Dental crown feature point extraction method, device, equipment and medium

Abstract

The invention relates to the technical field of dental crown data processing, and discloses a dental crown feature point extraction method, device and equipment and a medium, and the method comprises the steps: obtaining dental jaw point cloud data of a target patient, carrying out the point cloud grid reconstruction of the dental jaw point cloud data, and generating a three-dimensional dental crown digital model; performing dental region segmentation on the three-dimensional dental crown digital model to obtain an independent dental crown model of each tooth, and constructing a local coordinate system corresponding to the independent dental crown model; performing spatial registration on the independent dental crown model according to the local coordinate system to obtain a target dental crown model, and extracting an initial dental crown feature matrix according to the target dental crown model; performing adaptive sampling on the initial dental crown feature matrix to obtain dental crown sampling vertexes, and performing double-branch offset prediction according to the dental crown sampling vertexes to obtain an offset point cluster; and calculating a point cluster clustering center point of the offset point cluster, and determining a dental crown feature point according to the point cluster clustering center point. According to the method, the extraction efficiency and accuracy of the dental crown feature points can be improved.

Description

Technical Field

[0001] This invention relates to the field of dental crown data processing technology, and in particular to a method, apparatus, device and medium for extracting dental crown feature points. Background Technology

[0002] In the field of computer-aided dental treatment, constructing a high-precision three-dimensional digital model of the crown is the key to achieving personalized treatment and precise restoration. The accurate extraction and positioning of crown feature points, as the most representative anatomical landmarks on the tooth surface, directly affects the model construction, restoration design, and orthodontic treatment outcomes.

[0003] Currently, the commonly used methods for extracting crown feature points mainly rely on manual annotation. However, this method is not only limited by the differences in the understanding of crown structure among technicians, making it difficult to guarantee data consistency and accuracy, but also the manual annotation process is time-consuming and labor-intensive, which cannot meet the needs of large-scale clinical data processing. Furthermore, it is prone to introducing human error, affecting the accuracy and stability of the construction of three-dimensional crown digital models, thereby restricting the formulation and implementation of dental restoration and orthodontic treatment plans.

[0004] Therefore, improving the efficiency and accuracy of crown feature point extraction has become an urgent problem to be solved. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and medium for extracting crown feature points, the main purpose of which is to solve the problems of low extraction efficiency and inaccurate extraction of crown feature points.

[0006] Firstly, to achieve the above objectives, the present invention provides a method for extracting crown feature points, comprising: Acquire the dental and jaw point cloud data of the target patient, reconstruct the point cloud mesh from the dental and jaw point cloud data, and generate a three-dimensional digital model of the dental crown; The three-dimensional digital model of the tooth crown is segmented into the jaw region to obtain an independent tooth crown model for each tooth, and a local coordinate system corresponding to the independent tooth crown model is constructed. The independent crown model is spatially registered according to the local coordinate system to obtain the target crown model, and the initial crown feature matrix is ​​extracted according to the target crown model. Adaptive sampling is performed on the initial crown feature matrix to obtain multiple crown sampling vertices, and bi-branch offset prediction is performed based on the crown sampling vertices to obtain an offset point cluster; Calculate the cluster center point of the offset point cluster, and determine the crown feature points based on the cluster center point.

[0007] Secondly, the present invention also provides a crown feature point extraction device, comprising: The point cloud model generation module is used to acquire the dental and jaw point cloud data of the target patient, reconstruct the point cloud mesh of the dental and jaw point cloud data, and generate a three-dimensional digital model of the dental crown. The local coordinate system construction module is used to segment the jaw region of the three-dimensional crown digital model to obtain the independent crown model of each tooth, and to construct the local coordinate system corresponding to the independent crown model. The model space registration module is used to perform spatial registration of the independent crown model according to the local coordinate system to obtain the target crown model, and to extract the initial crown feature matrix according to the target crown model; The crown offset prediction module is used to adaptively sample the initial crown feature matrix to obtain multiple crown sampling vertices, and perform bi-branch offset prediction based on the crown sampling vertices to obtain an offset point cluster; The crown feature point determination module is used to calculate the cluster center point of the offset point cluster and determine the crown feature points based on the cluster center point.

[0008] Thirdly, the present invention also provides an electronic device, the electronic device comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the above-described method for extracting crown feature points.

[0009] Fourthly, the present invention also provides a computer-readable storage medium storing at least one computer program, which is executed by a processor in an electronic device to implement the above-described method for extracting crown feature points.

[0010] In this embodiment of the invention, a three-dimensional digital model of a dental crown is generated by reconstructing the grid of dental point cloud data. Data normalization and outlier removal eliminate equipment errors and noise interference, improve data consistency, reduce the consumption of computing resources by redundant data, and significantly improve the efficiency and accuracy of computer processing of dental data. Constructing a local coordinate system accurately describes the morphological characteristics of a single tooth, facilitating subsequent comparisons of different tooth models, positional adjustments in virtual orthodontics, and size matching in prosthesis design, thereby improving the accuracy and efficiency of computer-aided oral diagnosis and treatment. Based on the registered target dental crown model, rapid spatial alignment of the crown model is achieved, ensuring that the input data for subsequent neural network processing has standardized spatial distribution characteristics, effectively solving the problem of inconsistent feature space caused by differences in the acquisition angle of the original point cloud data. Using bi-branch offset prediction improves the accuracy and stability of offset prediction. Determining crown feature points using the cluster center point comprehensively reflects the distribution characteristics of crown feature points, enabling rapid and accurate location of key parts of the crown, providing important feature identifiers for subsequent crown analysis, modeling, and restoration operations, and improving processing efficiency and accuracy. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating a method for extracting crown feature points according to an embodiment of the present invention. Figure 2 A schematic diagram of a three-dimensional digital model of a dental crown provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the local coordinate system corresponding to an independent crown model provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of an offset point cluster provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the orientation fitting network structure provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of a crown feature point provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of FACC feature points provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of a point clustering center provided in an embodiment of the present invention; Figure 9This is a schematic diagram of a tooth crown feature point extraction device according to an embodiment of the present invention; The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0013] To enable those skilled in the art to better understand the technical solutions of this disclosure, and to fully understand and implement the process of how this disclosure applies technical means to solve technical problems and achieve corresponding technical effects, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. The embodiments of this disclosure and the various features within them can be combined with each other without conflict, and the resulting technical solutions are all within the protection scope of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort should fall within the protection scope of this disclosure.

[0014] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0015] This application provides a method for extracting dental crown feature points. The execution entity of this method includes, but is not limited to, at least one electronic device that can be configured to execute the device provided in this application, such as a server or a terminal. In other words, the method for extracting dental crown feature points can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0016] Reference Figure 1 The diagram shown is a flowchart illustrating a method for extracting crown feature points according to an embodiment of the present invention. In this embodiment, the method for extracting crown feature points includes: S1. Obtain the dental point cloud data of the target patient, reconstruct the point cloud mesh from the dental point cloud data, and generate a three-dimensional digital model of the dental crown.

[0017] In this embodiment of the invention, dental point cloud data refers to a massive set of discrete points on the surface of teeth, alveolar bone, and jawbone obtained through three-dimensional scanning technology. This set of discrete points contains three-dimensional coordinate information and can accurately reconstruct the three-dimensional morphology of dental tissues.

[0018] In detail, the acquisition of data commonly uses intraoral laser scanning technology (such as an intraoral scanner that uses laser or optical imaging to quickly collect data directly in the patient's mouth), cone-beam computed tomography (CBCT, which generates a three-dimensional point cloud containing the hard tissues of the teeth and jaws through X-ray tomography) or model scanning technology (first creating a plaster model, and then using a desktop 3D scanner to scan the model to obtain the point cloud). Finally, the original point cloud data is optimized through point cloud processing algorithms to provide a precise three-dimensional digital model basis for dental restoration, orthodontics, etc.

[0019] In this embodiment of the invention, the step of reconstructing the point cloud mesh from the dental point cloud data to generate a three-dimensional digital model of the dental crown includes: The dental point cloud data is normalized to obtain the first dental point cloud data, and outlier points are removed from the first dental point cloud data to obtain the target dental point cloud data. Perform normal vector analysis on the target tooth and jaw point cloud data to obtain the normal direction corresponding to the target tooth and jaw point cloud data; The target dental point cloud data is smoothed according to the normal direction to obtain smoothed dental point cloud data. Calculate the surface center point of the smoothed dental point cloud data, and generate a surface support domain based on the surface center point; Based on the surface support domain, the smooth dental point cloud data is reconstructed to obtain a three-dimensional crown surface; A three-dimensional model of the three-dimensional crown surface is obtained by performing a three-dimensional modeling.

[0020] In detail, a coordinate standardization method is adopted. By translation (moving the center of the point cloud to the origin of the coordinate system) and scaling (unifying the scale of the point cloud to a preset range, such as [-1,1]), the scale differences caused by different scanning devices or scanning conditions are eliminated to obtain the first dental point cloud data under a unified coordinate system. Outlier removal commonly uses statistical filtering (calculating the average distance between each point and its k nearest neighbors and removing points whose distance is far beyond the mean) or radius filtering (deleting isolated points with insufficient neighboring points within the radius) to filter out noise points and obtain the target dental point cloud data.

[0021] In this process, a local neighborhood, such as a k-nearest neighbor or a fixed radius neighborhood, is constructed for each point in the target dental point cloud data. The covariance matrix of the neighborhood points is calculated by principal component analysis (PCA), and the eigenvector corresponding to the smallest eigenvalue is the normal vector direction of that point. Then, the normal vector direction of the adjacent points is adjusted by global optimization (such as a graph-based propagation algorithm) to ensure the consistency of the normal vectors (same direction or opposite direction) and obtain the normal direction of each point.

[0022] Furthermore, based on the normal direction, the moving least squares (MLS) method is used to fit a quadratic surface along the normal direction in the local neighborhood of each point. The original point is projected onto the fitted surface, while preserving features such as tooth edges. This smooths noise while avoiding edge blurring, thus obtaining smooth dental point cloud data.

[0023] Specifically, the local surface center point of each point in the smooth dental point cloud data is calculated. The surface support domain is adaptively generated based on the local curvature and point density of the center point by weighted averaging of the coordinates of neighboring points. For example, the support domain is smaller in areas with large curvature to preserve details, and larger in areas with gentle curvature to ensure smoothness. It is usually a spherical or polygonal region around the center point, which is used for subsequent local surface reconstruction.

[0024] In detail, based on the curved support domain, the Poisson surface reconstruction algorithm is used to transform the point and normal information within the support domain into the gradient field of the indicator function. An implicit surface is constructed by solving the Poisson equation, and then isosurfaces are extracted to obtain a continuous three-dimensional crown surface. Mesh optimization is performed on the three-dimensional crown surface (e.g., simplifying redundant triangles and repairing cavities). The Catmull-Clark subdivision algorithm is used to improve mesh accuracy. Through topology adjustment (ensuring model closure and no flipped surfaces) and feature enhancement (highlighting anatomical structures such as cusps and fissures), a three-dimensional crown digital model (usually in STL or PLY format) is finally generated that can be used for clinical analysis or restoration fabrication. Specifically, the three-dimensional crown digital model is as follows: Figure 2 As shown.

[0025] In this embodiment of the invention, a three-dimensional digital model of a dental crown is generated by reconstructing the grid of dental point cloud data. With the help of data normalization and outlier removal, equipment errors and noise interference can be eliminated, data consistency can be improved, the consumption of computing resources by redundant data can be reduced, and the efficiency and accuracy of computer processing of dental data can be greatly improved.

[0026] S2. The three-dimensional digital crown model is segmented into dentition regions to obtain independent crown models for each tooth, and a local coordinate system corresponding to the independent crown model is constructed.

[0027] In this embodiment of the invention, the three-dimensional crown digital model is segmented to obtain an independent three-dimensional digital model of each tooth, thereby accurately extracting the geometric information of each tooth. Preferably, a segmentation algorithm based on surface features and topology is used to ensure the continuity and accuracy of the segmentation boundary. For each independent crown model after segmentation, a local coordinate system is established, with its origin located at the centroid of the crown model, to improve the accuracy of subsequent data processing.

[0028] In this embodiment of the invention, the step of segmenting the three-dimensional digital crown model into dentition regions to obtain independent crown models for each tooth includes: The three-dimensional crown digital model is smoothed and filtered to obtain the target crown digital model; Global features of the target crown digital model are extracted, and a region growing algorithm is used to initially divide the target crown digital model into regions based on the global features to obtain multiple initial tooth regions; Obtain the initial region boundaries corresponding to multiple initial tooth regions, and extract the local features of multiple initial tooth regions; Based on the local features, the boundary of the initial region is optimized to obtain the target tooth region; Morphological fitting is performed on the target tooth region to obtain an initial independent crown model; The initial independent crown model is post-processed to obtain an independent crown model.

[0029] In detail, filtering algorithms are used to denoise the three-dimensional crown digital model to improve the model quality. Gaussian filtering smooths high-frequency noise by weighting each vertex and its neighboring vertices on the model surface (the weight decreases as the distance increases), thereby eliminating burrs and uneven noise introduced by scanning or reconstruction on the surface of the three-dimensional model. At the same time, it avoids feature loss caused by over-smoothing, providing a more stable geometric basis for subsequent segmentation.

[0030] Specifically, the global features include the curvature distribution of the model (such as average curvature, Gaussian curvature), topological structure (such as connected components), size ratio, etc. The feature distribution is obtained by calculating the curvature value of each vertex and clustering (such as K-means). The region growing algorithm uses high curvature points (near the interdental gaps) as seed points and iteratively expands the region according to criteria such as "the curvature difference between adjacent vertices is less than a threshold" and "the normal vector direction is similar", so as to initially divide the model into multiple initial tooth regions and achieve preliminary separation.

[0031] Furthermore, the initial region boundary is determined by calculating the gradient change between vertices in the initial tooth region. The local features include the local curvature of vertices near the boundary, the density of neighboring points, and edge sharpness (such as the size of the dihedral angle). The geometric parameters of a small area around the boundary are calculated by a sliding window, which can accurately locate the region boundary. The extracted local features provide fine-grained basis for boundary optimization and avoid the global features from ignoring local details.

[0032] Among them, the boundary optimization based on local features adopts the active contour model (Snakes algorithm). Starting from the initial boundary, the contour is iteratively adjusted by combining local features (such as applying contraction force to areas with high edge sharpness and applying expansion force to areas with smooth edges) to make the boundary fit the real interdental gap, forming the target tooth region, making the region boundary fit the tooth anatomy more closely, and improving the segmentation accuracy.

[0033] Specifically, a parametric shape model is used to match the vertex coordinates of the target tooth region with the template model. The parameters are iteratively optimized to minimize the geometric error between the fitting result and the target region, resulting in an initial independent crown model. The model post-processing includes cavity filling (using Poisson surface reconstruction or neighborhood topology-based interpolation algorithms to repair small holes on the model surface), mesh simplification (reducing redundant vertices through edge folding algorithms to reduce model complexity while ensuring accuracy), and topology repair (correcting non-manifold meshes to ensure that the model surface is closed and free of self-intersections). This eliminates geometric defects in the model, optimizes mesh quality, and enables the independent crown model to meet the accuracy and computational efficiency requirements of clinical applications (such as virtual orthodontics and prosthesis design).

[0034] In this embodiment of the invention, constructing the local coordinate system corresponding to the independent crown model includes: Obtain the vertex coordinate data of the independent crown model and calculate the mean of the vertex coordinate data; The mean value is used to calculate the difference between the vertex coordinate data to obtain the centered coordinate data. A covariance matrix is ​​constructed based on the centralized coordinate data, and the eigenvalues ​​and corresponding eigenvectors of the covariance matrix are solved. The eigenvector corresponding to the largest eigenvalue among the eigenvalues ​​is selected as the first initial principal direction, the eigenvector corresponding to the second largest eigenvalue among the eigenvalues ​​is selected as the second initial principal direction, and the eigenvector corresponding to the smallest eigenvalue among the eigenvalues ​​is selected as the third initial principal direction. The first initial principal direction, the second initial principal direction, and the third initial principal direction are validated to obtain the target principal direction, and the target principal direction is determined as the first approximate coordinate axis. Extract the crown geometric center, crown occlusal plane, buccal and lingual planes, and crown mesiodistal direction of the independent crown model, and perform least squares fitting on the crown occlusal plane to obtain the target occlusal plane; Align the horizontal axis of the first approximate coordinate axis with the mesial-distal direction of the crown, align the vertical axis of the first approximate coordinate axis with the buccal-lingual plane direction, and make the vertical axis of the first approximate coordinate axis perpendicular to the target occlusal plane. Generate a second coordinate axis based on the alignment result. With the geometric center of the crown as the origin, a local coordinate system corresponding to the independent crown model is constructed based on the origin and the second coordinate axis.

[0035] In detail, the three-dimensional coordinates of all vertices are extracted from the mesh data of the three-dimensional crown digital model to form a vertex coordinate dataset. The mean of all vertex coordinates is calculated by arithmetic mean to provide a benchmark for subsequent coordinate centering and eliminate the interference of the overall model translation on geometric analysis. The coordinates of each vertex are interpolated and the original coordinates of each vertex are subtracted from the mean to obtain the coordinate data centered on the origin after translation, avoiding geometric feature analysis errors caused by the overall positional shift of the model.

[0036] Specifically, the centered coordinate data is substituted into the covariance formula to calculate the covariance in three-dimensional space, forming a 3×3 covariance matrix. The covariance matrix is ​​then solved using eigenvalue decomposition (EVD) or singular value decomposition (SVD) algorithms to obtain eigenvalues ​​and corresponding unit eigenvectors. The covariance matrix reflects the distribution correlation of vertex coordinates in three-dimensional space, and the eigenvalues ​​and eigenvectors can quantify the degree of dispersion of data in different directions.

[0037] The process involves sorting eigenvalues ​​by size and directly mapping them to corresponding eigenvectors. The largest eigenvalue corresponds to the first initial principal direction (the direction where data distribution is most dispersed), the second largest eigenvalue corresponds to the second initial principal direction, and the smallest eigenvalue corresponds to the third initial principal direction (the direction where data distribution is most concentrated). This avoids the subjectivity of manual definition and ensures that the principal directions reflect the overall morphological extension trend of the crown model (such as the long axis direction of the crown). The rationality of the initial principal directions is verified through geometric constraints. For example, the angle between the initial principal directions and the approximate anatomical directions of the crown (such as the mesiodistal direction and the buccal-lingual direction) is calculated. If the angle is too large (e.g., exceeding 30°), the direction is finely adjusted through rotation transformation. Furthermore, the orthogonality of the three principal directions is verified (by calculating the vector dot product, and correcting it if it deviates from 0), ensuring that the orthogonality requirement of the coordinate system is met. Finally, three orthogonal target principal directions that conform to geometric meaning are obtained.

[0038] The geometric center of the crown can be the mean of the vertex coordinates or the center of volume (calculated by grid integration); the occlusal surface region of the crown is the region with higher coordinate values ​​(or greater curvature) extracted from the model through threshold segmentation (the occlusal surface is usually the apex of the crown); the buccal-lingual plane direction is determined by calculating the maximum span direction of the vertices on the left and right sides (buccal and lingual); the mesiodistal direction is determined by the maximum span direction of the vertices at the front and rear ends of the crown (mesiodistal is the side closer to the midline, and distal is the side farther from the midline).

[0039] Furthermore, the least squares method is used to fit the plane to the vertices of the occlusal surface region to ensure that the sum of the squares of the distances between the plane and the vertices of the occlusal surface is minimized. Specifically, the vertices of the occlusal surface can be extracted from the top of the model by a region growing algorithm, and the least squares formula can be substituted into the overdetermined system of equations (using matrix inversion or SVD method) to obtain the target occlusal surface plane.

[0040] In detail, the target principal direction (first approximate coordinate axis) is aligned with the anatomical direction through rotational transformation. The horizontal axis (X-axis) is aligned with the mesiodistal direction, i.e., the angle between the target principal direction horizontal axis and the mesiodistal direction vector is calculated and adjusted through Euler angle rotation or quaternion rotation transformation so that the horizontal axis coincides with the mesiodistal direction. The vertical axis (Y-axis) is aligned with the buccal-lingual direction. Similarly, the vertical axis is adjusted to be consistent with the normal or maximum span direction of the buccal-lingual plane. The vertical axis (Z-axis) is perpendicular to the target occlusal plane. According to the equation of the occlusal plane, the vertical axis is adjusted to be in the same direction as the normal vector (or in the opposite direction, which must conform to the right-hand coordinate system rule) so that the coordinate axis conforms to the anatomical definition of the tooth crown (such as mesiodistal, buccal-lingual, perpendicular to the occlusal plane), solving the problem that the initial principal direction may deviate from the anatomical direction and ensuring that the coordinate system has directional consistency in clinical diagnosis and treatment.

[0041] Furthermore, using the extracted geometric center of the crown as the origin, and combining it with the second coordinate axis, a local coordinate system is defined according to the right-hand coordinate system rule. The origin is the geometric center, the X-axis is the mesiodistal direction, the Y-axis is the buccal-lingual direction, and the Z-axis is perpendicular to the occlusal plane. Combining the origin coordinates with the three orthogonal coordinate axes forms a local coordinate system describing the spatial morphology of a single crown, as detailed below. Figure 3 As shown, it can accurately describe the morphological characteristics of a single tooth, facilitating subsequent comparisons of different tooth models, positional adjustments in virtual orthodontics, and size matching in prosthesis design, thereby improving the accuracy and efficiency of computer-aided oral diagnosis and treatment.

[0042] S3. Spatial registration of the independent crown model according to the local coordinate system to obtain the target crown model, and extraction of the initial crown feature matrix according to the target crown model; In this embodiment of the invention, the independent crown models are spatially registered according to the local coordinate system to standardize their orientation and spatial position. Each crown model is placed in the reference coordinate system, that is, each crown model is automatically rotated to align with the principal axis of the reference system and translated to the preset origin position to ensure that all crowns have the same orientation and standardized position, thereby improving the accuracy and robustness of crown comparison and analysis, and further enhancing the stability and accuracy of model registration.

[0043] In this embodiment of the invention, the step of spatially registering the independent crown model according to the local coordinate system to obtain the target crown model includes: Extract feature points in the preset reference coordinate system and the local coordinate system, and calculate coordinate system transformation parameters based on the feature points; Based on the coordinate system transformation parameters, an initial rigid body matrix transformation is performed to obtain the coordinate system rotation matrix and the coordinate system translation vector. The target coordinate system is obtained by rigidly registering the local coordinate system using the coordinate system rotation matrix and the coordinate system translation vector. The independent crown model is rotated into the target coordinate system to obtain the target crown model.

[0044] In detail, the point cloud form corresponding to the target crown model is shown in the following formula: in, This represents the point cloud representation of the target crown model. Indicates the first The three-dimensional coordinates of each feature point Indicates the first The surface normal vector direction of each feature point Indicates the first The normalized 3D coordinates of each feature point after normalization processing Indicates the first One-dimensional defect features of a number of feature points Indicates the first The feature of the number of mesh patches per feature point.

[0045] In detail, in the field of 3D model processing, feature points refer to points on the model that have significant characteristics and are easy to identify and locate, such as corner points, edge points, and curvature maxima. Through specific algorithms, these feature points are found on independent crown models under a preset reference coordinate system and the current local coordinate system to be processed. Feature descriptors (a mathematical representation used to describe the local geometric characteristics around a feature point) are used to match the corresponding feature points in the two coordinate systems, thereby obtaining a set of corresponding feature point pairs. Feature point pairs can reflect the spatial correspondence between the two coordinate systems, providing a basis for subsequent calculation of coordinate system transformation parameters.

[0046] Specifically, based on feature point pairs, a spatial geometric transformation parameter calculation method is used. By utilizing the spatial positional relationship between these feature points and analyzing their distribution and differences in three-dimensional space, the transformation relationship between two coordinate systems is determined. This ensures that after the feature points in the local coordinate system undergo this transformation, they can coincide as much as possible with the corresponding feature points in the preset reference coordinate system. This transformation typically includes operations such as rotation and translation. By analyzing the displacement and angle changes of the corresponding point pairs, the corresponding transformation parameters can be calculated. These parameters describe the transformation method from the local coordinate system to the preset reference coordinate system.

[0047] Furthermore, rigid body transformation refers to a transformation in three-dimensional space where an object only undergoes rotation and translation without deformation (such as scaling, twisting, etc.). The initial rigid body matrix contains all the information about the coordinate system transformation. Through a matrix decomposition algorithm, the initial rigid body matrix is ​​decomposed into two parts: the coordinate system rotation matrix and the coordinate system translation vector. The rotation matrix describes the rotation angle and direction of the coordinate system in space. It is a 3×3 matrix, and the elements in the matrix can determine the rotation of the coordinate system around each coordinate axis. The translation vector represents the translation distance of the coordinate system in the three coordinate axis directions. It is a three-dimensional vector, corresponding to the translation amount in the x, y, and z directions, respectively.

[0048] Specifically, rigid registration refers to the process of aligning one coordinate system with another coordinate system through rotation and translation without changing the model's orientation. The local coordinate system is rotated using a rotation matrix so that its direction is consistent with the direction of the preset reference coordinate system. Then, the rotated coordinate system is translated according to the translation vector so that its position coincides with the position of the preset reference coordinate system. After such transformation, the target coordinate system is obtained.

[0049] In detail, after obtaining the target coordinate system, it is necessary to transform the independent crown model from the original local coordinate system to the target coordinate system. Specifically, for each point on the independent crown model, a corresponding coordinate transformation is performed according to the coordinate system rotation matrix and translation vector. That is, first, the points on the model are rotated according to the rotation matrix so that their direction is consistent with the target coordinate system, and then translated according to the translation vector so that their position is aligned with the target coordinate system. Through this transformation, the independent crown model is rotated into the target coordinate system, thus obtaining the target crown model.

[0050] In this embodiment of the invention, an initial crown feature matrix is ​​extracted from the target crown model. For the vertices of the crown surface, key features are extracted from the geometric level, namely, the normal direction reflects the surface orientation, and the curvature describes the degree of surface bending. The various features of each vertex are combined into feature vectors according to certain rules, and then the feature vectors of all vertices are arranged in an orderly manner to construct a matrix form, which is the initial crown feature matrix.

[0051] In this embodiment of the invention, rapid spatial alignment of the crown model is achieved by establishing a local coordinate system. An initial crown feature matrix is ​​generated based on the registered target crown model, so that the input data of the subsequent neural network processing has standardized spatial distribution characteristics, effectively solving the problem of inconsistent feature space caused by differences in the acquisition angle of the original point cloud data.

[0052] S4. Adaptively sample the initial crown feature matrix to obtain multiple crown sampling vertices, and perform bi-branch offset prediction based on the crown sampling vertices to obtain an offset point cluster.

[0053] In this embodiment of the invention, an adaptive sampling strategy is used to divide the initial crown feature matrix to generate multiple crown sampling vertices. The number of sampling points m can be selected as 2048, 4096, 8192, etc., depending on the actual computing resource configuration. Adaptive sampling can not only efficiently manage video memory resources and reduce the single computation load during training and inference, but also ensure that the neural network can fully learn and express large-scale point cloud data under limited video memory conditions, thereby balancing computational efficiency and feature extraction performance.

[0054] Specifically, after adaptive sampling of the initial crown feature matrix, an improved deep learning network (e.g., a Transformer architecture or a PointNet++ variant) is used to predict the offset distance and offset scale of each sampled vertex relative to the crown feature points, in order to generate offset point clusters. These offset point clusters are specifically as follows: Figure 4 As shown.

[0055] In this embodiment of the invention, the adaptive sampling of the initial crown feature matrix to obtain multiple crown sampling vertices includes: Extract the geometric and texture parameters of the initial crown feature matrix; The geometric feature parameters and the texture feature parameters are normalized, and the normalized feature parameters are mapped to the initial crown feature matrix to obtain a standardized feature matrix. Calculate the importance weight of each feature parameter in the standardized feature matrix; Feature parameters whose importance weight is greater than a preset first sampling threshold are determined as high sampling density parameters, and feature parameters whose importance weight is less than or equal to the first sampling threshold are determined as low sampling density parameters. The normalized feature matrix is ​​subjected to hierarchical sampling based on the high sampling density parameter and the low sampling density parameter to obtain a preliminary sampled vertex set; Calculate the Euclidean distance between adjacent vertices in the initial sampled vertex set, and determine the vertex pairs whose Euclidean distance is less than a preset distance threshold as crown sampling vertices.

[0056] In detail, in 3D model processing, geometric feature parameters describe the spatial attributes of the model, such as its shape, size, and curvature, including vertex coordinates, normal direction, and curvature value. Texture feature parameters, on the other hand, reflect the texture information of the model's surface, such as color distribution and texture intensity. Through specific algorithms, the initial crown feature matrix is ​​analyzed and processed to extract these geometric and texture feature parameters.

[0057] In this process, parameter normalization is performed to eliminate the influence of differences in dimensions and numerical ranges between different feature parameters, making each feature parameter comparable. Commonly used normalization methods include linear normalization and Z-score normalization. These methods map the values ​​of each feature parameter to a uniform range (such as between 0 and 1). After normalization, the normalized feature parameters are remapped back to the initial crown feature matrix to obtain the standardized feature matrix.

[0058] Furthermore, by analyzing the contribution of each feature parameter in the standardized feature matrix to the overall features of the crown model, their importance weights are determined. For example, statistical methods, such as analysis of variance, are used to measure the impact of each feature parameter on the data distribution and rank them by importance, thereby obtaining the importance weight of each feature parameter. The importance weight reflects the degree of importance of the feature parameter in describing the features of the crown model. The larger the weight, the stronger the feature parameter's discriminative power and representativeness of the model.

[0059] Specifically, each feature parameter in the standardized feature matrix is ​​compared with this threshold. If the importance weight of a feature parameter is greater than the threshold, it is identified as a high sampling density parameter; otherwise, if the importance weight is less than or equal to the threshold, it is identified as a low sampling density parameter. Through this classification method, feature parameters can be divided into two categories, each corresponding to a different sampling density, providing a basis for subsequent stratified sampling.

[0060] In detail, for regions corresponding to high sampling density parameters, a higher sampling density is used for sampling, that is, more sampling points are selected in the region to capture the features of the region in more detail; for regions corresponding to low sampling density parameters, a lower sampling density is used for sampling, that is, fewer sampling points are selected in the region. Through this hierarchical sampling method, while ensuring the sampling accuracy of important feature regions, unnecessary sampling points are reduced, and sampling efficiency is improved. Finally, all the sampled vertices are combined to obtain a preliminary sampled vertex set.

[0061] Furthermore, Euclidean distance is a method to measure the straight-line distance between two points in space. By calculating the coordinate difference between two vertices in a three-dimensional coordinate system, the Euclidean distance between them can be obtained. Then, a preset distance threshold is set, and vertex pairs with Euclidean distances less than the threshold are filtered out and determined as the final crown sampling vertices. This can further optimize the sampling vertex set, remove overly dense redundant vertices, and obtain more reasonable and effective crown sampling vertices.

[0062] In this embodiment of the invention, the step of performing bi-branch offset prediction based on the sampling vertices of the tooth crown to obtain an offset point cluster includes: An initial vertex feature matrix is ​​constructed based on the sampled vertices of the tooth crown, and multiple convolutions are performed on the initial vertex feature matrix to obtain a high-dimensional vertex feature matrix; Based on the preset directional branch network, the high-dimensional vertex feature matrix is ​​subjected to directional offset analysis to obtain the unit offset direction vector of each vertex in the high-dimensional vertex feature matrix; Distance offset analysis is performed on the high-dimensional vertex feature matrix according to the preset scale branch network to obtain the offset distance of each vertex in the high-dimensional vertex feature matrix; The offset of each vertex is calculated based on the unit offset direction vector and the offset distance; Obtain the initial three-dimensional coordinates of each vertex in the high-dimensional vertex feature matrix, and add the initial three-dimensional coordinates to the offset to obtain the offset point cluster.

[0063] In detail, the initial feature matrix of the vertex is shown in the following formula: in, Represents the initial characteristic matrix of the vertices. Indicates the sampling vertex of the crown. The first matrix represents the first sampled vertex of the first crown. The second matrix represents the sampling vertex of the second crown. Indicates the first The first sampling vertex of the crown matrix, This indicates transpose.

[0064] In detail, for each tooth crown sampling vertex, the features of the preset dimensions are arranged and combined in a certain order to construct the initial vertex feature matrix. Then, the convolution operation technique in the convolutional neural network (CNN) is used to perform multi-layer convolution on the initial vertex feature matrix. That is, multiple learnable convolution kernels (also called filters) are used to slide on the input matrix. In each convolution layer, the convolution kernel will perform element-wise multiplication and summation with the local region of the input matrix. Then, the result is processed by the activation function (such as the ReLU function, which can introduce non-linearity and enhance the expressive power of the network) to obtain the output of the layer.

[0065] Through the stacking of multiple convolutional layers, each convolutional layer can extract features at different levels in the input matrix. Lower-level convolutions may mainly capture some simple local features, such as edges and corners. As the number of convolutional layers increases, higher-level convolutions can gradually extract more complex and abstract global features. After multiple convolutions, the features in the initial vertex feature matrix are continuously refined and sublimated, and finally a high-dimensional vertex feature matrix is ​​obtained, which helps to make more accurate predictions of vertex offsets in the future.

[0066] Furthermore, directional offset analysis is performed on the high-dimensional vertex feature matrix according to a preset directional branch network. The directional branch network is a neural network structure specifically designed to predict vertex offset directions, typically composed of multiple fully connected layers or convolutional layers. These layers are connected by nonlinear activation functions to increase the network's nonlinear expressive power. In the directional branch network, after a series of neuron calculations and feature transformations, the network outputs a set of vectors with the same dimension as the number of vertices. To obtain unit offset direction vectors, these output vectors are normalized. The purpose of normalization is to adjust the length of the vectors to 1, so that each vector only represents directional information and is not affected by the vector length. In this way, the unit offset direction vectors of each vertex in the high-dimensional vertex feature matrix can be obtained. These vectors indicate the direction in which the vertex may offset in space.

[0067] Furthermore, the construction of the scale branch network is similar to that of the directional branch network. The scale branch network is also a specially designed neural network used to predict the offset distance of vertices. It also consists of multiple neuron layers connected by a non-linear activation function to learn the complex relationship between the input high-dimensional vertex feature matrix and the vertex offset distance. After processing the input high-dimensional vertex feature matrix, the scale branch network outputs a set of values ​​equal to the number of vertices. Each value represents the offset distance of the corresponding vertex. These offset distances indicate the distance that the vertex needs to move in the predicted offset direction and are one of the important parameters for determining the final offset position of the vertex.

[0068] Specifically, the offset is shown in the following formula: in, Indicates the offset. The horizontal axis represents the local coordinate system. The vertical axis represents the local coordinate system. The vertical axis represents the local coordinate system. Indicates the offset distance. Represents the unit offset direction vector. This represents the first offset distance of the first crown sampling vertex. Indicates the first The first sampling vertex of the crown Offset distance, This represents the first offset direction vector of the first crown sampling vertex. Indicates the first The first sampling vertex of the crown Offset direction vector.

[0069] In detail, the cluster of offset points is shown in the following formula: in, Indicates an offset point cluster, Indicates the sampling vertex of the crown. This represents the initial three-dimensional coordinates corresponding to the sampling vertex of the tooth crown. This indicates the offset.

[0070] Specifically, for each vertex, its unit offset direction vector is multiplied by the corresponding offset distance. Since the length of the unit offset direction vector is 1, the result of the scalar multiplication is to scale the direction vector according to the offset distance to obtain a vector with a corresponding length in the specified direction. This vector is the offset of the vertex. The offset combines the direction information and the distance information to accurately determine the offset of each vertex in space.

[0071] In detail, the initial coordinate information of each vertex in three-dimensional space can be directly obtained from the original crown sampling vertex data. Using vector addition, the initial three-dimensional coordinate vector of each vertex is added to the corresponding offset vector to obtain a new vector. The new vector represents the final position coordinates of the vertex after the offset. Combining the offset coordinates of all vertices together, the offset point cluster is obtained. This point cluster describes the new position distribution of the crown vertices after offset prediction, and can more accurately reflect the morphological changes of the crown.

[0072] Optionally, to optimize the offset prediction results, the network structure is fitted according to the orientation (see [reference]). Figure 5 As shown, an offset direction loss function and an offset distance loss function are set up, and the training is controlled by the collaboration of the two; wherein, the offset direction loss function uses cosine similarity to measure the difference between the unit offset direction vector and the actual offset direction, and its definition is as follows: in, This represents the offset direction loss function. Represents the unit offset direction vector. This is the true offset vector. The total number of vertices sampled for the crown.

[0073] Specifically, the offset direction loss function is used to ensure that the unit offset direction vector is as consistent as possible with the true offset direction, so as to improve the directional stability of the offset estimation; Optionally, the offset distance loss function uses Euclidean distance to measure the distance error between the offset distance and the true offset distance, and its definition is as follows: in, This represents the offset distance loss function. Indicates the offset distance. Indicates the actual offset distance. The total number of vertices sampled for the crown.

[0074] Specifically, the distance loss function is used to optimize the positional accuracy of the offset point to ensure the reasonableness of the offset amount; Finally, the total loss function is formed by weighting the offset direction loss function and the offset distance loss function according to preset weights. The specific total loss function is shown in the following formula: in, This represents the offset direction loss function. This indicates the preset offset direction loss weight coefficient. This represents the offset distance loss function. This represents the preset offset distance loss weighting coefficient. In this embodiment of the invention, a dual constraint is adopted, namely, an offset direction loss function based on cosine similarity and a distance loss function based on Euclidean distance, to optimize the offset, thereby improving the accuracy and stability of offset prediction.

[0075] S5. Calculate the cluster center point of the offset point cluster, and determine the crown feature points based on the cluster center point.

[0076] In this embodiment of the invention, the offset point clusters are iteratively clustered using a clustering algorithm under a preset number of clusters to automatically divide them into several clusters. Each cluster corresponds to a region on the crown surface that is similar in shape and concentrated in distribution. During the clustering process, the cluster center is initialized in a random or density-aware manner. Then, the cluster to which each point belongs and the position of the cluster center are continuously updated based on Euclidean distance until the movement of the cluster center is lower than a set threshold or the maximum number of iterations is reached to achieve convergence. Finally, the geometric centroid of each cluster is used as the crown feature point of the crown model.

[0077] Among them, the crown feature points (see Figure 6 (As shown) includes FACC feature points (see...) Figure 7 As shown), the FACC feature points include two points, located at the top and bottom of the crown, one point each. The mesial-distal feature points consist of three points: the mesial point, the distal point, and the center point, one point each. The number of cusps is not fixed, and is specifically set as follows: two cusps are set at the premolars, four cusps are set at the molars, and no cusps are set at the incisors and canines.

[0078] In this embodiment of the invention, calculating the cluster center point of the offset point cluster includes: The offset point clusters are clustered to obtain multiple initial clusters, and the offset vertices in the initial clusters with the same number of clusters as the preset number of clusters are selected as the initial cluster centers. Calculate the Euclidean distance from each offset vertex in the initial cluster to the initial cluster center, and assign the offset vertex to the initial cluster to which the nearest initial cluster center belongs based on the Euclidean distance, thereby obtaining an initial sub-cluster set; Calculate the mean three-dimensional coordinates of all offset vertices within the initial sub-cluster set, and use the mean three-dimensional coordinates as the first cluster center of the corresponding initial sub-cluster; The change in cluster position is determined based on the first cluster center and the initial cluster center. When the change in cluster position is less than a preset convergence threshold, the first cluster center is determined as the cluster center point of the point cluster.

[0079] In detail, the offset point clusters are grouped according to the initial random partitioning technique in the clustering algorithm to obtain multiple initial clusters. This can quickly distribute the data into different groups, providing a basis for subsequent cluster center selection and optimization. According to the preset number of clusters (i.e. how many clusters the data is expected to be divided into), the same number of offset vertices as the number of clusters are randomly selected from the multiple initial clusters as the initial cluster centers. The subsequent clustering adjustment and optimization will be based on the initial cluster centers.

[0080] Furthermore, the Euclidean distance metric is used to measure the distance between the offset vertex and the initial cluster center. The straight-line distance between the two points in space is calculated. In three-dimensional space, for each offset vertex and the initial cluster center, the difference between them on the three coordinate axes is calculated. Then, the square root of the sum of the squares of these differences is performed to obtain the Euclidean distance between them. Based on the calculated Euclidean distance, each offset vertex is assigned to the initial cluster to which the nearest initial cluster center belongs. In this way, each offset vertex can find its "closest" cluster center, thus forming an initial set of subclusters. This process is based on the nearest neighbor principle, ensuring that each vertex is reasonably assigned to the corresponding cluster.

[0081] For each initial sub-cluster set, the three-dimensional coordinate information of all offset vertices in the set is collected. Then, the coordinate values ​​on the x, y, and z axes are summed. The summation result is then divided by the number of vertices in the sub-cluster to obtain the average coordinates of all offset vertices in the sub-cluster on the three axes. Combining these three average coordinates forms the first cluster center of the corresponding initial sub-cluster.

[0082] Specifically, the spatial difference between the first cluster center and the initial cluster center is calculated. For each cluster, the coordinate values ​​of its first cluster center and the initial cluster center on the x, y, and z axes are compared, and the difference between them is calculated. These differences are then combined to determine the change in cluster position. A convergence threshold is preset. When the calculated change in cluster position is less than this threshold, it indicates that the position of the cluster center has basically stabilized and no longer moves significantly. At this point, the current first cluster center can be determined as the final cluster center point. The cluster center point is as follows: Figure 8 As shown, by setting a convergence threshold to determine whether the clustering process has ended, the number of iterations of the algorithm can be effectively controlled, unnecessary calculations can be avoided, and more accurate clustering results can be obtained.

[0083] In this embodiment of the invention, the crown feature points are determined by using the cluster center points. This fully utilizes the typicality represented by the cluster centers. The cluster centers are the core positions of each sub-cluster in the offset cluster, and can comprehensively reflect the distribution characteristics of the vertices within the sub-cluster. In crown data, these center points often correspond to key positions in the crown morphology, such as cusps, pits, and other significant structures. By directly identifying the cluster center points as crown feature points, the key parts of the crown can be located quickly and accurately, providing important feature identifiers for subsequent crown analysis, modeling, restoration, and other operations, thereby improving processing efficiency and accuracy.

[0084] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0085] like Figure 9 The diagram shown is a functional block diagram of a crown feature point extraction device provided in an embodiment of the present invention.

[0086] In this embodiment of the disclosure, a crown feature point extraction device is provided, which corresponds one-to-one with the crown feature point extraction method described in the above embodiment. For example... Figure 4 As shown, the crown feature point extraction device 100 can be installed in an electronic device. According to its functions, the crown feature point extraction device 100 includes a point cloud model generation module 101, a local coordinate system construction module 102, a model space registration module 103, a crown offset prediction module 104, and a crown feature point determination module 105. Detailed descriptions of each functional module are as follows: The point cloud model generation module 101 is used to acquire the dental and jaw point cloud data of the target patient, reconstruct the point cloud mesh of the dental and jaw point cloud data, and generate a three-dimensional digital model of the dental crown. The local coordinate system construction module 102 is used to segment the dentition region of the three-dimensional crown digital model to obtain the independent crown model of each tooth, and to construct the local coordinate system corresponding to the independent crown model. The model space registration module 103 is used to perform spatial registration of the independent crown model according to the local coordinate system to obtain the target crown model, and to extract the initial crown feature matrix according to the target crown model; The crown offset prediction module 104 is used to adaptively sample the initial crown feature matrix to obtain multiple crown sampling vertices, and perform bi-branch offset prediction based on the crown sampling vertices to obtain an offset point cluster; The crown feature point determination module 105 is used to calculate the cluster center point of the offset point cluster and determine the crown feature points based on the cluster center point.

[0087] In this invention, the specific limitations of the crown feature point extraction device can be found in the above-described limitations of the crown feature point extraction method, and will not be repeated here. Each module in the aforementioned crown feature point extraction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0088] In the several embodiments provided by this invention, it should be understood that the disclosed devices and apparatuses can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0089] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0090] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0091] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Furthermore, any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory.

[0092] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0093] In the embodiments provided in this disclosure, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0094] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

[0095] It should be noted that if any software tools or components not belonging to our company appear in the embodiments of this application, they are merely for illustrative purposes and do not represent actual use.

Claims

1. A method for extracting feature points of a tooth crown, characterized in that, The method includes: Acquire the dental point cloud data of the target patient, reconstruct the point cloud mesh from the dental point cloud data, and generate a three-dimensional digital model of the dental crown; The three-dimensional digital model of the tooth crown is segmented into the jaw region to obtain an independent tooth crown model for each tooth, and a local coordinate system corresponding to the independent tooth crown model is constructed. The independent crown model is spatially registered according to the local coordinate system to obtain the target crown model, and the initial crown feature matrix is ​​extracted according to the target crown model. Adaptive sampling is performed on the initial crown feature matrix to obtain multiple crown sampling vertices, and bi-branch offset prediction is performed based on the crown sampling vertices to obtain an offset point cluster; Calculate the cluster center point of the offset point cluster, and determine the crown feature points based on the cluster center point.

2. The method for extracting crown feature points as described in claim 1, characterized in that, The step of reconstructing the point cloud data of the jaws to generate a three-dimensional digital model of the tooth crown includes: The dental point cloud data is normalized to obtain the first dental point cloud data, and outlier points are removed from the first dental point cloud data to obtain the target dental point cloud data. Perform normal vector analysis on the target tooth and jaw point cloud data to obtain the normal direction corresponding to the target tooth and jaw point cloud data; The target dental point cloud data is smoothed according to the normal direction to obtain smoothed dental point cloud data. Calculate the surface center point of the smoothed dental point cloud data, and generate a surface support domain based on the surface center point; Based on the surface support domain, the smooth dental point cloud data is reconstructed to obtain a three-dimensional crown surface; A three-dimensional model of the three-dimensional crown surface is obtained by performing a three-dimensional modeling.

3. The method for extracting crown feature points as described in claim 1, characterized in that, The step of segmenting the three-dimensional digital model of the tooth crown into dentition regions to obtain independent crown models for each tooth includes: The three-dimensional crown digital model is smoothed and filtered to obtain the target crown digital model; Global features of the target crown digital model are extracted, and a region growing algorithm is used to initially divide the target crown digital model into regions based on the global features, resulting in multiple initial tooth regions; Obtain the initial region boundaries corresponding to multiple initial tooth regions, and extract the local features of multiple initial tooth regions; Based on the local features, the boundary of the initial region is optimized to obtain the target tooth region; Morphological fitting is performed on the target tooth region to obtain an initial independent crown model; The initial independent crown model is post-processed to obtain an independent crown model.

4. The method for extracting crown feature points as described in claim 1, characterized in that, The construction of the local coordinate system corresponding to the independent crown model includes: Obtain the vertex coordinate data of the independent crown model and calculate the mean of the vertex coordinate data; The mean value is used to calculate the difference between the vertex coordinate data to obtain the centered coordinate data. A covariance matrix is ​​constructed based on the centralized coordinate data, and the eigenvalues ​​and corresponding eigenvectors of the covariance matrix are solved. The eigenvector corresponding to the largest eigenvalue among the eigenvalues ​​is selected as the first initial principal direction, the eigenvector corresponding to the second largest eigenvalue among the eigenvalues ​​is selected as the second initial principal direction, and the eigenvector corresponding to the smallest eigenvalue among the eigenvalues ​​is selected as the third initial principal direction. The first initial principal direction, the second initial principal direction, and the third initial principal direction are validated to obtain the target principal direction, and the target principal direction is determined as the first approximate coordinate axis. Extract the crown geometric center, crown occlusal plane, buccal and lingual planes, and crown mesiodistal direction of the independent crown model, and perform least squares fitting on the crown occlusal plane to obtain the target occlusal plane; Align the horizontal axis of the first approximate coordinate axis with the mesial-distal direction of the crown, align the vertical axis of the first approximate coordinate axis with the buccal-lingual plane direction, and make the vertical axis of the first approximate coordinate axis perpendicular to the target occlusal plane. Generate a second coordinate axis based on the alignment result. With the geometric center of the crown as the origin, a local coordinate system corresponding to the independent crown model is constructed based on the origin and the second coordinate axis.

5. The method for extracting crown feature points as described in claim 1, characterized in that, The step of spatially registering the independent crown model according to the local coordinate system to obtain the target crown model includes: Extract feature points in the preset reference coordinate system and the local coordinate system, and calculate coordinate system transformation parameters based on the feature points; Based on the coordinate system transformation parameters, an initial rigid body matrix transformation is performed to obtain the coordinate system rotation matrix and the coordinate system translation vector. The target coordinate system is obtained by rigidly registering the local coordinate system using the coordinate system rotation matrix and the coordinate system translation vector. The independent crown model is rotated into the target coordinate system to obtain the target crown model.

6. The method for extracting crown feature points as described in claim 1, characterized in that, The adaptive sampling of the initial crown feature matrix yields multiple crown sampling vertices, including: Extract the geometric and texture parameters of the initial crown feature matrix; The geometric feature parameters and the texture feature parameters are normalized, and the normalized feature parameters are mapped to the initial crown feature matrix to obtain a standardized feature matrix. Calculate the importance weight of each feature parameter in the standardized feature matrix; Feature parameters whose importance weight is greater than a preset first sampling threshold are determined as high sampling density parameters, and feature parameters whose importance weight is less than or equal to the first sampling threshold are determined as low sampling density parameters. The normalized feature matrix is ​​subjected to hierarchical sampling based on the high sampling density parameter and the low sampling density parameter to obtain a preliminary sampled vertex set; Calculate the Euclidean distance between adjacent vertices in the preliminary sampled vertex set, and determine the vertex pairs whose Euclidean distance is less than a preset distance threshold as crown sampling vertices.

7. The method for extracting crown feature points as described in claim 1, characterized in that, The step of performing bi-branch offset prediction based on the sampled vertices of the crown to obtain an offset point cluster includes: An initial vertex feature matrix is ​​constructed based on the sampled vertices of the tooth crown, and multiple convolutions are performed on the initial vertex feature matrix to obtain a high-dimensional vertex feature matrix; Based on the preset directional branch network, the high-dimensional vertex feature matrix is ​​subjected to directional offset analysis to obtain the unit offset direction vector of each vertex in the high-dimensional vertex feature matrix; Distance offset analysis is performed on the high-dimensional vertex feature matrix according to the preset scale branch network to obtain the offset distance of each vertex in the high-dimensional vertex feature matrix; The offset of each vertex is calculated based on the unit offset direction vector and the offset distance; Obtain the initial three-dimensional coordinates of each vertex in the high-dimensional vertex feature matrix, and add the initial three-dimensional coordinates to the offset to obtain the offset point cluster.

8. A device for extracting dental crown feature points, characterized in that, The device includes: The point cloud model generation module is used to acquire the dental and jaw point cloud data of the target patient, reconstruct the point cloud mesh of the dental and jaw point cloud data, and generate a three-dimensional digital model of the dental crown. The local coordinate system construction module is used to segment the jaw region of the three-dimensional crown digital model to obtain the independent crown model of each tooth, and to construct the local coordinate system corresponding to the independent crown model. The model space registration module is used to perform spatial registration of the independent crown model according to the local coordinate system to obtain the target crown model, and to extract the initial crown feature matrix according to the target crown model; The crown offset prediction module is used to adaptively sample the initial crown feature matrix to obtain multiple crown sampling vertices, and perform bi-branch offset prediction based on the crown sampling vertices to obtain an offset point cluster; The crown feature point determination module is used to calculate the cluster center point of the offset point cluster and determine the crown feature points based on the cluster center point.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform a method for extracting crown feature points as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements a method for extracting crown feature points as described in any one of claims 1 to 7.

Images