A method, apparatus and computer program product for generating three-dimensional mesh sequences

By constructing an intermediate mesh with a fixed topology and a continuous-time convolution kernel, the interpolation problem of a three-dimensional mesh sequence with a variable topology is solved, achieving continuous interpolation in both working conditions and time dimensions, suppressing jitter and volume distortion, and ensuring the morphological continuity of the three-dimensional structure.

CN123118974APending Publication Date: 2026-11-17SHANGHAI QIMENG NETWORK TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611545816.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-08
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing 3D mesh frame deformation interpolation techniques cannot achieve continuous interpolation in both the working condition and time dimensions between topologically variable 3D mesh sequences, resulting in jitter, trailing, and volume distortion in the interpolation results, and making it impossible to generate 3D mesh frames at any intermediate time.

Method used

By constructing an intermediate mesh with a fixed topology, and using interpolation coefficients based on operating conditions and continuous-time convolution kernels, a unified vertex identity and time trajectory weighting of the 3D mesh sequence are achieved, generating a smooth 3D mesh sequence and suppressing jitter and volume distortion.

Benefits of technology

It achieves continuous interpolation of working conditions and time in two dimensions between topologically variable 3D mesh sequences, suppresses the jitter and trailing of interpolation results, and ensures the continuity and physical authenticity of the 3D structure morphology.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118974A_ABST
    Figure CN123118974A_ABST
Patent Text Reader

Abstract

This application discloses a method, device, and computer program product for generating three-dimensional mesh sequences, belonging to the field of computer graphics and three-dimensional geometry processing technology. The method includes: constructing an intermediate mesh based on a preset number of vertices; using the intermediate mesh to perform single-frame spherical mapping on each three-dimensional mesh frame in a first and second sequence to obtain the target position of each fixed vertex on the surface of each three-dimensional mesh frame; performing weighted interpolation on the target positions based on specified working condition interpolation coefficients to obtain a candidate three-dimensional mesh sequence; performing temporal convolution on the time trajectory to obtain the vertex positions of each fixed vertex at specified consecutive time points; and obtaining a smooth three-dimensional mesh sequence based on triangulation relationships and vertex positions. This application can achieve bicontinuous interpolation of working conditions and time for topologically variable three-dimensional mesh sequences, effectively suppressing interpolation result jitter, trailing, and volumetric distortion.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of computer graphics and three-dimensional geometry processing, and in particular to a method, apparatus and computer program product for generating three-dimensional mesh sequences. Background Technology

[0002] In applications such as Computational Fluid Dynamics (CFD) simulation, structural simulation, medical scanning, and 3D reconstruction, 3D dynamic structures are typically presented as multi-time-step 3D mesh sequences. Each 3D mesh sequence consists of multiple 3D mesh frames, and each 3D mesh frame corresponds to one time step. Due to limitations in simulation computing power, storage resources, and acquisition equipment, engineering projects can usually only obtain 3D mesh sequences corresponding to a limited number of discrete working conditions and discrete time nodes. Furthermore, the 3D mesh frames in each working condition's 3D mesh sequence are mostly generated independently from the simulation data or scan data of the current time step, resulting in differences in the number of vertices, vertex numbers, and triangulation relationships among different 3D mesh frames within the same 3D mesh sequence. It is also difficult to directly establish vertex-by-vertex correspondences between 3D mesh sequences corresponding to different parameter conditions.

[0003] Existing 3D mesh frame deformation interpolation techniques, based on establishing a corresponding mapping relationship between static source 3D mesh frames and target 3D mesh frames, can only achieve shape gradual interpolation for a single pair of static 3D mesh frames. This cannot adapt to the overall interpolation requirements of two sets of topologically variable 3D mesh sequences. Furthermore, existing 3D mesh frame deformation interpolation techniques cannot establish a unified vertex identity association across frames, making it difficult to smoothly optimize the temporal motion trajectory of 3D mesh frame vertices. This easily leads to problems such as high-frequency jitter, local flickering, and shape discontinuities in the interpolated animation. Simultaneously, traditional fixed convolution smoothing methods cannot adapt to the convergence characteristics of the first and last 3D mesh frames in a 3D mesh sequence, easily resulting in deformation tailing.

[0004] Furthermore, existing 3D mesh frame deformation interpolation techniques can only output the 3D mesh frame shape corresponding to the original discrete time nodes, failing to fill the gaps in the inter-frame shape under large time step intervals, making it difficult to achieve continuous interpolation in the time dimension and unable to directly generate 3D mesh frames at arbitrary intermediate times. Conventional linear interpolation schemes also cannot guarantee that the volume of closed 3D mesh frames changes regularly with the operating parameters, and the independent volume correction method for each 3D mesh frame will cause abrupt changes in the inter-frame correction amount, failing to meet the high-precision simulation interpolation requirements of physical objects such as fluid interfaces like bubbles and droplets, and soft structures. Therefore, how to achieve continuous interpolation in both the operating condition and time dimensions between topologically variable 3D mesh sequences remains an urgent problem to be solved. Summary of the Invention

[0005] The purpose of this application is to overcome the shortcomings of the prior art and provide a method, device and computer program product for generating three-dimensional mesh sequences. By constructing an intermediate mesh with a fixed topology, and using the working condition interpolation coefficients, continuous interpolation of the working condition dimension and the time dimension is achieved when each three-dimensional mesh frame in the input three-dimensional mesh sequence has different number of vertices, vertex numbers and triangulation relationships. This also suppresses the jitter, trailing and volume distortion of the interpolation results.

[0006] In a first aspect, this application provides a method for generating a three-dimensional mesh sequence, comprising the following steps: An intermediate mesh is constructed based on a preset number of vertices, and the numbering, spherical coordinates, and triangulation relationships of each fixed vertex of the intermediate mesh are frozen. Using the intermediate mesh, single-frame spherical mapping is performed on each three-dimensional mesh frame in the first and second sequences to obtain the target position of each fixed vertex of the intermediate mesh on the surface of each three-dimensional mesh frame; Based on the specified interpolation coefficients, the target position is weighted and interpolated to obtain a candidate 3D mesh sequence; the candidate positions of the same fixed vertex in each 3D mesh frame in the candidate 3D mesh sequence constitute the time trajectory of the fixed vertex; Temporal convolution is performed on the time trajectory to obtain the vertex positions of each fixed vertex at a specified consecutive time. The temporal convolution keeps the sampling time of each three-dimensional mesh frame unchanged, and continuous temporal reconstruction is performed by moving the center of the continuous temporal convolution kernel to the specified consecutive time. Based on the triangular connection relationship and the vertex position, the interpolation coefficients for the specified working condition and the smooth three-dimensional mesh sequence corresponding to the specified continuous time are obtained.

[0007] By freezing fixed vertex numbers, spherical coordinates, and triangulation relationships, the problem of inconsistent topology and lack of unified vertex identity among input 3D mesh frames is solved. Using specified interpolation coefficients, the target positions of the same fixed vertex on each of the first and second 3D mesh frames are weighted to obtain candidate 3D mesh sequences, thus unifying the output to a fixed topology beforehand.

[0008] Optionally, a temporal convolution is performed on the time trajectory to obtain the vertex positions of each fixed vertex at specified consecutive time points. The temporal convolution maintains the sampling time of each 3D mesh frame unchanged, and continuous temporal reconstruction is performed by moving the center of the continuous temporal convolution kernel to the specified consecutive time points, including: Keep the sampling time of each 3D mesh frame in the candidate 3D mesh sequence unchanged, and configure continuous-time convolution kernels; The center of the continuous-time convolution kernel is moved to a specified continuous time interval to obtain the center position of the convolution kernel. The corresponding continuous support radius is determined based on the position of the specified continuous time in the candidate 3D mesh sequence, thereby obtaining the support range of the continuous-time convolution kernel; wherein the weight of the continuous-time convolution kernel is zero at the support boundary; Based on the support range, select 3D mesh frames participating in temporal convolution from the candidate 3D mesh sequence, delete 3D mesh frames that exceed the temporal range boundary of the candidate 3D mesh sequence, and obtain a set of effective 3D mesh frames; Based on the distance between each three-dimensional mesh frame in the effective three-dimensional mesh frame set and the center position of the convolution kernel, the convolution weights of each three-dimensional mesh frame in the effective three-dimensional mesh frame set are determined, and the convolution weights are normalized to obtain normalized convolution weights. The time trajectory of the fixed vertex is weighted according to the normalized convolution weights to obtain the vertex position of the fixed vertex at the specified consecutive time points.

[0009] By keeping the sampling time of each 3D mesh frame constant, the center of the continuous-time convolution kernel is moved to a specified continuous time step. Normalized convolution weights are determined based on the distance between each 3D mesh frame and the kernel center position, as well as the continuous support radius, and weighted on the temporal trajectory. This reconstructs the vertex positions of continuous time steps across discrete 3D mesh frames, reducing high-frequency jitter caused by frame-by-frame correspondence errors. Moving the center of the continuous-time convolution kernel allows for the reconstruction of the mesh at any real time step using fixed discrete frame positions, supplementing the geometry between adjacent simulation output time steps.

[0010] Optionally, an effective set of 3D mesh frames : in, The total number of 3D mesh frames in the candidate 3D mesh sequence. This is the initial set of 3D mesh frames. Indicates the 3D mesh frame index; The effective three-dimensional mesh frame set The smallest 3D mesh frame index actually retained in the middle and the maximum 3D mesh frame index They are respectively: ; The initial three-dimensional mesh frame index set : in, express t The continuous support radius at any given time. Represents a set of integers.

[0011] Optionally, based on the distance between each 3D grid frame and the center position of the convolution kernel. and t Continuous support radius at time step The unnormalized convolution weights for each 3D mesh frame are calculated using the unnormalized weight function of the continuous-time convolution kernel. : ; For the effective set of 3D mesh frames The unnormalized convolution weights of all 3D mesh frames are summed, and the sum is used to normalize the convolution weights of each 3D mesh frame to obtain normalized convolution weights. : in, Represents the set of valid 3D mesh frames The index is s The convolution weights of the 3D mesh frame, This represents the unnormalized weight function. This represents the index of the three-dimensional mesh frame.

[0012] Optionally, the continuous support radius adopts a fixed width mode or a variable width mode based on its position in the candidate 3D mesh sequence at specified consecutive times; in the fixed width mode, the continuous support radius... In variable width mode, based on the width envelope Determine the continuous support radius ,in, is the maximum discrete convolution radius.

[0013] By continuously supporting the radius of the candidate 3D mesh sequence at specified consecutive time points, and by setting the weight of the continuous kernel to zero at the support boundary, the first and last 3D mesh frames converge to the candidate positions that have not undergone temporal convolution, thus avoiding tailing. 3D mesh frames that exceed the time range boundary of the sequence are deleted and re-normalized to correctly handle the boundary.

[0014] Optionally, the vertex position of each fixed vertex at specified consecutive times. The expression is: in, Indicates the first i The fixed vertex is in the... j Candidate positions on a 3D grid frame r Indicates the interpolation coefficient for a specified operating condition. Indicates the first j Normalized convolution weights for each 3D grid frame, This represents the set of valid 3D mesh frames.

[0015] Optionally, a volume constraint step is also included before outputting the smoothed 3D mesh sequence: The volume correction amount is determined along the normal direction of a fixed vertex based on the volume deviation between the target volume and the actual volume of the intermediate grid at a specified continuous time. The volume correction amount is time-series smoothed to obtain the smoothed volume correction amount; The vertex position is updated along the normal direction of the fixed vertex based on the smoothed volume correction amount until the volume deviation between the actual volume and the target volume meets the preset error range.

[0016] The target volume is determined by using the same normalized convolution weights as the temporal convolution. The volume is iteratively corrected along the fixed vertex normal direction based on the volume deviation, and the volume correction amount is smoothed along the time axis so that the actual volume converges to the target volume error range. The temporal smoothing of the correction distance reduces the inter-frame jump introduced by the volume correction, thus solving the problems of volume distortion and correction jump.

[0017] Optionally, the three-dimensional mesh sequence generation method further includes: Obtain the first three-dimensional mesh sequence and the second three-dimensional mesh sequence under different discrete working conditions; The first three-dimensional mesh sequence and the second three-dimensional mesh sequence are time-aligned to obtain a time-matched first sequence and a second sequence.

[0018] In a second aspect, this application also provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs; and, when the one or more programs are executed by the one or more processors, causing the one or more processors to perform the steps of the three-dimensional mesh sequence generation method as described in any one aspect.

[0019] Thirdly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the three-dimensional mesh sequence generation method as described in any one of the first aspects.

[0020] This application provides a method, device, and computer program product for generating three-dimensional mesh sequences. By constructing an intermediate mesh with unchanged vertex numbers and triangular connections, and obtaining the target positions of fixed vertices of the intermediate mesh on each three-dimensional mesh frame of the first and second three-dimensional mesh sequences through single-frame spherical mapping, a candidate three-dimensional mesh sequence is obtained by weighted interpolation of the target positions according to the working condition interpolation coefficients. For the time trajectory of the same fixed vertex in the candidate three-dimensional mesh sequence, while keeping the sampling time of each three-dimensional mesh frame constant, the center of the continuous temporal convolution kernel is moved to a specified continuous time. Normalized convolution weights are determined based on the distance between each three-dimensional mesh frame and the center position of the convolution kernel, and temporal convolution is performed. Combined with the triangular connections, a smooth three-dimensional mesh sequence is output. This method can achieve dual continuous interpolation in both the working condition and time dimensions even when the topology of each three-dimensional mesh frame in the input three-dimensional mesh sequence is inconsistent. It effectively suppresses interpolation result jitter, trailing, and volume distortion, ensuring the continuity and physical realism of the temporal changes in the three-dimensional structure morphology.

[0021] To make the above-mentioned features and advantages of the invention more apparent and understandable, specific embodiments are described below, and detailed descriptions are provided in conjunction with the accompanying drawings. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of a three-dimensional mesh sequence generation method provided in one embodiment of this application.

[0024] Figure 2 A flowchart of step S6 in the three-dimensional mesh sequence generation method provided in this application.

[0025] Figure 3 This is an internal structural diagram of an electronic device in yet another embodiment of this application. Detailed Implementation

[0026] To make the objectives and technical solutions of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the described embodiments of this application without creative effort are within the scope of protection of this application.

[0027] In fields such as computational fluid dynamics simulation, structural mechanics simulation, medical 3D scanning, and digital twin visualization, 3D dynamic structures are typically presented as multi-time-step 3D mesh sequences. A 3D mesh sequence refers to a time series of 3D surface morphology obtained by sampling the surface or interface of a dynamic physical object using triangular meshes step-by-step. The physical object corresponding to the 3D mesh sequence can be an underwater cavitation gas-liquid interface, droplet, or free liquid surface in computational fluid dynamics simulation; a deformable surface in structural simulation; an organ or soft tissue surface in medical scanning; or a flexible component surface in digital twins. The dynamic morphology of the physical object under different discrete operating conditions is represented by the corresponding 3D mesh sequence.

[0028] The discrete operating conditions are characterized by operating condition parameters, which include, but are not limited to, temperature, pressure, inlet gas volumetric flow rate, load, material parameters, or physiological parameters.

[0029] In one embodiment, see Figure 1 This application provides a method for generating a three-dimensional mesh sequence, which may include the following steps: steps S1 to S7.

[0030] Step S1: Obtain the first three-dimensional mesh sequence and the second three-dimensional mesh sequence under different discrete working conditions.

[0031] Step S2: Time-align the first three-dimensional mesh sequence and the second three-dimensional mesh sequence to obtain a time-matched first sequence and second sequence.

[0032] Step S3: Construct an intermediate mesh according to a preset number of vertices, and freeze the numbering, spherical coordinates, and triangular connection relationships of each fixed vertex of the intermediate mesh.

[0033] Step S4: Using the intermediate mesh, perform single-frame spherical mapping on each three-dimensional mesh frame in the first sequence and the second sequence to obtain the target position of each fixed vertex of the intermediate mesh on the surface of each three-dimensional mesh frame.

[0034] Step S5: Based on the specified interpolation coefficients, perform weighted interpolation on the target position to obtain a candidate 3D mesh sequence; the candidate positions of the same fixed vertex in each 3D mesh frame in the candidate 3D mesh sequence constitute the time trajectory of the fixed vertex.

[0035] Step S6: Perform temporal convolution on the time trajectory to obtain the vertex positions of each fixed vertex at a specified continuous time. The temporal convolution keeps the sampling time of each three-dimensional mesh frame unchanged, and performs continuous temporal reconstruction by moving the center of the continuous temporal convolution kernel to the specified continuous time.

[0036] Step S7: Based on the triangular connection relationship and the vertex position, obtain the interpolation coefficients for the specified working condition and the smooth three-dimensional mesh sequence corresponding to the specified continuous time.

[0037] This application enables continuous interpolation across two dimensions—working condition and time—between topologically variable 3D mesh sequences. The key lies in transforming the interpolation problem between topologically variable input 3D mesh sequences into how the position of the same fixed vertex on a fixed-topology intermediate mesh continuously changes with working condition and time. By setting an intermediate mesh and providing unified fixed vertex numbers and fixed triangulation relationships, differences in the number, numbering, and triangulation relationships of vertices in each input 3D mesh frame no longer hinder interpolation. Single-frame spherical mapping and its inverse mapping find a unique target position for each fixed vertex on each input 3D mesh frame, thus establishing a cross-frame and cross-working-condition correspondence under a unified vertex identity. Based on this, the interpolation coefficients for the specified working conditions are used to weight the target position of the same fixed vertex on the corresponding 3D mesh frames of the two endpoint sequences, generating a candidate 3D mesh sequence for any intermediate working condition, thus making the working condition dimension continuous. Temporal convolution, under the same vertex identity, weights the temporal trajectory of the same fixed vertex, and the center of the continuous temporal convolution kernel can be moved to any specified continuous time. The weights change continuously with the distance between each 3D mesh frame and the center position of the kernel. Therefore, the vertex position at any time can be reconstructed between discrete 3D mesh frames, making the temporal dimension continuous. By using temporal convolution instead of deforming or smoothing the two endpoint 3D mesh sequences independently frame by frame, high-frequency jumps in adjacent candidate 3D mesh frames that do not conform to the main motion trend can be weighted and smoothed, while preserving the main morphological change trend of the sequence, thereby suppressing surface jitter, flickering and local deformation discontinuities. Furthermore, by employing variable-width continuous-time convolution kernels, the convergence of the first and last three-dimensional mesh frames is ensured, and the weights continuously return to zero when frames enter or leave the support range, avoiding jumps and trailing. Volume constraints are also used to ensure that the actual volume of the intermediate form changes according to a set rule based on the interpolation coefficients for specified operating conditions, avoiding jumps and volume distortions in inter-frame correction caused by independent volume correction at specified consecutive time intervals, thus guaranteeing consistent physical volume. Therefore, this application achieves dual continuous interpolation by unifying the interpolation operation to fixed vertices of a fixed topology and continuously weighting along both the operating condition and time dimensions. The specified operating condition interpolation coefficients and specified continuous time can be set independently, allowing discrete operating conditions and discrete time to be simultaneously expanded into two continuously queryable dimensions.

[0038] In step S1, please refer to Figure 1 In step S1, the first three-dimensional mesh sequence and the second three-dimensional mesh sequence under different discrete working conditions are obtained.

[0039] This method obtains a three-dimensional mesh sequence of the same physical object under different discrete working conditions, including a first three-dimensional mesh sequence corresponding to the first parameter condition and a second three-dimensional mesh sequence corresponding to the second parameter condition.

[0040] Each 3D mesh sequence consists of multiple 3D mesh frames, each corresponding to a time step, representing the 3D form of the physical object at the corresponding time step under the discrete condition. Each 3D mesh frame is a closed triangular mesh with a consistent orientation; when spherical mapping is subsequently used, each 3D mesh frame is a connected, orientable, zero-genus surface that is homeomorphic to a sphere. Different 3D mesh frames within the same 3D mesh sequence, as well as between two 3D mesh sequences, are allowed to have different numbers of vertices, vertex numbers, triangular connections, and mesh densities.

[0041] It is understood that this method is not limited to two three-dimensional mesh sequences. When there are three or more discrete working conditions, the three-dimensional mesh sequences corresponding to each discrete working condition can be obtained separately, and interpolation can be performed between any two discrete working conditions or between multiple discrete working conditions in the same way.

[0042] In step S2, please refer to Figure 1 In step S2, the first three-dimensional mesh sequence and the second three-dimensional mesh sequence are time-aligned to obtain a time-matched first sequence and second sequence.

[0043] As an example, when the input first and second 3D mesh sequences have the same frame rate and the same number of frames, they are directly aligned according to the same frame index; when the frame rate or number of frames is different, the two sequences are resampled to a common time step based on the normalized simulation time, physical time step, or a predetermined event time. T At each time position, a first sequence A and a second sequence B with temporal matching are obtained. After time alignment, both the first sequence A and the second sequence B contain... T There are 3D mesh frames, denoted as follows: and .

[0044] It should be noted that the time alignment is used to make the first sequence A's first... k A 3D mesh frame The second sequence B's first k A 3D mesh frame This represents a comparable evolutionary stage under two parameter conditions, without altering the original mesh topology of any input 3D mesh frame.

[0045] In step S3, please refer to Figure 1 In step S3, an intermediate mesh is constructed according to a preset number of vertices, and the numbering, spherical coordinates, and triangular connection relationships of each fixed vertex of the intermediate mesh are frozen.

[0046] Specifically, the system receives a user-specified preset number of vertices N, and uses this preset number of vertices N as the final number of vertices in the intermediate mesh. The preset number of vertices N is not limited by the icosahedral subdivision level and can be set to thousands, tens of thousands, or even larger values ​​depending on the required precision.

[0047] Furthermore, based on the preset number of vertices N, N approximately uniformly distributed unit spherical vertices are generated, resulting in a unit spherical vertex set. .

[0048] In one implementation, the unit spherical vertex set can be generated using the Fibonacci spherical point set. U Let the golden angle be... Generate the generation sequence number of each unit spherical vertex. i The fixed vertices of the intermediate mesh are numbered. For the unit sphere vertex set... U The Middle i Each unit spherical vertex Calculate vertical height Horizontal projection radius Horizontal rotation angle And thus obtain the first i Coordinates of the vertices of a unit sphere .

[0049] As an example, the first i Vertical height of each unit sphere vertex The expression is: in, N This indicates the number of vertices on a unit sphere.

[0050] As an example, the first i The horizontal projection radius of a unit sphere vertex The expression is: .

[0051] As an example, the first i Horizontal rotation angle of each unit spherical vertex The expression is: in, It is the golden corner.

[0052] Furthermore, for the unit spherical vertex set U Calculate the 3D convex hull ,in This represents the convex hull operation. The three-dimensional convex hull... H The boundary is composed of multiple triangular faces, and the three vertices of each triangular face are the vertex set of the unit sphere.U The unit spherical vertices in the diagram are numbered according to the fixed vertex numbering system, i.e., the first vertex is the first unit spherical vertex. i The unit spherical vertex corresponds to the first i A fixed vertex.

[0053] As an example, the specific steps for constructing a 3D convex hull include: selecting four non-coplanar unit spherical vertices to form an initial tetrahedron, and determining the orientation coefficient based on the normal. Set the four current convex hull faces of the initial tetrahedron to face outwards; record the unit spherical vertex located outside each current convex hull face as the outer point of that current convex hull face, thus obtaining the outer point set; select the unit spherical vertex with the largest distance from the current convex hull face from the current convex hull faces with outer points as the expansion point; delete all current convex hull faces visible from the expansion point, and keep the current convex hull faces that have not been deleted as the retained convex hull faces, and extract the horizon edge between the visible region and the retained region; connect the expansion point to each horizon edge to generate a new triangle face; update the adjacency relationship of the triangle face and the outer point set; repeat the above process until all convex hull faces have no outer points, thus obtaining the final three-dimensional convex hull.

[0054] Ultimately, each triangle on the boundary of the 3D convex hull consists of three fixed vertex numbers, and these numbered triplets form the set of fixed triangles in the intermediate mesh. F 3. If the final 3D convex hull contains polygonal faces, then the fixed vertex with the smallest number in the polygonal face is used as the center of the sector, and deterministic triangulation is performed according to the boundary order of the polygonal faces, without adding any new fixed vertices.

[0055] As an example, for the fixed triangular facet set F The triangular face in 3 Calculate the normal orientation determination coefficient: , in, , , Represent the unit spherical point set respectively U The Middle i The, the j The, the k The coordinates of the nth unit spherical point, i.e., the nth unit of the middle grid. i The, the j The, the k The coordinates of a fixed vertex. If the normal orientation determination coefficient... Then exchange the first j The fixed vertex and the first k The numbering order of the fixed vertices makes the triangle face The normal to the sphere points outwards from the unit sphere.

[0056] Furthermore, based on the fixed triangular facet set F 3. Establish a set of undirected edges E 3. The intermediate mesh is then validated. Validation conditions include: the set of undirected edges. E Each edge in step 3 is connected to exactly two triangular faces; the intermediate mesh contains no isolated vertices, repeating faces, or non-manifold edges; the intermediate mesh has only one connected component; and the intermediate mesh satisfies Euler's formula. In a normal position, the intermediate grid also satisfies , ,in N This indicates the fixed number of vertices in the middle grid. Indicates the number of edges in the middle grid. This indicates the number of triangles in the middle grid.

[0057] If the above verification conditions are met, the intermediate mesh construction is deemed successful. The numbers, spherical coordinates, and triangular connections of each fixed vertex in the intermediate mesh are frozen. The triangular connections include the set of triangular faces, the set of edges, and the one-ring adjacency relationship of each fixed vertex. Subsequent steps only update the 3D coordinates of the fixed vertices and do not change any triangular connections. If the above verification conditions are not met, the intermediate mesh construction is deemed unsuccessful. The steps for constructing the intermediate mesh are repeated, adjusting the preset number of vertices or regenerating the unit spherical point set, until the constructed intermediate mesh meets the above verification conditions.

[0058] As an example, the intermediate grid is the reference intermediate grid of a spherical fixed topology.

[0059] In step S4, please refer to Figure 1 In step S4, the intermediate mesh is used to perform single-frame spherical mapping on each three-dimensional mesh frame in the first and second sequences to obtain the target position of each fixed vertex of the intermediate mesh on the surface of each three-dimensional mesh frame.

[0060] Specifically, for each three-dimensional mesh frame in the first sequence A and the second sequence B, a single-frame spherical mapping from the surface of that three-dimensional mesh frame to a normal unit sphere is constructed, expressed as: in, This represents a single-frame spherical mapping of the first sequence. This represents a single-frame spherical mapping of the second sequence. Represents the first sequence of the first sequence k A 3D mesh frame surface, Represents the second sequence B, the first... k A 3D mesh frame surface, The standard unit is the sphere.

[0061] As an example, the single-frame spherical mapping is a flip-free one-to-one mapping, which can employ conformal spherical parameterization, area-preserving spherical parameterization, or weighted spherical parameterization with angular and area distortion as optimization terms. This application does not limit the specific algorithm for the single-frame sphere; as long as a flip-free one-to-one mapping from the surface of a three-dimensional mesh frame to a standard unit sphere can be established, it falls within the protection scope of this application.

[0062] Furthermore, for each fixed vertex of the intermediate mesh, based on the coordinates of the fixed vertex... The target positions of each fixed vertex on the surface of each 3D mesh frame are obtained through inverse mapping, and the expression is: in, The middle grid represents the first i The fixed vertex in the first sequence k Target position on a 3D grid frame The middle grid represents the first i The fixed vertex in the second sequence k Target position on a 3D grid frame Represents the inverse mapping of the first sequence. This represents the inverse mapping of the second sequence.

[0063] Specifically, for the first sequence of the first sequence k A 3D mesh frame and the second sequence of the first k A 3D mesh frame After being mapped onto a single-frame spherical surface, the corresponding mapping results are obtained on a standard unit sphere; for the intermediate grid, the first... i Fixed vertex coordinates The mapping result includes the coordinates of the fixed vertex. The mapped triangular mesh surface is calculated, and the coordinates of the fixed spherical surface are calculated. Relative to the centroid coordinates of the mapped triangular mesh facet; in the first k On the corresponding triangular mesh facets in each 3D mesh frame, the 3D position is recovered according to the centroid coordinates to obtain the fixed vertex coordinates. In the k The target position on a 3D grid frame.

[0064] The inverse mapping result is recorded as the number and centroid coordinates of the triangular mesh facet in the corresponding three-dimensional mesh frame, and then the three-dimensional position is recovered from the three vertices of the triangular mesh facet.

[0065] Step S4 establishes the target position relationship between the fixed vertices of the intermediate mesh and the surface of each 3D mesh frame only for each 3D mesh frame, without performing smoothing processing across 3D mesh frames; the obtained target positions are used to generate candidate 3D mesh sequences based on the working condition interpolation coefficients, and smoothing across 3D mesh frames is completed by subsequent temporal convolution.

[0066] By establishing a one-to-one, non-flipped mapping from each 3D mesh frame to a standard unit sphere, and then inversely mapping the fixed spherical coordinates of the fixed vertices, the target position of the fixed vertex on each input 3D mesh frame is obtained, thus solving the problem of vertex correspondence across frames and sequences.

[0067] In step S5, please refer to Figure 1 In step S5, the target position is weighted and interpolated based on the specified working condition interpolation coefficients to obtain a candidate three-dimensional mesh sequence; the candidate positions of the same fixed vertex in each three-dimensional mesh frame in the candidate three-dimensional mesh sequence constitute the time trajectory of the fixed vertex.

[0068] Specifically, set the interpolation coefficients for the specified operating conditions. r The specified interpolation coefficients for the operating conditions r The value range is [0,1], which is used to represent the position of the intermediate working condition to be generated between the first parameter condition and the second parameter condition.

[0069] For each fixed vertex and each time frame of the intermediate mesh, obtain the position of the fixed vertex in the first sequence A. k Target position on a 3D grid frame And the fixed vertex in the second sequence B k Target position on a 3D grid frame According to the specified interpolation coefficients for the operating conditions. r For the target location and Perform weighted interpolation to obtain the fixed vertex at the th k Candidate locations under a 3D grid frame The expression is: .

[0070] Weighted interpolation is performed on all fixed vertices and all time frames of the intermediate mesh to obtain the interpolation coefficients for the working condition. r The candidate 3D mesh sequence is a sequence of candidate 3D meshes. The candidate 3D mesh sequence consists of multiple candidate 3D mesh frames, each containing the candidate positions of all fixed vertices of the intermediate mesh within that time frame. The same fixed vertex in the candidate 3D mesh sequence... i The candidate positions on each 3D mesh frame are arranged in time frame order to form the fixed vertices. i Time trajectory : in, , , These represent the interpolation coefficients under the specified operating conditions. r Below, the first of the intermediate grids i Candidate positions of a fixed vertex on the 0th, 1st, and T-1th 3D mesh frames.

[0071] The time trajectory This is used for subsequent temporal convolution. It should be noted that the temporal convolution is applied to the same fixed vertex in the candidate 3D mesh sequence. i Time trajectory However, it does not apply separately to the first sequence A and the second sequence B; and all adjacent time frames read in the same temporal convolution use the same specified interpolation coefficients. r The generated candidate positions ensure the consistency of temporal convolution in both the operational and temporal dimensions.

[0072] In step S6, please refer to Figure 1 In step S6, the time trajectory is subjected to temporal convolution to obtain the vertex positions of each fixed vertex at a specified continuous time. The temporal convolution keeps the sampling time of each three-dimensional mesh frame unchanged, and continuous temporal reconstruction is performed by moving the center of the continuous temporal convolution kernel to the specified continuous time.

[0073] For example, please refer to Figure 2 Step S6 may include the following steps: Step S61 to Step S66.

[0074] Step S61: Keep the sampling time of each 3D mesh frame in the candidate 3D mesh sequence unchanged, and configure the continuous-time convolution kernel.

[0075] Step S62: Move the center of the continuous-time convolution kernel to a specified continuous time interval to obtain the center position of the convolution kernel.

[0076] Step S63: Determine the corresponding continuous support radius according to the position of the specified continuous time in the candidate 3D mesh sequence to obtain the support range of the continuous-time convolution kernel; wherein the weight of the continuous-time convolution kernel is zero at the support boundary.

[0077] Step S64: Select 3D mesh frames participating in temporal convolution from the candidate 3D mesh sequence according to the support range, and delete 3D mesh frames that exceed the time range boundary of the candidate 3D mesh sequence to obtain a set of effective 3D mesh frames.

[0078] Step S65: Determine the convolution weights of each three-dimensional mesh frame in the effective three-dimensional mesh frame set based on the distance between each three-dimensional mesh frame in the effective three-dimensional mesh frame set and the center position of the convolution kernel, and normalize the convolution weights to obtain normalized convolution weights.

[0079] Step S66: Weight the time trajectory of the fixed vertex according to the normalized convolution weights to obtain the vertex position of the fixed vertex at the specified consecutive time.

[0080] Specifically, in step S61, the candidate positions of each three-dimensional grid frame in the candidate three-dimensional grid sequence are regarded as discrete sampling points fixed on the original sampling time coordinates of the three-dimensional grid frame. In the subsequent temporal convolution process, the original sampling time coordinates of each three-dimensional grid frame are not changed, each three-dimensional grid frame is not moved, no new three-dimensional grid frames are added, and the candidate three-dimensional grid sequence is not resampled in time.

[0081] As an example, the original sampling time coordinates can be integer time coordinates. It can also be the physical time coordinates corresponding to each three-dimensional mesh frame.

[0082] A continuous-time convolution kernel is configured, which is a continuous-time convolution kernel function with a finite support range, whose center can be continuously moved to a specified continuous time step. The continuous-time convolution kernel is used to assign corresponding weights to discrete sampling points within the support range according to the distance between discrete sampling points and the center of the convolution kernel, and to continuously reduce the weights to zero at the support boundary, so as to avoid the vertex position jump of the fixed vertex due to the addition or deletion of discrete sampling points.

[0083] The continuous-time convolution kernel is a variable-length convolution kernel, which provides strong smoothness in the middle and maintains the original candidate shape in the first and last frames, thereby avoiding deformation trailing.

[0084] As an example, setting the maximum discrete convolution radius Maximum effective sampling width and Gaussian scale The unnormalized weight function of the continuous-time convolution kernel is denoted as... The expression is: in x This represents the time distance between the time coordinates of discrete sampling points and the specified consecutive time points where the center of the convolution kernel is located. This represents the current supported radius. The unnormalized weight function... Non-negative, about Symmetry, in When the time is positive, in Take zero at time, and in The value continuously decreases to zero. Therefore, during the continuous movement of the center of the convolution kernel in the continuous time, when a 3D mesh frame in the candidate 3D mesh sequence enters or leaves the support range, the convolution weight corresponding to the 3D mesh frame continuously increases or decreases to zero from zero; the position of the fixed vertex at the specified continuous time will not change due to the direct addition or deletion of 3D mesh frames with non-zero convolution weights.

[0085] As an example, the continuous-time convolution kernel can be a continuous Gaussian window function with zero weights at the support boundaries, or it can be a double-weighted kernel, a cubic spline kernel, or other continuous-time convolution kernel functions that satisfy the condition of zero weights at the support boundaries and continuous center movement. This application does not limit the specific functional form of the continuous-time convolution kernel, as long as it has a finite support range, its weights continuously decrease to zero at the support boundaries, and its support range can change according to its position in the candidate 3D mesh sequence at the specified consecutive time points.

[0086] Further, in step S62, the center of the continuous-time convolution kernel is moved to a specified continuous time interval, so that the center position of the convolution kernel corresponds to the specified continuous time interval. Through this movement, the center position of the convolution kernel is no longer limited to the integer time frame position of the candidate 3D mesh sequence, but can be located at any specified continuous time interval. Therefore, the distance between each 3D mesh frame in the candidate 3D mesh sequence and the specified continuous time interval can be calculated based on the center position of the convolution kernel, thereby determining the 3D mesh frames participating in temporal convolution and their convolution weights. Since the original sampling time coordinates of each 3D mesh frame remain unchanged, the center of the continuous-time convolution kernel can continuously move to any specified continuous time interval, thus providing a basis for continuous-time reconstruction between discrete time frames without increasing the number of input 3D mesh frames.

[0087] Further, in step S63, the continuous support radius of the continuous-time convolution kernel is determined according to the specified continuous time intervals, and the support range of the continuous-time convolution kernel is obtained from the continuous support radius. The support range is centered on the center position of the convolution kernel and has the continuous support radius as its radius; within the support range, the unnormalized weight function of the continuous-time convolution kernel takes a positive value; outside the support range, the unnormalized weight function takes a zero value; at the boundary of the support range, i.e., the support boundary, the unnormalized weight function continuously decreases to zero.

[0088] The continuous support radius is determined using either a fixed-width or variable-width mode based on the position of the continuous support radius within the candidate 3D mesh sequence at specified consecutive time points. In the fixed-width mode, the continuous support radius... In variable width mode, set the width transition length. Then the maximum discrete convolution radius satisfy ,in This represents the total number of 3D mesh frames in the candidate 3D mesh sequence. For any given consecutive time interval... Calculate the width envelope and according to the width envelope Calculate the continuous support radius At the first and last time steps, the width envelope... Continuous support radius In the middle of the sequence, the width envelope Continuous support radius In the transition zone between the beginning and end, the continuous support radius Follow t Continuously increasing or decreasing. For the midpoint of an integer time. ,satisfy The three-dimensional mesh frame is the first The first to the second The number of effective samples in the three-dimensional grid frames is exactly the maximum effective sampling width. .

[0089] Further, in step S64, for a specified continuous time... t The center position of the continuous-time convolution kernel is placed t Time. Without considering the time range boundaries of the candidate 3D mesh sequence, the 3D mesh frame indices within the support range of the candidate 3D mesh sequence are determined according to the support range, resulting in an initial set of 3D mesh frame indices. : in, Indicates the 3D mesh frame index. express t The continuous support radius at any given time. Represents a set of integers.

[0090] Furthermore, the initial set of three-dimensional mesh frames is deleted. The set of valid 3D mesh frames that actually participate in temporal convolution is obtained by identifying 3D mesh frames whose temporal index exceeds the time range boundary of the candidate 3D mesh sequence. : in, This represents the total number of 3D mesh frames in the candidate 3D mesh sequence. The time range boundary of the candidate 3D mesh sequence is defined by the minimum time index. and the maximum value of the time index limited.

[0091] As an example, the effective set of 3D mesh frames The smallest 3D mesh frame index actually retained in the middle and the maximum 3D mesh frame index They are respectively: .

[0092] Therefore, the effective three-dimensional mesh frame set .

[0093] When deleting 3D mesh frames that exceed the time range boundary of the candidate 3D mesh sequence, this method does not set virtual 3D mesh frames outside the candidate 3D mesh sequence, does not perform zero-filling, copy-filling, or mirror-filling, and does not change the original sampling time coordinates of any existing 3D mesh frames in the candidate 3D mesh sequence.

[0094] Further, in step S65, for the effective set of three-dimensional mesh frames... Each 3D mesh frame Calculate the distance between it and the center position of the convolution kernel. : in, Indicates the 3D mesh frame index. This indicates the specified consecutive time intervals.

[0095] According to the distance and t Continuous support radius at time step The unnormalized convolution weights of the three-dimensional mesh frame are calculated using the unnormalized weight function of the continuous-time convolution kernel. : .

[0096] For the effective three-dimensional mesh frame set The unnormalized convolution weights of all 3D mesh frames are summed, and the sum is used to normalize the convolution weights of each 3D mesh frame to obtain normalized convolution weights. : in, Represents the set of valid 3D mesh frames The index is s The unnormalized convolution weights of the 3D mesh frame.

[0097] Due to the continuous support radius The specified continuous timet The distance between the candidate 3D mesh and its nearest integer 3D mesh frame index in the sequence does not exceed 0.5, therefore the effective 3D mesh frame set is... There exists at least one 3D mesh frame with positive unnormalized convolution weights, and the normalization denominator is greater than zero, so the normalization operation can be performed normally.

[0098] Furthermore, in step S66, for each fixed vertex of the intermediate mesh, the effective 3D mesh frame set is utilized. Normalized convolution weights of each 3D mesh frame The candidate positions of each fixed vertex in the time trajectory of the corresponding 3D mesh frame are weighted and summed to obtain the vertex position of each fixed vertex at the specified consecutive time points. The expression is: in, Indicates the first i The fixed vertex is in the... j Candidate positions on a 3D grid frame r This indicates the interpolation coefficient for a specified operating condition.

[0099] In the above weighting process, the first i The fixed vertex is in the... j Candidate positions on a 3D grid frame The sampling time remains unchanged, that is, the sampling time of the first... j Each three-dimensional grid frame remains unchanged; only the center position of the continuous-time convolution kernel moves continuously, i.e., only continuous time intervals are specified. t Continuous movement. During the specified continuous time intervals... t The effective set of 3D mesh frames when moving from one integer time frame to the next integer time frame. Normalized convolution weights of each 3D mesh frame The continuous change allows us to directly obtain the vertex positions at any time between two original 3D mesh frames without increasing the number of input 3D mesh frames.

[0100] Integer time steps are a special case of the continuous-time reconstruction described above. The candidate vertex positions after temporal convolution are recorded. In the first moment of the variable width mode t= 0, only the 0th 3D mesh frame has a positive weight; at the final time... t =T-1, only the (T-1)th 3D mesh frame has a positive weight. Therefore, , ,in, Indicates the first i A fixed vertex at a specified consecutive time interval tVertex position after time convolution with =0 Indicates the first i Candidate positions of a fixed vertex in the 0th time frame (i.e., time frame index k=0). Indicates the first i The position of a fixed vertex after time convolution at a specified consecutive time t=T-1. Indicates the first i The candidate positions of fixed vertices in the (T-1)th time frame (i.e., time frame index k=T-1) are thus obtained. Therefore, the middle of the candidate 3D mesh sequence obtains the maximum temporal smoothing intensity, while the first and last 3D mesh frames strictly maintain the unsmoothed candidate positions.

[0101] When each 3D mesh frame in the candidate 3D mesh sequence uses an equal-interval physical time When requesting physical time The corresponding consecutive frame coordinates are ,in, This represents the physical time corresponding to the k-th 3D mesh frame. This represents the physical time corresponding to the 0th 3D mesh frame. This represents the physical time interval between two adjacent 3D mesh frames. When the input time interval is not uniform, the actual physical time of each 3D mesh frame is retained. , distance Replace with And set physical time in the same time unit. Continuous support radius The remaining trimming, weighting, and normalization processes remain unchanged.

[0102] As an example, the candidate vertex positions of neighboring 3D mesh frames read in the same temporal convolution are all generated using the same specified interpolation coefficients.

[0103] By continuously supporting the change of the radius of the candidate 3D mesh in the sequence of candidate 3D meshes at specified continuous time intervals, smaller values ​​are taken at the beginning and end of the 3D mesh frames and larger values ​​are taken in the middle. Only the frame at the beginning and end of the sequence has positive weights, so that the fixed vertices converge to the candidate positions that have not been convolved in time. This avoids the mixing of the shape of the previous 3D mesh frame into the end of the sequence by the fixed width convolution. The weights are zero at the support boundary by the continuous time convolution kernel. When the 3D mesh frame enters or leaves the support range, the weights continuously increase or decrease from zero to zero. 3D mesh frames that exceed the time range boundary of the candidate 3D mesh sequence are deleted and renormalized, which can avoid deformation tails and position jumps caused by abrupt weight changes at the boundary.

[0104] In step S7, please refer to Figure 1In step S7, based on the triangular connection relationship and the vertex position, the specified interpolation coefficients for the specified working condition and the smooth three-dimensional mesh sequence corresponding to the specified continuous time are obtained.

[0105] As an example, the fixed vertices of the intermediate mesh are... t The vertex position at time step S3 is combined with the triangle connection relationship frozen in step S3 to obtain... t A 3D mesh frame at a given time. The 3D mesh frame uses the fixed vertex numbering and triangulation relationships of the intermediate mesh, with only the vertex positions of each fixed vertex varying according to the specified consecutive time points. t and the specified interpolation coefficients r Update. Perform the above combination at multiple specified consecutive time points to obtain a smooth 3D mesh sequence consisting of multiple 3D mesh frames.

[0106] When the interpolation coefficient for the specified working condition is given When, the output is the first sequence of the intermediate grid pair. t Continuous time reconstruction at a given moment; interpolation coefficients for specified operating conditions. When, the output is the intermediate grid pair of the second sequence in t Continuous time reconstruction at a given moment; interpolation coefficients for specified operating conditions. When the output is the intermediate parameter condition between the first parameter condition and the second parameter condition, t The form of time. The interpolation coefficients for the specified operating condition. r and the specified continuous time t The values ​​can be taken independently to obtain a smooth three-dimensional mesh sequence that is continuous in both the working condition dimension and the time dimension.

[0107] As an example, based on a user-specified set of output time points, a single 3D mesh frame representing a specified consecutive time period can be output, or a smooth 3D mesh sequence consisting of multiple 3D mesh frames representing specified consecutive time periods can be output at specified time sampling intervals. The output smooth 3D mesh sequence includes the fixed vertex numbers of the intermediate mesh, the triangulation relationships, and the vertex positions at each specified consecutive time period.

[0108] In an optional implementation, before outputting the smoothed 3D mesh sequence in step S7, a volume constraint step is further included: determining a volume correction amount along the normal direction of the fixed vertex based on the volume deviation between the target volume and the actual volume of the intermediate mesh at specified consecutive times; performing temporal smoothing on the volume correction amount to obtain a smoothed volume correction amount; updating the vertex position along the normal direction of the fixed vertex based on the smoothed volume correction amount until the volume deviation between the actual volume and the target volume meets a preset error range.

[0109] Specifically, for a closed triangular mesh, its volume can be determined based on the triangular connection relationships of the intermediate meshes and the set of fixed triangular faces. F 3. Confirmed: It can also be written as ,in, , and Each is a set of fixed triangular faces. F 3 triangular faces f The centroid, unit outward normal, and area of ​​the facet. Represents a set of fixed triangular faces F Three triangular faces, A, B, and C are triangular faces. The three fixed vertices.

[0110] The volume of each 3D mesh frame in the first sequence is pre-calculated. Volume of each 3D mesh frame in the second sequence .for t At time, the same set of effective 3D mesh frames as the temporal convolution in step S6 is used. and normalized convolution weights Reconstruct the continuous time volume of the two end conditions respectively: in, Indicates the condition of the first parameter. t The continuous time volume at a given moment. Indicates the condition of the second parameter. t The continuous time volume at a given moment.

[0111] Based on the specified interpolation coefficients r ,Sure t Target volume at time 1 The expression is: .

[0112] Therefore, the target volume With fixed vertices t The vertex positions at each time step use the same continuous time coordinates and the same set of normalized convolution weights to avoid the problem that the target volume still takes a certain integer three-dimensional mesh frame even though the shape has moved to the inter-frame time step.

[0113] Furthermore, based on the triangulation relationships of the intermediate mesh and the current vertex position, the following calculations are performed. t Actual volume at time And calculate the actual volume. With the target volume Volume deviation between For each fixed vertex of the intermediate mesh, according to the volume deviation... And the area weights of the adjacent faces of the fixed vertex are used to determine the original corrected distance of the fixed vertex along its normal direction. Specifically, let... t Fixed vertex at time i The set of adjacent faces is Its area weight for: Then fix the vertex i The original normal correction distance is : in, Represents a set of fixed triangular faces F 3 triangular faces f area, This represents the traversal index of a fixed vertex in the intermediate grid. Indicates the first The set of adjacent faces of a fixed vertex.

[0114] For example, suppose the user requests the output of a specified set of consecutive time points as follows: Each of them All times are within the time range of the candidate 3D mesh sequence. The real numbers within. First, obtain the candidate grids for each specified consecutive time step according to step S6, and then... m In each iteration, the calculation is based on the candidate grid state from the previous round. Actual volume at time And calculate the actual volume. With the target volume Volume deviation between .right Fixed vertex at time i Area weight of adjacent face set Original normal correction distance ,in, Indicates in Triangle facet at time f The area.

[0115] Subsequently, the original normal was corrected at a distance. Smooth along the time axis. For t l Time, in the specified set of consecutive time points Retaining satisfaction The specified consecutive times form a smoothing set of correction amounts. : in, t s Represents a specified set of consecutive time points The Middle s M represents a specified set of consecutive moments. The total number of consecutive moments specified in the text. Indicates in t l The continuous support radius at any given time.

[0116] Using the same continuous-time convolution kernel, continuous support radius, continuous-time convolution kernel function with zero support boundary weights as in step S6, boundary clipping rules, and weight normalization rules, the correction weights are calculated and normalized. The expression is: in, Represents the smooth set of corrections The Middle s a specified consecutive time period t s Compared to The unnormalized correction weight at time step Represents the smoothing set of corrections. The weighted sum of the unnormalized correction values ​​for all specified consecutive time intervals is calculated.

[0117] Corrected distance of the original normal to the same fixed vertex Perform time smoothing to obtain the smoothed volume correction. The expression is: .

[0118] Therefore, both the candidate vertex position and volume corrections use a center-movable continuous-time convolution kernel, the same support radius variation rule, the same support boundary weight zero rule, the same boundary clipping rule, and the same weight normalization rule. When only a single specified continuous time step is requested, the correction smoothing set is applied. Only retain At time, the smoothed volume correction amount Do not mix with non-existent output moments; when batch outputting multiple specified consecutive moments, the smoothed volume correction amount for each specified consecutive moment. It changes smoothly along the time axis.

[0119] Furthermore, based on the smoothed volume correction amount Update the vertex position along the fixed vertex i The normal direction is used with the smoothed volume correction amount. Update the fixed vertices i exist vertex position at time ,in, This indicates the vertex position before the update. This represents the current vertex normal. After updating, the vertex normal and actual volume are recalculated, and the volume deviation is also recalculated. When the following conditions are met... When, end The volume correction iteration is performed at time 1; otherwise, the next round continues. ε represents the volume error threshold, which can be set to 1. The maximum number of iterations can be set simultaneously to prevent the iteration from failing to converge. The triangular connection relationships of the intermediate mesh and the fixed set of triangular faces remain unchanged during the iteration process. F 3.

[0120] When volume constraints are enabled, the updated vertex positions are combined with the triangulation relationships of the intermediate mesh to obtain and output a smooth 3D mesh sequence corresponding to the specified working condition interpolation coefficients and the specified continuous time. When volume constraints are not enabled, the vertex positions obtained in step S6 are combined with the triangulation relationships to obtain and output the smooth 3D mesh sequence.

[0121] In the 3D mesh sequence generation method of this application, a globally topologically frozen intermediate mesh is constructed as a unified mapping carrier. After mapping two sets of heterogeneous dynamic mesh sequences to this intermediate mesh, the working condition interpolation is first performed to obtain candidate sequences. Then, a variable-width temporal convolution is used to smooth the temporal trajectory of fixed vertices. By moving the continuous temporal convolution kernel, continuous mesh queries in both working condition and time dimensions are realized. A small number of discrete parameter conditions and relatively sparse temporal output results can be used to generate continuous geometric morphology sequences under intermediate parameter conditions and intermediate time moments. Furthermore, the volume constraint and geometric interpolation reuse the same convolution weight system to ensure that physical quantities and temporal morphology are synchronous and continuous, thus solving the technical defects of topological mismatch, inter-frame jitter, and temporal misalignment of physical constraints during heterogeneous dynamic mesh interpolation.

[0122] In an optional implementation, the three-dimensional mesh sequence generation method of this application can also be configured as follows: when the input three-dimensional mesh frame splits, merges, disappears, or is perforated, it can be processed by segmenting according to homeomorphic time periods or connected components, or expanded into multiple fixed topological intermediate three-dimensional meshes, and corresponding smooth three-dimensional mesh sequences can be generated according to the method of this application. Single-frame spherical mapping is a pre-correspondence establishment method, which can be replaced by other methods that can obtain the relationship between fixed spherical coordinates and target points on the surface of the input three-dimensional mesh frame. This application does not limit the specific algorithm of single-frame spherical mapping. Continuous-time reconstruction uses a non-negative, finite-support continuous-time convolution kernel with zero weights at the support boundaries. The continuous-time convolution kernel can be the product of a Gaussian function and a zero-boundary-window function, a double-weighted kernel, a cubic spline kernel, or other kernel functions that satisfy the above conditions. When determining the continuous support radius, a fixed-width mode or a variable-width mode can be selected. In the fixed-width mode, the continuous support radius is the same at any specified continuous time; in the variable-width mode, the continuous support radius changes with the position of the specified continuous time in the candidate three-dimensional mesh sequence. The output time can be a specified continuous moment corresponding to the original integer time frame, or any real moment within the time range of the candidate 3D mesh sequence. A single 3D mesh frame within a specified continuous moment can be queried, or a smooth 3D mesh sequence can be generated in batches according to user-specified time intervals. The volume constraint step can be enabled or disabled according to application requirements. When enabled, it outputs a smooth 3D mesh sequence that meets the preset volume error range; when disabled, it directly outputs the smooth 3D mesh sequence obtained through time convolution.

[0123] In one specific embodiment, the application of the three-dimensional mesh sequence generation method described in this application is illustrated using underwater bubble dynamic morphology interpolation in computational fluid dynamics simulation as an example. The parameters in this embodiment are for process description purposes only and are not considered as experimental performance data.

[0124] For a single underwater bubble generated by the same nozzle, two computational fluid dynamics (CFD) conditions were applied, one with a first parameter and the other with a second parameter. The first parameter condition, denoted as Condition 1, had an inlet gas flow rate of 20 mL / min; the second parameter condition, denoted as Condition 2, had an inlet gas flow rate of 40 mL / min. Both conditions output the gas-liquid interface evolution from bubble formation to 0.2 seconds, with an output time interval of 0.002 seconds. Therefore, each condition's corresponding 3D mesh sequence contains 101 3D mesh frames, with time frame indices from 0 to 100.

[0125] The gas-liquid interface of each 3D mesh frame is extracted from the volume fraction isosurface. In the first 3D mesh sequence corresponding to Working Condition 1, the number of vertices in each 3D mesh frame varies between 8,000 and 13,000; in the second 3D mesh sequence corresponding to Working Condition 2, the number of vertices in each 3D mesh frame varies between 9,000 and 15,000. Each 3D mesh frame is a triangular mesh with outward-facing normals, connected, closed, and of zero genus. Within the selected time period, the bubble does not split, merge, or perforate, satisfying the topological requirements of the 3D mesh frame for subsequent single-frame spherical mapping.

[0126] The processing parameters used in this embodiment are as follows: the number of output frames T of the first and second sequences after time alignment is 101; the preset number of vertices N of the intermediate grid is 20,000; and the specified interpolation coefficients are used. r 0.4; Maximum discrete convolution radius n max The maximum effective sampling width is 5. L max The value is 11; the width transition length b is 10; the Gaussian scale σ is 2; the volume error threshold ε is 1% of the target volume; and the maximum number of volume iterations is 10.

[0127] In this embodiment, the interpolation coefficients for the specified operating conditions are specified. r =0.4 indicates that the intermediate working condition corresponding to the generated smooth 3D mesh sequence is located between working condition one and working condition two, and is close to working condition one. The specified continuous time can be any real number between 0 and 100, so that the 3D mesh frame of the underwater bubble gas-liquid interface at any time can be obtained within the physical time range of 0 seconds to 0.2 seconds, and multiple 3D mesh frames with specified continuous time can be combined to form a continuous time underwater bubble dynamic morphology sequence.

[0128] After acquiring the first and second 3D mesh sequences, check whether the triangular mesh faces in each 3D mesh frame have a consistent outward normal orientation; check whether each mesh edge in each 3D mesh frame belongs to exactly two triangular mesh faces; check whether each 3D mesh frame has only one connected component; calculate the Euler characteristic number of each 3D mesh frame to confirm that the surface to be processed is homeomorphic to a sphere; use the physical time output by computational fluid dynamics as a unified time coordinate to set the first 3D mesh sequence to the second 3D mesh sequence. k The first 3D mesh frame and the second 3D mesh sequence k Each 3D mesh frame corresponds to the same physical time. Since the two operating conditions use the same output time, this embodiment does not require additional time resampling. If a certain operating condition lacks a 3D mesh frame for a specified time, the corresponding 3D mesh frame is generated within that operating condition first, and then the subsequent steps are performed. After time alignment, a first sequence A and a second sequence B with time matching are obtained, both of which contain T=101 3D mesh frames.

[0129] Based on the user-specified preset number of vertices N=20000, 20000 approximately uniformly distributed unit spherical points are generated, resulting in a unit spherical vertex set U. Specifically, the unit spherical vertex set is generated using the Fibonacci spherical point set, while maintaining the generation number of each unit spherical point. Let the golden angle be... ,for Calculate vertical height Horizontal projection radius Horizontal rotation angle , obtained the i Coordinates of a unit spherical point Generate the generation sequence number of each unit spherical point. i As the first in the middle grid i The fixed vertex number of a fixed vertex.

[0130] Four non-coplanar unit spherical vertices are selected from the unit spherical vertex set U to form an initial tetrahedron; the orientation coefficient of the normal is used as the basis for the determination. The initial tetrahedron's four initial convex hull faces are all directed towards the outside of a unit sphere. In each iteration, the triangular face currently constituting the boundary of the 3D convex hull is designated as the current convex hull face. For each current convex hull face, the vertex of the unit sphere located outside the current convex hull face is designated as the outside point of the current convex hull face. From the current convex hull faces with outside points, the vertex of the unit sphere with the largest distance from the current convex hull face is selected as the expansion point. All current convex hull faces visible from the expansion point are deleted, and the current convex hull faces that are not deleted are designated as retained convex hull faces. The horizon edge between the visible convex hull faces and the retained convex hull faces is extracted. The expansion point is connected to each of the horizon edges to generate new triangular faces. The adjacency relationship of the triangular faces and the set of outside points are updated. The above process is repeated until all current convex hull faces have no outside points, resulting in the final 3D convex hull. Each triangular face on the boundary of the final 3D convex hull consists of three fixed vertex numbers. These numbered triplets constitute the fixed triangular face set of the fixed topology intermediate 3D mesh. F 3. If the final 3D convex hull contains polygonal faces, then the fixed vertex with the smallest number in the polygonal face is used as the center of the sector, and deterministic triangulation is performed according to the boundary order of the polygonal face, without adding any new fixed vertices.

[0131] According to the fixed triangular face set F 3. Establish a set of undirected edges E 3. The intermediate grid is then validated. After validation, the fixed topology data includes: , The system stores the spherical coordinates of 20,000 fixed vertices, 59,994 undirected edges, 39,996 triangles, and the adjacency relationships of each fixed vertex. A checksum is calculated for the vertex number array of the triangles, and the same array is used in all output 3D mesh frames; the triangle connections are not reconstructed in any subsequent steps.

[0132] This embodiment performs discrete conformal spherical parameterization on each of the 202 input 3D mesh frames. After parameterization of each 3D mesh frame, it checks whether all triangular mesh faces mapped to the normalized unit sphere maintain a consistent orientation; 3D mesh frames with flipped faces are not further processed and their parameterization needs to be recalculated. For each fixed vertex of the intermediate mesh, its fixed spherical coordinates are... Search for elements containing [the specified element] in the spherical triangular mesh mapped from the current input 3D mesh frame. Calculate the triangular face. Relative to the three barycentric coordinates of the triangular face, and using the same set of three barycentric coordinates, the 3D position is reconstructed on the corresponding triangular mesh facet of the original input 3D mesh frame, thus obtaining the target position of the fixed vertex on the surface of the input 3D mesh frame. For each 3D mesh frame... k and each fixed vertex i Save separately: In the first sequence corresponding to working condition one, input the triangular mesh patch number, the three centroid coordinates, and the target position. In the second sequence corresponding to working condition two, input the triangular mesh patch number, the three centroid coordinates, and the target position. The target location table is obtained, where the rows correspond to fixed time frame indices and the columns correspond to fixed vertex numbers, thus not depending on the vertex numbers of the input 3D mesh frame itself.

[0133] Interpolation coefficients for specified operating conditions r =0.4, calculate the unsmoothed candidate positions for all 101 time frames and all 20,000 fixed vertices: After calculation, a candidate 3D mesh sequence containing 101 3D mesh frames is obtained. Each 3D mesh frame in the candidate 3D mesh sequence has 20,000 fixed vertices and the same triangular face connection table of 39,996. The candidate positions of the same fixed vertex in each 3D mesh frame constitute the temporal trajectory of that fixed vertex.

[0134] This embodiment uses the maximum discrete convolution radius. Width transition length The value is 10, therefore the maximum effective sampling width at the middle integer time step is... The maximum continuous support radius is 6. The width envelope is calculated based on the position within the candidate 3D mesh sequence at specified consecutive time points. and continuous support radius Some of the results are shown in Table 1.

[0135] Table 1 Width envelope at some specified consecutive time points and continuous support radius When the specified continuous time interval t is not an integer, the center of the continuous-time convolution kernel is located between two adjacent 3D mesh frames, and the number of integer 3D mesh frames covered on the left and right sides of the support range can differ by 1. When the change in the number of 3D mesh frames occurs when a certain 3D mesh frame is located at the support boundary, since the weight of the continuous-time convolution kernel at the support boundary is zero, the addition or deletion of this 3D mesh frame does not cause a jump in vertex position.

[0136] This embodiment uses a Gaussian scale. Continuous Gaussian window function: At a specified continuous time t=0, the continuous support radius The distance between the 0th 3D mesh frame and the center of the convolution kernel is 0, and its unnormalized convolution weights are greater than zero; the distance between the 1st 3D mesh frame and the center of the convolution kernel is equal to the continuous support radius, and its unnormalized convolution weights are zero; negative temporal frame indices are outside the time range of the candidate 3D mesh sequence and are deleted. Therefore, the normalized convolution weights... At a specified continuous time t=1, the continuous support radius The calculations are performed on the 0th, 1st, and 2nd three-dimensional mesh frames. The calculations are performed separately for each frame. , and 3D mesh frames outside the time range of the candidate 3D mesh sequence are deleted, and then normalized using the sum of three unnormalized convolution weights. At a specified continuous time t=100, the continuous support radius... Only the 100th 3D mesh frame has a positive weight. Therefore, the outputs at the first and last time steps are strictly equal to the first and last candidate positions without time smoothing.

[0137] If switched to fixed width mode, the radius is continuously supported. At a specified consecutive time t=0, the uncropped set still contains negative time frame indices; all negative time frame indices are deleted without padding, and the positive weights of the 0th to 5th 3D mesh frames are renormalized. The resulting fixed-width result is used for comparison with the variable-width mode and is not the final output of this embodiment.

[0138] For each integer time point t = k The set of valid 3D mesh frames is determined based on their positions in the candidate 3D mesh sequence at specified consecutive time points. and normalized convolution weights Calculate the candidate vertex positions after temporal convolution for all 20,000 fixed vertices. All fixed vertices share the same set of valid 3D mesh frames and normalized convolution weights for the same specified consecutive time intervals, but each vertex is weighted and summed for its own candidate position temporal trajectory. After calculation, the results are recalculated based on the fixed set of triangles. F 3. Calculate the face normals and vertex normals for each 3D mesh frame.

[0139] For a specified continuous time period t =12.3, corresponding to the physical time: seconds, at this time the width envelope and continuous support radius The center of the continuous-time convolution kernel moves to 12.3, while the original 101 candidate 3D mesh frames remain fixed at their respective integer time positions. This satisfies... The effective 3D mesh frame indices are 7 to 18. The set of effective 3D mesh frames frame by frame. The time distance of each 3D grid frame relative to the center position of the convolution kernel Unnormalized convolution weights Normalized convolution weights, continuous-time convolution kernels at specified consecutive time steps t The calculation results of the convolution weights at 12.3 are shown in Table 2.

[0140] Table 2 Calculation results of continuous-time convolution kernels at specified continuous time intervals t Convolution weight calculation results at 12.3 Calculate the specified consecutive time intervals for each fixed vertex. t =12.3 vertex position The resulting set of 20,000 fixed vertices is then compared with the set of fixed triangles. F The combination of 3 elements yields a 3D mesh frame with an interpolation coefficient of 0.4 and a physical time of 0.0246 seconds for the specified operating condition. Since this frame is not a raw output 3D mesh frame from any computational fluid dynamics system, it demonstrates that this method can directly perform continuous-time interpolation in addition to operating condition interpolation.

[0141] Furthermore, volume constraint correction is performed, and the volume of each 3D mesh frame in the first sequence is calculated and saved. and the volume of each three-dimensional mesh frame in the second sequence For any given consecutive time intervals t First, use the set of valid 3D mesh frames at that moment. and normalized convolution weights Calculate the continuous time volume of both ends of the working condition and Then calculate the specified consecutive time intervals. t Corresponding target volume .

[0142] For integer output moments, use As the vertex position of the 0th volume iteration; for non-integer output times, use This is used as the vertex position in round 0. Each round executes the following steps sequentially: calculate the actual volume and volume deviation for all requested output times; calculate the original normal correction distance for each fixed vertex at each output time based on the area weights of adjacent faces; smooth the correction distance along the output time axis using the same continuous temporal convolution kernel, variable width, boundary clipping, and weight normalization method as temporal convolution; apply the smoothed volume correction along the current vertex normal; recalculate the vertex normal and actual volume; if the volume error is less than 1% of the target volume, the specified consecutive time interval reaches the termination condition; volume correction ends when all requested output times reach the termination condition or when 10 iterations are completed. Volume correction only changes the vertex position of the fixed vertex and does not change the fixed triangle face set. F 3.

[0143] The final output includes topology data with 20,000 fixed vertex numbers and 39,996 fixed triangles; fixed vertex 3D coordinates generated according to the user-specified output time set, where each output time can be an integer or non-integer time; recalculated vertex normals corresponding to each output time; specified interpolation coefficients, convolution parameters, volume error threshold, and actual number of iterations; and the target volume, uncorrected volume, and corrected volume for each output time, used for result checking. When the user outputs according to the original computational fluid dynamics time interval, specifying the continuous time set as {0,1,…,100} yields 101 sets of vertex coordinates. When the user requests output at one-tenth of the original time interval, specifying the continuous time as 0,0.1,0.2,…,100 yields 1001 sets of vertex coordinates, without needing to re-execute the computational fluid dynamics simulation or re-establish the spherical mapping relationship.

[0144] It should be noted that the three-dimensional mesh sequence generation method described in this application is not limited to computational fluid dynamics bubble simulation, but can also be applied to the following scenarios: three-dimensional mesh sequences of fluid interfaces such as droplet collisions, sprays, free liquid surfaces, waves, and cavitation interfaces; three-dimensional mesh sequences of structural deformation under different loads, temperatures, material parameters, or boundary conditions; three-dimensional mesh sequences of organs and soft tissues under different breathing depths, heart rates, physiological parameters, or treatment stages; three-dimensional mesh sequences of stamping, forging, injection molding, blow molding, thermoforming, foaming, and sintering processes; digital twin three-dimensional mesh sequences of flexible pipelines, seals, diaphragms, soft robots, and flexible mechanical components; animated three-dimensional mesh sequences under different motion intensities, facial expression weights, character body shapes, or interaction conditions; three-dimensional scan reconstruction three-dimensional mesh sequences under different postures, expressions, time states, or growth stages; and scientific research visualization three-dimensional mesh sequences of cells, material microstructures, particle clusters, and geological interfaces.

[0145] The above application scenarios are for illustrative purposes only and do not constitute a limitation on the scope of protection of this application. In each of the above application scenarios, as long as the input data consists of several discrete parameter conditions and several discrete-time samples, it can be interpolated using specified working condition interpolation coefficients. r and specified consecutive times t The query examines intermediate parameter conditions and the geometry at intermediate time points. This application does not require time continuity for computational fluid dynamics data, nor does it require uniform time intervals for each 3D mesh frame in the input 3D mesh sequence.

[0146] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the accompanying drawings may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0147] In another embodiment, the present invention also provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs; wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the three-dimensional mesh sequence generation method described in the above embodiments.

[0148] As an example, the electronic device could be a server, and its internal structure diagram could be as follows: Figure 3As shown, the electronic device includes a processor 31, a memory 32, a communication interface 33, and a bus 34. The processor 31, memory 32, and communication interface 33 communicate with each other via the bus 34. The memory 32 stores computer program instructions; the processor 31 executes the computer program instructions stored in the memory 32 to implement the three-dimensional mesh sequence generation method described in the above embodiments; the communication interface 33 is used for data interaction with external devices or networks.

[0149] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the electronic device to which the present invention is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0150] In another embodiment, this application also provides a computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, wherein the computer program, when executed by a processor, implements the various steps of the three-dimensional mesh sequence generation method provided in the above embodiments.

[0151] Computer programs used to implement the methods of this application may be written in any combination of one or more programming languages. These computer programs may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are performed. The computer programs may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or electronic device.

[0152] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0153] Although this application has been disclosed above with reference to embodiments, it is not intended to limit this application. Anyone skilled in the art can make some modifications and refinements without departing from the spirit and scope of this application.

Claims

1. A method for generating a three-dimensional mesh sequence, characterized in that, Includes the following steps: An intermediate mesh is constructed based on a preset number of vertices, and the numbering, spherical coordinates, and triangulation relationships of each fixed vertex of the intermediate mesh are frozen. Using the intermediate mesh, single-frame spherical mapping is performed on each three-dimensional mesh frame in the first and second sequences to obtain the target position of each fixed vertex of the intermediate mesh on the surface of each three-dimensional mesh frame; Based on the specified interpolation coefficients, the target position is weighted and interpolated to obtain a candidate 3D mesh sequence; the candidate positions of the same fixed vertex in each 3D mesh frame in the candidate 3D mesh sequence constitute the time trajectory of the fixed vertex; Temporal convolution is performed on the time trajectory to obtain the vertex positions of each fixed vertex at a specified consecutive time. The temporal convolution keeps the sampling time of each three-dimensional mesh frame unchanged, and continuous temporal reconstruction is performed by moving the center of the continuous temporal convolution kernel to the specified consecutive time. Based on the triangular connection relationship and the vertex position, the interpolation coefficients for the specified working condition and the smooth three-dimensional mesh sequence corresponding to the specified continuous time are obtained.

2. The method for generating a three-dimensional mesh sequence according to claim 1, characterized in that, Temporal convolution is performed on the time trajectory to obtain the vertex positions of each fixed vertex at specified consecutive time points. The temporal convolution maintains the sampling time of each 3D mesh frame unchanged, and continuous temporal reconstruction is performed by moving the center of the continuous temporal convolution kernel to the specified consecutive time points, including: Keep the sampling time of each 3D mesh frame in the candidate 3D mesh sequence unchanged, and configure continuous-time convolution kernels; The center of the continuous-time convolution kernel is moved to a specified continuous time interval to obtain the center position of the convolution kernel. The corresponding continuous support radius is determined based on the position of the specified continuous time in the candidate 3D mesh sequence, thereby obtaining the support range of the continuous-time convolution kernel; wherein the weight of the continuous-time convolution kernel is zero at the support boundary; Based on the support range, select 3D mesh frames participating in temporal convolution from the candidate 3D mesh sequence, delete 3D mesh frames that exceed the temporal range boundary of the candidate 3D mesh sequence, and obtain a set of effective 3D mesh frames; Based on the distance between each three-dimensional mesh frame in the effective three-dimensional mesh frame set and the center position of the convolution kernel, the convolution weights of each three-dimensional mesh frame in the effective three-dimensional mesh frame set are determined, and the convolution weights are normalized to obtain normalized convolution weights. The time trajectory of the fixed vertex is weighted according to the normalized convolution weights to obtain the vertex position of the fixed vertex at the specified consecutive time points.

3. The method for generating a three-dimensional mesh sequence according to claim 2, characterized in that, Effective 3D mesh frame set : in, The total number of 3D mesh frames in the candidate 3D mesh sequence. This is the initial set of 3D mesh frames. Indicates the 3D mesh frame index; The effective three-dimensional mesh frame set The smallest 3D mesh frame index actually retained in the middle and the maximum 3D mesh frame index They are respectively: ; The initial three-dimensional mesh frame index set : in, express t The continuous support radius at any given time. Represents a set of integers.

4. The method for generating a three-dimensional mesh sequence according to claim 2, characterized in that, Based on the distance between each 3D grid frame and the center position of the convolution kernel and t Continuous support radius at time step The unnormalized convolution weights for each 3D mesh frame are calculated using the unnormalized weight function of the continuous-time convolution kernel. : ; For the effective set of 3D mesh frames The unnormalized convolution weights of all 3D mesh frames are summed, and the sum is used to normalize the convolution weights of each 3D mesh frame to obtain normalized convolution weights. : in, Represents the set of valid 3D mesh frames The index is s The convolution weights of the 3D mesh frame, This represents the unnormalized weight function. This represents the index of the three-dimensional mesh frame.

5. The method for generating a three-dimensional mesh sequence according to claim 4, characterized in that, The continuous support radius is determined by its position within the candidate 3D mesh sequence at specified consecutive time points, employing either a fixed-width mode or a variable-width mode; in the fixed-width mode, the continuous support radius... In variable width mode, based on the width envelope Determine the continuous support radius ,in, is the maximum discrete convolution radius.

6. The method for generating a three-dimensional mesh sequence according to claim 2, characterized in that, Vertex positions of each fixed vertex at specified consecutive times The expression is: in, Indicates the first i The fixed vertex is in the... j Candidate positions on a 3D grid frame r Indicates the interpolation coefficient for a specified operating condition. Indicates the first j Normalized convolution weights for each 3D grid frame, This represents the set of valid 3D mesh frames.

7. The method for generating a three-dimensional mesh sequence according to claim 1, characterized in that, Before outputting the smoothed 3D mesh sequence, a volume constraint step is also included: The volume correction amount is determined along the normal direction of a fixed vertex based on the volume deviation between the target volume and the actual volume of the intermediate grid at a specified continuous time. The volume correction amount is time-series smoothed to obtain the smoothed volume correction amount; The vertex position is updated along the normal direction of the fixed vertex based on the smoothed volume correction amount until the volume deviation between the actual volume and the target volume meets the preset error range.

8. The method for generating a three-dimensional mesh sequence according to claim 1, characterized in that, It also includes, Obtain the first three-dimensional mesh sequence and the second three-dimensional mesh sequence under different discrete working conditions; The first three-dimensional mesh sequence and the second three-dimensional mesh sequence are time-aligned to obtain a time-matched first sequence and a second sequence.

9. An electronic device, comprising: One or more processors; Storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the three-dimensional mesh sequence generation method as described in any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the three-dimensional mesh sequence generation method according to any one of claims 1 to 8.