties of the input 3D model. By extracting and converting the geometric features, the complex structure of the simplified mesh model M is quantitatively extracted, including fixed-point normal vectors, surface normal vectors and curvature. The geometric features are then converted into feature vector geometry using a data processing method, including the normal vector of each vertex, the normal vector of each face and the curvature of each vertex, marked as feature vector geometry F = {Nv(i), Nf(j), Kv(i)}, where Nv(i) represents the normal vector of the i-th vertex, Nf(j) represents the j-th face normal vector, and Kv(i) represents the curvature value of the i-th vertex;

[0065] The quantitative extraction process includes S121, S122 and S123.

[0066] S121, the normal vector Nv(i) is specifically obtained by weighted averaging the surface normal vectors adjacent to the vertex to obtain the normal vector Nv(i) of the i-th vertex, providing feedback of a local surface direction and reflecting the local geometric shape of the position of the vertex;

[0067] The fixed-point normal vector Nv(i) is obtained by the following calculation formula:

[0068]

[0069] Where k(i) is the number of faces adjacent to the i-th vertex, Nface(j) is the normal vector of the j-th face, and w(j) is the weight of the adjacent face, which is set according to the area and angle difference of the face.

[0070] S122, the surface normal vector Nf(j) is used to describe the direction of the plane where each face is located. For each triangular face, the surface normal vector can reflect the relative direction of the triangular face to other faces in space, and is specifically obtained by calculating the vertex coordinates of the triangular face;

[0071] The surface normal vector Nf(j) is obtained by the following calculation formula:

[0072]

[0073] Wherein, P1, P2 and P3 represent vertices, specifically the coordinates of vertex P1 (x1, y1, z1), vertex P2 (x2, y2, z2) and vertex P3 (x3, y3, z3);

[0074] S123, the curvature value Kv(i) is used to describe the index of the curvature of the surface, specifically used to reflect the details and shape of the 3D model, and also reflects the curvature of the surface around a point, including the principal curvature and Gaussian curvature. Specifically, the curvature value Kv(i) is obtained by estimating the local curvature of the mechanical energy based on the distance metric of the vertex neighborhood;

[0075] Among them, the principal curvature reflects the overall curvature of the surface at a certain point;

[0076] Gaussian curvature reflects the average degree of curvature at a point;

[0077] The curvature value Kv(i) is obtained by the following calculation formula:

[0078]

[0079] Where Pi represents the coordinates of the i-th vertex, Pj represents the coordinates of the vertices adjacent to the vertex coordinate Pi, and k(i) is the number of faces adjacent to the i-th vertex.

[0080] In the present embodiment, the complexity of the 3D model is simplified by edge folding and vertex aggregation methods, while retaining the geometric shape and curvature features of the model, ensuring that the simplified mesh model M can still effectively reflect the geometric information of the input model. In particular, in the simplification process, by weighted calculation of the new vertex position, excessive deformation is avoided, so that the geometric features of the model are retained as accurately as possible. Subsequently, by extracting geometric features from the simplified mesh model M, including fixed-point normal vectors Nv(i), face normal vectors Nf(j) and curvature values ​​Kv(i), feature vector geometry F is generated, providing high-precision geometric feature data for subsequent similarity calculations. These features can not only accurately reflect the local geometric characteristics of each vertex and face, but also effectively describe the overall shape and surface structure of the model. By this method, while ensuring that the simplified model structure is as complete as possible, the processing of redundant data is reduced, the calculation efficiency of the subsequent duplicate checking stage is greatly improved, and the misjudgment and calculation delay caused by geometric morphological changes are avoided.

[0081] Example 3

[0082] This embodiment is explained in Example 2. Please refer to Figure 1 , specifically: S2 includes S21 and S22;

[0083] S21. Performing a local area matching stage, dividing the simplified network model M into multiple independent local areas by geometrically dividing the simplified mesh model M. Specifically, the division is performed according to the topological structure and geometric distribution of the simplified network model M, so that the boundaries of the multiple independent local areas are continuous and non-overlapping, and multiple local areas are obtained. The multiple local areas are integrated to generate multiple local area sets {A1, A2, ..., AG}, and each local area is composed of a fixed number of triangular faces, and G represents the total number of local areas.

[0084] S22, extracting geometric feature information from each local area in the multiple local area sets, including the local normal vector JNv(i) of the i-th vertex in the local area, the local curvature value JKv(i) of the j-th surface normal vector in the local area, and the curvature value JNf(j) of the i-th vertex in the local area, to form a local feature set FA={JN(i), JKv(i), JNf(j)}, and calculating the similarity score S(A, B) between adjacent local areas;

[0085] The similarity score S(A, B) is obtained by the following calculation formula:

[0086]

[0087] Where S(A, B) represents the similarity score between local area A and local area B, JNv(i, A) and JNv(i, B) represent the normal vector of the i-th vertex in local area A and local area B, respectively, and the similarity of vertex normal vectors is measured by calculating the dot product, JNf(j, A) and JNf(j, B) represent the normal vector of the j-th face in local area A and local area B, respectively, and the similarity of face normals is measured by calculating the dot product, JKv(q, A) and JKv(q, B) represent The curvature value of the qth vertex in the local area A and the local area B is used to measure the similarity of the curvature by calculating the difference. s1, s2 and s3 represent weight coefficients. Specifically, s1 represents the weight coefficient of the vertex normal vector, s2 represents the weight coefficient of the surface normal vector, s3 represents the weight coefficient of the vertex curvature value, n, m and p represent the number of vertex normal vectors, the number of surface normal vectors and the number of vertex curvature values ​​in the local area A and the local area B, respectively. Specifically, the number of features in the local area is unified by using interpolation and normalization methods.

[0088] In this embodiment, by geometrically segmenting the simplified mesh model M, the model is divided into a plurality of local area sets {A1, A2, ..., AG}, and each local area is ensured to be composed of a fixed number of triangular faces, thereby providing a more uniform and controllable local area structure. The geometric features of the local area, such as vertex normal vectors, face normal vectors and curvature values, are fully extracted at this stage to form a local feature set FA = {JNv(i), Nf(j), Kv(i)}, which provides high-dimensional geometric information for subsequent similarity calculations. By calculating the similarity score S(A, B) between each local area and introducing appropriate weight coefficients to balance the influence of different features, it is ensured that the geometric similarity of each local area can be comprehensively evaluated. In particular, the number of features in each local area is unified through interpolation and normalization methods, avoiding the calculation deviation caused by inconsistent local area features, thereby making the entire similarity score S(A, B) more stable and reliable. It effectively avoids the computational bottleneck caused by the high complexity of the model, and improves the accuracy of similarity matching by optimizing the feature extraction and scoring mechanism, further improving the efficiency and accuracy of the 3D model duplication checking process.

[0089] Example 4

[0090] This embodiment is explained in Example 3, please refer to Figure 1 , specifically: the S3 includes S31;

[0091] S31, perform global geometric similarity matching, specifically match the divided local feature set, obtain the local similarity score Savg(A, B) by fitting the similarity score S(A, B) of each area in the local area set, reflect the geometric feature similarity between the local areas, and then perform fitting processing with the feature vector geometry F based on the local similarity score Savg(A, B) to obtain the global similarity score Sglobal(D, F) between the model D and the model F, reflect the global geometric structure similarity of the model, and evaluate the structural similarity between different models;

[0092] The global similarity score Sglobal(D, F) is obtained by the following calculation formula:

[0093]

[0094] Where Savg(D, F) represents the global similarity score between model D and model F, G and H represent the total number of local regions in model D and model F respectively, g represents the g-th local region, and w(A, B) represents the weight coefficient of local region A and local region B in model D and model F.

[0095] Through the global geometric similarity matching stage, the similarity score of the local area is combined with the fitting processing of the feature vector geometry, and the overall geometric structure similarity between different models can be effectively evaluated. First, the similarity score S(A, B) of the local area is used to calculate the local similarity score Savg(A, B), so as to ensure that the geometric features of each local area are fully considered. Then, by further fitting the local similarity score with the feature vector geometry F, the global similarity score Sgloba l(D, F) between model D and model F is finally obtained. This score reflects the global geometric structure similarity between the models.

[0096] The core advantage of this process is that by carefully fitting and weighting the similarities of local regions, the differences and similarities of the model at various geometric detail levels can be captured more accurately. In addition, the calculation of the global similarity score Sgloba l(D,F) not only relies on the simple comparison of local regions, but also considers the contribution of each local region to the geometric features of the entire model, ensuring that the evaluation results can fully reflect the overall geometric structure characteristics of the model. This method provides an efficient and accurate tool for geometric matching and similarity evaluation of complex models, and is particularly suitable for large-scale model duplication detection and automated analysis tasks.

[0097] The S4 includes S41 and S42;

[0098] S41, compare the preset similarity evaluation threshold Sthe with the global similarity score Sglobal(D, F) to obtain a model similarity evaluation result, reflecting the overall geometric similarity of model D and model F;

[0099] When the global similarity score Sgloba l(D, F) ≥ the similarity evaluation threshold Sthe, it means that the model D and the model F are similar in geometric structure, and the model similarity evaluation result is a repeated result;

[0100] When the global similarity score Sglobal(D, F) is less than the similarity evaluation threshold Sthe, it means that the model D and the model F are not similar in geometric structure, and the model similarity evaluation result is a non-repetitive result;

[0101] S42. Mark and classify the models according to the model similarity evaluation results. When the model similarity evaluation result is a duplicate result, mark the association relationship between model D and model F as a duplicate relationship, including model D∈model F, and simultaneously classify model D and model F into∈candidate duplicate model list Cdup. After marking and classifying all input 3D models, a candidate duplicate model list Cdup is formed.

[0102] In this embodiment, by fitting the similarity score S(A, B) of the local area, the local similarity score Savg(A, B) is calculated, and further combined with the feature vector geometry F, the global similarity score Sglobal(D, F) between model D and model F is obtained, thereby providing a quantitative index that comprehensively reflects the similarity of the geometric structure of the model. On the basis of the global similarity score Sglobal(D, F), it is compared with the preset similarity evaluation threshold Sthe, and the models with similar geometric structures and dissimilar models are successfully distinguished, ensuring accurate similarity evaluation results. Through label classification, the geometrically similar models D and model F are classified into the candidate duplicate model list Cdup, further improving the efficiency of 3D model duplicate checking. Compared with traditional methods, the present invention can perform similarity analysis at a more fine-grained local area level, and through a global evaluation mechanism, it ensures the high accuracy and efficiency of the duplicate checking results, greatly reduces the misjudgment rate, and is particularly suitable for the duplicate checking and management of large-scale 3D model libraries.

[0103] Example 5

[0104] This embodiment is explained in Example 4. Please refer to Figure 1 , specifically: the S5 includes S51 and S52;

[0105] S51. Optimize and iteratively evaluate the candidate repeated model list Cdup, and obtain the adjusted, optimized and iterated similarity fluctuation average score Sfina l(D, F) by adjusting the similarity calculation model, wherein the adjustment and optimization includes proportionally adjusting the weight coefficient in the similarity score S(A, B) calculation formula, the weight coefficient in the global similarity score Sgloba l(D, F) calculation formula and the simplification coefficient α in the simplified grid model M for optimization, wherein the proportional adjustment includes enlarging and reducing the weight coefficient and the simplification coefficient α by 5% twice respectively, and iteratively calculating the global similarity score Sgloba l(D, F) four times on the similarity score S(A, B) and the global similarity score Sgloba l(D, F) after proportional adjustment, and performing statistics to obtain the average value AVG, judging the iteration volatility, including marking the difference between the average value AVG and the global similarity score Sgloba l(D, F) obtained by the initial calculation as the iteration volatility, and obtaining the similarity fluctuation average score Sfina l(D, F).

[0106] S52, comparing the similarity fluctuation average score Sfina l(D, F) with the preset duplication determination threshold Cthe to obtain the input 3D model duplication check result;

[0107] When the similarity fluctuation average score Sfina l(D, F) ≥ the duplicate determination threshold Cthe, model D and model F are determined to be duplicate models and are added back to the candidate duplicate model list Cdup;

[0108] When the similarity fluctuation average score Sfina l(D, F) is less than the duplicate determination threshold Cthe, model D and model F are determined to be non-duplicate models and are removed from the candidate duplicate model list Cdup.

[0109] In the present embodiment, similarity score S (A, B) and global similarity score Sgloba l (D, F) are repeatedly adjusted and optimized by iterative optimization mode, and the parameters such as weight coefficient and simplification coefficient are adjusted to further improve the precision of similarity calculation. In this process, similarity score and global similarity score are adjusted by proportion, and then combined with iterative calculation and volatility assessment, similarity fluctuation average score Sfina l (D, F) is obtained, so that the final duplicate checking result has stronger robustness and credibility. Compared with traditional methods, the present invention can reduce errors in the judgment of small differences between models by multiple iterations, improves the precision of duplicate checking results, especially when processing large-scale model libraries with complex geometric structures, shows excellent stability and high efficiency, ensures that the repeatability identification of 3D models is more accurate, and avoids the misjudgment caused by simple calculations.

[0110] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An efficient 3D geometry duplicate checking method, characterized by: The following steps are involved: S1. Obtain a simplified mesh model by performing cloud point generation and simplification processing on the input 3D model, and extract geometric features of the 3D model, including fixed-point normal vectors, surface normal vectors, and curvature, and then convert the geometric features into feature vector geometry using a data processing method, and mark them as feature vector geometry; S2, performing local region matching, by segmenting the simplified grid model to obtain multiple local regions, then extracting features of the local regions to obtain a local feature set, and calculating similarity scores between adjacent local regions; S3, performing a global geometric similarity matching phase, obtaining a local similarity score by fitting the similarity score of each region, and then fitting it with the feature vector geometry to obtain a global similarity score; S4. Compare the preset similarity evaluation threshold with the global similarity score to obtain a model similarity evaluation result, and mark and classify the model according to the model similarity evaluation result to obtain a candidate duplicate model list; S5. Optimize and iteratively evaluate the candidate duplicate model list, adjust the similarity calculation model, obtain the average similarity fluctuation score after adjustment, optimization and iteration, and compare it with the preset duplication detection threshold to obtain the input 3D model duplication detection result.

2. An efficient 3D geometry duplicate checking method according to claim 1, characterized in that: Said S1 includes S11 and S12; S11, performing cloud point generation processing and simplification processing on the input 3D model, wherein the cloud point generation processing is performed by setting up at least three coordinate axes at the center point of the 3D model, and forming a plurality of satellite circles around each coordinate axis along a direction perpendicular to the coordinate axis and passing through the center point, and then photographing the geometric center point of the 3D model to generate a plurality of positions of the 3D model and a plurality of multi-angle 2D depth images, and the simplification processing is performed by reducing the number of faces obtained by photographing the input 3D model, while retaining the geometric features of the 3D model, including using edge folding and vertex aggregation methods for simplification processing to obtain a simplified mesh model; Among them, edge folding is performed by selecting the edges with the farthest distance between the two ends of the 3D model for folding, and the new vertex positions after folding are obtained through weighted calculation, so as to achieve the purpose of simultaneously retaining the shape and curvature-related geometric characteristics of the 3D model.

3. An efficient 3D geometry duplicate checking method according to claim 2, characterized in that: S12, extracting and processing geometric features of the simplified mesh model to reflect the shape, surface structure and local geometric properties of the input 3D model, and performing a quantitative extraction process on the complex structure of the simplified mesh model by extracting and converting the geometric features, including fixed-point normal vectors, surface normal vectors and curvature, and then converting the geometric features into feature vector geometry using a data processing method, including the normal vector of each vertex, the normal vector of each face and the curvature of each vertex, marked as feature vector geometry; The quantitative extraction process includes S121, S122 and S123.

4. An efficient 3D geometry duplicate checking method according to claim 3, characterized in that: S121, the normal vector is specifically obtained by weighted averaging the surface normal vectors adjacent to the vertex to provide feedback on a local surface direction and reflect the local geometric shape of the location of the vertex; S122, the surface normal vector is used to describe the direction of the plane where each face is located. For each triangular face, the surface normal vector can reflect the relative direction of the triangular face to other faces in space, and is specifically obtained by calculating the vertex coordinates of the triangular face; S123. The curvature value is an indicator used to describe the curvature of the surface, specifically used to reflect the details and shape of the 3D model. It also reflects the curvature of the surface around a point, including the principal curvature and Gaussian curvature. The curvature value is obtained by estimating the local curvature of the mechanical energy based on the distance metric of the vertex neighborhood.

5. The efficient 3D geometry duplicate checking method according to claim 1, characterized in that: The S2 includes S21 and S22; S21. Performing a local area matching stage, dividing the simplified network model into multiple independent local areas by geometrically dividing the simplified mesh model. Specifically, the division is performed according to the topological structure and geometric distribution of the simplified network model, so that the boundaries of the multiple independent local areas are continuous and non-overlapping. Multiple local areas are obtained, and the obtained multiple local areas are integrated to generate multiple local area sets, and each local area is composed of a fixed number of triangular faces.

6. An efficient 3D geometry duplicate checking method according to claim 5, characterized in that: S22. Extract geometric feature information for each local area in the multiple local area sets, including local normal vectors of vertices in the local area, local curvature values ​​of surface normal vectors in the local area, and curvature values ​​of vertices in the local area, to form a local feature set, and calculate similarity scores between adjacent local areas.

7. An efficient 3D geometry duplicate checking method according to claim 1, characterized in that: The S3 includes S31; S31, perform global geometric similarity matching stage, specifically match the divided local feature set, obtain local similarity score by fitting the similarity score of each area in the local area set, reflect the geometric feature similarity between local areas, and then fit the feature vector geometry based on the local similarity score to obtain the global similarity score between models, reflect the global geometric structure similarity of the models, which is used to evaluate the structural similarity between different models.

8. An efficient 3D geometry duplicate checking method according to claim 1, characterized in that: The S4 includes S41 and S42; S41, comparing the preset similarity evaluation threshold with the global similarity score to obtain a model similarity evaluation result, reflecting the overall geometric similarity between the models; When the global similarity score is ≥ the similarity evaluation threshold, it means that the models are similar in geometric structure, and the model similarity evaluation result is a repeated result; When the global similarity score is less than the similarity evaluation threshold, it means that the models are not similar in geometric structure, and the model similarity evaluation result is a non-repetitive result; S42. Mark and classify the models according to the model similarity evaluation results. When the model similarity evaluation results are duplicate results, mark the association relationship between the models as a duplicate relationship, and simultaneously classify the models into a candidate duplicate model list. After marking and classifying all input 3D models, a candidate duplicate model list is formed.

9. An efficient 3D geometry duplicate checking method according to claim 1, characterized in that: The S5 includes S51 and S52; S51. Optimize and iteratively evaluate the candidate repeated model list, and obtain the average similarity fluctuation score after adjustment, optimization and iteration by adjusting the similarity calculation model, wherein the adjustment optimization includes optimizing the proportional adjustment similarity score, the global similarity score and the simplified grid model, and iteratively calculate the proportional adjustment similarity score and the global similarity score at the same time, determine the iteration volatility, and obtain the average similarity fluctuation score.

10. The efficient 3D geometry duplicate checking method according to claim 1, characterized in that: S52, comparing the similarity fluctuation average score with a preset duplication detection threshold to obtain a duplication detection result of the input 3D model; When the average similarity fluctuation score is ≥ the duplicate checking threshold, the models are determined to be duplicate models and are re-substituted into the candidate duplicate model list; When the similarity fluctuation average score is less than the duplicate determination threshold, the models are determined to be non-duplicate models and are removed from the candidate duplicate model list.

Images