A registration and segmentation method based on dental oral scan and CBCT data

Abstract

The application discloses a registration and segmentation method based on dental oral scanning and CBCT data, and comprises the following steps: S1, CBCT dental data and oral scanning crown data are acquired to obtain target point cloud and source point cloud; S2, a sampling consistency initial registration algorithm is used to coarsely register the target point cloud and the source point cloud in the step S1, and an optimal transformation input matrix of the source point cloud to the target point cloud is obtained; S3, according to the optimal transformation input matrix of the source point cloud to the target point cloud in the step S2, an improved ICP algorithm is used to finely register the target point cloud and the source point cloud, and a scaling factor, a rotation matrix and a translation vector are obtained; S4, according to the scaling factor, the rotation matrix and the translation vector in the step S3, a K-D tree nearest neighbor algorithm is used to delete outliers to obtain a crown part and point cloud near the crown; S5, the step S3 and the step S4 are repeated to obtain accurate crown registration and segmentation results; and the method improves the accuracy of the crown data registration and segmentation results.

Description

Technical Field

[0001] This invention relates to the field of medical multimodal data processing technology, specifically to a registration and segmentation method based on dental intraoral scan and CBCT data. Background Technology

[0002] Today, with the improvement of living standards, more and more people are paying attention to their dental health and aesthetics. Similarly, medical imaging technology is developing rapidly, and dental CT has become the most popular imaging technology in clinical dentistry, widely used in orthodontic treatment and dental implants. Dental CBCT has low radiation dose and high spatial resolution, objectively reflecting the three-dimensional spatial structure of maxillofacial tissues. However, due to the low precision of CT scans, the thickness of the slices, and the influence of metal artifacts, the surface precision of the crowns is insufficient. Furthermore, CT scans generate volume data of the patient's entire skull; although the tooth portion can be extracted from the volume data to generate a surface mesh, the extracted surface mesh is coarser than that obtained from laser scanning. The goal of dental implants and orthodontic treatment is to reconstruct teeth and balance occlusal contact; inaccurate crown dimensions will lead to incorrect orthodontic diagnoses.

[0003] By scanning the internal oral cavity with a laser scanner, a relatively accurate mesh of the crown and gingival surfaces can be obtained. Furthermore, the crown surface can be extracted from the entire mesh using specialized segmentation techniques. Although optically scanned tooth models have high image resolution and the reconstructed models are highly accurate, they only contain data about the crown and cannot obtain information about the tooth root.

[0004] Current point cloud registration technologies typically employ two stages: coarse registration and fine registration. However, this approach fails to address the issue of varying data scales acquired by different devices. Furthermore, traditional crown segmentation methods utilize only intraoral scanning crown point cloud information, neglecting data acquired by other devices such as CBCT. Summary of the Invention

[0005] To address the aforementioned shortcomings in the prior art, this invention provides a registration and segmentation method based on dental intraoral scan and CBCT data.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0007] A registration and segmentation method based on dental intraoral scan and CBCT data includes the following steps:

[0008] S1. Acquire CBCT tooth data and oral scan crown data to obtain target point cloud and source point cloud;

[0009] S2. The sample consistency initial registration algorithm is used to perform coarse registration between the target point cloud and the source point cloud in step S1 to obtain the optimal transformation input matrix from the source point cloud to the target point cloud.

[0010] S3. Based on the optimal transformation input matrix from the source point cloud to the target point cloud in step S2, the improved ICP algorithm is used to perform fine registration between the target point cloud and the source point cloud to obtain the scaling factor, rotation matrix and translation vector.

[0011] S4. Based on the scaling factor, rotation matrix, and translation vector in step S3, the KD tree nearest neighbor algorithm is used to delete outliers and obtain the point cloud of the crown and the area near the crown.

[0012] S5. Repeat steps S3 and S4 to obtain accurate crown registration and segmentation results.

[0013] Furthermore, step S1 specifically includes:

[0014] S11. Acquire CBCT tooth data and segment it using a data segmentation method to obtain the target point cloud;

[0015] S12. Obtain oral scan data of the crown and segment it using the data segmentation method to obtain the source point cloud.

[0016] Furthermore, step S2 specifically includes:

[0017] S21. Use the source point cloud and the target point cloud as inputs to the sample consistency initial registration algorithm, and set the minimum sampling distance and the number of pairings between the source point cloud data and the target point cloud data.

[0018] S22. Set the minimum distance between sampling points to be greater than the minimum sampling distance set in step S21, and randomly collect n sampling points in the source point cloud;

[0019] S23. Based on the n sampling points randomly collected in step S22, use the KD tree nearest neighbor search method in the target point cloud to find a group of points similar to the FPFH features of the sampling points, and randomly select a point from the similar points as the matching point.

[0020] S24. Use the singular value decomposition method to solve the rotation and translation transformation matrix of the matching point, and perform error iteration by calculating the error corresponding to the matching point to obtain the minimum error function of the matching point.

[0021] S25. Take the rotation and translation matrix corresponding to the minimum error function of the matching points in step S24 as the optimal transformation input matrix from the source point cloud to the target point cloud, that is:

[0022] (R best , t best )

[0023] Among them, R best Let t represent the optimal rotation matrix. best This represents the optimal translation vector.

[0024] Furthermore, the minimum error function for the matching points in step S24 is:

[0025]

[0026] Where D(·) represents the minimum error function of the matching point, e i t represents the error of the i-th group of corresponding matching points after matrix transformation. e This represents a pre-set value.

[0027] Furthermore, step S3 specifically includes:

[0028] S31. Based on the optimal transformation input matrix from the source point cloud to the target point cloud in step S2, introduce a scaling factor and establish an objective function based on the rotation matrix and translation vector.

[0029] S32. Calculate the centroid-free coordinates of the target point cloud and the source point cloud, i.e.:

[0030]

[0031] Among them, u g u o These represent the centroid coordinates of the target point cloud data and the source point cloud data, respectively. Let i represent the i-th target point cloud. Let R represent the i-th source point cloud, and let R represent the rotation matrix;

[0032] S33. Based on the centroid coordinates of the target point cloud and the source point cloud in step S32, minimize the objective function based on the rotation matrix and translation vector established in step S31 to obtain the minimized objective and the minimized objective function.

[0033] S34. The ICP algorithm is used to solve the minimization objective function established in step S33 to obtain the scaling factor, rotation matrix and translation vector.

[0034] Furthermore, the objective function established in step S31 based on the rotation matrix and translation vector is:

[0035]

[0036] Where E(R,t) represents the objective function based on the rotation matrix and translation vector, t represents the translation vector, and N p The target point cloud and the source point cloud represent the number of points in the point cloud, where i represents the number of points, s represents the scaling factor, and p represents the number of points in the target point cloud. g Represents the set of target point clouds. This represents the first target point cloud. This represents the second target point cloud. Let p represent the nth target point cloud. g Represents the set of source point clouds. This represents the first source point cloud. This represents the second source point cloud. This represents the nth source point cloud.

[0037] Furthermore, the minimization objective and the minimization objective function obtained in step S33 are as follows:

[0038]

[0039] Among them, R * N represents the objective function to be minimized. p The target point cloud and the source point cloud represent the number of points in the point cloud, s represents the scaling factor, t represents the translation vector, argmin(·) represents finding the minimum value, i represents the number of points in the point cloud, and T represents the transpose. This represents the centroid-free coordinates of the i-th source point cloud. This represents the centroid coordinates of the i-th target point cloud.

[0040] Furthermore, step S34 specifically includes:

[0041] S341. Differentiate the scaling factor to obtain the value of the scaling factor when minimizing the extremum of the objective function, i.e.:

[0042]

[0043] Where S represents the scaling factor, N p This represents the number of points in the target point cloud and the source point cloud, where i represents the number of points. This represents the centroid-free coordinates of the i-th target point cloud. Let T represent the centroid-free coordinates of the i-th source point cloud, T represent the transpose, P represent the numerator, and M represent the denominator.

[0044] S342. Substitute the scaling factor value from step S341 into the minimization objective function in step S32 and regress the scale-free solution to obtain the updated minimization objective function, i.e.:

[0045]

[0046] Among them, R * This represents minimizing the objective function, and argmax represents finding the maximum value.

[0047] S343. Based on the updated minimization objective function in step S342, calculate the H matrix of the newly generated target point cloud and the source point cloud, and decompose the H matrix using singular value decomposition to obtain the decomposed H matrix, U matrix, and V matrix, i.e.:

[0048]

[0049] Where H represents the H matrix of the newly generated target point cloud and the source point cloud, and U represents HH T The matrix formed by the eigenvectors of H, where ∑ represents a diagonal matrix and V represents H. T The matrix formed by the eigenvectors of H, where S represents intermediate temporary parameters and has no practical meaning, and c t c s Let I represent the center of the target point cloud and the source point cloud, respectively; let det(V) represent the determinant of matrix V; let det(U) represent the determinant of matrix U; and let diag(1,1,-1) represent the diagonal matrix.

[0050] S344. Based on the H matrix, U matrix, and V matrix decomposed in step S343, the scaling factor, rotation matrix, and translation vector are obtained using the following formulas:

[0051]

[0052] Where R represents the rotation matrix, s represents the scaling factor, and t represents the translation vector.

[0053] Furthermore, step S4 specifically includes:

[0054] S41. Using the KD tree nearest neighbor algorithm, the source point cloud is taken as input, and a data structure is established to store the nearest neighbor index of the query point in the source point cloud and the distance to the nearest neighbor point.

[0055] S42. Traverse the target point cloud and use the scaling factor, rotation matrix and translation vector obtained in step S3 to find outliers in the target point cloud relative to the source point cloud, and add the outliers to the deletion queue to obtain the point cloud of the crown part and the area near the crown.

[0056] The present invention has the following beneficial effects:

[0057] The present invention proposes a registration and segmentation method based on dental intraoral scan and CBCT data. By effectively combining CBCT dental data and intraoral scan crown data, the method uses data segmentation to transform them into target point clouds and source point clouds. After coarse registration using the sample consistency initial registration algorithm, a modified ICP algorithm is used for fine registration. By introducing a scaling factor and iterating continuously, the method effectively solves the problem of scale difference in multimodal data acquired by multiple devices, thereby improving the accuracy of crown data registration and segmentation results. Attached Figure Description

[0058] Figure 1 This is a flowchart illustrating a registration and segmentation method based on dental intraoral scan and CBCT data proposed in this invention.

[0059] Figure 2 This is a schematic diagram showing the result of the CBCT tooth data and the oral scan crown data being basically aligned after the first fine registration.

[0060] Figure 3 This is a schematic diagram showing the precise registration and segmentation results of the tooth crown after multiple fine registrations. Detailed Implementation

[0061] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0062] like Figure 1 As shown, a registration and segmentation method based on dental intraoral scan and CBCT data includes the following steps S1-S5:

[0063] S1. Acquire CBCT tooth data and oral scan crown data to obtain target point cloud and source point cloud.

[0064] In this embodiment, CBCT equipment is used to acquire raw tooth data in DICOM format, and a medical image segmentation algorithm is used to segment the raw tooth data to obtain complete tooth STL data including the root crown. The complete tooth STL data including the root crown is then converted into target point cloud data using a data segmentation method, i.e., medical image software. An optical intraoral scanner is used to acquire crown STL data, and the data segmentation method, i.e., medical image software, is used to convert it into source point cloud data.

[0065] Specifically, step S1 includes S11-S12:

[0066] S11. Acquire CBCT tooth data and segment it using a data segmentation method to obtain the target point cloud.

[0067] S12. Obtain oral scan data of the crown and segment it using the data segmentation method to obtain the source point cloud.

[0068] S2. The sample consistency initial registration algorithm is used to perform coarse registration between the target point cloud and the source point cloud in step S1 to obtain the optimal transformation input matrix from the source point cloud to the target point cloud.

[0069] In this embodiment, the source point cloud and the target point cloud are used as inputs, and a sample consistency initial registration algorithm is used to perform coarse registration to obtain the optimal transformation input matrix from the source point cloud to the target point cloud, i.e., (R best , t best ), where R best Let t represent the optimal rotation matrix. best This represents the optimal translation vector.

[0070] Specifically, step S2 includes S21-S25:

[0071] S21. Use the source point cloud and the target point cloud as inputs to the initial registration algorithm for sample consistency, and set the minimum sampling distance and the number of pairings between the source point cloud data and the target point cloud data.

[0072] In this embodiment, the number of pairings between source point cloud data and target point cloud data is set to n, and the minimum sampling distance is set to d. min .

[0073] S22. Set the minimum distance between sampling points to be greater than the minimum sampling distance set in step S21, and randomly collect n sampling points in the source point cloud.

[0074] In this embodiment, n sampling points are randomly sampled in the source point cloud, and the minimum distance between the sampling points is guaranteed to be greater than the minimum sampling distance.

[0075] S23. Based on the n sampling points randomly collected in step S22, use the KD tree nearest neighbor search method in the target point cloud to find a group of points with FPFH features similar to the sampling points, and randomly select a point from the similar points as the matching point.

[0076] In this embodiment, for each sampling point, a set of points with similar FPFH features to the sampling point are found in the target point cloud using the KD-tree nearest neighbor search method. Then, a point is randomly selected from these similar points as a matching point, and n sets of matching points are obtained. Here, FPFH stands for Fast Point Feature Histogram, which is constructed by calculating the normal vectors between each point and its surrounding neighborhood to describe the geometry of the point within a local area.

[0077] S24. Use the singular value decomposition method to solve the rotation and translation transformation matrix of the matching point. By calculating the error corresponding to the matching point, perform error iteration to obtain the minimum error function of the matching point.

[0078] Specifically, the minimum error function of the matching points in step S24 is:

[0079]

[0080] Where D(·) represents the minimum error function of the matching point, e i t represents the error of the i-th group of corresponding matching points after matrix transformation. e This represents a pre-set value.

[0081] In this embodiment, the singular value decomposition (SVD) method is used to find the rotation and translation transformation matrix of the matching points. By calculating the error corresponding to the matching points and continuously iterating the error, the minimum error function of the matching points is obtained, which is the optimal transformation input matrix from the source point cloud to the target point cloud. Here, the parameter t... e As the parameter is being tuned, the optimal t may vary depending on the specific data during implementation.

[0082] S25. Take the rotation and translation matrix corresponding to the minimum error function of the matching points in step S24 as the optimal transformation input matrix from the source point cloud to the target point cloud, that is:

[0083] (R best , t best )

[0084] Among them, R best Let t represent the optimal rotation matrix. best This represents the optimal translation vector.

[0085] S3. Based on the optimal transformation input matrix from the source point cloud to the target point cloud in step S2, the improved ICP algorithm is used to perform fine registration between the target point cloud and the source point cloud to obtain the scaling factor, rotation matrix and translation vector.

[0086] In this embodiment, an improved ICP algorithm is adopted, which introduces a scaling factor. Based on the optimal transformation input matrix from the source point cloud to the target point cloud, an objective function based on the rotation matrix and translation vector is established, and the objective function is minimized. The scaling factor, rotation matrix and translation vector are obtained by solving the objective function. The whole process is the fine registration process between the target point cloud and the source point cloud.

[0087] Specifically, step S3 includes S31-S34:

[0088] S31. Based on the optimal transformation input matrix from the source point cloud to the target point cloud in step S2, introduce a scaling factor and establish an objective function based on the rotation matrix and translation vector.

[0089] Specifically, the minimization objective and the minimization objective function obtained in step S33 are as follows:

[0090]

[0091] Among them, R * N represents the objective function to be minimized. pThe target point cloud and the source point cloud represent the number of points in the point cloud, s represents the scaling factor, t represents the translation vector, argmin(·) represents finding the minimum value, i represents the number of points in the point cloud, and T represents the transpose. This represents the centroid-free coordinates of the i-th source point cloud. This represents the centroid coordinates of the i-th target point cloud.

[0092] S32. Calculate the centroid-free coordinates of the target point cloud and the source point cloud, i.e.:

[0093]

[0094] Among them, u g u o These represent the centroid coordinates of the target point cloud data and the source point cloud data, respectively. Let i represent the i-th target point cloud. Let R represent the i-th source point cloud, and let R represent the rotation matrix;

[0095] S33. Based on the centroid coordinates of the target point cloud and the source point cloud in step S32, minimize the objective function based on the rotation matrix and translation vector established in step S31 to obtain the minimized objective and the minimized objective function.

[0096] In this embodiment, the minimization objective is The simplification process of minimizing the objective function is as follows:

[0097]

[0098] Among them, let u g -sRu o If -t = 0, then:

[0099]

[0100] Since the rotation matrix R is an orthogonal matrix, R T If R = E, then:

[0101]

[0102] Removing irrelevant terms from the above formula yields the minimized objective function.

[0103] Specifically, the minimization objective and the minimization objective function obtained in step S33 are as follows:

[0104]

[0105] Among them, R * N represents the objective function to be minimized. pThe target point cloud and the source point cloud represent the number of points in the point cloud, s represents the scaling factor, t represents the translation vector, argmin(·) represents finding the minimum value, i represents the number of points in the point cloud, and T represents the transpose. This represents the centroid-free coordinates of the i-th source point cloud. This represents the centroid coordinates of the i-th target point cloud.

[0106] S34. The ICP algorithm is used to solve the minimization objective function established in step S33 to obtain the scaling factor, rotation matrix and translation vector.

[0107] Specifically, step S34 includes S341-S344:

[0108] S341. Differentiate the scaling factor to obtain the value of the scaling factor when minimizing the extremum of the objective function, i.e.:

[0109]

[0110] Where s represents the scaling factor, N p This represents the number of points in the target point cloud and the source point cloud, where i represents the number of points. This represents the centroid-free coordinates of the i-th target point cloud. Let T represent the centroid-free coordinates of the i-th source point cloud, T represent the transpose, P represent the numerator, and M represent the denominator.

[0111] S342. Substitute the scaling factor value from step S341 into the minimization objective function in step S32 and regress the scale-free solution to obtain the updated minimization objective function, i.e.:

[0112]

[0113] Among them, R * This represents minimizing the objective function, and argmax represents finding the maximum value.

[0114] S343. Based on the updated minimization objective function in step S342, calculate the H matrix of the newly generated target point cloud and the source point cloud, and decompose the H matrix using singular value decomposition to obtain the decomposed H matrix, U matrix, and V matrix, i.e.:

[0115]

[0116] Where H represents the H matrix of the newly generated target point cloud and the source point cloud, and U represents HH T The matrix formed by the eigenvectors of H, where ∑ represents a diagonal matrix and V represents H. T The matrix formed by the eigenvectors of H, where S represents intermediate temporary parameters and has no practical meaning, and c t c sLet I represent the center of the target point cloud and the source point cloud, respectively. Let I represent the identity matrix, det(V) represent the determinant of matrix V, det(U) represent the determinant of matrix U, and diag(1,1,-1) represent the diagonal matrix.

[0117] In this embodiment, step S343 is a singular value decomposition process, the purpose of which is to obtain the decomposed H matrix, U matrix, and V matrix so that they can be used to solve for the scaling factor, rotation matrix, and translation vector in the subsequent process.

[0118] S344. Based on the H matrix, U matrix, and V matrix decomposed in step S343, the scaling factor, rotation matrix, and translation vector are obtained using the following formulas:

[0119]

[0120] Where R represents the rotation matrix, s represents the scaling factor, and t represents the translation vector.

[0121] like Figure 2 As shown, Figure 2 This is a schematic diagram showing the result of the CBCT tooth data and the oral scan crown data being basically aligned after the first fine registration. Figure 2 The center dot represents intraoral scan data, the square represents CBCT data, and the "+" in the center indicates the center of view. From Figure 2 As can be seen, after the first fine registration, the CBCT tooth data and the oral scan crown data are basically aligned.

[0122] S4. Based on the scaling factor, rotation matrix, and translation vector in step S3, the KD tree nearest neighbor algorithm is used to remove outliers and obtain the point cloud of the crown and the area near the crown.

[0123] In this embodiment, the KD tree nearest neighbor algorithm is used to delete outliers in the target point cloud relative to the source point cloud, that is, to delete the part of the point cloud that does not overlap with the source point cloud. The remaining point cloud, which is the overlapping point cloud, is the segmentation result of the crown part and the point cloud near the crown.

[0124] Specifically, step S4 includes S41-S42:

[0125] S41. Using the KD tree nearest neighbor algorithm, the source point cloud is taken as input, and a data structure is established to store the nearest neighbor index of the query point in the source point cloud and the distance to the nearest neighbor point.

[0126] S42. Traverse the target point cloud and use the scaling factor, rotation matrix and translation vector obtained in step S3 to find outliers in the target point cloud relative to the source point cloud, and add the outliers to the deletion queue to obtain the point cloud of the crown part and the area near the crown.

[0127] S5. Repeat steps S3 and S4 to obtain accurate crown registration and segmentation results.

[0128] In this embodiment, the purpose of repeating steps S3-S4 is to obtain more accurate crown registration and segmentation results so that the target point cloud and the source point cloud completely overlap.

[0129] like Figure 3 As shown, Figure 3 This is a schematic diagram showing the precise registration and segmentation results of the tooth crown after multiple fine registrations. From... Figure 3 As can be seen, the crown portion completely overlaps with the point cloud near the crown, and registration is complete.

[0130] This invention acquires tooth data using CBCT and optical intraoral scanning equipment, and obtains target and source point cloud data through data processing. A sample consistency initial registration algorithm (SAC-IA) is used for preliminary coarse registration of the target and source point clouds. An improved ICP algorithm, incorporating a scaling factor, is then used for fine registration. Based on the matching results—that is, using the scaling factor, rotation matrix, and translation vector—outliers in the target point cloud relative to the source point cloud are removed by judging nearest neighbors. The remaining point cloud represents the crown portion and its vicinity. To obtain more accurate crown registration and segmentation results, the fine registration of the target and source point clouds is repeated, and outliers are removed. This invention employs a registration and segmentation method based on dental intraoral scanning and CBCT data, which incorporates a scaling factor into the ICP algorithm and performs iterative solving, effectively addressing the scale difference problem of multimodal data acquired from multiple devices and improving the accuracy of crown data registration and segmentation results.

[0131] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

[0132] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A registration and segmentation method based on dental intraoral scan and CBCT data, characterized in that, Includes the following steps: S1. Acquire CBCT tooth data and oral scan crown data to obtain target point cloud and source point cloud; S2. The sample consistency initial registration algorithm is used to perform coarse registration between the target point cloud and the source point cloud in step S1 to obtain the optimal transformation input matrix from the source point cloud to the target point cloud. S3. Based on the optimal transformation input matrix from the source point cloud to the target point cloud in step S2, the improved ICP algorithm is used to perform fine registration between the target point cloud and the source point cloud, obtaining the scaling factor, rotation matrix, and translation vector, specifically: S31. Based on the optimal transformation input matrix from the source point cloud to the target point cloud in step S2, introduce a scaling factor and establish an objective function based on the rotation matrix and translation vector. S32. Calculate the centroid-free coordinates of the target point cloud and the source point cloud, i.e.: in, , These represent the centroid coordinates of the target point cloud data and the source point cloud data, respectively. Indicates the first A target point cloud, Indicates the first Individual point clouds, Represents the rotation matrix; S33. Based on the centroid-free coordinates of the target point cloud and the source point cloud obtained in step S32, minimize the objective function established in step S31 based on the rotation matrix and translation vector to obtain the minimized objective and the minimized objective function, i.e.: in, This represents minimizing the objective function. This represents the number of points in the target point cloud and the source point cloud. Indicates the scaling factor. Represents the translation vector. This indicates finding the minimum value. Represents the number of point clouds. Indicates transpose. Indicates the first Centroid coordinates of a source point cloud Indicates the first Centroid coordinates of a target point cloud; S34. The ICP algorithm is used to solve the minimization objective function established in step S33 to obtain the scaling factor, rotation matrix and translation vector. S4. Based on the scaling factor, rotation matrix, and translation vector in step S3, the KD tree nearest neighbor algorithm is used to delete outliers and obtain the point cloud of the crown and the area near the crown. S5. Repeat steps S3 and S4 to obtain accurate crown registration and segmentation results.

2. The registration and segmentation method based on dental intraoral scan and CBCT data according to claim 1, characterized in that, Step S1 specifically includes: S11. Acquire CBCT tooth data and segment it using a data segmentation method to obtain the target point cloud; S12. Obtain oral scan data of the crown and segment it using the data segmentation method to obtain the source point cloud.

3. The registration and segmentation method based on dental intraoral scan and CBCT data according to claim 1, characterized in that, Step S2 specifically includes: S21. Use the source point cloud and the target point cloud as inputs to the sample consistency initial registration algorithm, and set the minimum sampling distance and the number of pairings between the source point cloud data and the target point cloud data. S22. Set the minimum distance between sampling points to be greater than the minimum sampling distance set in step S21, and randomly collect n sampling points in the source point cloud; S23. Based on the n sampling points randomly collected in step S22, use the KD tree nearest neighbor search method in the target point cloud to find a group of points similar to the FPFH features of the sampling points, and randomly select a point from the similar points as the matching point. S24. Use the singular value decomposition method to solve the rotation and translation transformation matrix of the matching point, and perform error iteration by calculating the error corresponding to the matching point to obtain the minimum error function of the matching point. S25. Take the rotation and translation matrix corresponding to the minimum error function of the matching points in step S24 as the optimal transformation input matrix from the source point cloud to the target point cloud, that is: in, Represents the optimal rotation matrix. This represents the optimal translation vector.

4. The registration and segmentation method based on dental intraoral scan and CBCT data according to claim 3, characterized in that, The minimum error function for the matching points in step S24 is: in, The minimum error function representing the matching points. Represents the matrix transformation after the first transformation. The error of the matching points corresponding to the group. This represents a pre-set value.

5. The registration and segmentation method based on dental intraoral scan and CBCT data according to claim 1, characterized in that, The objective function established in step S31 based on the rotation matrix and translation vector is: in, This represents the objective function based on the rotation matrix and translation vector. Represents the translation vector. This represents the number of points in the target point cloud and the source point cloud. Represents the number of point clouds. Indicates the scaling factor. Represents the set of target point clouds. This represents the first target point cloud. This represents the second target point cloud. Indicates the first A target point cloud, Represents the set of source point clouds. This represents the first source point cloud. This represents the second source point cloud. Indicates the first Individual point clouds.

6. The registration and segmentation method based on dental intraoral scan and CBCT data according to claim 1, characterized in that, Step S34 specifically includes: S341. Differentiate the scaling factor to obtain the value of the scaling factor when minimizing the extremum of the objective function, i.e.: in, Indicates the scaling factor. This represents the number of points in the target point cloud and the source point cloud. Represents the number of point clouds. Indicates the first Centroid coordinates of a target point cloud Indicates the first Centroid coordinates of a source point cloud Indicates transpose. Indicates the molecular part. Indicates the denominator; S342. Substitute the scaling factor value from step S341 into the minimization objective function in step S32 and regress the scale-free solution to obtain the updated minimization objective function, i.e.: in, This represents minimizing the objective function. This indicates finding the maximum value; S343. Based on the updated minimization objective function in step S342, calculate the H matrix of the newly generated target point cloud and the source point cloud, and decompose the H matrix using singular value decomposition to obtain the decomposed H matrix. matrix, Matrix, that is: in, The H matrix represents the newly generated target point cloud and the source point cloud. express The matrix formed by the eigenvectors of , Represents a diagonal matrix. express The matrix formed by the eigenvectors of , This represents an intermediate, temporary parameter with no practical meaning. , These represent the centers of the target point cloud and the source point cloud, respectively. Represents the identity matrix. Representation matrix The determinant, Representation matrix The determinant, Represents a diagonal matrix; S344. Based on the H matrix decomposed in step S343, matrix, The matrix, using the following formula, yields the scaling factor, rotation matrix, and translation vector: in, Represents the rotation matrix. Indicates the scaling factor. This represents the translation vector.

7. The registration and segmentation method based on dental intraoral scan and CBCT data according to claim 1, characterized in that, Step S4 specifically includes: S41. Using the KD tree nearest neighbor algorithm, the source point cloud is taken as input, and a data structure is established to store the nearest neighbor index of the query point in the source point cloud and the distance to the nearest neighbor point. S42. Traverse the target point cloud and use the scaling factor, rotation matrix and translation vector obtained in step S3 to find outliers in the target point cloud relative to the source point cloud, and add the outliers to the deletion queue to obtain the point cloud of the crown part and the area near the crown.

Images