A virtual collision implementation method and device

Abstract

The application discloses a virtual collision implementation method and device, the method comprises the following steps: based on the space attribute of the four-dimensional Gaussian asset, a proxy boundary field of the potential contact area of the grid physical asset is constructed; wherein the proxy boundary field comprises a density gradient field, a deformation tensor field and a topologically adaptive proxy grid; according to the field value of each face sheet in the proxy grid in the density gradient field and the deformation tensor field, a target collision area is determined; the face sheet of the proxy grid in the target collision area and the face sheet of the grid physical asset are intersected and detected; for each contact pair detected, a collision energy constraint function of the contact pair is constructed and the contact force of the contact pair is solved; the contact force of the contact pair is inversely mapped into the four-dimensional Gaussian asset, so that the four-dimensional Gaussian asset is rendered and adjusted by using the inverse mapping result, and a collision effect is obtained.

Description

Technical Field

[0001] This application relates to the fields of computer graphics and physical simulation technology, and in particular to a method and apparatus for implementing virtual collision. Background Technology

[0002] The character and special effects assets (usually called 4D Gaussian assets) generated by the new generation of 4D Gaussian Splatting (4DGS) high-fidelity dynamic rendering technology cannot directly achieve collision interaction in the physics engine because there is an insurmountable format gap between the underlying data representation and traditional Mesh assets (polygonal mesh physical assets, which can be simply referred to as mesh physical assets).

[0003] To achieve collision interaction between the two, three main methods are currently employed. The first method converts the 4DGS point cloud into a dense voxel mesh in each frame and utilizes a standard collision detection pipeline between voxels and mesh assets to achieve collision interaction. The second method extracts the surface mesh assets from the 4DGS point cloud and then uses a collision solver between mesh assets to achieve collision interaction. The third method treats each Gaussian sphere as an independent physical particle, allowing point-to-face intersection detection to be performed on the mesh triangles, thus achieving collision interaction.

[0004] However, existing methods, whether voxelization, full reconstruction, or point-by-point testing, all require computation on the global data of 4DGS assets, resulting in a large computational load and low efficiency in implementing collision interaction, failing to effectively guarantee real-time collision interaction. Furthermore, converting 4DGS assets to mesh or voxels leads to the loss of the continuity and viewpoint-dependent characteristics of Gaussian representation, making it impossible to fully maintain high-fidelity rendering quality. Summary of the Invention

[0005] In view of the shortcomings of the prior art, this application provides a virtual collision implementation method and apparatus to solve the problem that the prior art cannot achieve high-fidelity real-time collision interaction between four-dimensional Gaussian assets and mesh physical assets.

[0006] To achieve the above objectives, this application provides the following technical solution:

[0007] The first aspect of this application provides a method for implementing virtual collision, including:

[0008] Based on the spatial properties of four-dimensional Gaussian assets, a proxy boundary field is constructed for the potential contact region with the mesh physical asset; wherein, the proxy boundary field includes a density gradient field, a deformation tensor field, and a topology-adaptive proxy mesh;

[0009] The target collision region is determined based on the field values ​​of each patch in the proxy mesh in the density gradient field and the deformation tensor field.

[0010] Intersection detection is performed between the facets of the proxy mesh and the facets of the mesh physical asset within the target collision region;

[0011] For each detected contact pair, a collision energy constraint function for the contact pair is constructed and the contact force of the contact pair is solved.

[0012] The contact force of the contact pair is inversely mapped to the four-dimensional Gaussian asset, and the result of the inverse mapping is used to render and adjust the four-dimensional Gaussian asset to obtain the collision effect.

[0013] Optionally, in the above-described virtual collision implementation method, the construction of a proxy boundary field for the potential contact region with the mesh physical asset based on the spatial properties of the four-dimensional Gaussian asset includes:

[0014] Intersection detection is performed on the axial bounding boxes of the four-dimensional Gaussian asset and the mesh physical asset to obtain potential contact areas;

[0015] A density gradient field is generated based on the density contribution of each first Gaussian sphere; wherein the first Gaussian sphere is a Gaussian sphere in the four-dimensional Gaussian asset within the potential contact region;

[0016] The spatial gradient of each first Gaussian sphere is calculated based on its inter-frame displacement to form a deformation tensor field.

[0017] Each sampling vertex is determined based on the density gradient field, and after the deformation displacement bias is applied to each sampling vertex based on the deformation tensor field, the sampling vertices are connected in patches to form a topology-adaptive proxy mesh.

[0018] The density gradient field, the deformation tensor field, and the proxy mesh are combined to form the proxy boundary field.

[0019] Optionally, in the above-described virtual collision implementation method, generating the density gradient field based on the density contribution of each first Gaussian sphere includes:

[0020] Sampling is performed within the potential contact area at a set step size to obtain multiple sampling vertices;

[0021] The density contribution of each of the first Gaussian spheres in the neighborhood of each of the sampling vertices is accumulated to obtain the scalar density field;

[0022] The scalar density field is solved by spatial gradient calculation to obtain the density gradient field.

[0023] Optionally, in the above-described virtual collision implementation method, the step of determining each sampling vertex based on the density gradient field, applying deformation displacement bias to each sampling vertex based on the deformation tensor field, and then connecting each sampling vertex with patches to form a topology-adaptive proxy mesh includes:

[0024] The isosurface is determined based on the density gradient field, and vertex sampling is performed on the isosurface according to the density gradient field to obtain the initial sampled vertices;

[0025] Based on the deformation tensor field, each of the sampling vertices is displaced and biased along the dominant deformation direction;

[0026] Based on a preset algorithm, each of the biased sampled vertices is connected in patches to form a topology-adaptive proxy mesh.

[0027] Optionally, the above-described virtual collision implementation method further includes:

[0028] If the inter-frame increment of the maximum eigenvalue of any target collision region in the deformation tensor field exceeds a preset increment threshold, the proxy mesh is reconstructed by performing center weight interpolation on the vertices of the current proxy mesh.

[0029] Optionally, in the above-described virtual collision implementation method, determining the target collision region based on the field magnitude values ​​of each facet in the proxy mesh in the density gradient field and the deformation tensor field includes:

[0030] Calculate the face center position of each face based on the vertex coordinates of each face in the proxy mesh;

[0031] The field values ​​of the face center positions of each of the face patches in the density gradient field and the deformation tensor field are compared with the corresponding thresholds to classify each face patch into an inert face or an active face.

[0032] The active surfaces are clustered according to their spatial connectivity to form target collision regions.

[0033] Optionally, in the above-described virtual collision implementation method, the step of constructing a collision energy constraint function for each detected contact pair and solving for the contact force of the contact pair includes:

[0034] For each detected contact pair, a dynamically weighted collision energy constraint function for the contact pair is constructed;

[0035] The collision energy constraint function of the contact pair at the contact point is solved to obtain the contact force of the contact pair.

[0036] Optionally, in the above-described virtual collision implementation method, the step of inversely mapping the contact force of the contact pair to the four-dimensional Gaussian asset, and then using the result of the inverse mapping to render and adjust the four-dimensional Gaussian asset to obtain a collision effect, includes:

[0037] The contact force of the contact pair is inversely mapped to the four-dimensional Gaussian asset to obtain the parameter correction amount of each Gaussian sphere in the corresponding location area;

[0038] By using the parameter correction values ​​of each Gaussian sphere, the rendering of each Gaussian sphere is adjusted to obtain a collision effect.

[0039] Optionally, in the above-described virtual collision implementation method, the step of inversely mapping the contact force of the contact pair to the four-dimensional Gaussian asset to obtain the parameter correction amount of each Gaussian sphere in the corresponding location neighborhood includes:

[0040] Determine the density contribution weight of each second Gaussian sphere; wherein, the second Gaussian sphere is the Gaussian sphere within the neighborhood of the contact point location of the contact pair;

[0041] According to the density contribution weight of each second Gaussian sphere, the contact force of the contact pair is distributed to each second Gaussian sphere, and the position displacement of each second Gaussian sphere is calculated based on the distributed contact force.

[0042] The deformation tensor at the contact point of the contact pair is determined based on the deformation tensor field, and the covariance adjustment of each second Gaussian sphere is determined using the deformation tensor.

[0043] The position displacement and covariance adjustment of each second Gaussian sphere are time-series smoothed and used as the parameter correction amount for each second Gaussian sphere.

[0044] A second aspect of this application provides a virtual collision implementation apparatus, comprising:

[0045] A proxy construction unit is used to construct a proxy boundary field for potential contact regions with the mesh physical asset based on the spatial properties of the four-dimensional Gaussian asset; wherein the proxy boundary field includes a density gradient field, a deformation tensor field, and a topology-adaptive proxy mesh;

[0046] The collision region determination unit is used to determine the target collision region based on the field values ​​of each patch in the proxy mesh in the density gradient field and the deformation tensor field.

[0047] A collision detection unit is used to perform intersection detection on the facets of the proxy mesh and the facets of the mesh physical asset within the target collision area;

[0048] The force analysis unit is used to construct the collision energy constraint function of each detected contact pair and solve for the contact force of the contact pair.

[0049] The rendering adjustment unit is used to inversely map the contact force of the contact pair to the four-dimensional Gaussian asset, so as to use the result of the inverse mapping to render and adjust the four-dimensional Gaussian asset to obtain the collision effect.

[0050] Optionally, in the above-described virtual collision implementation apparatus, the proxy construction unit includes:

[0051] A coarse-grained detection unit is used to perform intersection detection on the axial bounding boxes of the four-dimensional Gaussian asset and the mesh physical asset to obtain potential contact areas;

[0052] A first generation unit is configured to generate a density gradient field based on the density contribution of each first Gaussian sphere; wherein the first Gaussian sphere is a Gaussian sphere in the four-dimensional Gaussian asset within the potential contact region;

[0053] The second generation unit is used to calculate the spatial gradient of each of the first Gaussian spheres based on their inter-frame displacement to form a deformation tensor field.

[0054] The third generation unit is used to determine each sampling vertex according to the density gradient field, and after performing deformation displacement bias on each sampling vertex based on the deformation tensor field, connect each sampling vertex with patches to form a topology-adaptive proxy mesh.

[0055] The constituent unit is used to combine the density gradient field, the deformation tensor field, and the proxy mesh to form the proxy boundary field.

[0056] Optionally, in the above-described virtual collision implementation apparatus, the first generation unit includes:

[0057] The first sampling unit is used to sample within the potential contact area at a set step size to obtain multiple sampling vertices;

[0058] An accumulation unit is used to accumulate the density contribution of each of the first Gaussian spheres in the neighborhood of each of the sampling vertices to obtain a scalar density field.

[0059] The gradient solving unit is used to solve the spatial gradient of the scalar density field to obtain the density gradient field.

[0060] Optionally, in the above-described virtual collision implementation apparatus, the third generation unit includes:

[0061] The second sampling unit is used to determine the isosurface according to the density gradient field, and to sample the vertices on the isosurface according to the density gradient field to obtain the initial sampling vertices;

[0062] The bias unit is used to offset each of the sampling vertices along the dominant deformation direction based on the deformation tensor field.

[0063] Vertex connection unit is used to perform patch connection on each of the biased sampled vertices based on a preset algorithm to form a topology-adaptive proxy mesh.

[0064] Optionally, the virtual collision implementation device described above further includes:

[0065] The reconstruction unit is used to reconstruct the proxy mesh by performing center weight interpolation on the vertices of the current proxy mesh when the inter-frame increment of the maximum eigenvalue of any target collision region in the deformation tensor field exceeds a preset increment threshold.

[0066] Optionally, in the above-described virtual collision implementation apparatus, the collision region determination unit includes:

[0067] The face center determination unit is used to calculate the face center position of each face based on the vertex coordinates of each face in the proxy mesh.

[0068] A classification unit is used to compare the field values ​​of the face center position of each of the face patches in the density gradient field and the deformation tensor field with the corresponding thresholds, and to classify each of the face patches into inert face patches or active face patches.

[0069] Clustering units are used to cluster the active surfaces according to their spatial connectivity to form target collision regions.

[0070] Optionally, in the above-described virtual collision implementation device, the force analysis unit includes:

[0071] The function construction unit is used to construct a dynamically weighted collision energy constraint function for each detected contact pair;

[0072] The function solving unit is used to solve the collision energy constraint function of the contact pair at the contact point location to obtain the contact force of the contact pair.

[0073] Optionally, in the above-described virtual collision implementation apparatus, the rendering adjustment unit includes:

[0074] The inverse mapping unit is used to inversely map the contact force of the contact pair to the four-dimensional Gaussian asset to obtain the parameter correction amount of each Gaussian sphere in the corresponding location area;

[0075] The Gaussian sphere rendering unit is used to adjust the rendering of each Gaussian sphere by utilizing the parameter correction amount of each Gaussian sphere to obtain a collision effect.

[0076] Optionally, in the above-described virtual collision implementation device, the inverse mapping unit includes:

[0077] A weighting determination unit is used to determine the density contribution weight of each second Gaussian sphere; wherein, the second Gaussian sphere is a Gaussian sphere within the neighborhood of the contact point location of the contact pair;

[0078] The displacement determination unit is used to distribute the contact force of the contact pair to each of the second Gaussian spheres according to the density contribution weight of each of the second Gaussian spheres, and to calculate the position displacement of each of the second Gaussian spheres based on the distributed contact force.

[0079] The adjustment amount determination unit is used to determine the deformation tensor at the contact point position of the contact pair based on the deformation tensor field, and to determine the covariance adjustment amount of each second Gaussian sphere using the deformation tensor.

[0080] The smoothing unit is used to perform time-series smoothing on the position displacement and covariance adjustment of each of the second Gaussian spheres, and then use the result as the parameter correction amount for each of the second Gaussian spheres.

[0081] A third aspect of this application provides a computer storage medium for storing a computer program, which, when executed by a processor, is used to implement the virtual collision implementation method as described in any of the preceding claims.

[0082] This application provides a virtual collision implementation method that constructs a proxy boundary field for the potential contact region between a four-dimensional Gaussian asset and a mesh physical asset, based on the spatial properties of the Gaussian asset. The proxy boundary field includes a density gradient field, a deformation tensor field, and a topology-adaptive proxy mesh. This proxy boundary field acts as a proxy for the four-dimensional Gaussian asset for collision calculations, eliminating the need for transformation of the Gaussian asset and performing calculations only on the potential contact region. Therefore, it maintains high-fidelity rendering quality while significantly reducing computational load. Then, based on the field values ​​of each facet in the proxy mesh in the density gradient field and deformation tensor field, the target collision region is determined. This allows for precise location of the collision region based on the density gradient and deformation tensor, enabling accurate collision analysis and further reducing computational load. Finally, intersection detection is performed on the faces of the proxy mesh and the faces of the mesh physical asset within the target collision region. For each detected contact pair, a collision energy constraint function is constructed, and the contact force is solved, thus determining the contact force during collision based on energy conservation constraints. Next, the contact forces of the contact pairs are inversely mapped onto the four-dimensional Gaussian asset, thereby determining the influence of the contact forces on the Gaussian spheres in the four-dimensional Gaussian asset through inverse mapping. Therefore, the results of the inverse mapping are used to adjust the rendering of the four-dimensional Gaussian asset, thereby rendering the effect affected by the contact forces, that is, the collision effect. Thus, high-fidelity real-time collision interaction is achieved. Attached Figure Description

[0083] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0084] Figure 1 A flowchart illustrating a virtual collision implementation method provided in this application embodiment;

[0085] Figure 2 A flowchart illustrating a method for constructing a proxy boundary field provided in an embodiment of this application;

[0086] Figure 3 A flowchart illustrating a method for generating a density gradient field as provided in an embodiment of this application;

[0087] Figure 4 A flowchart illustrating a method for generating a proxy mesh, as provided in an embodiment of this application;

[0088] Figure 5 A flowchart illustrating a method for determining a target collision region provided in an embodiment of this application;

[0089] Figure 6 A flowchart illustrating a method for calculating the contact force of a contact pair, provided as an embodiment of this application;

[0090] Figure 7 A flowchart illustrating a method for inverse mapping of contact force provided in this application embodiment;

[0091] Figure 8 This is a schematic diagram of the architecture of a virtual collision implementation device provided in an embodiment of this application. Detailed Implementation

[0092] 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 embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0093] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0094] This application provides a method for implementing virtual collision, such as... Figure 1 As shown, it includes the following steps:

[0095] S101. Based on the spatial properties of four-dimensional Gaussian assets, construct a proxy boundary field for potential contact areas with grid physical assets.

[0096] The surrogate boundary field includes the density gradient field, the deformation tensor field, and the topology-adaptive surrogate mesh.

[0097] It should be noted that, in this embodiment, instead of converting 4DGS assets into Mesh assets or voxels before implementing collision interaction, a multi-level lightweight proxy structure, namely a proxy boundary field, is constructed in real time based on the spatial properties of the 4D Gaussian assets. The proxy boundary field includes a density gradient field, a deformation tensor field, and a topology-adaptive proxy mesh, representing the density gradient, deformation tensor, and mesh of the 4D Gaussian assets, respectively. Therefore, physical collision can be achieved by proxying the 4D Gaussian assets through the proxy boundary field. All physical collision calculations are first performed on this proxy boundary field, and then the structure of the proxy calculation is inversely mapped back to the Gaussian parameters of the 4DGS assets to obtain the Gaussian parameters of the 4DGS assets when collision interaction occurs. This allows for the rendering of the collision effect between 4DGS assets and Mesh assets, thus maintaining high-fidelity rendering quality. Furthermore, by avoiding the conversion of a large number of 4DGS assets, computational efficiency is effectively improved.

[0098] Specifically, the density gradient field (implicit surface normal) can be extracted from the 4DGS Gaussian mixture distribution, the deformation tensor feature field (deformation trend) can be extracted from the Gaussian parameter changes of adjacent frames, and a topology adaptive proxy mesh can be generated based on the joint of the two to obtain the proxy boundary field.

[0099] To reduce computational load and improve computational efficiency, thereby effectively ensuring real-time collision interaction, this embodiment analyzes the potential collision regions between the four-dimensional Gaussian asset and the mesh physical asset, and constructs only proxy boundary fields for these potential collision regions, thus calculating only local potential collision regions. Optionally, coarse-grained collision detection can be performed on the four-dimensional Gaussian asset and the mesh physical asset to obtain their potential collision regions.

[0100] Optionally, in another embodiment of this application, one specific implementation of step S101 is as follows: Figure 2 As shown, it includes the following steps:

[0101] S201. Perform intersection detection on the axial bounding boxes of the 4D Gaussian asset and the mesh physical asset to obtain the potential contact area.

[0102] Specifically, a coarse-grained intersection pre-check is performed on the axial bounding boxes of 4DGS assets and Mesh assets. If an intersection is detected between the bounding boxes of the two assets, the intersection region is extended outward by a safety margin to identify a potential contact region. Optionally, since the extended safety margin is used to compensate for potential missed detections caused by high-speed inter-frame motion, the safety margin can be the product of the Mesh motion speed and the frame interval.

[0103] In the 4DGS asset, the density contribution of each Gaussian sphere to any point in space follows a three-dimensional Gaussian distribution function. Therefore, the density contribution can be determined based on this function, that is, the degree of influence of each Gaussian sphere on the density of any point can be determined.

[0104] S202. Generate a density gradient field based on the density contribution of each first Gaussian sphere.

[0105] Among them, the first Gaussian sphere is the Gaussian sphere in the four-dimensional Gaussian asset within the potential contact area.

[0106] Specifically, based on the density contribution of each first Gaussian sphere, the density distribution information of Gaussian spheres at various locations within the potential contact area can be determined. Then, based on the changes in density distribution, the density gradient of Gaussian spheres at various locations within the potential contact area can be solved, forming the density gradient field of the potential contact area.

[0107] Optionally, in another embodiment of this application, one specific implementation of step S202 is as follows: Figure 3 As shown, it includes:

[0108] S301. Sample within the potential contact area with a set step size to obtain multiple sampling vertices.

[0109] Specifically, a uniform sampling grid is arranged within the contact area with a set step size to obtain multiple sampling vertices.

[0110] S302. The density contribution of each first Gaussian sphere in the neighborhood of each sampled vertex is accumulated to obtain the scalar density field.

[0111] Specifically, based on the center position and covariance matrix of each first Gaussian sphere, the density contribution of each first Gaussian sphere in the neighborhood of each sampling vertex is calculated using the standard three-dimensional Gaussian distribution function. These density contributions are then summed to obtain the scalar density field. Therefore, the scalar density field can be specifically represented as:

[0112] .

[0113] in, To sample vertices The set of Gaussian spheres within the neighborhood of the center, specifically the largest Gaussian sphere within the region with radius [missing information]. A set of Gaussian spheres within a certain range. and The center position and covariance matrix of the Gaussian sphere are obtained from the intrinsic parameters of the 4DGS asset. G is the standard three-dimensional Gaussian distribution function.

[0114] S303. Solve for the spatial gradient of the scalar density field to obtain the density gradient field.

[0115] Specifically, the density gradient field is obtained by taking the partial derivative of the scalar density field. Therefore, the specific calculation method of the density gradient field is as follows:

[0116] .

[0117] S203. Calculate the spatial gradient of each first Gaussian sphere based on its inter-frame displacement to form a deformation tensor field.

[0118] Specifically, if the current frame is... i The positions of the Gaussian spheres are The previous frame was Then the inter-frame displacement of the Gaussian sphere is Therefore, within the potential contact region, the spatial gradient of the displacement field composed of inter-frame displacements is calculated, i.e., partial derivatives are taken with respect to the three axes to obtain the deformation tensor field. :

[0119] .

[0120] S204. Determine each sampling vertex based on the density gradient field, and after applying deformation displacement bias to each sampling vertex based on the deformation tensor field, connect each sampling vertex with patches to form a topology-adaptive proxy mesh.

[0121] Specifically, vertices are adaptively sampled based on the density gradient field, and the positional offset of each sampled vertex due to deformation is determined based on the deformation tensor field, thereby displacing each sampled vertex. Then, the sampled vertices are connected to form patches, and the patches together form a proxy mesh that achieves topological adaptation based on the density gradient field and the deformation tensor field.

[0122] Optionally, in another embodiment of this application, one specific implementation of step S204 is as follows: Figure 4 As shown, it includes:

[0123] S401. Determine the isosurface based on the density gradient field, and sample the vertices on the isosurface according to the density gradient field to obtain the initial sampled vertices.

[0124] Specifically, the density gradient in the density gradient field The specified isosurface is the seed surface. The isosurface condition is... , Set a preset isosurface gradient threshold. Apply the gradient to this isosurface. The size of the sampled vertices is adaptively selected. Among them, Larger location sampling vertex density Small position sampling vertices are sparse.

[0125] S402. Based on the deformation tensor field, each sampling vertex is shifted and biased along the dominant deformation direction.

[0126] Specifically, each sample vertex Along the deformation tensor The displacement is offset according to the dominant deformation direction, specifically expressed as:

[0127] .

[0128] in, The preset expansion coefficient is used to control the degree of deviation between the proxy mesh and the surface of the 4DGS asset; Let Frobenius norm be the deformation tensor at the sampling vertex; for The dominant eigenvector direction (unit vector) at the vertex is the dominant deformation direction. Therefore, this calculation method enables the proxy mesh to expand outward in areas of severe deformation to reserve deformation tolerance space, and to adhere closely to the asset surface in areas of gentle deformation, thus achieving topology-adaptive mesh generation.

[0129] S403. Based on a preset algorithm, each sampled vertex after biasing is connected to form a topology-adaptive proxy mesh.

[0130] Optionally, the preset algorithm can be a triangulation algorithm. Specifically, the biased sampled vertices can be connected to form a patch network based on constrained Delaunay triangulation (connecting only vertex pairs with distances within a threshold to avoid incorrect connections across gaps), thus forming a topology-adaptive proxy mesh.

[0131] S205. The density gradient field, deformation tensor field, and proxy mesh are combined to form a proxy boundary field.

[0132] S102. Determine the target collision region based on the field values ​​of each facet in the density gradient field and deformation tensor field.

[0133] It should be noted that the potential contact area is only a coarse-grained defined region, so it still contains a large number of redundant faces located inside the asset that have never made contact. If these faces are not screened, the subsequent surface intersection detection will perform meaningless collision detection using the GJK algorithm and the EPA algorithm on these invalid faces.

[0134] Due to the impact of collisions, the density gradient and deformation tensor of the patches within the collision region are greater than those of the patches in the non-collision region. Therefore, based on the field values ​​of each patch in the surrogate mesh in the density gradient field and deformation tensor field, i.e., the density gradient and deformation tensor of each patch, the patches belonging to the collision region can be determined, thus forming at least one target collision region. Then, subsequent steps can be performed separately for each target collision region.

[0135] Optionally, in another embodiment of this application, one specific implementation of step S102 is as follows: Figure 5 As shown, it includes:

[0136] S501. Calculate the face center position of each face based on the vertex coordinates of each face in the proxy mesh.

[0137] It should be noted that, in the embodiments of this application, the density gradient and deformation tensor at the face center position of the face are used as the density gradient and deformation tensor of the face. Therefore, the face center position of each face is calculated based on the vertex coordinates of each face in the proxy mesh.

[0138] Specifically, since the face is a triangular face, for the k-th triangular face in the proxy mesh... Calculate its face center position :

[0139] .

[0140] in, , , Let be the coordinates of the three vertices of the triangular facet.

[0141] S502. Compare the field values ​​of the center position of each facet in the density gradient field and deformation tensor field with the corresponding thresholds to classify each facet into an inert facet or an active facet.

[0142] Specifically, in the center of the face The density gradient at that location is obtained through density gradient field interpolation. Similarly, the maximum deformation eigenvalue is obtained by eigendecomposition of the deformation tensor field. Then, the obtained density gradient and maximum deformation eigenvalue are compared with the corresponding thresholds to classify the patches into inert patches with small changes or active patches with large changes. Specifically, if both the density gradient and the maximum deformation eigenvalue are not less than the corresponding threshold, the patch is classified as an active patch; otherwise, it is classified as an inert patch. This can be expressed as follows:

[0143] .

[0144] .

[0145] in, Density gradient threshold; The threshold value is the deformation characteristic value.

[0146] S503. Cluster the active surfaces according to their spatial connectivity to form target collision regions.

[0147] Since active surfaces are the surfaces that come into contact, we cluster these active surfaces according to their spatial connectivity, that is, we cluster connected surfaces to form collision regions for each target. This can be represented as follows:

[0148] .

[0149] in, The connected component analysis algorithm uses whether faces share edges as the basis for determining connectivity (i.e., if two active triangles share at least one vertex, they belong to the same connected component). For the first The target collision area includes All active aspects of it.

[0150] S103. Perform intersection detection on the facets of the proxy mesh and the facets of the mesh physical asset within the target collision area.

[0151] Specifically, for each target collision region, the triangular facets of the proxy mesh and the spatially adjacent triangular facets of the mesh assets are used to perform standard mesh asset-to-mesh asset intersection detection using the existing GJK-EPA collision detection pipeline. The collision geometry is constructed by adding an inward offset along the density gradient direction to the vertices of the proxy mesh's triangular facets. The offset is taken as the average value of the Gaussian spheres at the vertices of the proxy mesh's triangular facets. This value allows the contact depth of the collision detection to approximate the actual physical boundary of the Gaussian sphere.

[0152] S104. For each detected contact pair, construct the collision energy constraint function of the contact pair and solve for the contact force of the contact pair.

[0153] It should be noted that intersection detection can detect each contact pair. Each contact pair consists of the collision geometry of a proxy mesh and the triangle of a Mesh asset, i.e., the collision pair of the two assets.

[0154] Since the energy during a collision conforms to physical constraints, namely the conservation of the total momentum of the contact system before and after the collision response, a corresponding collision energy constraint function can be constructed for the contact pair. Then, the contact energy constraint function can be solved to obtain the contact force of the contact pair.

[0155] Optionally, in another embodiment of this application, one specific implementation of step S104 is as follows: Figure 6 As shown, it includes:

[0156] S601. For each detected contact pair, construct a dynamically weighted collision energy constraint function for the contact pair.

[0157] The collision energy constraint function is a weighted function of the anti-penetration term, the elastic deformation term, and the momentum conservation term according to the degree of deformation.

[0158] Specifically, the anti-penetration term's function models the proxy mesh penetration restoring force as a spring and damping system. The deeper the penetration and the faster the penetration speed, the stronger the restoring force, preventing 4DGS assets from penetrating the mesh. This can be expressed as:

[0159] .

[0160] in, for The depth to which a facet penetrates the mesh surface is output by GJK / EPA collision detection (intersection depth). The penalty stiffness coefficient is a preset parameter of the physical material and is determined by the physical engine's contact material configuration or system preset parameters. The damping coefficient is determined jointly by the physics engine frame interval and the contact surface recovery coefficient. The rate of change of penetration depth is calculated from the penetration depth difference between frames, specifically as follows: .

[0161] The elastic deformation term function automatically reduces the elastic constraint stiffness in the contact area where the 4DGS asset is undergoing severe deformation. This allows the collision response to adapt to the asset's own deformation trend rather than forcibly bouncing back, avoiding unnatural and stiff rebound effects. Specifically, it is expressed as follows:

[0162] .

[0163] in, The basic elastic stiffness of the contact area of ​​the contact pair is determined by... Estimate the average covariance eigenvalues ​​of the Gaussian spheres at the vertices of the facet; The deformation sensitivity coefficient is a preset constant that controls the degree of influence of deformation on stiffness. It is calculated from the Frobenius norm of the deformation tensor field at the contact location.

[0164] Correspondingly, the energy contributed by the elastic deformation term is:

[0165] .

[0166] in, The displacement gradient at the contact surface of the contact pair is determined by adjacent frames. Vertex displacement difference calculation.

[0167] The energy conservation term is based on the velocity difference before and after the collision and the effective mass at the point of contact. A standard momentum conservation constraint is constructed to ensure that the total momentum and total angular momentum of the system remain unchanged, specifically expressed as follows:

[0168] .

[0169] in:

[0170] .

[0171] .

[0172] in, In order to contact China The equivalent mass of the contact surface (estimated by the average mass of the Gaussian sphere at the vertex of the surface). The equivalent quality of the contact surface of the Mesh asset in the contact pair (calculated from the density and area of ​​the Mesh physical material); The relative velocity of the point of contact before the collision (read from the velocity field of the physics engine); Let be the relative velocity of the contact point of the contact pair after the collision (solved by the momentum conservation constraint); The collision impulse (calculated iteratively by the constraint solver); The collision normal vector (the contact normal direction is output by GJK / EPA).

[0173] To ensure the conservation of total momentum of the contact system before and after the collision response, the velocity change after the collision is constrained by effective mass weighting, preventing energy from being created or lost out of thin air. Specifically, the three terms are summed using adaptive weights to form the collision energy constraint function:

[0174] .

[0175] The adaptive weights are:

[0176] .

[0177] in, For the deformation and weight mapping coefficients (preset constants); It is calculated from the Frobenius norm of the deformation tensor field at the contact location.

[0178] Therefore, the three energy constraints are adaptively weighted and combined according to the degree of deformation in the contact region. As the deformation is more severe in the region (… (Increase) Automatically reduce elasticity weight and increase momentum weight, prioritizing the maintenance of motion trends in deforming assets. In static contact areas, elasticity weight dominates, achieving a natural physical response.

[0179] S602. Solve the collision energy constraint function of the contact pair at the contact point to obtain the contact force of the contact pair.

[0180] It should be noted that, in this embodiment, the spatial gradient of the collision energy constraint function at the contact point of the contact pair is transformed into a vector of contact force, ensuring that the direction of the contact force always points towards the direction of penetration recovery / energy minimization. Therefore, the contact force of the contact pair is obtained by solving the collision energy constraint function of the contact pair, specifically as follows:

[0181] .

[0182] in, The location of the contact point (output by GJK / EPA collision detection); the symbols in the formula indicate the direction of force reduction (restoring force direction).

[0183] S105. Inverse map the contact force of the contact pair to the four-dimensional Gaussian asset, and use the result of the inverse mapping to render and adjust the four-dimensional Gaussian asset to obtain the collision effect.

[0184] It should be noted that the contact force of the contact pair is calculated on the proxy boundary field and cannot be directly reflected in the four-dimensional Gaussian asset. Therefore, the contact force of the contact pair is inversely mapped back to the corresponding location in the four-dimensional Gaussian asset to obtain the inverse mapping result, which is the influence of the contact force of the contact pair on the four-dimensional Gaussian asset. In other words, the inverse mapping of the contact force in this embodiment is not simply a parameter transformation of the contact force, but rather a determination of its influence on the four-dimensional Gaussian asset when it acts on it. Therefore, by using the result of the inverse mapping to adjust the rendering of the four-dimensional Gaussian asset, the effect of the contact force of the contact pair generated when a collision occurs can be rendered, thus obtaining the collision effect.

[0185] Optionally, in another embodiment of this application, one specific implementation of step S105 includes:

[0186] The contact forces of the contact pairs are inversely mapped to the four-dimensional Gaussian asset to obtain the parameter corrections of each Gaussian sphere in the corresponding location area. The parameter corrections of each Gaussian sphere are then used to render and adjust each Gaussian sphere to obtain the collision effect.

[0187] It's important to note that since the contact force of a contact pair only affects a certain range, the process involves inversely mapping the contact force of the contact pair back to the corresponding Gaussian spheres in the neighborhood of the four-dimensional Gaussian asset. Specifically, this mapping back to each Gaussian sphere in the neighborhood of the contact point of the contact pair in the four-dimensional Gaussian asset. This causes the contact force of the contact pair to act on each Gaussian sphere in the corresponding neighborhood of the four-dimensional Gaussian asset, changing the parameters of each Gaussian sphere and thus obtaining the parameter correction amount for each Gaussian sphere. Then, the rendering is adjusted according to the parameter correction amounts of each Gaussian sphere, resulting in the rendered Gaussian spheres after the impact, thus rendering the collision effect. In detail, the parameter correction amounts of each Gaussian sphere can be written to the GPU, adjusting the parameters of each Gaussian sphere so that the next frame is rendered according to the adjusted parameters of each Gaussian sphere, thereby rendering the collision effect.

[0188] Alternatively, in another embodiment of this application, a specific implementation of inverse mapping of contact force is as follows: Figure 7 As shown, it includes:

[0189] S701. Determine the density contribution weight of each second Gaussian sphere.

[0190] The second Gaussian sphere is the Gaussian sphere within the area of ​​the contact point of the contact pair.

[0191] It should be noted that the Gaussian sphere with a larger density contribution weight obviously experiences greater force. Therefore, the density contribution weight of each second Gaussian sphere is determined in order to distribute the contact force of the contact pair and obtain the force of each second Gaussian sphere.

[0192] Specifically, regarding contact force Contact points of action The density contribution weights of each neighboring Gaussian sphere are:

[0193] .

[0194] in, The index of the Gaussian sphere currently assigned contact force; To traverse the contact point locations The set of Gaussian spheres in the neighborhood of , that is to Center, radius (Take the largest Gaussian sphere within the contact area) All Gaussian spheres within a radius of ( ); and The center position of the corresponding Gaussian sphere is read from the inherent parameters of the 4DGS asset; The covariance matrix of the corresponding Gaussian sphere is also read from the intrinsic parameters of the 4DGS asset; It is a standard three-dimensional Gaussian distribution function used to calculate the density contribution.

[0195] S702. According to the density contribution weight of each second Gaussian sphere, the contact force of the contact pair is distributed to each second Gaussian sphere, and the position displacement of each second Gaussian sphere is calculated based on the distributed contact force.

[0196] Specifically, the density contribution weight of the second Gaussian sphere is multiplied by the contact force of the contact pair to obtain the contact force allocated to the second Gaussian sphere. Then, the contact force of the second Gaussian sphere is converted into its actual spatial displacement. This displacement is directly proportional to the contact force and inversely proportional to the spatial expansion of the Gaussian sphere (larger Gaussian spheres have greater inertia), achieving a collision displacement that aligns with physical intuition. Therefore, the calculation method for the displacement is as follows:

[0197] .

[0198] in, The physical engine frame interval (system default); For the first Trace of the covariance matrix of a Gaussian sphere (read from the 4DGS parameter buffer). This is a reference density constant (system preset, related to the global density mean of 4DGS assets).

[0199] S703. Determine the deformation tensor at the contact point of the contact pair based on the deformation tensor field, and use the deformation tensor to determine the covariance adjustment of each second Gaussian sphere.

[0200] Specifically, the local compression or stretching caused by the collision is transferred to the shape parameters of the Gaussian sphere space. Therefore, strain is applied to the covariance matrix based on the deformation tensor, specifically expressed as:

[0201] .

[0202] in, This is the small strain tensor at the contact point. It is an identity matrix.

[0203] S704. The position displacement and covariance adjustment of each second Gaussian sphere are time-series smoothed and used as the parameter correction amount for each second Gaussian sphere.

[0204] Specifically, the position displacement and covariance adjustment of each second Gaussian sphere are the parameter corrections obtained from the initial calculation. To smoothly transition the deformation, the parameter corrections are calculated using an exponential moving average between the current frame and the previous frame, achieving temporal smoothing. The smoothing factor and... Eigenvalue correlation: the more severe the deformation, the less smoothness.

[0205] Alternatively, in another embodiment of this application, the following may be further performed:

[0206] If the inter-frame increment of the maximum eigenvalue of any target collision region in the deformation tensor field exceeds a preset increment threshold, the proxy mesh is reconstructed by performing center weight interpolation on the vertices of the current proxy mesh.

[0207] Optionally, when topological separation is detected, each separated subset can be independently marked as a new target collision region and a boundary closure patch can be added.

[0208] That is, in a certain within the area When the inter-frame increment of the maximum feature value exceeds a preset threshold, a local proxy mesh is triggered. The reconstruction process does not involve re-extracting... isosurface, but for the current The vertex performs center weight interpolation, specifically: for each vertex Based on its location Tensor projection interpolation calculates weights .in, For larger vertices, insert new vertices proportionally within their neighborhoods and re-delaunay partition; Smaller vertices remain stationary. Furthermore, upon detecting a topological separation, the system independently labels each subset as a new one. Sub-regions are processed and boundary closure patches are added. This allows for "local weighted sampling" instead of "full reconstruction," effectively reducing computational load.

[0209] This application provides a virtual collision implementation method. Based on the spatial properties of a four-dimensional Gaussian asset, a proxy boundary field is constructed for the potential contact region with the mesh physical asset. The proxy boundary field includes a density gradient field, a deformation tensor field, and a topology-adaptive proxy mesh. This proxy boundary field acts as a proxy for the four-dimensional Gaussian asset for collision calculations, eliminating the need for transformation of the four-dimensional Gaussian asset and performing calculations only on the potential contact region. Therefore, high-fidelity rendering quality is maintained while significantly reducing computational load. Then, based on the field values ​​of each facet in the proxy mesh in the density gradient field and deformation tensor field, the target collision region is determined. This allows for precise location of the collision region based on the density gradient and deformation tensor, enabling accurate collision analysis and further reducing computational load. Next, intersection detection is performed on the faces of the proxy mesh and the faces of the mesh physical asset within the target collision region. For each detected contact pair, a collision energy constraint function is constructed, and the contact force is solved, thereby determining the contact force during collision based on energy conservation constraints. Next, the contact forces of the contact pairs are inversely mapped onto the four-dimensional Gaussian asset, thereby determining the influence of the contact forces on the Gaussian spheres in the four-dimensional Gaussian asset through inverse mapping. Therefore, the results of the inverse mapping are used to adjust the rendering of the four-dimensional Gaussian asset, thereby rendering the effect affected by the contact forces, that is, the collision effect. Thus, high-fidelity real-time collision interaction is achieved.

[0210] Another embodiment of this application provides a virtual collision implementation device, such as... Figure 8 As shown, it includes:

[0211] The proxy building unit 801 is used to construct a proxy boundary field for potential contact regions with the mesh physical asset based on the spatial properties of the four-dimensional Gaussian asset. The proxy boundary field includes a density gradient field, a deformation tensor field, and a topology-adaptive proxy mesh.

[0212] The collision region determination unit 802 is used to determine the target collision region based on the field values ​​of each facet in the density gradient field and deformation tensor field in the proxy mesh.

[0213] The collision detection unit 803 is used to perform intersection detection between the facets of the proxy mesh and the facets of the mesh physical asset within the target collision area.

[0214] The force analysis unit 804 is used to construct the collision energy constraint function of each detected contact pair and solve for the contact force of the contact pair.

[0215] The rendering adjustment unit 805 is used to inversely map the contact force of the contact pair to the four-dimensional Gaussian asset, so as to use the result of the inverse mapping to render and adjust the four-dimensional Gaussian asset to obtain the collision effect.

[0216] Optionally, in another embodiment of the virtual collision implementation apparatus provided in this application, the proxy construction unit includes:

[0217] The coarse-grained detection unit is used to perform intersection detection on the axial bounding boxes of the 4D Gaussian asset and the mesh physical asset to obtain potential contact areas.

[0218] The first generation unit is used to generate a density gradient field based on the density contribution of each first Gaussian sphere. Here, the first Gaussian sphere is a Gaussian sphere in the four-dimensional Gaussian asset within the potential contact region.

[0219] The second generation unit is used to calculate the spatial gradient of each first Gaussian sphere based on its inter-frame displacement, forming a deformation tensor field.

[0220] The third generation unit is used to determine each sampling vertex based on the density gradient field, and after applying deformation displacement bias to each sampling vertex based on the deformation tensor field, it connects each sampling vertex with patches to form a topology-adaptive proxy mesh.

[0221] The constituent unit is used to combine the density gradient field, deformation tensor field, and surrogate mesh to form a surrogate boundary field.

[0222] Optionally, in another embodiment of the virtual collision implementation apparatus provided in this application, the first generation unit includes:

[0223] The first sampling unit is used to sample within the potential contact area at a set step size to obtain multiple sampling vertices.

[0224] The accumulation unit is used to accumulate the density contribution of each first Gaussian sphere in the neighborhood of each sampled vertex to obtain the scalar density field.

[0225] The gradient solving unit is used to solve for the spatial gradient of the scalar density field to obtain the density gradient field.

[0226] Optionally, in another embodiment of the virtual collision implementation apparatus provided in this application, the third generation unit includes:

[0227] The second sampling unit is used to determine the isosurface based on the density gradient field, and to sample the vertices on the isosurface according to the density gradient field to obtain the initial sampling vertices.

[0228] The bias unit is used to offset each sampling vertex along the dominant deformation direction based on the deformation tensor field.

[0229] Vertex connection units are used to connect the biased sampled vertices into patches based on a preset algorithm to form a topology-adaptive proxy mesh.

[0230] Optionally, in another embodiment of the virtual collision implementation device provided in this application, the device further includes:

[0231] The reconstruction unit is used to reconstruct the proxy mesh by performing center weight interpolation on the vertices of the current proxy mesh when the inter-frame increment of the maximum eigenvalue of any target collision region in the deformation tensor field exceeds a preset increment threshold.

[0232] Optionally, in another embodiment of the virtual collision implementation apparatus provided in this application, the collision region determination unit includes:

[0233] The face center determination element is used to calculate the face center position of each face based on the vertex coordinates of each face in the proxy mesh.

[0234] The classification unit is used to compare the field values ​​of the face center position of each facet in the density gradient field and the deformation tensor field with the corresponding thresholds, and classify each facet into an inert facet or an active facet.

[0235] Clustering units are used to cluster active surfaces according to spatial connectivity to form target collision regions.

[0236] Optionally, in another embodiment of the virtual collision implementation device provided in this application, the force analysis unit includes:

[0237] The function building unit is used to construct a dynamically weighted collision energy constraint function for each detected contact pair.

[0238] The function solving unit is used to solve the collision energy constraint function of the contact pair at the contact point to obtain the contact force of the contact pair.

[0239] Optionally, in another embodiment of the virtual collision implementation apparatus provided in this application, the rendering adjustment unit includes:

[0240] The inverse mapping unit is used to inversely map the contact force of the contact pair to the four-dimensional Gaussian asset to obtain the parameter correction of each Gaussian sphere in the corresponding location area.

[0241] The Gaussian sphere rendering unit is used to adjust the rendering of each Gaussian sphere by utilizing the parameter corrections of each Gaussian sphere to obtain collision effects.

[0242] Optionally, in another embodiment of the virtual collision implementation apparatus provided in this application, the inverse mapping unit includes:

[0243] The weighting determination unit is used to determine the density contribution weight of each second Gaussian sphere. Here, the second Gaussian sphere is the Gaussian sphere within the neighborhood of the contact point location of the contact pair.

[0244] The displacement determination unit is used to distribute the contact force of the contact pair to each second Gaussian sphere according to the density contribution weight of each second Gaussian sphere, and calculate the position displacement of each second Gaussian sphere based on the distributed contact force.

[0245] The adjustment amount determination unit is used to determine the deformation tensor at the contact point location of the contact pair based on the deformation tensor field, and to determine the covariance adjustment amount of each second Gaussian sphere using the deformation tensor.

[0246] The smoothing unit is used to perform time-series smoothing on the position displacement and covariance adjustment of each second Gaussian sphere, and then use it as the parameter correction amount for each second Gaussian sphere.

[0247] Another embodiment of this application provides a computer storage medium for storing a computer program, which, when executed by a processor, is used to implement the virtual collision implementation method provided in any of the above embodiments.

[0248] Computer storage media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0249] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0250] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for implementing virtual collision, characterized in that, include: Based on the spatial properties of four-dimensional Gaussian assets, a proxy boundary field is constructed for the potential contact region with the mesh physical asset; wherein, the proxy boundary field includes a density gradient field, a deformation tensor field, and a topology-adaptive proxy mesh; The target collision region is determined based on the field values ​​of each patch in the proxy mesh in the density gradient field and the deformation tensor field. Intersection detection is performed between the facets of the proxy mesh and the facets of the mesh physical asset within the target collision region; For each detected contact pair, a collision energy constraint function for the contact pair is constructed and the contact force of the contact pair is solved. The contact force of the contact pair is inversely mapped to the four-dimensional Gaussian asset, and the result of the inverse mapping is used to render and adjust the four-dimensional Gaussian asset to obtain the collision effect.

2. The method according to claim 1, characterized in that, The construction of a proxy boundary field for potential contact areas with the grid physical asset based on the spatial properties of the four-dimensional Gaussian asset includes: Intersection detection is performed on the axial bounding boxes of the four-dimensional Gaussian asset and the mesh physical asset to obtain potential contact areas; A density gradient field is generated based on the density contribution of each first Gaussian sphere; wherein the first Gaussian sphere is a Gaussian sphere in the four-dimensional Gaussian asset within the potential contact region; The spatial gradient of each first Gaussian sphere is calculated based on its inter-frame displacement to form a deformation tensor field. Each sampling vertex is determined based on the density gradient field, and after the deformation displacement bias is applied to each sampling vertex based on the deformation tensor field, the sampling vertices are connected in patches to form a topology-adaptive proxy mesh. The density gradient field, the deformation tensor field, and the proxy mesh are combined to form the proxy boundary field.

3. The method according to claim 2, characterized in that, The generation of the density gradient field based on the density contribution of each first Gaussian sphere includes: Sampling is performed within the potential contact area at a set step size to obtain multiple sampling vertices; The density contribution of each of the first Gaussian spheres in the neighborhood of each of the sampling vertices is accumulated to obtain the scalar density field; The scalar density field is solved by spatial gradient calculation to obtain the density gradient field.

4. The method according to claim 2, characterized in that, The step of determining each sampling vertex based on the density gradient field, applying deformation displacement bias to each sampling vertex based on the deformation tensor field, and then connecting each sampling vertex with patches to form a topology-adaptive proxy mesh includes: The isosurface is determined based on the density gradient field, and vertex sampling is performed on the isosurface according to the density gradient field to obtain the initial sampled vertices; Based on the deformation tensor field, each of the sampling vertices is displaced and biased along the dominant deformation direction; Based on a preset algorithm, each of the biased sampled vertices is connected in patches to form a topology-adaptive proxy mesh.

5. The method according to claim 1, characterized in that, Also includes: If the inter-frame increment of the maximum eigenvalue of any target collision region in the deformation tensor field exceeds a preset increment threshold, the proxy mesh is reconstructed by performing center weight interpolation on the vertices of the current proxy mesh.

6. The method according to claim 1, characterized in that, The step of determining the target collision region based on the field values ​​of each patch in the proxy mesh in the density gradient field and the deformation tensor field includes: Calculate the face center position of each face based on the vertex coordinates of each face in the proxy mesh; The field values ​​of the face center positions of each of the face patches in the density gradient field and the deformation tensor field are compared with the corresponding thresholds to classify each face patch into an inert face or an active face. The active surfaces are clustered according to their spatial connectivity to form target collision regions.

7. The method according to claim 1, characterized in that, For each detected contact pair, constructing a collision energy constraint function for the contact pair and solving for the contact force of the contact pair includes: For each detected contact pair, a dynamically weighted collision energy constraint function for the contact pair is constructed; The collision energy constraint function of the contact pair at the contact point is solved to obtain the contact force of the contact pair.

8. The method according to claim 1, characterized in that, The step of inversely mapping the contact force of the contact pair to the four-dimensional Gaussian asset, and then using the result of the inverse mapping to render and adjust the four-dimensional Gaussian asset to obtain a collision effect, includes: The contact force of the contact pair is inversely mapped to the four-dimensional Gaussian asset to obtain the parameter correction amount of each Gaussian sphere in the corresponding location area; By using the parameter correction values ​​of each Gaussian sphere, the rendering of each Gaussian sphere is adjusted to obtain a collision effect.

9. The method according to claim 8, characterized in that, The step of inversely mapping the contact force of the contact pair to the four-dimensional Gaussian asset to obtain the parameter corrections for each Gaussian sphere within the corresponding location area includes: Determine the density contribution weight of each second Gaussian sphere; wherein, the second Gaussian sphere is the Gaussian sphere within the neighborhood of the contact point location of the contact pair; According to the density contribution weight of each second Gaussian sphere, the contact force of the contact pair is distributed to each second Gaussian sphere, and the position displacement of each second Gaussian sphere is calculated based on the distributed contact force. The deformation tensor at the contact point of the contact pair is determined based on the deformation tensor field, and the covariance adjustment of each second Gaussian sphere is determined using the deformation tensor. The position displacement and covariance adjustment of each second Gaussian sphere are time-series smoothed and used as the parameter correction amount for each second Gaussian sphere.

10. A virtual collision implementation device, characterized in that, include: A proxy construction unit is used to construct a proxy boundary field for potential contact regions with the mesh physical asset based on the spatial properties of the four-dimensional Gaussian asset; wherein the proxy boundary field includes a density gradient field, a deformation tensor field, and a topology-adaptive proxy mesh; The collision region determination unit is used to determine the target collision region based on the field values ​​of each patch in the proxy mesh in the density gradient field and the deformation tensor field. A collision detection unit is used to perform intersection detection on the facets of the proxy mesh and the facets of the mesh physical asset within the target collision area; The force analysis unit is used to construct the collision energy constraint function of each detected contact pair and solve for the contact force of the contact pair. The rendering adjustment unit is used to inversely map the contact force of the contact pair to the four-dimensional Gaussian asset, so as to use the result of the inverse mapping to render and adjust the four-dimensional Gaussian asset to obtain the collision effect.

Images