Three-dimensional digital human mesh model projection simplification method

By constructing Kd-tree and the minimum heap for edge collapse and combining the edge flip operation, the problem of difficulty in taking into account speed and accuracy in the existing technology is solved, and efficient projection simplification of three-dimensional digital human grid model is achieved, which significantly improves the quality of the triangle and geometric accuracy, suitable for mobile devices.

CN119090708APending Publication Date: 2024-12-06XI AN JIAOTONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411196330.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing mesh simplification methods are difficult to balance between speed and accuracy, the vertex clustering method has poor geometric accuracy, while the QEM and re-meshing methods can maintain high accuracy, but the time cost is high, and the triangle quality is poor.

Method used

The projection simplified method of three-dimensional digital human grid model is adopted. By constructing Kd-tree to store vertex and surface information, the projection point of the middle point of the edge is calculated as the collapse point, and the edge collapse is used to combine the edge flip operation to improve the quality of the triangle. Reduce calculation complexity.

Benefits of technology

It significantly improves the quality and geometric accuracy of triangles, reduces simplified time, and achieves efficient grid simplification, suitable for mobile devices, ensuring smooth operation and high-quality display of digital human models.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090708A_ABST
    Figure CN119090708A_ABST
Patent Text Reader

Abstract

A three-dimensional digital human mesh model projection simplification method comprises the following steps: step 1, acquiring vertex and surface information of a three-dimensional digital human mesh model, storing gravity centers of vertexes and surfaces in a Kd-tree, and storing corresponding vertex normal and surface normal in a vector container; 2, calculating a projection point of a side midpoint on the surface of the original digital human grid as a collapse point vnew, calculating the sum of volumes of tetrahedrons formed by the vnew and a neighborhood surface of the side as a collapse cost, storing the vnew, the cost and a starting point of the side in a structural body struct, and storing the vnew, the cost and the starting point of the side in a minimum heap H taking the cost as an arrangement basis; 3, selecting an edge with the minimum cost from the minimum heap for collapsing, updating vertex coordinates and a neighborhood connection relation thereof, calculating a collapsed minimum heap of an edge formed by a new vertex and a neighborhood point thereof, and updating the minimum heap; and 4, turning over the non-sharp edge of the approximate plane area to improve the display effect of the digital human model. According to the method, the problem that speed and precision are difficult to take into account during grid simplification is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of mesh simplification in computer graphics, and in particular to a projection simplification method for a three-dimensional digital human mesh model. Background Art

[0002] Mesh simplification technology is an important research topic in the field of computer graphics and computer-aided design. Although high-resolution mesh models provide rich detailed information, a large number of polygons put forward higher requirements for model rendering, which has become a bottleneck for improving the performance of real-time interactive applications such as video games, virtual reality and augmented reality. With the rapid progress and widespread application of artificial intelligence technology, virtual digital humans have penetrated into many fields such as intelligent customer service, video entertainment, and personalized virtual images. However, these digital human models usually have millions of faces when modeling. If they are to be deployed on mobile terminals, they must be mesh simplified to run smoothly. Mesh simplification reduces the complexity of the mesh model, which not only improves the rendering efficiency of the mesh and reduces the computational load of the model, but also the simplified mesh model occupies less memory, which is especially important for mobile devices or embedded systems with limited memory resources.

[0003] Current mesh simplification methods still face challenges in balancing speed and accuracy. Specifically, methods such as vertex clustering can quickly generate low-resolution mesh models, but their geometric accuracy is poor; while edge collapse methods represented by QEM (Surface simplification using quadric error metrics) can maintain good geometric accuracy in a relatively short time, the polygon quality of the mesh is poor; although re-meshing methods can generate high-quality triangles, their geometric accuracy is often difficult to surpass QEM, and because such methods usually require iterative calculations, the time cost is significantly increased. Therefore, developing a mesh simplification method that can not only ensure high geometric accuracy and triangle quality, but also achieve breakthroughs in speed is an issue that the academic community urgently needs to solve. Summary of the invention

[0004] In order to overcome the defects of the above-mentioned prior art, the purpose of the present invention is to provide a projection simplification method for a three-dimensional digital human mesh model, which can not only surpass the QEM method in geometric accuracy, significantly improve the quality of triangles, and improve the display effect of the digital human model, but also maintain a low time consumption, solving the problem of difficulty in balancing speed and accuracy when simplifying the mesh.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for simplifying projection of a three-dimensional digital human mesh model comprises the following steps:

[0007] Step 1: Construct Kd-tree based on the original mesh information and store the corresponding normal information:

[0008] Get the vertex and face information of the 3D digital human mesh model, store the centroids of the vertices and faces in the Kd-tree, and store the corresponding vertex normals and face normals in the vector container;

[0009] Step 2: Traverse each edge of the original digital human mesh and calculate the projection point of the edge midpoint on the surface of the original digital human mesh as the collapse point v of the edge. new , and calculate v new The sum of the volumes of the tetrahedrons formed by the neighboring faces of the edge is taken as the collapse cost, and v new The cost and the starting point of the edge are stored in the structure struct, and these structure instances are stored in a minimum heap H arranged based on cost, that is, the one with the smallest cost is placed at the top of the heap, so that all data are arranged in order from small to large;

[0010] Step 3: Collapse of loop edges and update of vertices: Select the edge with the smallest cost from the minimum heap to collapse, update the vertex coordinates and its neighborhood connection relationship, calculate the collapsed minimum heap of the edge composed of the new vertex and its neighborhood points, and update the minimum heap; repeat the above until the required number of simplified faces is met;

[0011] Step 4: Edge flipping: Flipping sharp edges will affect the model's feature contours, so only non-sharp edges in the approximate plane area are flipped to reduce the appearance of narrow triangles and improve triangle quality. Triangle quality directly affects the display effect of the digital human model. Poor triangle quality will lead to map distortion and action binding deformation.

[0012] Preferably, in step 1, the center of gravity of the vertex and the face are both used as the Kd-tree of the query point, which expands the nearest point search range and improves the search accuracy. Assuming that the three vertices of face f are v0, v1 and v2, the center of gravity of face f is expressed as:

[0013]

[0014] Preferably, in step 2, the projection point of the edge midpoint on the original grid is selected as the candidate collapse point v new , the specific calculation method of the projection point is as follows:

[0015] Let the two endpoints of edge e be v0 and v1, then the midpoint Using knn to search for v mid The closest point v in the Kd-tree closest , and get v according to the index value closestThe normalized normal vector n of mid Projection distance to the original digital human mesh

[0016] d=(v mid -v ctosest )·n

[0017] This gives the projection point

[0018] v pro =v mid -d*n.

[0019] Preferably, in step 2, the projection point v is calculated pro The sum of the volumes of the tetrahedrons formed by the neighboring faces of edge e is taken as the collapse cost cost. The specific steps are as follows:

[0020]

[0021] Where faces(e) represents the set of all faces in the neighborhood of edge e, area(f) represents the area of ​​face f, and dis(f) represents the projection point v pro The distance to the surface f. Suppose the three vertices of surface f are v0, v1 and v2 respectively. Then

[0022]

[0023] Assume that the unit normal vector of face f is n, the plane equation of face f is ax+by+cz+d=0, and the constant d is calculated as follows:

[0024] d=-n·v0

[0025] Then the projection point v pro Distance to surface f

[0026] dis=(n·v pro )+d

[0027]

[0028] Here, ω is the penalty weight parameter. Here, setting ω to 2.06 can achieve the best effect, thereby avoiding the situation where the model is concave during the simplification process.

[0029] Preferably, in step 3, firstly select the edge with the smallest cost from the minimum heap, determine whether the edge has collapsed, and if not, perform the edge collapse operation, and convert the v corresponding to the cost into new As a new vertex, and set both vertices of the edge to the deleted state, then traverse the points in the 1-ring neighborhood of the new vertex and calculate the relationship between these points and v new The midpoint of the edge is projected onto the original grid, and the new v is calculated using the method in step 2. newAnd cost, update the minimum heap, each edge collapse operation will delete two faces, and exit the loop when the number of deleted faces meets the requirement.

[0030] Preferably, in step 4, the sharp edge recognition and edge flip condition determination method is as follows:

[0031] Assume that the unit normal vectors of the two adjacent faces of edge e are n1 and n2 respectively. When the dihedral angle θ of these two faces satisfies the following conditions:

[0032]

[0033] If the edge is sharp, the edge flipping operation will affect the model feature contour. Edge flipping will delete the original edge. If the feature edge is flipped, the model feature will be missing or weakened.

[0034] Furthermore, the lengths of the two sides before and after the edge flip are compared. If the length of the side after flipping is less than the length of the side before flipping, the edge flipping operation is performed. This is because the longer the side, the more likely it is to form an obtuse triangle. If the side becomes shorter after the edge flipping, it means that it will be closer to an acute triangle.

[0035] Preferably, in order to better measure the geometric error of the mesh model before and after simplification, a measurement standard maximum projection distance MPD is proposed, and its calculation method is as follows:

[0036] MPD=max(({d pro (v i ,M j ), v i ∈X i},{d pro (v j ,M i ),v j ∈X j})

[0037] X i and X j Represents the grid model M i and M j The set of sampled points, d pro (v i ,M j ) indicates v i Click to M j projection distance.

[0038] The Hausdorff distance calculates the maximum value of the distance between the closest points of the two models, which is greatly affected by the number of sampling points. The projection distance proposed by this method is not affected by this and can measure the approximate error between the two models more scientifically than the Hausdorff distance.

[0039] Beneficial effects of the present invention:

[0040] The mesh simplification of the present invention has high geometric accuracy. Because of the use of the projection method, the result is superior to the classical QEM algorithm in indicators such as chamfer distance, Hausdorff distance and the projection distance proposed by the present invention; the sum of the volumes of each tetrahedron formed by the projection point of the edge midpoint and the edge neighborhood surface is used as the collapse cost to perform edge collapse mesh simplification, which reduces the calculation complexity and improves the simplification speed; the edge flipping operation is performed on non-sharp edges, which reduces the occurrence of narrow and long triangles and improves the triangle quality. At the same time, it is dozens of times faster than the iterative re-gridding method.

[0041] Even if there are unpopular edges and faces or internal boundaries in the digital human mesh model, it can be successfully simplified to ensure that the model's "appearance" does not drop. After binding the animation, the movement is still natural and smooth without "explosion". At the same time, the simplified model can run smoothly on the mobile terminal, which expands the application scenarios of digital humans and enriches people's life and entertainment methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Flowchart of the simplified method for mesh projection.

[0043] Figure 2 Schematic diagram of edge collapse operation.

[0044] Figure 3 Schematic diagram of edge flipping operation.

[0045] Figure 4 The results of the method of the present invention and the QEM method on the rabbit model, block model and digital tetrahedron model simplified to a model with 1000 faces. (a is the original model, b is QEM, and c is the present method).

[0046] Figure 5 This is a picture showing the effect of successfully loading a digital human mesh model with 800,000 faces on a mobile phone after being simplified by the method of the present invention. DETAILED DESCRIPTION

[0047] The present invention is further described in detail below in conjunction with embodiments.

[0048] The present invention discloses a method for simplifying the projection of a three-dimensional digital human mesh model. The specific implementation flowchart is as follows: Figure 1 As shown, the steps are as follows:

[0049] Step 1: Construct the nearest point query Kd-tree to facilitate the subsequent calculation of projection points and speed up the nearest point search.

[0050] The specific implementation steps of step 1 are:

[0051] First, obtain the vertex and face information of the 3D digital human mesh model, store the centroids of the vertices and faces in the Kd-tree, and store the corresponding vertex normals and face normals in the vector container. Suppose the three vertices of face f are v0, v1, and v2 respectively;

[0052] The calculation formula of the center of gravity of the surface is:

[0053]

[0054] If the read mesh has no normal information, you need to calculate the vertex normals and face normals first.

[0055] Step 2: Construct an edge-collapse minimum heap; prepare and sort candidate collapse points and collapse costs for later simplification.

[0056] The specific implementation steps of step 2 are:

[0057] First, traverse each edge of the original digital human mesh, and calculate the projection point of the edge midpoint on the original digital human mesh based on the normal vector information as the candidate collapse point v new , the specific calculation method of the projection point is as follows:

[0058] Let the two endpoints of edge e be v0 and v1, then the midpoint Using knn to search for v mid The closest point v in the Kd-tree closest , and get v in the vector container according to the index value closest The normalized normal vector n of mid Projection distance to the original mesh

[0059] d=(v mid -v closest )·n

[0060] This gives the projection point

[0061] v pro =v mid -d*n

[0062] Then calculate the projection point v pro The sum of the volumes of the tetrahedrons formed by the neighboring faces of edge e is taken as the collapse cost cost. The specific steps are as follows:

[0063]

[0064] Where faces(e) represents the set of all faces in the neighborhood of edge e, area(f) represents the area of ​​face f, and dis(f) represents the projection point v pro The distance to the face f. Suppose the three vertices of face f are v0, v1 and v2, then

[0065]

[0066] Assume that the unit normal vector of face f is n, the plane equation of face f is ax+by+cz+d=0, and the constant d is calculated as follows:

[0067] d=-n·v0

[0068] Then the projection point v pro Distance to surface f

[0069] dis=(n·v pro )+d

[0070]

[0071] Where ω is the penalty weight parameter to avoid model concavity during simplification.

[0072] Finally, v new The cost and the starting point of the edge are stored in a structure struct, and then the struct is stored in a minimum heap H arranged by cost.

[0073] Step 3: Collapse of loop edges and update of vertices; the core step of simplification, deleting points and faces;

[0074] The specific implementation steps of step 3 are:

[0075] After obtaining the edge-collapsed minimum heap H in step 2, select the edge with the smallest cost from the minimum heap to collapse, update the vertex coordinates and its neighborhood connection relationship, calculate the collapsed minimum heap of the edge composed of the new vertex and its neighborhood points, and update the minimum heap; repeat the above until the required number of simplified faces is met.

[0076] The specific method is:

[0077] First, select the edge with the smallest cost from the minimum heap and determine whether the edge has collapsed. If it has not collapsed, execute the following Figure 2 The edge collapse operation shown above converts the v corresponding to the cost into new As a new vertex, and set both vertices of the edge to the deleted state. Then traverse the points in the 1-ring neighborhood of the new vertex and calculate the differences between these points and v new The midpoint of the edge is projected onto the original grid, and the new v is calculated using the method in step 2. new And cost, update the minimum heap. Each edge collapse operation will delete two faces, and exit the loop when the number of deleted faces meets the requirement.

[0078] Step 4: Edge flipping; used to improve triangle quality, and finally most triangles become regular triangles.

[0079] The specific implementation steps of step 4 are:

[0080] The simplified grid obtained in step 3 is Figure 3 The edge flipping operation shown in the figure does not necessarily perform edge flipping on all edges. Sharp edges, as the most important contour features of the model, should be retained. And the quality of triangles will not necessarily improve after edge flipping, so it is necessary to determine whether the edge flipping operation is beneficial before performing edge flipping.

[0081] First, traverse each edge of the mesh to identify whether it is a sharp edge. Assume that the unit normal vectors of the two adjacent faces of edge e are n1 and n2 respectively, then when the dihedral angle θ of these two faces satisfies the following conditions:

[0082]

[0083] Then the edge is a sharp edge, and performing an edge flipping operation will affect the feature contour of the model.

[0084] Then, the lengths of the two edges before and after the edge flipping are compared. If the length of the edge after flipping is less than the length of the edge before flipping, the edge flipping operation can be performed.

[0085] In addition, if Figure 4 a. Figure 4 b. Figure 4 As shown in Figure c, the results of the method of the present invention and the QEM method are the results of the rabbit model, block model and digital tetrahedron model simplified to a model with 1000 faces. It can be seen from the figure that the triangle quality of the method of the present invention is significantly better than that of QEM, demonstrating the versatility and robustness of the method of the present invention, and also predicting the practical value of the method of the present invention in lightweight display of CAD models.

[0086] The present invention obtains the position of the simplified vertex by the method of edge midpoint reprojection. The algorithm takes into account the geometric accuracy and time efficiency of mesh simplification. Specifically, for each edge of the mesh, the projection point of its midpoint on the original mesh surface is calculated, and this is used as the new vertex after the edge collapses. The sum of the volumes of the tetrahedron formed by the new vertex and the adjacent faces of the original edge is used as the cost of edge collapse, and edge collapse simplification is performed according to the cost minimum heap. In order to further improve the quality of triangles, the algorithm flips the non-sharp edges to reduce the occurrence of low-quality triangles. Finally, the application introduces the average projection distance and the maximum projection distance to more intuitively evaluate the geometric error before and after the simplification of the mesh model. The method of the present invention can not only reduce the geometric error of mesh simplification, but also further improve the quality of the simplified mesh.

[0087] The following table shows the comparison of QEM / the method of the present invention in terms of geometric error and simplification time, with / before the QEM method and / after the present invention method. The geometric error comparison indicators include chamfer distance (CD), Hausdorff distance (HD) and maximum projection distance (MPD), and the best result is bolded. It can be seen that the method of the present invention achieves geometric accuracy that exceeds the QEM method at a lower time cost.

[0088] Table 1 Comparison of QEM / method of the present invention

[0089]

[0090] Figure 5 This is the effect diagram of successfully loading the 800,000-facet digital human mesh model on a mobile phone after being simplified by the method of the present invention. To facilitate action binding, the model is generally designed in the shape of "A" or "T". Figure 5 It shows the effect of the digital human model performing other actions on the mobile phone after mesh simplification and animation binding.

Claims

1. A three-dimensional digital human mesh model projection simplification method, characterized in that: The steps include: Step 1: Construct a Kd-tree based on the original mesh information and store the corresponding normal information: obtain the vertex and face information of the 3D digital human mesh model, store the centroid points of the vertices and faces in the Kd-tree, and store the corresponding vertex normals and face normals in the vector container; Step 2: Traverse each edge of the original digital human mesh and calculate the projection point of the edge midpoint on the surface of the original digital human mesh as the collapse point v of the edge. new , and calculate v new The sum of the volumes of the tetrahedrons formed by the neighboring faces of the edge is taken as the collapse cost, and v new The cost and the starting point of the edge are stored in a structure struct, and these structures are stored in a minimum heap H arranged based on cost; Step 3: Collapse of loop edges and update of vertices: Select the edge with the smallest cost from the minimum heap to collapse, update the vertex coordinates and its neighborhood connection relationship, calculate the collapsed minimum heap of the edge composed of the new vertex and its neighborhood points, and update the minimum heap; repeat the above until the required number of simplified faces is met; Step 4: Edge flipping operation: Flip the non-sharp edges of the approximate plane area to reduce the appearance of narrow triangles and improve the display effect of the digital human model.

2. A three-dimensional digital human mesh model projection simplification method according to claim 1, characterized in that: In step 1, the vertex and the center of gravity of the face are both used as the Kd-tree of the query point, and the nearest point search range is expanded. Assuming that the three vertices of face f are v0, v1 and v2, the center of gravity of face f is expressed as:

3. The method for simplifying the projection of a three-dimensional digital human mesh model according to claim 2, characterized in that: In step 2, the projection point of the edge midpoint on the original mesh is selected as the candidate collapse point v new , the specific calculation method of the projection point is as follows: Let the two endpoints of edge e be v0 and v1, then the midpoint Using knn to search for v mid The closest point v in the Kd-tree closest , and get v according to the index value closest The normalized normal vector n of mid Projection distance to the original digital human mesh d=(v mid -v closest )·n This gives the projection point v pro =v mid -d*n。 4. The method for simplifying the projection of a three-dimensional digital human mesh model according to claim 3, characterized in that: In step 2, calculate the projection point v pro The sum of the volumes of the tetrahedrons formed by the neighboring faces of edge e is taken as the collapse cost cost. The specific steps are as follows: Where faces(e) represents the set of all faces in the neighborhood of edge e, area(f) represents the area of ​​face f, and dis(f) represents the projection point v pro The distance to the surface f. Suppose the three vertices of surface f are v0, v1 and v2 respectively. Then Assume that the unit normal vector of face f is n, the plane equation of face f is ax+by+cz+d=0, and the constant d is calculated as follows: d=-n·v0 Then the projection point v pro Distance to surface f dis=(n·v pro )+d Where ω is the penalty weight parameter.

5. The method for simplifying the projection of a three-dimensional digital human mesh model according to claim 1, characterized in that: In step 3, first select the edge with the smallest cost from the minimum heap, and determine whether the edge has collapsed. If it has not collapsed, perform the edge collapse operation and replace the v corresponding to the cost with the new As a new vertex, and set both vertices of the edge to the deleted state, then traverse the points in the new vertex 1 ring neighborhood, calculate the relationship between these points and v new The midpoint of the formed edge is the projection point of the original grid, and the new v is calculated using the method in step 2 new And cost, update the minimum heap, each edge collapse operation will delete two faces, and exit the loop when the number of deleted faces meets the requirement.

6. The method for simplifying the projection of a three-dimensional digital human mesh model according to claim 1, characterized in that: In step 4, the sharp edge recognition and edge flip condition determination method is as follows: Assume that the unit normal vectors of the two adjacent faces of edge e are n1 and n2 respectively. When the dihedral angle θ of these two faces satisfies the following conditions: Then the edge is a sharp edge, and edge flipping will affect the model feature contour. Edge flipping will delete the original edge. If the feature edge is flipped, the model feature will be missing or weakened.

7. The method for simplifying the projection of a three-dimensional digital human mesh model according to claim 6, characterized in that: Compare the lengths of the two edges before and after the edge flip. If the length of the edge after flipping is less than the length of the edge before flipping, perform the edge flipping operation.

8. The method for simplifying the projection of a three-dimensional digital human mesh model according to claim 1, characterized in that: To measure the geometric error of the mesh model before and after simplification, the maximum projection distance MPD is proposed, and its calculation method is as follows: MPD=max({d pro (v i ,M j ),v i ∈X i },{d pro (v j ,M i ),v j ∈X j }) X i and X j Represents the grid model M i and M j The set of sampled points, d pro (v i ,M j ) indicates v i Click to M j projection distance.