Method and system for adjusting point cloud boundary line of dental cast by fusing regional features

Abstract

The invention relates to the technical field of stomatology, in particular to a dental cast point cloud boundary line adjusting method and system fusing regional features, based on dental cast point cloud data and a generated initial boundary line, the normal vector direction of the point cloud is adjusted to enable the normal vector direction to face outwards consistently, and points on the boundary line are sorted and mapped to obtain a point cloud point cloud boundary line; generating a target point cloud area and an expansion point set; calculating a local curvature and a normal vector change entropy of each point in the target point cloud area based on the extension point set, and obtaining a local risk weight of the point after fusion; constructing a multi-constraint optimization objective function, and performing iterative optimization on the boundary point set in combination with fidelity, smoothness and anti-mold penetration constraints; and according to the optimized boundary point set, sequentially connecting to generate a final boundary line. According to the method, through risk assessment fusing the local curvature and the normal vector entropy and combining multi-constraint optimization, high-precision generation of the dental cast boundary line is achieved, and it is ensured that the boundary line is tightly attached to the geometrical shape of the dental cast, smooth, natural and free of penetration.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of oral medicine, and in particular to a dental cast point cloud boundary line adjustment method and system fusing regional features. BACKGROUND

[0002] In dental cast point cloud processing, boundary line generation is a key step for defining the outline of the dental cast (such as the gum line or tooth boundary). In dental cast point cloud processing, the "frenulum region" on the dental cast refers to the mucosal fold structure connecting the gum with the lip, tongue, or cheek inside the oral cavity (such as the lingual frenulum, labial frenulum). These regions exhibit the following high mutation characteristics on the three-dimensional dental cast:

[0003] (1) Sharp protrusions: The frenulum region usually presents a sudden bulge or sharp edge with a dramatic change in geometry.

[0004] (2) High curvature change: The curvature of the surface in this region may suddenly change from flat to highly curved within a very small range, resulting in local geometric discontinuity.

[0005] When an algorithm generates a boundary line (such as a gum line), if these mutation characteristics are not adequately considered, the boundary points may incorrectly penetrate the dental cast surface (i.e., "penetration") due to the dramatic change in curvature or normal vector, resulting in unreasonable geometric penetration phenomena. This can lead to structural distortion in subsequent 3D printing or modeling (such as the virtual boundary line sinking into the model), affecting accuracy and safety.

[0006] Traditional methods often only adjust the boundary through local curvature or other geometric features, making it difficult to identify mutation regions (such as frenulum), leading to problems only discovered during subsequent processing, which affects optimization effectiveness. Many existing methods do not comprehensively consider all important factors in the objective function design. Many boundary adjustment methods use static models that cannot be adjusted in real time according to changes in regional features, leading to boundary adjustments that do not conform to actual conditions. Some techniques rely too much on global optimization, ignoring the impact of local features, making it difficult to handle complex local changes or mutations. SUMMARY

[0007] To solve the problem of traditional methods only adjusting the boundary based on local curvature, failing to identify high mutation regions such as frenulum, and leading to boundary penetration and imbalance between smoothness and fidelity, the present application proposes a dental cast point cloud boundary line adjustment method and system fusing regional features. Through ordered mapping and expansion of boundary points, fusion of local curvature and normal vector change entropy to identify high-risk regions, and construction of a multi-constraint optimization model, high-precision generation of dental cast boundary lines is achieved, ensuring that the boundary line closely follows the geometric shape of the dental cast, is smooth and natural, and has no penetration, significantly improving the accuracy and safety of 3D printing or modeling.

[0008] To achieve the above object, the technical scheme adopted is:

[0009] The application provides a dental model point cloud boundary line adjustment method fusing regional features, comprising the following steps:

[0010] Step 1: based on the dental model point cloud data and the generated initial boundary line, the normal vector direction of the point cloud is adjusted to be consistent outward, and the points on the boundary line are sorted and mapped to generate a target point cloud region and an extended point set;

[0011] Step 2: based on the extended point set, the local curvature and the normal vector change entropy of each point in the target point cloud region are calculated, and the local risk weight of the point is obtained after fusion;

[0012] Step 3: a multi-constraint optimization objective function is constructed, and the boundary point set is iteratively optimized in combination with the fidelity, smoothness and anti-penetration constraints;

[0013] Step 4: according to the optimized boundary point set, the final boundary line is generated by being connected in turn.

[0014] According to the dental model point cloud boundary line adjustment method fusing regional features, further, the dental model point cloud data in step 1 is obtained from the oral cavity hard tissue or soft tissue surface data acquired by an intraoral scanner, a cone beam CT or a structured light scanning device.

[0015] According to the dental model point cloud boundary line adjustment method fusing regional features, further, step 1 specifically comprises:

[0016] The normal vector of the dental model point cloud is calculated using the PCA algorithm, and the normal vector is adjusted based on the viewpoint direction to be consistent outward;

[0017] The points on the boundary line are sequentially sampled and mapped into the dental model point cloud, and the nearest points on the dental model corresponding to the points on the boundary line are found through KDtree accelerated search, and the corresponding points on the boundary line are taken as the point set to generate the target point cloud region;

[0018] Taking the points in the target point cloud region as the center, a neighborhood expansion is performed by setting a radius r, and the points within the radius of all points in the point set are taken as the extended point set.

[0019] According to the dental model point cloud boundary line adjustment method fusing regional features, further, the calculation formula of the local curvature in step 2 is:

[0020]

[0021] Wherein, k i is the curvature estimation value of point p i , k is the number of neighboring points, N i is the curvature estimation value of point p iThe neighbor set of point p, n i The normal vector of point p, n i The normal vector of neighbor point p j The normal vector of neighbor point p j The normal vector of neighbor point p i -n j || represents the Euclidean distance of two normal vectors, the larger the value, the greater the direction difference; ||p i -p j || -1 represents the inverse of the Euclidean distance of two points, the closer the distance, the greater the weight.

[0022] According to the boundary line adjustment method of the dental model point cloud based on the fusion of regional features, further, the calculation formula of the normal vector change entropy in step 2 is:

[0023]

[0024] Wherein, E i is the normal vector direction entropy of point p i , n i ·n j is the dot product of the normal vectors of point p i and neighbor point p j , is the cosine similarity of two normal vectors, and log is the natural logarithm.

[0025] According to the boundary line adjustment method of the dental model point cloud based on the fusion of regional features, further, the calculation formula of the local risk weight in step 2 is:

[0026] w i =σ(αk i +βE i )

[0027] Wherein, w i is the penetration risk value of point p i , that is, the local risk weight of the point; α and β are respectively the feature fusion coefficients of curvature estimation value k i and normal vector direction entropy E i , and σ(x) is a sigmoid function.

[0028] According to the boundary line adjustment method of the dental model point cloud based on the fusion of regional features, further, the multi-constraint optimization objective function in step 3 is:

[0029]

[0030] In the formula, p i is the original boundary point; Q is the boundary point set to be optimized, wherein q i is the variable to be optimized, q i-1 , qi , q i+1 are three adjacent points in the boundary point set to be optimized; λ S is a smoothing term weight coefficient, λ p is a penetration prevention term weight coefficient, φ(q i , S) is a penetration penalty function, S is a dental cast point cloud surface, represents an impenetrable geometric plane, and N is the total number of points in the original boundary point set.

[0031] According to the dental cast point cloud boundary line adjustment method fusing regional features, further, the penetration penalty function φ(q i , S) is defined as the penetration degree of the point q i relative to the dental cast point cloud surface S, and the expression is:

[0032]

[0033] wherein v i is the nearest neighbor point of the point q i on the surface S, n i is the normal vector of the point v i ; when the point q i is located outside the surface S, there is no penalty; when the point q i penetrates the surface S, the penalty is based on the square of the penetration depth.

[0034] According to the dental cast point cloud boundary line adjustment method fusing regional features, further, the step 3 adopts a gradient descent algorithm to iteratively optimize the objective function until a set convergence condition is reached.

[0035] Further, the present application also provides a dental cast point cloud boundary line adjustment system fusing regional features, comprising:

[0036] A point cloud region acquisition module is configured to adjust the point cloud normal vector direction to be consistent outward based on the dental cast point cloud data and the generated initial boundary line, sort and map the points on the boundary line, and generate a target point cloud region and an extended point set;

[0037] A risk weight calculation module is configured to calculate the local curvature and the normal vector change entropy of each point in the target point cloud region based on the extended point set, and fuse the local risk weight of the point after calculation;

[0038] A boundary point set optimization module is configured to construct a multi-constraint optimization objective function, combine the fidelity, smoothness and penetration prevention constraints, and iteratively optimize the boundary point set;

[0039] A boundary line generation module is configured to sequentially connect and generate a final boundary line according to the optimized boundary point set.

[0040] The beneficial effects achieved by the above technical solution are:

[0041] 1. Guaranteeing boundary mapping accuracy and topological continuity

[0042] The present application forms an accurate one-to-one mapping between the boundary points and the dental model point cloud by means of boundary point sampling, ordering and extended point set generation, effectively maintaining the continuity of topological structures such as closed contours. At the same time, the radius extended point set covers the neighborhood information around the boundary line, avoiding feature loss caused by single point mapping error, enhancing the noise resistance, and providing support for the subsequent calculation of local curvature and normal vector change entropy.

[0043] 2. Accurate identification of high-risk areas to improve optimization robustness

[0044] The present application can accurately quantify the bending degree and geometric change of the target point cloud area by calculating the local curvature and normal vector change entropy. Among them, the curvature can capture the macro bending degree, and the normal vector change entropy can quantify the normal chaos degree to reflect the microscopic complexity, and the two complement each other to identify high penetration risk areas (such as the tie area), significantly improving the robustness and noise resistance of the boundary optimization process.

[0045] 3. Balancing multiple optimization objectives to improve boundary quality

[0046] The present application constructs a target function containing multiple constraints such as fidelity, smoothness and anti-penetration, balancing different optimization needs - fidelity ensures that the boundary line fits the original shape of the dental model, smoothness avoids sharp changes or broken lines in the boundary line, and the anti-penetration term prevents the boundary points from penetrating the model surface. This design avoids the overfitting problem that may be caused by single objective optimization, improves the accuracy, smoothness and anti-penetration effect of the boundary point set, and can better cope with complex geometric scenes (such as ties and corners).

[0047] 4. Improve the practicality in application scenarios

[0048] The boundary line generated by the present application is smooth and does not penetrate while fitting the original shape of the dental model, effectively solving the problems of easy "penetration" of the boundary line in traditional methods and single influence of local or global features on optimization effect, significantly improving the accuracy and safety in 3D printing or modeling. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application will be briefly introduced in the following. Among them, the drawings are only used to show some embodiments of the present application, and not to limit all embodiments of the present application to this.

[0050] Figure 1 is the flowchart of the dental model point cloud boundary line adjustment method of the fusion area feature of the embodiment of the present application. DETAILED DESCRIPTION

[0051] The example schemes of the embodiments of the present application will be described in detail below with reference to the accompanying drawings. Unless otherwise defined, technical terms or scientific terms used in the present application should be interpreted as their common meanings understood by those skilled in the art.

[0052] The embodiment discloses a dental model point cloud boundary line adjustment method fusing regional features, as shown in the figure, comprising the following steps: Figure 1

[0053] Step S101, based on the dental model point cloud data and the generated initial boundary line thereof, adjust the point cloud normal vector direction to be consistent outward, and sort and map the points on the boundary line to generate a target point cloud region and an extended point set.

[0054] The dental model point cloud data of the embodiment of the present application is derived from the oral cavity hard tissue or soft tissue surface data obtained by an intraoral scanner, a cone beam CT or a structured light scanning device.

[0055] (1) Calculate the normal vector of the dental model point cloud using the PCA algorithm, and adjust the direction of the normal vector to be consistent outward based on the viewpoint direction.

[0056] The normal vector is an arrow perpendicular to the surface of the model (such as a dental model), which is used to represent the "orientation" of the surface. In the dental model, the normal vector should be uniformly directed to the "outside" (i.e. away from the inside of the tooth or gum). If the normal vector is inward, subsequent processing (such as boundary generation) may result in penetration or incorrect results.

[0057] The adjustment method is: first set an observation angle (such as the front of the dental model). Then check whether each normal vector is roughly directed at the observer (i.e. outward), if the normal vector is directed away from the observer (inward), then reverse it, if it is directed towards the observer (outward), then keep it unchanged.

[0058] For example: if a person stands in front of the teeth (viewpoint direction = looking at the teeth from the outside), check the normal vector of point A, if it is directed towards the inside of the tooth (away from the person), then reverse it to be directed outward. Finally, all normal vectors are uniformly directed outward, which facilitates subsequent processing (such as boundary generation).

[0059] (2) According to the generated boundary line, first sort the boundary points, then sample the points on the boundary line in any direction and in order along the boundary line, to obtain the points of the boundary line that have been sorted in order.

[0060] ​(3) Calculate the Euclidean distance of each point on the boundary line independently in the tooth model point cloud, use KDtree to accelerate the search, find the nearest point on the tooth model corresponding to the nearest point on the boundary line, and generate the target point cloud region by taking all the corresponding points on the boundary line as the point set. At this time, the points of the ordered boundary line and the points of the target point cloud region are one-to-one corresponding and are also arranged in order.

[0061] (4) Further expand the point set of all points by setting a radius r, and calculate the points within the radius of all points in the point set, taking the points within the radius of all points in the point set as the expanded point set.

[0062] At this point, the normal vector of each point in the point cloud is obtained; the normal vector of each point in the point cloud is outward; the boundary point; the target point cloud region; the expanded point set.

[0063] The effect achieved by this step is as follows:

[0064] The ordered nature of the accurate mapping of the boundary point and the tooth model point is maintained: the boundary point is sampled in spatial order and mapped to the tooth model point cloud, preserving the topological structure (such as the continuity of the closed contour). Robustness of region expansion, noise resistance: expand the point set by radius r to cover the neighborhood information around the boundary line, avoiding feature loss due to single-point mapping error. Feature compatibility: the expanded point set contains both the accurate position of the boundary and the local surface context information, supporting the calculation of curvature / entropy later.

[0065] Step S102, based on the expanded point set, calculating the local curvature and normal vector change entropy of each point in the target point cloud region, and fusing to obtain the local risk weight of the point.

[0066] In order to quantify the bending degree of the local surface of the point cloud in the target point cloud region and measure the degree of confusion of the normal vector distribution, so as to calculate the local curvature and normal vector change entropy of the target point cloud region, in order to identify the high penetration risk area. Further, according to the fusion of the calculated local curvature and normal vector change entropy of the point cloud in the target point cloud region, a risk assessment function is constructed, and finally the local risk weight of the point is obtained.

[0067] (1) Local curvature, used to quantify the bending degree of the local surface of the point cloud, the core idea is: by comparing the normal vector difference and spatial distance of the current point and its neighbor points, to judge whether the region is flat or sharp.

[0068] According to the coordinate values of all points in the target point cloud region and the coordinate values of all points in the expanded region calculated in step S101, the normal vectors corresponding to all points in the target point cloud region and the normal vectors corresponding to all points in the expanded point cloud region, the local curvature of each point is calculated. For each point in the target point cloud region, the calculation process is as follows:

[0069] ① Obtain point p i coordinates and normal vector n i .

[0070] ② Set the number of neighbors k. The value of k can be adjusted according to the actual situation.

[0071] ③ Take point p i Centered on a point, find its k nearest neighbors in the extended point cloud region, forming a neighbor set N. i This leads to the neighbor index set.

[0072] ④ For each neighboring point, calculate the Euclidean distance of the normal vector, calculate the reciprocal of the spatial distance, and multiply them to obtain the contribution of a single neighbor.

[0073] ⑤ Add up the contributions of all neighbors, then divide by the number of neighbors to obtain the local curvature of the current point.

[0074] The following formulas are given:

[0075]

[0076] Where, k i For point p i The curvature estimate (a larger value indicates a more curved surface); k is the number of nearest neighbors, N i For point p i The neighbor set (containing the indices of the k nearest nodes), n i For point p i The normal vector, n j Point p for the neighbor j The normal vector; ∑ is the summation.

[0077] ||n i -n j || represents the Euclidean distance between two normal vectors, which essentially compares the surface orientation changes between the current point and its neighboring points. If the normal vector of the neighboring point is n j With n i If the directional differences are large (such as the sharp edges of the frenulum region), then ||n i -n j A large value indicates high curvature; if the direction is consistent (such as in a flat region), the value approaches 0.

[0078] ||p i -p j || -1 It represents the reciprocal of the Euclidean distance between two points, and the weighted influence of neighboring points. Points that are closer together contribute more, and the geometric features of nearby points have a greater impact on curvature, while the influence of distant points is smaller (reciprocal decay).

[0079] (2) Normal vector variation entropy: used to quantify the degree of microscopic geometric disorder of the local surface of the point cloud, especially good at capturing complex details that curvature cannot describe (such as wrinkles, laces, etc. small undulations). The calculation process of the normal vector variation entropy is as follows:

[0080] ① Get the normal vector of point p i and the neighbor set N i .

[0081] ② For each neighbor point, calculate the cosine similarity of the current point.

[0082] ③ Calculate the normal vector direction entropy of the current point.

[0083] The following formula is used:

[0084]

[0085] Where E i is the normal vector direction entropy of point p i (the larger the value, the more chaotic the direction); n i · n j is the dot product of the normal vectors of point p i and neighbor point p j . cos represents the cosine similarity of the two normal vectors (the value range is [-1, 1], and the absolute value or normalized to [0, 1] is taken to avoid negative values).

[0086] The cosine similarity represents the direction consistency of the two normal vectors, =1 indicates that the directions are completely the same (such as flat surfaces), ≈0 indicates that the directions are perpendicular (such as edges), and =-1 indicates that the directions are opposite (such as the back of a wrinkle). In actual calculation, the absolute value or mapping to [0, 1] is taken to avoid negative values affecting the physical meaning of entropy.

[0087] When the entropy is low (E i ≈0), the neighbor normal vectors are consistent in direction, indicating that the surface is flat or uniformly curved (such as the smooth part of the gum); when the entropy is high (E i is large), the neighbor normal vectors are chaotic in direction, indicating that the surface has microscopic undulations or mutations (such as laces, wrinkles).

[0088] (3) In order to fuse the curvature and entropy features to generate a risk map, a continuous and differentiable risk assessment function is established, the features are fused and activated, the linear weighting is calculated, the Sigmoid function is applied to map the value to 0-1, and the following formula is constructed and calculated:

[0089] w i =σ(αk i +βE i )

[0090]

[0091] where w i is the risk value of the point p i , i.e., the local risk weight of the point; and alpha and beta are feature fusion coefficients of curvature estimation value k i and normal vector direction entropy E i , respectively, and sigma(x) is a sigmoid function.

[0092] The effect realized by the step is as follows:

[0093] The local curvature of the embodiment of the application indicates that there is a sharp geometric bending or direction mutation (such as a belt edge or a corner) near the point, and is sensitive to large-scale undulation. The normal vector change entropy is a measure of the disorder or chaos degree of the neighborhood normal distribution, and is sensitive to small-scale jitter and wrinkles. The mutation degree of different regions is comprehensively evaluated, the feature value on the expanded point cloud is calculated according to the points in the target point cloud region, the neighborhood information is added, and finally the comprehensive consideration is realized.

[0094] The innovation of curvature estimation, the fusion of normal difference and spatial proximity, the normal distance captures the direction mutation of the surface (such as the edge), the spatial distance reciprocal strengthens the influence of the near neighbor points and weakens the interference of the distant points. The supplementary value of the normal vector direction entropy, the entropy formula measures the randomness of the normal distribution, and the high-entropy region (such as the belt and the wrinkle) indicates the micro complexity of the surface, which is independent of the macro curvature. The curvature is sensitive to the explicit bending, and the entropy is sensitive to the micro undulation, and the two are complementary.

[0095] In step S103, a multi-constraint boundary point position optimization model is constructed to unify the fidelity, smoothness and anti-penetration requirements, a target function is constructed and optimized, and a multi-constraint optimization framework is established to balance the fidelity, smoothness and anti-penetration, and an optimized target function is obtained.

[0096] According to the target point cloud region calculated in step S101, the local risk weight of the point in the target point cloud region calculated in step S102, the smoothness item weight coefficient lambda S , and the anti-penetration item weight coefficient lambda p , a target function is constructed, and through continuous iterative calculation, an optimized boundary line point set is finally obtained.

[0097] The specific calculation process is as follows:

[0098] ①The original boundary point set p (the mapping point set of the original boundary line on the dental model point cloud), the boundary point set Q to be optimized (its initial value comes from the mapped original boundary point set p), the weight coefficients lambda S and lambda p , the local risk weight w i , and the dental model point cloud surface S are taken as inputs.

[0099] ② Set the learning rate and the number of iterations.

[0100] ③ According to the set parameters, use the gradient descent algorithm to iteratively calculate the constructed objective function until the specified number of iterations is reached or the set convergence threshold is less than the set convergence threshold. At this time, the points of the boundary point set to be optimized best fit the dental model and will not penetrate the model. The setting of the learning rate and the number of iterations, λ S and the setting of the weight coefficient λ p are adjusted according to the actual situation. For example, the learning rate is generally set between 0.01 and 0.05.

[0101] The calculation formula of the multi-constrained optimization objective function is:

[0102]

[0103] In the formula, p is the original boundary point set, where p i is a point in three-dimensional space; Q is the boundary point set to be optimized, where q i is the variable to be optimized; S is the point cloud surface (point cloud representation) of the dental model, representing an impenetrable geometric surface; λ S is the weight coefficient of the smoothing term (λ S > 0), which controls the smoothing strength of the boundary line; λ p is the weight coefficient of the anti-penetration constraint (λ p > 0), which controls the global strength of the anti-penetration constraint; φ(q i , S) is the penetration penalty function, which defines the penetration degree of point q i relative to the surface S, and N is the total number of points in the original boundary point set p (and the corresponding boundary point set to be optimized Q)

[0104] λ p controls the strength of the anti-penetration constraint, the larger the value, the more strictly the optimization process will punish the behavior of the boundary point penetrating the surface, ensuring that q i is always located outside the dental model, the smaller the value, the weaker the influence of the anti-penetration constraint, which may allow slight penetration. λ S is used to control the strength of the boundary line smoothing degree, the larger the value, the smoother the optimized boundary line, the smaller the value, the weaker the influence of the smoothing constraint.

[0105] is the data fidelity term, which ensures that the optimized point q i is close to the original point p i , avoiding excessive deviation from the input data, thereby preserving the geometric characteristics of the original mapping and preventing the optimized boundary line from deviating excessively from the actual shape of the dental model.

[0106] is the smoothing term, which ensures the smoothness of the boundary point sequence; q i-1 , q i, q i+1 are three adjacent points in the set of boundary points to be optimized, q i is the current point, q i-1 , q i+1 are its predecessor and successor points, ||q i-1 - 2q i + q i+1 || 2 is the second-order difference term, which is used to approximate the local curvature of the boundary line, and its physical meaning is to measure the bending degree between three consecutive points; if the value of this term is small, it means that the three points are arranged in a straight line, and the boundary line is smooth in this area; if the value is large, it means that there is a sharp turning or fluctuation, and the smoothness will be enhanced by penalizing this term during optimization.

[0107] is the anti-penetration term, which prevents the point q i from penetrating the surface S, and the weight w i makes the constraint stronger in the mutation area (high w i ) and weaker in the flat area (low w i ).

[0108] The penetration penalty function φ(q i , S) is defined as:

[0109]

[0110] where v i is the nearest neighbor point of q i on the surface S (obtained by nearest neighbor search); n i is the normal vector (unit vector) of point v i , which represents the local geometric direction of the surface S at v i .

[0111] When the point q i is located on the outside of the surface S (i.e. the normal vector direction) (n i · (q i - v i ) ≥ 0), there is no penalty; when the point q i penetrates the surface S (inside), the penalty is based on the square of the penetration depth (projected onto the normal vector direction).

[0112] When the penetration penalty function detects that a boundary point q i is incorrectly optimized to the inside of the dental model (i.e. n i · (q i - v i ) < 0), it will move q i outward along the normal vector direction n i through the feedback mechanism of gradient descent until it satisfies n i · (q​i -v i )≥0. This process essentially re-corrects q i that were misjudged as "internal points" to points on the boundary line, ensuring that they strictly adhere to the outside of the dental model surface, where it is understood that boundary points must be located on the outside of the dental model surface (normal vector direction) to represent the true anatomical profile (e.g., gum line).

[0113] It should be noted that when the original boundary point set and the boundary point set to be optimized are input into the objective function for calculation, the original boundary point set and the boundary point set to be optimized are in one-to-one correspondence in order, one original boundary point set corresponds to one boundary point set to be optimized, and one local risk weight w i .

[0114] For example: for the first iteration optimization of the objective function, the boundary point set to be optimized is initialized, and other parameters are fixed after being set. In the iteration of the objective function, only the boundary point set to be optimized is continuously optimized. When calculating the value of the objective function for the first time, the initial boundary point set to be optimized is used as the initial parameter to calculate the value of the objective function (using the existing gradient descent algorithm). In the calculation process, the i-th point p i in the original boundary point solution corresponds to the i-th point q i in the boundary point set to be optimized, and the local risk weight w i of the i-th point. After the point set is updated, the boundary point set to be optimized is updated to a new boundary point set to be optimized, which is input into the objective function to calculate the value of the objective function, and the iteration is repeated until a specified number of iterations is reached.

[0115] The effects achieved by this step are as follows:

[0116] Multiple objectives are optimized comprehensively: By constructing the objective function, multiple constraints (fidelity, smoothness, and anti-penetration) are combined together to ensure that each objective is reasonably balanced during the optimization process. This can avoid excessive pursuit of a certain target while ignoring other targets, ensuring that the final position of the boundary point meets the design requirements and avoids the phenomenon of penetration.

[0117] Fidelity guarantee: Fidelity ensures that the generated boundary line is as faithful as possible to the original shape of the dental model. The fidelity term in the objective function ensures that the generated boundary line can closely follow the original contour of the dental model.

[0118] Smoothness guarantee: The smoothness constraint ensures that the generated boundary line during the optimization process will not have sharp changes or broken lines, making the generated boundary line more natural and smooth, consistent with the surface characteristics of the real object. Through the smoothness term, irregular jumps or fluctuations of the boundary points are avoided, making the result more smooth and stable.

[0119] Prevent the mold from being penetrated: The introduction of the anti-penetration item ensures that the boundary points will not penetrate the dental mold surface during the optimization process. This is a key step to avoid generating unrealistic surfaces or model errors, especially when performing three-dimensional modeling or 3D printing, the effect of anti-penetration is crucial.

[0120] Multi-constraint processing of boundary point position optimization

[0121] Avoid local extremum: During the optimization process, the use of multi-constraint objective function can prevent the model from falling into local extremum, which is particularly important for complex surfaces or noisy point clouds. The multi-constraint objective function considers smoothness, fidelity and anti-penetration, avoiding unreasonable optimization results caused by single constraint.

[0122] Improve fault tolerance: The multi-constraint method makes the system have higher fault tolerance when facing noise and error, even if the point cloud data has certain noise, the system can still provide stable optimization results, so as to better adapt to actual application.

[0123] Robustness: In the process of dental mold point cloud processing, the robustness of the optimization method is particularly important, because the surface of the dental mold may have a more complex geometry. Through multi-constraint optimization, the optimized boundary points not only maintain accuracy, but also maintain smoothness and continuity, and are not affected by local noise or data problems.

[0124] Step S104, according to the optimized boundary point set, the final boundary line is generated by connecting in turn.

[0125] According to the boundary point set obtained after the iterative optimization is completed, the boundary points are connected in turn to finally obtain the boundary line, which realizes the conversion of the optimized discrete boundary points into smooth and continuous closed boundary line.

[0126] Corresponding to the above method, the embodiment also discloses a dental mold point cloud boundary line adjustment system fusing regional features, comprising:

[0127] A point cloud region acquisition module is used to adjust the direction of the point cloud normal vector to be consistent outward on the basis of the dental mold point cloud data and the generated initial boundary line, sort and map the points on the boundary line, and generate a target point cloud region and an extended point set;

[0128] A risk weight calculation module is used to calculate the local curvature and normal vector change entropy of each point in the target point cloud region based on the extended point set, and fuse to obtain the local risk weight of the point;

[0129] A boundary point set optimization module is used to construct a multi-constraint optimization objective function, combine fidelity, smoothness and anti-penetration constraints, and iteratively optimize the boundary point set;

[0130] The boundary line generation module is configured to sequentially connect the optimized boundary point set to generate a final boundary line.

[0131] The relative arrangement of components and steps, numerical expressions, and numerical values set forth in the examples are not intended to limit the scope of the application unless specifically stated otherwise.

[0132] The various embodiments described in the specification are progressive in nature, and each embodiment highlights the differences from other embodiments. The same or similar parts between various embodiments can be mutually referred to. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts are referred to the method part.

[0133] The units and method steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example are described in the above description in general. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation does not exceed the scope of the present application.

[0134] Those skilled in the art can understand that all or part of the steps of the above method can be instructed by a program to complete the relevant hardware, and the program can be stored in a computer readable storage medium, such as a read-only memory, a magnetic disk or an optical disk. Alternatively, all or part of the steps of the above embodiments can also be implemented using one or more integrated circuits, and accordingly, each module / unit in the above embodiments can be implemented in the form of hardware or in the form of a software function module. The present application is not limited to any specific form of combination of hardware and software.

[0135] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present application, which are used to illustrate the technical solutions of the present application, but not to limit it. The protection scope of the present application is not limited to this. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can make modifications or easily think of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed by the present application, or make equivalent replacements to some of the technical features. These modifications, changes or replacements do not cause the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of adjusting a boundary line of a dental model point cloud fused with a region feature, characterized by, The method comprises the following steps: Step 1: Based on the dental model point cloud data and the generated initial boundary line thereof, the direction of the point cloud normal vector is adjusted to be consistent outward, and the points on the boundary line are sorted and mapped to generate a target point cloud region and an extended point set; Step 2: Based on the extended point set, the local curvature and the normal vector change entropy of each point in the target point cloud region are calculated, and the local risk weight of the point is obtained after fusion; Step 3: A multi-constraint optimization objective function is constructed, and the boundary point set is iteratively optimized in combination with the fidelity, smoothness and anti-penetration constraints; Step 4: According to the optimized boundary point set, the final boundary line is generated by sequentially connecting.

2. The fusion region feature's dental model point cloud boundary line adjustment method according to claim 1, wherein, The dental model point cloud data in step 1 is obtained from the oral scanning instrument, cone beam CT or structured light scanning device.

3. The method of claim 1, wherein the method further comprises: The step 1 specifically comprises: The normal vector of the dental model point cloud is calculated using the PCA algorithm, and the normal vector is adjusted based on the viewpoint direction to be consistent outward; The points on the boundary line are sequentially sampled and mapped to the dental model point cloud, and the nearest points on the dental model corresponding to the points on the boundary line are found through KDtree accelerated search, and the corresponding points on the boundary line are taken as the point set to generate the target point cloud region; The points within the radius r of the points in the target point cloud region are set as the neighborhood expansion, and all points within the radius of the points in the point set are taken as the extended point set.

4. The method of claim 1, wherein the method further comprises: The calculation formula of the local curvature in step 2 is: where k i is the curvature estimate of point p i , k is the number of neighboring points, N i is the neighbor set of point p i , n i is the normal vector of point p i , n j is the normal vector of neighbor point p j , ||n i -n j || represents the Euclidean distance of two normal vectors, the larger the value represents the greater the difference in direction; ||p i -p j || -1 represents the inverse of the Euclidean distance of two points, the closer the distance, the greater the weight.

5. The method of claim 1, wherein the method further comprises: The calculation formula of the normal vector change entropy in step 2 is: where E i is the normal direction entropy of point p i , n i · n j is the dot product of the normal vectors of point p i and its neighbor point p j , and is the cosine similarity of two normal vectors, and log is the natural logarithm.

6. The method of claim 1, wherein the method further comprises: The calculation formula of the local risk weight in step 2 is: w i = σ(αk i + βE i ) where w i is the risk value of the point p i , i.e., the local risk weight of the point; α and β are the feature fusion coefficients of curvature estimation value k i and normal vector direction entropy E i , respectively, and σ(x) is a sigmoid function.

7. The fusion region feature's dental model point cloud boundary line adjustment method according to claim 1, wherein, The multi-constraint optimization objective function in step 3 is: where p i is the original boundary point; Q is the set of boundary points to be optimized, where q i is the variable to be optimized, q i-1 , q i , q i+1 are three adjacent points in the set of boundary points to be optimized; λ S is the weight coefficient of the smoothing term, λ p is the weight coefficient of the anti-penetration term, φ(q i , S) is the penetration penalty function, S is the dental mold point cloud surface, represents the non-penetrable geometric plane, and N is the total number of points in the original boundary point set.

8. The method of claim 7, wherein the method further comprises: The penetration penalty function φ(q i i The penetration degree with respect to the tooth model point cloud surface S is expressed as:​ Among them, v i For point q i The nearest neighbor point on surface S, n i For point v i The normal vector of point q; i There is no penalty when the point q is located outside the surface S; when the point q is outside the surface S, there is no penalty i When penetrating a surface S, the penalty is based on the square of the penetration depth.

9. The fusion region feature's dental model point cloud boundary line adjustment method according to claim 1, wherein, In step 3, the gradient descent algorithm is used to iteratively optimize the objective function until the set convergence condition is reached.

10. A system for adjusting a boundary line of a dental model point cloud fused with regional features, the system comprising: It comprises: A point cloud region acquisition module is used to adjust the direction of the point cloud normal vector to be consistent outward based on the dental model point cloud data and the generated initial boundary line thereof, and the points on the boundary line are sorted and mapped to generate a target point cloud region and an extended point set; A risk weight calculation module is used to calculate the local curvature and the normal vector change entropy of each point in the target point cloud region based on the extended point set, and the local risk weight of the point is obtained after fusion; A boundary point set optimization module is used to construct a multi-constraint optimization objective function, and the boundary point set is iteratively optimized in combination with the fidelity, smoothness and anti-penetration constraints; A boundary line generation module is used to generate a final boundary line by sequentially connecting according to the optimized boundary point set.

Images