A method and system for optimizing three-dimensional modeling of a simulated dinosaur

Abstract

The application discloses a kind of simulation dinosaur three-dimensional modeling optimization method and system, method constructs rigid body skeleton system based on Lagrange dynamics;According to the muscle layer of super-elastic material of anatomical attachment point, the displacement field is calculated using position-based dynamics method;Skin layer is constructed outside, Cauchy stress is projected to grid vertex, and wrinkle and scale texture are dynamically generated according to principal strain;The grid simplification degree and physical simulation accuracy of corresponding area are dynamically adjusted by calculating perceptual importance weight;Establish the quadratic programming problem with anatomical constraint to realize motion reorientation and angular momentum distribution;And construct deformation prediction cache model to carry out online constraint correction.The application realizes the unity of high-precision biological feature restoration and high-efficiency real-time rendering.

Description

Technical Field

[0001] This invention relates to the field of 3D modeling and computer simulation technology, specifically to a method and system for optimizing 3D modeling of simulated dinosaurs. Background Technology

[0002] With the rapid development of computer graphics and virtual reality technologies, 3D digital modeling has been widely applied in paleontology research, digital entertainment, and science education. Simulated dinosaur and other paleontological models not only need to present a rigorous anatomical structure in a static appearance, but also require them to exhibit biomechanically consistent skin and muscle deformation characteristics during dynamic interaction, while simultaneously meeting the performance requirements of real-time rendering. However, existing technologies face the following technical challenges when processing such complex biological models: Traditional modeling methods based on reverse engineering, such as the invention patent with authorization announcement number CN105224764B, obtain skeletal point clouds through CT scans and use discrete Coons volume interpolation and least squares method for surface reconstruction, which can obtain high-precision static skeletal geometric models. However, this method is highly dependent on complete solid references. For paleontology lacking soft tissue preservation, skeletal interpolation alone cannot deduce the nonlinear deformation relationship of muscles and skin during movement, resulting in anatomical distortion and topological tearing in dynamic scenes.

[0003] Chinese patent application CN117011455A discloses a method, device, and storage medium for real-time three-dimensional deformation simulation modeling of skin, belonging to the field of biotribology. The method includes: acquiring the dimensional information of the skin region; determining the positional information of each mass point in the simulation model based on the acquired dimensional information, and establishing a multi-layer mass spring physical model of skin-fat-muscle-bone; determining the spring coefficient in the established multi-layer mass spring physical model; establishing a skin viscosity and sliding motion model, acquiring the influence of viscosity and sliding friction between the object and the skin on the three-dimensional deformation of the skin during the lateral movement of the object on the skin; calculating the positional information of each mass point in the multi-layer mass spring physical model under the simultaneous action of normal force and viscosity-slip friction, performing surface fitting on the skin layer mass points, and acquiring the deformation information of the skin during movement.

[0004] Existing technologies generally employ distance-based discrete detail level switching strategies for computational resource scheduling, failing to consider the physiological characteristics of human visual perception. This results in unnecessary consumption of computational resources in non-visually sensitive areas, while potentially sacrificing visual quality due to oversimplification in highly perceptual areas. Current technologies lack a systematic solution that deeply integrates biological anatomical logic with digital modeling, making it impossible to achieve efficient computational resource scheduling while ensuring high-fidelity biological feature reconstruction. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for optimizing 3D modeling of simulated dinosaurs. By constructing a full-link physical simulation pipeline from skeleton to skin and combining it with a perception-driven adaptive optimization strategy, it achieves the unity of high-precision biological feature restoration and efficient real-time rendering.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for optimizing 3D modeling of simulated dinosaurs includes the following steps: Step S1: Obtain three-dimensional scanning data of biological skeletal fossils and generate a rigid skeletal system with mass attributes, rotational inertia attributes, and joint constraint attributes; establish the Lagrange dynamic equation in generalized coordinates, and calculate the motion state of the skeletal system under the action of generalized forces by solving the equation. The generalized forces include active muscle contraction force, passive ligament constraint force, and external contact force. Step S2: On the surface of the skeletal model, a muscle layer composed of tetrahedral meshes is constructed based on anatomical attachment points; the muscle tissue is defined as a hyperelastic material, and its deformation is obtained by solving for the minimum value of the strain energy density function; the displacement field of the muscle driven by skeletal motion is calculated in real time using a position-based dynamics method. Step S3: Construct a skin layer composed of triangular meshes outside the muscle layer; project the Cauchy stress tensor calculated from the muscle surface onto the vertices of the skin mesh as the driving force for skin deformation; dynamically generate wrinkle and scale texture features based on the principal strain direction and magnitude at each point on the skin surface. Step S4: Construct a multi-scale topology optimization model based on perceived importance The perceptual importance weight of each vertex of the skin mesh is calculated in real time. The perceptual importance weight is obtained by fusing geometric saliency factors, motion saliency factors and viewpoint saliency factors. The mesh simplification degree and physical simulation accuracy of the corresponding region are dynamically adjusted according to the perceptual importance weight. Step S5: Acquire external motion capture data and map the motion capture data to the rigid body skeleton system; establish a quadratic programming optimization problem with anatomical constraints to solve the optimal skeleton posture that satisfies the joint range of motion constraints; when the constraints are violated, redistribute the excess angular momentum to adjacent joints. Step S6: Perform offline physical simulation on representative motion sequences and record the skin mesh deformation sequence; use principal component analysis to reduce the dimensionality of the deformation sequence and extract characteristic deformation fundamental modes, each of which represents a typical deformation pattern; during online execution, predict the projection coefficient (i.e., deformation weight coefficient) of each characteristic deformation fundamental mode based on the current skeleton motion parameters. This coefficient describes the contribution of each fundamental mode in the current deformation; use the predicted weight coefficients to linearly combine the fundamental modes to reconstruct an approximate skin deformation, and perform a round of physical constraint iterative correction on the reconstruction result.

[0007] Furthermore, in step S1, the active muscle contraction force is obtained through the following sub-steps: A desired activation signal is generated based on a preset motion sequence, and the desired activation signal is decomposed into several intrinsic mode functions through Hilbert-Huang transform. The instantaneous frequency of the intrinsic mode function is calculated. Intrinsic mode functions with instantaneous frequencies higher than a preset threshold are assigned to fast-twitch muscle fibers to obtain fast-twitch muscle activation. Intrinsic mode functions with instantaneous frequencies lower than or equal to the preset threshold are assigned to slow-twitch muscle fibers to obtain slow-twitch muscle activation. Using an improved Hill-type muscle model, the contractile forces of fast-twitch and slow-twitch muscle fibers are calculated based on the activation levels of the fast-twitch and slow-twitch muscles, respectively. The generalized force is obtained by summing the torques generated by all muscle fibers crossing the joint.

[0008] Furthermore, the improved Hill-type muscle model represents muscle contractile force as the product of the muscle's maximum isometric contractile force, activation degree, force-length relationship function, and force-velocity relationship function.

[0009] Furthermore, in step S2, the muscle tissue is defined as a fiber-reinforced hyperelastic material, and the strain energy density function includes an isotropic matrix term and an anisotropic fiber term. The anisotropic fiber term includes Macaulay brackets such that the anisotropic fiber term contributes stiffness only under tension when the square of the elongation ratio along the fiber direction is greater than 1.

[0010] Furthermore, in step S3, the skin layer and the muscle layer are coupled at the interface through a nonlinear spring-damper system; The nonlinear spring-damper system applies an interfacial force between the skin apex and the muscle surface, the interfacial force being controlled by the spring stiffness coefficient and the damping coefficient, allowing the skin layer to move along the tangential direction of the muscle surface when subjected to tangential stress.

[0011] Furthermore, in step S3, the dynamically generated wrinkle features adopt a viscoelastic model based on strain-time history. The strain-time history-based viscoelastic model establishes differential equations including static elastic modulus, dynamic elastic modulus, and viscosity coefficient. The current strain tensor is obtained by numerically solving the differential equations, and the minimum principal strain is extracted to calculate the wrinkle amplitude.

[0012] Furthermore, in step S3, the dynamically generated scale texture features adopt an anisotropic sampling method based on principal stress trajectory lines. During the shading stage, anisotropic filtering sampling is performed on the high-resolution scale map according to the principal strain direction of the mesh vertices.

[0013] Furthermore, in step S4, the geometric saliency factor is obtained by calculating the Gaussian curvature of the skin mesh using the Voronoi area method in discrete differential geometry; the motion saliency factor is obtained by calculating the screen space motion velocity of the vertices of the skin mesh in two adjacent frames.

[0014] Furthermore, in step S4, the perceived importance weight also includes a physical salience factor; The physical saliency factor is obtained by calculating the sensitivity of the vertices of the skin mesh to the total strain energy of the system using the adjoint method.

[0015] Furthermore, in step S5, the redistribution of the excess angular momentum to adjacent joints specifically involves: calculating the excess angular momentum and distributing the angular momentum to adjacent hip, ankle, and spinal joints according to a preset ratio.

[0016] Furthermore, in step S6, the predicted deformation weighting coefficient is specifically as follows: Extract the joint rotation quaternion and displacement vector from the current skeletal motion parameters, and use a linear regression model to predict the deformation weight coefficient.

[0017] In addition, this invention also discloses a simulated dinosaur 3D modeling optimization system for implementing the method described above, the system comprising: The physics-driven engine includes a bone dynamics solution module, a muscle finite element analysis module, and a skin deformation mapping module, which are used to execute the physical simulation process described in steps S1 to S3. The perception optimization controller includes a geometric analysis unit, a motion analysis unit, and a viewpoint analysis unit, and is used to calculate the perception importance weights described in step S4. The motion redirection module is used to perform the motion redirection and anatomical constraint correction described in step S5. The deformation prediction cache module is used to perform the deformation prediction and caching described in step S6. An adaptive scheduler dynamically allocates computing resources based on perceived importance weights, reducing mesh simplification and physical iteration counts for low-importance regions, and maintaining full-precision computation for high-importance regions.

[0018] Furthermore, the perception optimization controller also includes a physical saliency analysis unit for calculating the physical saliency factor in the perception importance weight.

[0019] Furthermore, the adaptive scheduler adjusts the grid simplification error tolerance of the corresponding grid region in real time according to the numerical range of the perceived importance weight, so that the grid simplification error tolerance is a different multiple of the original QEM error threshold.

[0020] Compared with the prior art, the present invention has the following beneficial effects: This invention decomposes a three-dimensional model into a rigid skeletal system based on Lagrangian dynamics, a muscle layer based on hyperelastic materials, and a stress-driven skin layer. These layers are cascaded and driven through explicit physical quantity interfaces. The muscle layer incorporates a fiber-reinforced hyperelastic constitutive model, with its strain energy density function including Macaulay bracket terms that contribute stiffness only when the fibers are under tension, achieving anisotropic deformation with high stiffness along the fiber direction and low stiffness perpendicular to it. A nonlinear spring-damper system is used for interface coupling between the skin layer and the muscle layer, allowing the skin layer to slide relative to the muscle surface under tangential stress, overcoming the motion restriction problem caused by traditional rigid binding.

[0021] This invention employs the Hilbert-Huang transform to perform intrinsic mode decomposition of neural activation signals. Based on instantaneous frequency thresholds, signal components are allocated to fast-twitch and slow-twitch muscle fibers, and these components are then substituted into an improved Hilbert-type muscle model to calculate contractile force. This allows for differentiated muscle mechanical responses under fast and slow motion, resolving the dynamic distortion problem caused by the inability of traditional models to distinguish fiber types. This invention introduces the Praggle-Reuss viscoelastic model to describe the time history dependence of stress-strain in the skin layer. The current strain tensor is obtained through numerical solution of differential equations, allowing the formation and decay of wrinkles to exhibit creep and relaxation time-lag effects, overcoming the limitation of instantaneous strain models in simulating material memory properties. This invention integrates geometric saliency factors, motion saliency factors, viewpoint saliency factors, and physical saliency factors calculated through the adjoint method to quantify the sensitivity of each vertex to the total strain energy of the system. An adaptive scheduler dynamically adjusts the mesh simplification and physical iteration count based on the fusion weights, ensuring that deep load regions maintain physical accuracy even when visually insignificant, avoiding computational resource misallocation caused by purely visual-driven approaches.

[0022] This invention establishes a quadratic programming optimization problem with anatomical constraints, mapping external motion capture data to the target skeleton. When anatomical constraints are violated, angular momentum is redistributed to adjacent joints to ensure that the redirected posture is anatomically achievable. Simultaneously, by extracting deformation principal component basis modes offline and using online linear regression prediction, combined with single-round physical constraint iterative correction, the computational load of online physical simulation is significantly reduced.

[0023] This invention achieves enhanced biomechanical realism of dynamic deformation, ensures anatomical rationality of motion redirection, precise on-demand allocation of computing resources, and extended compatibility of system architecture with anatomical features of multiple species. It has broad application prospects in the fields of paleontology reconstruction, film and television special effects, virtual reality, and biomechanical simulation. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0025] Figure 1 This is an overall flowchart of the method described in this invention.

[0026] Figure 2 This is a simplified flowchart of the overall process of the 3D modeling optimization method for simulated dinosaurs according to the present invention.

[0027] Figure 3 This is a flowchart illustrating the collaborative process of perception optimization and motion redirection in this invention.

[0028] Figure 4 This is a partial schematic diagram of the dinosaur's skin after modeling according to the present invention. Detailed Implementation

[0029] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0030] The following is in conjunction with the appendix Figures 1-4 The embodiments of the present invention will be described in detail below.

[0031] Example 1: This example uses an adult Tyrannosaurus Rex as the subject and employs the method described in this invention to construct a high-precision simulation model for scientific research-level biomechanical analysis. The key technical problem addressed in this example is the limitation of existing technologies that rely solely on skeletal geometric interpolation, which cannot accurately recreate the nonlinear deformation of muscles and skin, leading to distortions in dynamic anatomy.

[0032] The specific method is as follows: Step S1 Implementation Details: Construct a rigid body skeleton system based on Lagrange dynamics; To address the problem that skeletal motion in existing technologies is merely a geometric transformation and lacks physical inertial response, this embodiment first constructs a skeletal system with physical properties.

[0033] Obtaining 3D scanning data of Tyrannosaurus Rex skeletal fossils: An industrial-grade laser scanner was used to perform a 360-degree scan of the complete Tyrannosaurus Rex skeletal fossil, with a sampling precision set to 0.02 mm. The raw point cloud data obtained from the scan contained approximately 850 million points. The point cloud was then reconstructed using the Poisson reconstruction algorithm, with an octree depth set to level 10, generating an initial skeletal geometric mesh with approximately 1.2 million triangular faces.

[0034] Based on paleontological anatomy databases, joint annotation was performed on skeletal meshes. The annotation process employed a local coordinate system alignment technique based on eigenvalue decomposition: a subset of local point clouds was extracted at each skeletal joint connection, the covariance matrix of this subset was calculated, and eigenvalue decomposition was performed. The largest eigenvector was determined as the principal extension direction of the bone, and the axis determined by the smallest eigenvector was defined as the principal axis of joint rotation. This method was used to annotate 68 major joints throughout the body, including the cervical, shoulder, elbow, hip, knee, ankle, and coccygeal joints.

[0035] Physical properties were assigned to each bone: mass was calculated by multiplying the scanned volume by the bone density, which was set at 1.8 grams per cubic centimeter according to paleontological literature (Currie et al., 2003). The moment of inertia tensor was calculated by integrating the triangular facets of the bone mesh, using the following formula: ,in Let be the distance vector from the vertex to the centroid. Let Kronecker function be used.

[0036] Establish generalized coordinates subscript The value ranges from 1 to 68, each This corresponds to one degree of freedom of a joint. Generalized forces are defined. It includes three types of forces: active muscle contraction force (generated based on the Hill-type muscle model, the contraction force and muscle activation are non-linearly related), passive ligament constraint force (generated based on the exponential passive torque-angle curve), and external contact force (including ground reaction force, calculated based on the spring-damped contact model).

[0037] Solve the Lagrange dynamic equations: in, For the first The generalized coordinates of each joint, in radians (rad). To and The corresponding generalized velocity is expressed in radians per second (rad / s). For the system's Lagrange quantity, The total kinetic energy of the system is expressed in joules (J). The total potential energy of the system is expressed in joules (J). For the action in the The generalized force on a joint is measured in Newton-meters (N·m).

[0038] Implicit Euler method is used for time integration, with time step set to... Seconds. During the solution process, a sparse matrix solver is used to process the Jacobian matrix of joint constraints, ensuring that the rigid body system maintains the integrity of joint connections during motion. Finally, a skeletal motion sequence is generated, outputting the rotation quaternions and displacement vectors of each joint at each time step. This step solves the "floating" problem in traditional skeletal animation, giving the motion physical inertial characteristics.

[0039] Step S2 Implementation Details: Construct a muscle layer deformation model based on hyperelastic materials and minimizing strain energy density; To address the issues of stiff muscle deformation and lack of volume conservation in existing technologies, this embodiment constructs a physically realistic muscle layer based on a skeletal model.

[0040] Based on paleontological anatomical data regarding muscle origins and insertions (Hutchinson et al., 2005), muscle layers were constructed between skeletal attachment points. A total of 350 major muscle groups were defined, covering the neck, back, forelimbs, hindlimbs, and tail. Each muscle group was discretized into a tetrahedral mesh, generated using the Delaunay tetrahedralization algorithm, with an average side length of 2 mm. For larger muscle groups such as the quadriceps femoris, the number of tetrahedra was approximately 8000; for smaller neck muscles, the number was approximately 1200.

[0041] Muscle tissue is defined as a hyperelastic material, and its constitutive relation is described by the strain energy density function: in, Strain energy density, in Pascals (Pa). Shear modulus, in Pascals (Pa). is the first invariant of the left Gauss-Green strain tensor, and is dimensionless; is the determinant of the deformation gradient, dimensionless, representing the rate of volume change; is Lamé constant, with the unit being Pascal (Pa).

[0042] The muscle material parameters were set according to biomechanical literature (Weis-Fogh & Alexander, 1977): thigh muscle shear modulus The value is 8 kPa, and the neck muscle shear modulus is... The value is 5 kPa, Lamé constant. The shear modulus is uniformly set to 20 kPa. With Lamé constant Together, they determine the stiffness characteristics of muscles under compression and tension.

[0043] When the skeletal motion calculated in step S1 causes displacement of the muscle attachment points, the muscle tetrahedral mesh deforms. The total strain energy is then calculated. The final morphology of the muscle is determined by minimizing the displacement field. A nonlinear finite element solver is used, employing the Newton-Raphson iterative method. In each iteration, the tangent stiffness matrix is ​​calculated and the system of linear equations is solved.

[0044] To ensure numerical stability, a linear search strategy is used to adjust the step size during the iteration process to ensure the volume ratio. Always greater than 0.01 to avoid grid flipping. This step causes the muscle to expand laterally when squeezed and contract cross-section when stretched, which conforms to the physical law of conservation of volume in biological tissues.

[0045] Step S3 Implementation Details: Construct a stress-driven skin deformation and microtexture generation model based on strain tensor; The skin layer uses a triangular mesh with approximately 2 million initial vertices. A distance-field-based wrapping algorithm is used to attach the skin mesh to the muscle layer surface. The distance-field function is defined as the minimum Euclidean distance from the skin vertex to the muscle surface, ensuring the skin adheres tightly to the muscle without penetrating it.

[0046] The Cauchy stress tensor of the muscle surface nodes is projected onto the skin mesh vertices through radial basis function interpolation. The radial basis function uses a thin plate spline function. During the interpolation process, the muscle nodes are used as the source points and the skin vertices as the target points. An interpolation matrix is ​​established and the weight coefficients are solved to obtain the stress vector of each skin vertex.

[0047] Microtextures are generated based on the strain state at various points on the skin surface. First, the deformation gradient tensor of the skin mesh is calculated. Through polar decomposition Obtain rotational components With stretching component From the stretching component Calculate the Green-Lagrange strain tensor Solve for its eigenvalues ​​and eigenvectors to obtain the maximum principal strain. With minimum principal strain Both are dimensionless quantities.

[0048] Set the compressive strain threshold It is -0.03. When the minimum principal strain... When the value is less than -0.03, the region is determined to be in a compressed state, and the system automatically generates a wrinkle texture. The wrinkle amplitude is calculated according to the following formula: in, The fold width is measured in millimeters (mm). This is an empirical proportionality coefficient, dimensionless, and in this embodiment, it is set to 1.2; The minimum principal strain is dimensionless. The wrinkle direction is perpendicular to the minimum principal strain direction (i.e., parallel to the principal compressive stress direction). The wrinkle effect is achieved by generating a displacement map with a resolution of 8192 pixels by 8192 pixels, where each pixel carries an offset in the normal direction. During rendering, the displacement map is mapped to the skin mesh through the GPU's subdivision surface stage, generating geometrically detailed wrinkles. This step enables the skin to automatically wrinkle in stressed areas and automatically smooth in stretched areas, achieving a physical coupling of "skin and bone linkage".

[0049] This embodiment is an offline simulation designed to generate the highest accuracy model for scientific research and analysis. It does not involve real-time interaction or computational resource optimization. Therefore, the perception optimization weight in step S4 is uniformly set to 1 (i.e., full-precision calculation), and the motion redirection in step S5 and the deformation prediction cache in step S6 are not enabled.

[0050] Simulation Results and Data Output: After 72 hours of computation, a complete simulation model containing bones, muscles, and skin was generated. The model's static geometric accuracy was compared with a 3D scan of a physical model hand-carved by paleontologists, with a root mean square error of 0.31 mm. In the dynamic motion sequences (including walking, running, and roaring sequences, each 10 seconds long, totaling 1200 frames), anatomical experts reviewed each frame and found no skin tears, volume abnormalities, or joint wear-through. The morphology of muscle bulges and skin folds conforms to paleontological expectations; for example, when the knee is flexed at 60 degrees, the height of the biceps femoris muscle bulge increases by approximately 12 mm compared to the resting state, and skin folds are evenly distributed along the joint creases.

[0051] Example 2: This example uses the same Tyrannosaurus Rex model as a base and employs the method described in this invention to construct a real-time simulation model that can be used for virtual reality interaction. The key technical problems addressed in this example are: the contradiction between high-precision models and real-time rendering performance in existing technologies, and the inability to directly map external motion capture data to paleontological skeletons.

[0052] This embodiment is a further optimization based on Embodiment 1. In this embodiment, the specific details of the real-time optimization of steps S1 and S2 are as follows: To address the issue of excessive computational load in physical simulation during real-time rendering, this embodiment simplifies the algorithms for steps S1 and S2.

[0053] Skeletal dynamics solution: fix the time step to The time complexity is measured in seconds (synchronized with the display refresh rate), using an explicit Euler method instead of an implicit Euler method. The computational complexity of the explicit Euler method is O(n). More implicit methods Significantly reduced. The discretization formula for the explicit Euler method is: , acceleration It can be calculated directly from the dynamic equations.

[0054] Muscle layer deformation solution: Instead of nonlinear finite element analysis, a position-based dynamic method is used. This method transforms the deformation problem of the muscle tetrahedral mesh into a constraint-solving problem. Two types of constraints are defined: Volume conservation constraint: ,in This represents the current tetrahedral volume, in cubic millimeters (mm³). Resting volume, in cubic millimeters (mm³).

[0055] Shape matching constraints: ,in and is the position vector of adjacent vertices, in millimeters (mm); This is the resting distance, measured in millimeters (mm).

[0056] The solution is performed in three iterations per frame. For each constraint, the position correction is calculated according to the standard position-based dynamics formula: in, As vertex Position correction vector, in millimeters (mm); As vertex The reciprocal of mass, expressed in kilograms (kg⁻¹). The constraint function is dimensionless. Constraint functions for vertices The gradient of position, expressed in millimeters (mm⁻¹). Muscle material parameters. and It remains consistent with Example 1, but is no longer used directly for energy calculation. Instead, it is reflected indirectly through constraint stiffness parameters.

[0057] The implementation details of real-time mapping in step S3 are as follows: In order to solve the computational load problem caused by high-precision skin mesh and texture in real-time rendering, this embodiment optimizes the skin layer.

[0058] The initial number of vertices in the skin layer was simplified to 500,000, approximately one-quarter of that in Example 1. Principal strain and The calculation is no longer obtained through the extreme decomposition of the deformation gradient tensor, but is approximated by the rate of change of the side lengths of the skin mesh. For each triangular facet, the rate of change of the lengths of the three sides is calculated, the average value is taken as the face strain, and then the vertex strain is obtained by averaging through the vertex neighborhood.

[0059] Compressive strain threshold Set to -0.02, wrinkle amplitude coefficient Set to 1.0. The wrinkle effect is achieved by dynamically modifying the normals through the vertex shader, without increasing geometric complexity. For the scale texture, an anisotropic sampling method based on principal stress trajectory lines is adopted: a set of high-resolution scale maps (8192 pixels by 8192 pixels) is pre-generated, and during the shading stage, the map is anisotropically filtered and sampled according to the principal strain direction of the vertex, so that the long axis direction of the scale is aligned with the principal stress direction, achieving a visual stress following effect.

[0060] The implementation details of the perception optimization in step S4 are as follows: In order to solve the problem of wasting computing resources in non-visually sensitive areas, this embodiment introduces multi-scale topology optimization based on perception importance.

[0061] Set the weights of the geometric significance factor The weight of the motion significance factor is 0.5. The viewpoint significance factor weight is 0.3. The value is 0.2. The constant in the motion significance factor. Set it to 0.2.

[0062] Perceived importance weight Calculate using the following formula: in, Vertices on the skin mesh; As vertex Geometric significance factor; As vertex The significant factors of motion; As vertex The viewpoint saliency factor; For normalized weight coefficients, satisfying .

[0063] Geometric significance factor ,in As vertex The Gaussian curvature, expressed in millimeters (mm⁻¹). The Gaussian curvature is calculated using the Voronoi area method in discrete differential geometry. in, As vertex The area of ​​Voronoi, in square millimeters (mm²). For the vertex A set of adjacent triangular facets; For dough pieces At the apex The interior angle at point A is expressed in radians.

[0064] To eliminate the noise impact of mesh quality on Gaussian curvature calculation, the Gaussian curvature field is smoothed using Laplace smoothing before calculating the geometric significance factor. in, As vertex The set of vertices in the ring neighborhood; For normalized weights, the value is [value to be filled in]. ; The smoothing factor is dimensionless, and in this embodiment, it is set to 0.5. The smoothed Gaussian curvature is used for subsequent gradient calculation of geometric significance factors.

[0065] Significant factors of motion ,in As vertex The speed of movement in screen space, measured in pixels per frame; is a constant, dimensionless, and in this embodiment, it is taken as 0.2. It is obtained by calculating the difference in screen coordinates of a vertex between two adjacent frames and dividing by the frame interval.

[0066] Viewpoint salience factor From vertex The projected area is determined by the distance from the center of the screen, and its value ranges from 0 to 1. The specific calculation method is as follows: calculate the vertex... Area projected onto screen space The unit is pixels; calculate the Euclidean distance from the projection center to the screen center. The unit is pixels; after normalization, we get ,in The maximum projected area (one-tenth of the total projected area of ​​the model), in pixels; It is half the length of the screen diagonal, and the unit is pixels.

[0067] The system calculates the value of each vertex in real time. This allows for dynamic level-of-detail switching within the rendering pipeline. when At that time, the mesh simplification error tolerance of the region where the vertex is located is increased to 3 times the original QEM error threshold, so that the number of mesh vertices is reduced to 20% of the original high-poly model vertices.

[0068] when At that time, the number of mesh vertices was simplified to 60% of the original high-poly vertices.

[0069] when At the same time, the original accuracy is maintained, and the number of physical iterations remains unchanged at 3 per frame.

[0070] The details of motion redirection in step S5 are as follows: External motion capture data source: The walking motion of human actors was captured using an optical motion capture system at a sampling rate of 120 frames per second, outputting the 3D position and rotation data of 45 joints.

[0071] Human skeletal joints are mapped to Tyrannosaurus Rex skeletons using a scaling method. The scaling is based on the bone length ratio: the ratio of the T. Rex femur length to the human femur length is set to 3.2, the tibia length ratio to 3.5, and so on for other bones.

[0072] Establish a quadratic programming optimization problem: Objective function: ,in This is a vector representation of the Tyrannosaurus Rex skeleton. This is a scaled-down vector of human skeletal pose. It is a weighted diagonal matrix, with higher weights assigned to major joints such as the thigh and spine.

[0073] Constraints: The permissible range of motion for each joint is pre-defined based on paleomechanics literature (Hutchinson & Garcia, 2002). For example, the range of knee flexion is 0 to 70 degrees (hyperextension is prohibited), the range of hip flexion is -30 to 120 degrees (-30 degrees is for posterior extension), and the range of shoulder flexion is -20 to 90 degrees.

[0074] When the optimization solver detects a joint angle violation (e.g., a calculated knee joint angle of 85 degrees, exceeding the 70-degree upper limit), the system performs angular momentum reallocation. The specific logic is as follows: Calculate the angular momentum of the excess portion. ,in The moment of inertia of the joint is expressed in kilograms per square meter (kg·m²). To optimize the solution, the desired angular velocity is expressed in radians per second (rad / s). The upper limit is defined as the angular velocity, expressed in radians per second (rad / s).

[0075] Angular velocity is obtained by the difference between the joint angle and time: ,in The simulation time step is expressed in seconds (s).

[0076] Distribute this angular momentum to adjacent joints: 50% to the hip joint (increasing hip flexion angle), 30% to the ankle joint (increasing ankle plantar flexion angle), and 20% absorbed by the spine (increasing lateral flexion of the spine). After resolving, the foot contact point remains unchanged, and all joints satisfy the constraints.

[0077] To address the issue of excessive computational load from performing a complete physical simulation for each frame, this embodiment introduces a deformation prediction cache based on motion mode decomposition.

[0078] Offline training phase: Full physical simulations were performed on four representative movement sequences of Tyrannosaurus Rex: walking, running, turning its head, and roaring. Each movement was simulated for 120 frames, and a total of 480 frames of skin mesh deformation data were collected.

[0079] The skin mesh deformation of each frame is represented as a vector. ,in The number of skin vertices is 500,000, and the vector elements are the displacement components of each vertex in millimeters (mm).

[0080] Calculate the mean of the deformation vector. ,in After centralization, the result is... .

[0081] Constructing the covariance matrix Perform eigenvalue decomposition and extract the eigenvectors corresponding to the top 20 largest eigenvalues. As a characteristic deformation basic mode.

[0082] For each frame of training data, calculate its projection coefficients onto the base mode. ,in .

[0083] Online prediction phase: Extract the skeletal motion parameters of the current frame It contains rotational quaternions (272 dimensions in total) and displacement vectors (204 dimensions in total) for 68 joints, totaling 476 dimensions.

[0084] Predict 20 weighted coefficients using a linear regression model: Among them, the regression coefficient Obtained through least-squares fitting of offline training data.

[0085] Reconstructing the deformation state: The unit is millimeters (mm).

[0086] A position-based dynamic iteration is performed on the reconstructed results to enforce volume conservation constraints and eliminate physical inconsistencies in the prediction results. Prediction accuracy is quantified by calculating the reconstructed deformation vector. Deformation vectors in real physical simulations root mean square error between The unit is millimeters (mm); simultaneously, the root mean square magnitude of the true deformation vector is calculated. The unit is millimeters (mm). Prediction accuracy is defined as... In this embodiment, the value is 92%, which means that the root mean square error of the predicted deformation accounts for only 8% of the root mean square amplitude of the actual deformation.

[0087] The operating environment and hardware configuration are as follows: Central Processing Unit: 24 cores, 3.2 GHz, supporting AVX-512 instruction set. Graphics Processing Unit: 32 GB of video memory, supporting DirectX 12 Ultimate and mesh shaders. System Memory: 64 GB DDR4 3200 MHz. Storage: 2 TB NVMe SSD, read speed 3500 MB / s.

[0088] Simulation results and data output: The system runs stably at 72 frames per second at 4K resolution (3840 pixels x 2160 pixels), with a peak memory usage of 5.2 gigabytes (GB). In interactive testing, users control the dinosaur to perform actions such as walking, turning its head, and roaring using a controller. The model's dynamic deformation is natural, with wrinkles automatically generated when the knee joints and neck bend. The textures are not stretched or distorted, and the orientation of the skin scales dynamically adjusts with muscle movement.

[0089] The perception optimization mechanism concentrates computing resources on the head, claws, and near-field areas. Statistical data shows that the average number of triangles rendered per frame is approximately 800,000, effectively controlling the rendering load and enabling the system to run stably at 72 frames per second at 4K resolution.

[0090] The motion redirection module successfully mapped human motion capture data onto the Tyrannosaurus Rex skeleton. In 20 repeated tests, the system automatically adjusted the hip and ankle joint angles at the knee joint's extreme position without any limb clipping or joint twisting, and the foot contact point position error was less than 2 mm.

[0091] The deformation prediction cache module achieved 92% accuracy in inter-frame deformation prediction, reducing the number of frames requiring full physical simulation from once every five frames, thus lowering the computational overhead of physical simulation. CPU utilization decreased from 78% to 42%, while GPU utilization remained around 85%.

[0092] Comparative Example 1: To verify the technical effects of the present invention, Comparative Example 1 was set up. The same hardware environment and Tyrannosaurus Rex model as in Example 2 were used, but the model was constructed using a traditional linear blending skinning method. This comparative example is used to compare and verify the technical advantages of the present invention in terms of dynamic deformation realism, rendering efficiency, and motion retargeting accuracy.

[0093] The modeling process is described in detail below: Skeletal model construction: The same skeletal geometry model and joint annotations (68 joints) as in Example 2 were used. However, skeletal motion was not solved dynamically, but driven by preset animation curves. The animation curves were manually keyframed by the artist and included four types of actions: walking, running, turning the head, and roaring, with each type of action lasting 5 seconds.

[0094] Skin binding and weight calculation: The skin mesh is directly bound to the skeleton, using a linear blending skinning algorithm. The position calculation formula for each skin vertex is as follows: ,in The position of the vertex after deformation, in millimeters (mm); The initial vertex position is in millimeters (mm). For the first A dimensionless transformation matrix that affects the skeleton; The corresponding weights are dimensionless and satisfy the following conditions: The weights are automatically calculated using a heatmap method: the closer the bone, the higher the weight. Joint areas are manually adjusted to ensure smooth deformation. Without a muscle layer, skin deformation is entirely determined by linear interpolation of bone weights, making it impossible to simulate muscle bulging and volume conservation.

[0095] Texture mapping: Static 8K resolution (8192 pixels by 8192 pixels) normal maps and diffuse maps are used. The maps are obtained through high-precision scanning and contain details such as skin scales and wrinkles. The texture mapping is static and does not change with movement, so obvious texture stretching artifacts will appear in areas of skin stretching (such as when joints are bent).

[0096] Level of detail strategy: a distance-based discrete switching mechanism. A high-poly model (500,000 polygons) is used when the distance is less than 5 meters, a medium-poly model (100,000 polygons) is used when the distance is between 5 and 15 meters, and a low-poly model (20,000 polygons) is used when the distance is greater than 15 meters. During switching, a sudden change in model size may cause a visual "jump" phenomenon.

[0097] Motion capture retargeting: Human motion capture data is mapped to the Tyrannosaurus Rex skeleton using a simple linear scaling method. The scaling ratio is calculated based on the bone length ratio, but without anatomical constraint correction. When the mapped joint angle exceeds physiological limits (e.g., the knee joint exceeds 70 degrees), the system directly outputs that angle, resulting in the model exhibiting an unphysiological posture.

[0098] The system achieved an average frame rate of 45 frames per second at 4K resolution, with a peak memory usage of 8.8 gigabytes (GB). The main reason for the low frame rate is that high-poly areas (at a distance of less than 5 meters) require rendering 500,000 polygons, and there is no perceptible optimization mechanism, resulting in computational resources being consumed evenly across all areas.

[0099] During large joint movements (such as knee flexion at 80 degrees during running or lumbar rotation at 45 degrees during turning), noticeable skin indentation and cross-cutting occur at the knee and neck. The dynamic deformation distortion rate, measured by the proportion of negative Jacobian determinants on the skin surface, is 4.2%. A negative Jacobian determinant indicates mesh flipping and is a quantitative indicator of skin distortion.

[0100] Specific detection method: In a 1200-frame motion sequence, the Jacobian determinant values ​​of all skin triangle patches were counted. The proportion of negative values ​​was 4.2% of the total number of triangles per frame. In comparison, this proportion was 0.03% in Example 2.

[0101] Texture stretching artifact: When the knee joint is bent, the scale texture on the inner side of the skin is stretched, with the maximum stretching rate reaching 28%, resulting in visually blurred texture and scale deformation. In Embodiment 2 of the present invention, the maximum texture stretching rate is 3.5%.

[0102] Motion redirection results: In 20 repeated tests, when the human motion capture data required a knee angle exceeding 70 degrees, the model directly presented an unphysiological posture of knee hyperextension, resulting in the thigh and lower leg crossing and severe skin tearing. The joint over-limit rate (the ratio of the number of joints exceeding the physiological range to the total number of joints) was 100%.

[0103] The key performance indicators of Example 2 and Comparative Example 1 are summarized in Table 1.

[0104] Table 1 Performance comparison between Example 2 and Comparative Example 1; As shown in Table 1, the present invention outperforms the traditional linear hybrid skinning method in terms of dynamic deformation realism, rendering efficiency, resource consumption, motion redirection accuracy, and texture integrity. Example 2 shows a 99.3% reduction in dynamic deformation distortion, indicating that the physically driven muscle-skin cascade model effectively avoids skin collapse; a 60% increase in rendering frame rate and a 40.9% reduction in video memory usage demonstrate that the perceptual optimization mechanism achieves on-demand allocation of computing resources; and a 0% joint over-limit rate during motion redirection indicates that the inverse dynamics constraint solution and angular momentum redistribution mechanism ensure the anatomical rationality of the redirected motion.

[0105] Example 3: This example uses the same Tyrannosaurus Rex model to address the problems in existing technologies, such as coarse modeling of muscle contraction force, inability to accurately simulate different nonlinear mechanical responses of muscles under fast and slow motion, and inaccurate coupling between dynamic skin texture generation and underlying stress field. This example focuses on a deeper physical decomposition and modeling of the muscle-skeleton-skin coupling system, constructing a complete physical simulation pipeline from neural activation signals to final visual presentation.

[0106] Step S1': Modeling muscle activation and Hill-type muscle contractile force based on time-frequency domain decomposition; To address the problem that existing technologies use simple linear or static models of muscle contraction force, which fail to reflect the different recruitment and contraction characteristics of muscle fibers under fast and slow movements, resulting in distorted dynamic response, this embodiment introduces a time-frequency decomposition method of muscle activation based on Hilbert-Huang transform (HHT) and combines it with a nonlinear Hilbert-type muscle model to construct a refined muscle contraction force.

[0107] S1'-1: Classification of muscle fiber types and definition of anatomical attachment points; Based on paleontological anatomy literature (Persons & Currie, 2011), the 350 major muscle groups throughout the body are divided into two main fiber types: Type I fibers (slow-twitch muscle fibers): accounting for approximately 40%, mainly distributed in muscle groups that maintain posture (such as deep neck muscles and erector spinae muscles). They are characterized by slow contraction speed, low sustained force, and strong resistance to fatigue.

[0108] Type II fibers (fast-twitch fibers): accounting for approximately 60%, mainly distributed in muscle groups involved in explosive movements (such as the quadriceps, gastrocnemius, and caudal sway muscles). They are characterized by rapid contraction speed, high explosive power, and easy fatigue.

[0109] For each muscle group, the precise coordinates of its origin and insertion points on the skeletal model were defined. These coordinates were determined using a three-dimensional coordinate system from a paleontological anatomy database, in millimeters (mm). In the local coordinate system of the skeletal model, the quadriceps femoris originates at the anterior border of the ilium (120.5, 450.2, 89.3) and inserts at the tibial tuberosity (140.8, 320.1, 95.7).

[0110] S1'-2: Simulation of neural activation signals and calculation of muscle activation; Before the simulation begins, the desired activation signals for the muscles are generated based on a preset motion sequence (such as walking or running). The signal is time. The function, which takes values ​​from 0 to 1, represents the desired activation level of the muscle.

[0111] To simulate the different control strategies of the nervous system on fast and slow muscle fibers, the desired activation signal is... Decomposition using the Hilbert-Huang transform: Empirical Mode Decomposition: Decomposed into a series of intrinsic mode functions and a residual : in, This represents oscillation components at different time scales. The decomposition process is achieved through an iterative screening process until the definition conditions of the IMF are met: throughout the entire data segment, the number of extreme points and the number of zero-crossing points are equal or differ by at most one; at any point, the mean of the envelope defined by the local maxima and local minima is zero. In this embodiment, the number of IMFs obtained from the decomposition is... .

[0112] Hilbert spectral analysis: for each Perform a Hilbert transform and calculate its instantaneous frequency. : in, Indicates the Cauchy principal value. Instantaneous frequency. The unit is radians per second (rad / s).

[0113] Frequency threshold decomposition: based on instantaneous frequency The intrinsic mode functions are assigned to fast and slow muscle fibers: Will Above the threshold radians per second The fraction is allocated to type II fast muscle fibers.

[0114] Will Less than or equal to of The components are allocated to type I slow-twitch muscle fibers. This yields the activation level of fast-twitch muscle fibers. and slow muscle activation : in, and All are dimensionless quantities, and their values ​​are normalized to [0,1].

[0115] S1'-3: Nonlinear Hill-type muscle contraction force calculation; For each muscle group, the contractile forces of type I and type II fibers were calculated separately. A modified Hill-type muscle model was used, which incorporates muscle contractile forces... Represented as activation level Muscle fiber length and contraction speed Functions: in, This represents muscle contraction force, measured in Newtons (N). This represents the maximum isometric contractile force of the muscle, measured in Newtons (N). This value is obtained by multiplying the physiological cross-sectional area (PCSA) of the muscle by an empirical stress constant. In this example, the quadriceps femoris muscle... N.

[0116] For muscle activation, specifically for type I fibers, For type II fibers, .

[0117] The force-length relationship function, dimensionless, describes the maximum active tension that a muscle can generate at different lengths, and is defined as: in, The current muscle fiber length is expressed in millimeters (mm). The optimal sarcomere length (corresponding to the length that generates the maximum force) is measured in millimeters (mm). This embodiment is set based on anatomical data. This is the curve width coefficient, which is dimensionless and is set to 0.5 in this embodiment. is the curve steepness coefficient, which is dimensionless, and is set to 2 in this embodiment.

[0118] The force-velocity function, dimensionless, describes the relationship between the force and velocity generated by a muscle during contraction or extension, and is defined as follows: in, This represents the contraction speed of muscle fibers, measured in millimeters per second (mm / s). Positive values ​​indicate concentric contraction, while negative values ​​indicate eccentric contraction. The maximum shrinkage speed is measured in millimeters per second (mm / s). This embodiment is set according to the fiber type (Type I: mm / s, Type II: mm / s); is the curvature constant, which is dimensionless, and is taken as 5 in this embodiment; The centrifugal / centripetal force ratio is dimensionless, and in this embodiment, it is taken as 1.5.

[0119] Finally, it acts on the first Generalized forces of each joint It is obtained by summing the torques generated by all the muscle fibers that cross the joint: in, For the first The total contractile force of the muscle group at the current moment (the sum of type I and type II fibers), in Newtons (N); The lever arm is the vertical distance from the line of action of the muscle to the center of rotation of the joint, and is measured in meters (m).

[0120] Muscle lever arm The calculation method is as follows: Define muscle pathways: For each muscle group, define the origin on the skeletal model. and stop point The spatial coordinates, and a set of winding points ( The winding points are used to simulate the path changes of muscles as they navigate around bony protrusions or pulleys. These winding points are preset based on paleontological anatomical data and maintain their relative position to the bone during the simulation.

[0121] Calculate the muscle line of action: The actual line of action of a muscle is a spatial polyline segment formed by connecting the origin, the point of rotation, and the insertion point in sequence. Project this polyline segment onto the plane containing the joint rotation axis to obtain the two-dimensional path.

[0122] Calculating the lever arm: lever arm The shortest distance from the joint rotation center to the muscle's line of action is calculated using the following formula: in, The coordinates of the joint rotation center are in meters (m). and These are the coordinates of the muscle's attachment point (or wrapping point) at the proximal and distal ends of the joint, respectively, in meters (m). Represents the cross product of three-dimensional vectors; This represents the Euclidean norm of a vector.

[0123] The lever arm changes in real time with the joint angle and is updated every frame.

[0124] This step will solve for the generalized force. Substituting the Lagrange dynamics equations from step S1 into the equations drives skeletal movement, thus achieving a complete physical drive from neural activation signals to skeletal movement.

[0125] Step S2': Modeling muscle layer deformation based on anisotropic hyperelastic constitutive and fiber-reinforced model; To address the problem that existing technologies simplify muscle tissue as an isotropic material, neglecting the significant influence of muscle fiber orientation on deformation and resulting in deformation patterns that do not match physiological realities, this embodiment introduces a fiber-reinforced hyperelastic material model to consider the anisotropic characteristics of muscle fibers during muscle layer deformation.

[0126] Based on the tetrahedral mesh in step S2, a fiber orientation field is defined for each muscle unit. This direction field is preset based on the anatomical fiber direction of the muscle. For example, the fiber direction of the quadriceps femoris is approximately parallel to the line connecting the muscle's origin and insertion points.

[0127] In this embodiment, the fiber orientation field is defined by a unit vector at the center of each tetrahedral element. This is achieved by spatial interpolation based on anatomical atlases.

[0128] Define the strain energy density function of fiber-reinforced hyperelastic materials The function contains an isotropic matrix term. and anisotropic fiber terms : in, Let be the right Cauchy-Green deformation tensor, which is dimensionless.

[0129] Isotropic matrix term The mechanical behavior of the matrix material was characterized using the method described in Example 2 above: Anisotropic fiber terms The fiber's resistance to tension is characterized using an exponential form: in, This refers to the fiber stiffness modulus, measured in Pascals (Pa). In this embodiment, it is set to [value missing] based on the literature. Pa; The fiber nonlinear hardening coefficient is dimensionless; in this embodiment, it is taken as [value missing]. ; The fourth invariant along the fiber direction is dimensionless, and represents the square of the elongation ratio along the fiber direction; Macaulay brackets are defined as follows: That is, only when the fiber is in tension ( When subjected to pressure, the fiber generates resistance, but does not contribute stiffness.

[0130] The total deformation of the muscle is solved to obtain the total strain energy. The minimum displacement field is obtained. A nonlinear finite element method is used. When calculating the tangent stiffness matrix in the Newton-Raphson iteration, the contributions of both isotropic and anisotropic components must be considered simultaneously. The total Cauchy stress tensor of the muscle is derived from the passive stress tensor. (from strain energy density function) (Extracted) and active stress tensor (Contractile force calculated using the Hill-type muscle model) (Distributed along the fiber direction) and stacked together. This step makes the muscle exhibit high stiffness when stretched along the fiber direction and low stiffness when compressed perpendicular to the fiber direction, more realistically simulating the mechanical behavior of muscle tissue. For example, the thigh muscle bulges along the fiber direction when contracting, rather than expanding isotropically.

[0131] Step S3': Texture generation based on nonlinear elastic constitutive and dynamic frictional contact skin-muscle interface coupling and strain-time history correlation; To address the issues in existing technologies where the simple binding of skin and muscle layers fails to simulate the relative sliding between them, and where texture generation is based solely on the current instantaneous strain, neglecting the time lag effect caused by material viscoelasticity, this embodiment introduces a nonlinear elastic interface coupling model and a texture generation mechanism related to strain-time history.

[0132] S3'-1: Skin-muscle interface coupling based on nonlinear spring-damper; The skin and muscle layers are treated as two independent continuums, and their interaction is simulated by applying a nonlinear spring-damper system between the skin apex and the muscle surface. For each skin apex... Find its nearest projection point on the muscle surface. and apply interfacial forces : in, This is the interfacial force between the skin apex and the muscle surface, measured in Newtons (N). This is the displacement vector from the skin apex to the muscle surface, in millimeters (mm). The relative velocity of the skin apex with respect to a point on the muscle surface is expressed in millimeters per second (mm / s). The spring stiffness coefficient is expressed in Newtons per millimeter (N / mm). In this embodiment, it is set based on the viscoelastic properties of skin and muscle. N / mm; The damping coefficient is expressed in Newton-seconds per millimeter (N·s / mm). In this embodiment, it is set to... N·s / mm.

[0133] This model allows for a certain degree of relative sliding between the skin and muscle. When the skin is subjected to tangential stress, the skin apex can move along the tangential direction of the muscle surface, overcoming the deficiency of no relative sliding between the skin and muscle in traditional binding methods. At the same time, the spring-damper system ensures that the skin closely follows the movement of the muscle, preventing penetration.

[0134] S3'-2: Texture generation based on strain-time history (Prag-Royce model). To address the issue that skin texture is determined solely by the current instantaneous strain, failing to simulate viscoelastic effects such as creep and relaxation in skin wrinkles, this embodiment introduces a strain-time history-based viscoelastic model (Prag-Royce model) to dynamically generate texture amplitude. This model simplifies the skin as a standard linear solid model consisting of a Maxwell element (a series of springs and dampers) and a spring in parallel, whose strain response... With stress The relationship can be described by a differential equation: in, Skin stress is measured in Pascals (Pa). For skin strain, dimensionless; It is the static elastic modulus, with the unit Pascal (Pa), representing the long-term stiffness of the skin; It is the dynamic elastic modulus, with the unit Pascal (Pa), representing the instantaneous stiffness of the skin; The viscosity coefficient, measured in Pascal-seconds (Pa·s), represents the skin's relaxation properties. This embodiment is based on biomechanical literature. Pa, Pa, Pa·s.

[0135] In the simulation, the stress tensor at various points on the skin surface is calculated in real time. The current strain tensor is obtained by numerically solving the above differential equation (e.g., using the backward Euler method). Then, the minimum principal strain is extracted. Substitute into the formula Calculate the wrinkle amplitude at the current moment. However, at this point, the amplitude... The inherent stress and strain information from history gives the formation and disappearance of wrinkles a "memory" effect: when the skin is stretched (increased strain) and then suddenly relaxed, the wrinkles do not disappear immediately, but will loosen over time, simulating the persistence of real skin wrinkles.

[0136] Step S4': Multi-scale optimization based on the fusion of visual saliency and physical saliency; Building upon step S4, a physical saliency factor is introduced, integrating visual and physical saliency to more scientifically guide the allocation of computational resources. Physical saliency is defined as the degree of influence of a region on the motion results in the overall physical simulation.

[0137] Physical significance factor The calculation method is as follows: the sensitivity of each skin vertex to a certain global physical target (such as the overall strain energy) is calculated using the adjoint method. Specifically, the global physical target is defined as the total strain energy. ,in The strain energy density is defined in step S2 or S3. The calculation is performed by solving the adjoint equation. For skin apex The gradient at location, i.e., sensitivity. The unit is Newtons (N). The magnitude of this sensitivity reflects the vertex... The degree of influence on the overall physical state of the system. Physical significance factor. The sensitivity was obtained by normalizing it, with a value ranging from 0 to 1.

[0138] Final perceived importance weight Incorporating visual salience and physical significance : in, and For normalized weight coefficients, satisfying In this embodiment , The adaptive scheduler is based on... The system dynamically adjusts computational resources. For example, for regions with high physical saliency but low visual saliency (such as deep, occluded muscle attachment point regions that bear the main load), the system will still allocate higher computational precision to ensure the accuracy of physical simulation and avoid the accumulation of physical errors that may be caused by relying solely on visual saliency.

[0139] The model constructed in this embodiment shows improvements in simulation accuracy and realism compared to Embodiment 2. In measuring muscle contractile force, the time-frequency decomposition-based model accurately simulates the explosive contractile force (peak value 4200N, rise time 0.1 seconds) generated by the quadriceps muscle at the moment of starting a run (rapid movement) and the sustained contractile force (average value 1500N) during constant-speed running (slow movement). In contrast, the traditional model cannot distinguish between fast and slow muscle groups, simulating a slow rise in contractile force (rise time 0.3 seconds) and a lower peak value (3600N).

[0140] In terms of skin wrinkle generation, texture generation methods based on viscoelastic models show a slow fading process of wrinkles when the joint remains stationary after rapid bending (relaxation time constant is about 0.4 seconds), while wrinkles based on instantaneous strain models disappear instantly, losing visual realism.

[0141] In terms of computational resource scheduling, the perception optimization mechanism that integrates physical saliency enables the simulation system to reduce the physical simulation error (defined as the relative error between the overall deformation energy and the full-precision simulation result) to 0.8% while maintaining the same visual quality, compared to only 2.1% in Example 2. This shows that this example achieves efficient resource utilization while ensuring physical accuracy.

[0142] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0143] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing 3D modeling of simulated dinosaurs, characterized in that, Includes the following steps: Step S1: Obtain three-dimensional scanning data of biological skeletal fossils and generate a rigid skeletal system with mass attributes, rotational inertia attributes, and joint constraint attributes; A Lagrange dynamic equation in generalized coordinates is established, and the motion state of the skeletal system under the action of generalized forces is calculated by solving the equation. The generalized forces include active muscle contraction force, passive ligament constraint force, and external contact force. Step S2: On the surface of the skeletal model, a muscle layer composed of tetrahedral meshes is constructed based on anatomical attachment points; the muscle tissue is defined as a hyperelastic material, and its deformation is obtained by solving for the minimum value of the strain energy density function; the displacement field of the muscle driven by skeletal motion is calculated in real time using a position-based dynamics method. Step S3: Construct a skin layer composed of triangular meshes outside the muscle layer; project the Cauchy stress tensor calculated from the muscle surface onto the vertices of the skin mesh as the driving force for skin deformation; dynamically generate wrinkle and scale texture features based on the principal strain direction and magnitude at each point on the skin surface. Step S4: Construct a multi-scale topology optimization model based on perceived importance The perceptual importance weight of each vertex of the skin mesh is calculated in real time. The perceptual importance weight is obtained by fusing geometric saliency factors, motion saliency factors and viewpoint saliency factors. The mesh simplification degree and physical simulation accuracy of the corresponding region are dynamically adjusted according to the perceptual importance weight. Step S5: Acquire external motion capture data and map the motion capture data to the rigid body skeleton system; Establish a quadratic programming optimization problem with anatomical constraints, and solve for the optimal skeletal posture that satisfies the joint range of motion constraints; When a constraint is violated, the excess angular momentum is redistributed to adjacent joints; Step S6: Perform offline physical simulation on representative motion sequences and record the skin mesh deformation sequence; Principal component analysis was used to reduce the dimensionality of the deformation sequence and extract the characteristic deformation basic modes. During online operation, the deformation weight coefficient is predicted based on the current skeletal motion parameters, and a round of physical constraint iterative correction is performed on the prediction results.

2. The method for optimizing 3D modeling of simulated dinosaurs according to claim 1, characterized in that, In step S1, the active muscle contraction force is obtained through the following sub-steps: A desired activation signal is generated based on a preset motion sequence, and the desired activation signal is decomposed into several intrinsic mode functions through Hilbert-Huang transform. The instantaneous frequency of the intrinsic mode function is calculated. Intrinsic mode functions with instantaneous frequencies higher than a preset threshold are assigned to fast-twitch muscle fibers to obtain fast-twitch muscle activation. Intrinsic mode functions with instantaneous frequencies lower than or equal to the preset threshold are assigned to slow-twitch muscle fibers to obtain slow-twitch muscle activation. Using an improved Hill-type muscle model, the contractile forces of fast-twitch and slow-twitch muscle fibers are calculated based on the activation levels of the fast-twitch and slow-twitch muscles, respectively. The generalized force is obtained by summing the torques generated by all muscle fibers crossing the joint.

3. The method for optimizing 3D modeling of simulated dinosaurs according to claim 2, characterized in that, The improved Hill-type muscle model expresses muscle contractile force as the product of the muscle's maximum isometric contractile force, activation level, force-length relationship function, and force-velocity relationship function.

4. The method for optimizing 3D modeling of simulated dinosaurs according to claim 1, characterized in that, In step S2, the muscle tissue is defined as a fiber-reinforced hyperelastic material, and the strain energy density function includes an isotropic matrix term and an anisotropic fiber term. The anisotropic fiber term includes Macaulay brackets such that the anisotropic fiber term contributes stiffness only under tension when the square of the elongation ratio along the fiber direction is greater than 1.

5. The method for optimizing 3D modeling of simulated dinosaurs according to claim 1, characterized in that, In step S3, the skin layer and the muscle layer are coupled at the interface through a nonlinear spring-damper system; The nonlinear spring-damper system applies an interfacial force between the skin apex and the muscle surface, the interfacial force being controlled by the spring stiffness coefficient and the damping coefficient, allowing the skin layer to move along the tangential direction of the muscle surface when subjected to tangential stress.

6. The method for optimizing 3D modeling of simulated dinosaurs according to claim 1, characterized in that, In step S3, the dynamically generated wrinkle features adopt a viscoelastic model based on strain-time history. The strain-time history-based viscoelastic model establishes differential equations including static elastic modulus, dynamic elastic modulus, and viscosity coefficient. The current strain tensor is obtained by numerically solving the differential equations, and the minimum principal strain is extracted to calculate the wrinkle amplitude.

7. The method for optimizing 3D modeling of simulated dinosaurs according to claim 1, characterized in that, In step S3, the dynamically generated scale texture features adopt an anisotropic sampling method based on principal stress trajectory lines. During the shading stage, the high-resolution scale map is anisotropically filtered and sampled according to the principal strain direction of the mesh vertices.

8. The method for optimizing 3D modeling of simulated dinosaurs according to claim 1, characterized in that, In step S4, the geometric saliency factor is obtained by calculating the Gaussian curvature of the skin mesh using the Voronoi area method in discrete differential geometry; the motion saliency factor is obtained by calculating the motion velocity of the vertices of the skin mesh in the screen space between two adjacent frames.

9. The method for optimizing 3D modeling of simulated dinosaurs according to claim 1, characterized in that, In step S4, the perceived importance weight also includes a physical salience factor; The physical saliency factor is obtained by calculating the sensitivity of the vertices of the skin mesh to the total strain energy of the system using the adjoint method.

10. A simulated dinosaur 3D modeling optimization system, characterized in that, The system for implementing the method according to any one of claims 1 to 9, the system comprising: The physics-driven engine includes a bone dynamics solution module, a muscle finite element analysis module, and a skin deformation mapping module, which are used to execute the physical simulation process described in steps S1 to S3. The perception optimization controller includes a geometric analysis unit, a motion analysis unit, and a viewpoint analysis unit, and is used to calculate the perception importance weights described in step S4. The motion redirection module is used to perform the motion redirection and anatomical constraint correction described in step S5. The deformation prediction cache module is used to perform the deformation prediction and caching described in step S6. An adaptive scheduler dynamically allocates computing resources based on perceived importance weights, reducing mesh simplification and physical iteration counts for low-importance regions, and maintaining full-precision computation for high-importance regions.

Images