3D animation model intelligent generation method based on image recognition

Abstract

The invention discloses a 3D animation model intelligent generation method based on image recognition, and relates to the technical field of 3D animation models.Physical accuracy and visual fidelity of mechanical arm animation are remarkably improved through a multi-view feature fusion and physical constraint fusion mechanism; the space-time convolution kernel module is combined with mechanical kinematics constraint to effectively inhibit action distortion under high-speed motion; a dynamic anti-aliasing algorithm and a speed-adaptive LOD strategy are subjected to collaborative optimization, and edge aliasing and fuzzy mismatching are eliminated while the calculation load is reduced; external vibration data are converted into microscopic deformation through a micro-amplitude displacement superposition technology, and the limitation of traditional deformation simulation is broken through.

Description

Technical Field

[0001] The present invention relates to the technical field of 3D animation models, and in particular to an intelligent generation method of 3D animation models based on image recognition. Background Art

[0002] In the field of industrial automation, 3D robotic arm animation models are widely used in scenarios such as production line design verification, operator training, and remote control system development.

[0003] Traditional generation methods rely on keyframe animation technology and physical engine simulation to visualize the robot arm's motion process by pre-setting joint motion trajectories and rigid body dynamics parameters. However, in high-speed and high-precision operation scenarios, such as precision welding and high-speed sorting, there are motion distortions and visual artifacts when processing the high-speed motion of the robot arm.

[0004] Although the keyframe interpolation algorithm used in current technology can generate basic motion trajectories, it is difficult to accurately restore dynamic details at high speeds, such as inertial jitter and flexible deformation of joints, resulting in deviations between the animation and actual physical behavior; at the same time, the rendering pipeline based on rigid models is prone to edge jaggedness and motion blur mismatch problems when representing high-speed rotating parts, affecting the operator's accurate judgment of the robotic arm's position; therefore, there is an urgent need for an intelligent generation method of 3D animation models based on image recognition to solve such problems. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides an intelligent generation method for 3D animation models based on image recognition to solve the problems of motion distortion, visual artifacts and physical deviations in high-speed robotic arm simulations in traditional methods. The optimization scheme is limited by computational efficiency and model simplification, and it is difficult to balance accuracy and real-time performance.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] An embodiment of the present invention provides a method for intelligently generating a 3D animation model based on image recognition, comprising: step S1, collecting a sequence of continuous frame images of a target object through a multi-viewing sensor, and extracting a multimodal feature set including surface texture features, motion trajectory features, and spatial posture features;

[0009] Step S2: constructing a spatiotemporal convolution kernel module, inputting the multimodal feature set into a neural network including a motion trajectory prediction layer to generate a joint motion vector field, wherein the motion trajectory prediction layer includes an attention weight allocation mechanism based on the kinematic constraints of a mechanical linkage mechanism;

[0010] Step S3, driving the three-dimensional skeleton model according to the joint motion vector field, and synchronously performing dynamic detail enhancement processing in the rendering pipeline, wherein the dynamic detail enhancement processing includes an anti-aliasing algorithm based on motion speed adaptation and real-time compensation of material reflectivity;

[0011] Step S4: outputting an animation model integrated with physical constraints, wherein the model surface detail layer and the skeleton motion layer adopt independent resolution control strategies.

[0012] As a preferred solution of the method for intelligently generating a 3D animation model based on image recognition described in the present invention, the image processing of the multi-viewing sensor in step S1 includes:

[0013] Construct a multi-view feature fusion network to combine the edge features of mechanical structures and motion blur features collected from different viewpoints Figure 1 The composite motion trajectory descriptor is generated by consistent alignment.

[0014] As a preferred solution of the method for intelligently generating a 3D animation model based on image recognition described in the present invention, the motion trajectory prediction layer in step S2 includes:

[0015] The inertia compensation submodule calculates the momentum conservation constraints of each joint through inverse kinematics to generate a smooth motion trajectory that conforms to the dynamic characteristics of the rigid body. The momentum conservation constraints include the center of mass offset threshold and the joint torque limit parameters.

[0016] As a preferred solution of the method for intelligently generating 3D animation models based on image recognition described in the present invention, in step S2, in order to quantify the degree to which the length of the i-th connecting rod deviates from the ideal state, a compatibility score s is introduced. i : Among them, s i represents the compatibility score of the i-th link, α is the sensitivity adjustment coefficient, l i is the actual length of the i-th connecting rod, is the expected length of the i-th connecting rod;

[0017] Map the compatibility scores to attention weights so that the total weights sum to 1:

[0018]

[0019] Among them, w i represents the attention weight of the i-th link, k is the link index variable, and N is the total number of links;

[0020] Map the joint angle offset of the j-th link to the center of mass displacement:

[0021] Where Δr represents the mass center offset of the jth connecting rod, l j is the length of the jth connecting rod, θ j is the current angle of the j-th joint, is the initial angle of the jth joint;

[0022] Bonus rigid body dynamics differential equation, integrating the effects of moment of inertia, damping and stiffness on joint motion, the equation is:

[0023] Among them, I j is the moment of inertia of the j-th joint, is the angular acceleration of the jth joint, b j is the damping coefficient of the jth joint, is the angular velocity of the jth joint, k j is the stiffness coefficient of the jth joint, θ j is the current angle of the jth joint, τ j is the external torque of the j-th joint;

[0024] Convert the angle equation into a smooth displacement equation through centroid mapping:

[0025]

[0026] in, is the acceleration of the center of mass offset, is the speed of mass center displacement, and Δr is the mass center displacement.

[0027] As a preferred solution of the method for intelligently generating a 3D animation model based on image recognition described in the present invention, the anti-aliasing algorithm specifically includes:

[0028] The edge sampling density is dynamically adjusted according to the joint angular velocity, non-uniform oversampling is implemented in the rotation axis area, and gradient domain filtering is performed in the direction of the motion vector while retaining high-frequency mechanical structure features.

[0029] As a preferred solution of the method for intelligently generating a 3D animation model based on image recognition described in the present invention, the dynamic detail enhancement processing further includes:

[0030] Based on the real-time movement speed of each component of the model, a multi-level level of detail (LOD) control strategy is automatically triggered. The model patch subdivision level of high-speed moving components is negatively correlated with the movement speed, and a simplified collision body calculation mode is enabled above the preset speed threshold.

[0031] As a preferred solution of the method for intelligently generating a 3D animation model based on image recognition described in the present invention, the real-time compensation of material reflectivity is specifically as follows:

[0032] According to the ambient light sensor data and the changes in the normal direction of the moving parts' surfaces, the anisotropic reflection parameters of the metal material are dynamically adjusted, where the highlight intensity is positively correlated with the joint angular acceleration.

[0033] As a preferred embodiment of the method for intelligently generating a 3D animation model based on image recognition described in the present invention, in step S3, during the process of performing non-uniform supersampling on the rotation axis region and performing gradient domain filtering in the direction of the motion vector: in the rotation axis region, the number of supersampling points is dynamically allocated according to the joint angular velocity, and then gradient domain filtering is applied in the direction of the motion vector, and a filter kernel is constructed according to an anisotropic weight allocation algorithm, and finally, the anisotropic reflection parameters of the metal material are adjusted in real time according to the angular acceleration;

[0034] In step S3, a piecewise function model of joint angular velocity-sampling point number is constructed:

[0035] At the jth joint, the angular velocity is defined as ω j ;

[0036] Number of sample points n j Expressed as:

[0037]

[0038] Among them, n j is the number of super sampling points in the j-th joint area, ω j is the angular velocity of the jth joint, ω1, ω2, ω3 are the angular velocity thresholds, n min ,n max are the minimum and maximum number of sampling points, respectively; m1 and m2 are the slope coefficients of the corresponding intervals; c1 and c2 are the intercept coefficients of the corresponding intervals;

[0039] In step S3, gradient domain filtering and anisotropic weight distribution in the direction of the motion vector are performed:

[0040] Assume that the unit vector of the motion vector direction at the jth joint is:

[0041]

[0042] Define the filter kernel:

[0043]

[0044] Where x is the pixel offset vector, v j is the motion vector, σ ||,j ,σ ⊥,j are the filtering standard deviations along the motion direction and the vertical direction, respectively;

[0045] Introducing the direction weight function w d (θ)=exp(βj cos(θ-φ j )), where θ is the pixel sampling point relative to u j The polar angle, φ j is the main filtering direction angle, which is u j Direction, β j is the anisotropic sensitivity coefficient;

[0046] In actual filtering, each sample offset x is multiplied by the weight w d (∠(x,u j )), then convolution kernel K j (x) Complete gradient domain filtering;

[0047] In step S3, the anisotropic reflection parameters of the metal material are dynamically adjusted:

[0048] The angular acceleration of the jth joint As input, define the highlight intensity S j The sigmoid map:

[0049] Among them, S j is the dynamic highlight intensity, R0 is the static reference highlight intensity, R max The maximum highlight intensity, κ is the curve steepness coefficient, is the angular acceleration trigger threshold, is the angular acceleration of the jth joint.

[0050] As a preferred solution of the method for intelligently generating a 3D animation model based on image recognition described in the present invention, the method further includes:

[0051] A real-time physical feedback interface is integrated into the rendering pipeline to receive external input device vibration sensor data, and a micro-displacement superposition algorithm is used to present the microscopic deformation characteristics of the mechanical transmission system in the animation model.

[0052] As a preferred solution of the method for intelligently generating 3D animation models based on image recognition described in the present invention, the rendering pipeline integrates a real-time physical feedback interface, and maps vibration signals into vertex-level micro-deformations through a micro-displacement superposition algorithm. The specific process includes:

[0053] Convert the external device vibration sensor data d[n] to the spectrum domain:

[0054]

[0055] Where D[k] is the complex amplitude of the kth frequency component, d[n] is the vibration amplitude of the nth sampling point, and N FFT is the discrete Fourier transform length, ι is the imaginary unit;

[0056] Energy is aggregated on the preset B frequency bands to obtain the spectrum feature sequence:

[0057]

[0058] Among them, s b [t] is the energy of the b-th frequency band at time t, is the frequency index set of frequency band b, |·| represents the modulus of the complex amplitude;

[0059] Construct a convolutional mapping network from vibration spectrum to vertex displacement, and calculate the displacement Δv of the i-th vertex i The calculation formula for [t] is:

[0060]

[0061] Where Δv i [t] is the displacement vector of the i-th vertex at time t, W i,b,τ is the mapping weight tensor, which represents the influence of frequency band b on vertex i at time delay τ, and T is the length of the temporal convolution kernel;

[0062] Add nonlinear transformation after convolution: Δv' i (t)=σ(Δv i [t]), where Δv' i (t) is the displacement vector after the activation function, and σ(·) is the nonlinear activation function;

[0063] Add a slight displacement to the base model vertices: Among them, p i (t) is the final position of the i-th vertex at time t, is its initial resting position.

[0064] The beneficial effects of the present invention are as follows: the present invention significantly improves the physical accuracy and visual fidelity of the robotic arm animation through the multi-perspective feature fusion and physical constraint fusion mechanism; the spatiotemporal convolution kernel module is combined with mechanical kinematic constraints to effectively suppress motion distortion under high-speed motion; the dynamic anti-aliasing algorithm and the speed-adaptive LOD strategy are synergistically optimized to eliminate edge aliasing and blur mismatch while reducing the computational load; the micro-displacement superposition technology converts external vibration data into microscopic deformation, breaking through the limitations of traditional deformation simulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0066] Figure 1 This is a flow chart of the method for intelligently generating a 3D animation model based on image recognition in Example 1. DETAILED DESCRIPTION

[0067] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0068] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0069] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0070] Example 1, reference Figure 1 This embodiment provides a method for intelligently generating a 3D animation model based on image recognition, comprising:

[0071] Step S1, collecting a continuous frame image sequence of the target object through a multi-view visual sensor, and extracting a multimodal feature set including surface texture features, motion trajectory features, and spatial posture features;

[0072] The image processing of the multi-viewing sensor in step S1 includes:

[0073] Construct a multi-view feature fusion network to combine the edge features of mechanical structures and motion blur features collected from different viewpoints Figure 1 Consistent alignment to generate a composite motion trajectory descriptor with artifact removal;

[0074] Step S2: construct a spatiotemporal convolution kernel module and input the multimodal feature set into a neural network containing a motion trajectory prediction layer to generate a joint motion vector field. The motion trajectory prediction layer includes an attention weight allocation mechanism based on the kinematic constraints of the mechanical linkage mechanism.

[0075] The motion trajectory prediction layer in step S2 includes:

[0076] The inertia compensation submodule calculates the momentum conservation constraints of each joint through inverse kinematics to generate a smooth motion trajectory that conforms to the dynamic characteristics of the rigid body. The momentum conservation constraints include the center of mass offset threshold and joint torque limit parameters;

[0077] In step S2, in order to quantify the degree to which the length of the i-th link deviates from the ideal state, the compatibility score s is introduced i : Among them, s i represents the compatibility score of the i-th link, α is the sensitivity adjustment coefficient, l i is the actual length of the i-th connecting rod, is the expected length of the i-th connecting rod;

[0078] Map the compatibility scores to attention weights so that the total weights sum to 1:

[0079]

[0080] Among them, w i represents the attention weight of the i-th link, k is the link index variable, and N is the total number of links;

[0081] Map the joint angle offset of the j-th link to the center of mass displacement:

[0082] Where Δr represents the mass center offset of the jth connecting rod, l j is the length of the jth connecting rod, θ j is the current angle of the j-th joint, is the initial angle of the jth joint;

[0083] Bonus rigid body dynamics differential equation, integrating the effects of moment of inertia, damping and stiffness on joint motion, the equation is:

[0084] Among them, I j is the moment of inertia of the j-th joint, is the angular acceleration of the jth joint, b j is the damping coefficient of the jth joint, is the angular velocity of the jth joint, k j is the stiffness coefficient of the jth joint, θ j is the current angle of the jth joint, τ j is the external torque of the j-th joint;

[0085] Convert the angle equation into a smooth displacement equation through centroid mapping:

[0086]

[0087] in, is the acceleration of the center of mass offset, is the speed of mass center displacement, Δr is the mass center displacement;

[0088] Step S3, driving the three-dimensional skeleton model according to the joint motion vector field, and synchronously performing dynamic detail enhancement processing in the rendering pipeline, the dynamic detail enhancement processing including an anti-aliasing algorithm based on motion speed adaptation and real-time compensation of material reflectivity;

[0089] The anti-aliasing algorithm specifically includes:

[0090] Dynamically adjust edge sampling density based on joint angular velocity, implement non-uniform oversampling on the rotation axis area, and perform gradient domain filtering in the direction of the motion vector while retaining high-frequency mechanical structure features.

[0091] Dynamic detail enhancement processing further includes:

[0092] Based on the real-time movement speed of each model component, a multi-level level of detail (LOD) control strategy is automatically triggered. The subdivision level of the model facets of high-speed moving components is negatively correlated with the movement speed, and a simplified collision body calculation mode is enabled above the preset speed threshold.

[0093] Real-time compensation of material reflectivity is as follows:

[0094] Dynamically adjust the anisotropic reflectance parameters of metal materials based on ambient light sensor data and changes in the normal direction of the moving parts' surfaces, with a positive correlation between highlight intensity and joint angular acceleration.

[0095] In step S3, non-uniform supersampling is performed on the rotation axis region, and gradient domain filtering is performed in the direction of the motion vector: in the rotation axis region, the number of supersampling points is dynamically allocated according to the joint angular velocity, and then gradient domain filtering is applied in the direction of the motion vector. The filter kernel is constructed according to the anisotropic weight allocation algorithm, and finally the anisotropic reflection parameters of the metal material are adjusted in real time according to the angular acceleration;

[0096] In step S3, a piecewise function model of joint angular velocity-sampling point number is constructed:

[0097] At the jth joint, the angular velocity is defined as ω j ;

[0098] Number of sample points n j Expressed as:

[0099]

[0100] Among them, n j is the number of super sampling points in the j-th joint area, ω j is the angular velocity of the jth joint, ω1, ω2, ω3 are the angular velocity thresholds, n min ,n max are the minimum and maximum number of sampling points, respectively; m1 and m2 are the slope coefficients of the corresponding intervals; c1 and c2 are the intercept coefficients of the corresponding intervals;

[0101] In step S3, gradient domain filtering and anisotropic weight distribution in the direction of the motion vector are performed:

[0102] Assume that the unit vector of the motion vector direction at the jth joint is:

[0103]

[0104] Define the filter kernel:

[0105]

[0106] Where x is the pixel offset vector, v j is the motion vector, σ ||,j ,σ ⊥,j are the filtering standard deviations along the motion direction and the vertical direction, respectively;

[0107] Introducing the direction weight function w d (θ)=exp(β j cos(θ-φ j )), where θ is the pixel sampling point relative to u j The polar angle, φ j is the main filtering direction angle, which is u j Direction, β j is the anisotropic sensitivity coefficient;

[0108] In actual filtering, each sample offset x is multiplied by the weight w d (∠(x,u j )), then convolution kernel K j (x) Complete gradient domain filtering;

[0109] In step S3, the anisotropic reflection parameters of the metal material are dynamically adjusted:

[0110] The angular acceleration of the jth joint As input, define the highlight intensity S j The sigmoid map:

[0111] Among them, S j is the dynamic highlight intensity, R0 is the static reference highlight intensity, R max The maximum highlight intensity, κ is the curve steepness coefficient, is the angular acceleration trigger threshold, is the angular acceleration of the jth joint;

[0112] Specifically, through a piecewise function model, the number of supersampling points is directly linked to the angular velocity of the joints, maintaining rendering efficiency at low speeds and enhancing details at high speeds. In the gradient domain filtering stage, anisotropic filter kernels in the direction of the motion vector and directional weight distribution are combined to effectively remove motion blur artifacts while preserving the high-frequency characteristics of the mechanical structure. The dynamic highlight mapping mechanism, based on the S-shaped function response of angular acceleration, can adjust the reflection intensity in real time during motion acceleration or deceleration, making the metal material present a more realistic highlight trailing effect when rotating rapidly.

[0113] Step S4, outputting an animation model integrated with physical constraints, wherein the model surface detail layer and the skeleton motion layer adopt independent resolution control strategies;

[0114] The intelligent animation model generation method also includes: integrating a real-time physical feedback interface into the rendering pipeline to receive external input device vibration sensor data, and using a micro-displacement superposition algorithm to present the microscopic deformation characteristics of the mechanical transmission system in the animation model;

[0115] A real-time physics feedback interface is integrated into the rendering pipeline, mapping vibration signals to vertex-level micro-deformations through a micro-displacement superposition algorithm. The specific process includes:

[0116] Convert the external device vibration sensor data d[n] to the spectrum domain:

[0117]

[0118] Where D[k] is the complex amplitude of the kth frequency component, d[n] is the vibration amplitude of the nth sampling point, and N FFT is the discrete Fourier transform length, ι is the imaginary unit;

[0119] Energy is aggregated on the preset B frequency bands to obtain the spectrum feature sequence:

[0120]

[0121] Among them, s b [t] is the energy of the b-th frequency band at time t, is the frequency index set of frequency band b, |·| represents the modulus of the complex amplitude;

[0122] Construct a convolutional mapping network from vibration spectrum to vertex displacement, and calculate the displacement Δv of the i-th vertex i The calculation formula for [t] is:

[0123]

[0124] Where Δv i [t] is the displacement vector of the i-th vertex at time t, W i,b,τis the mapping weight tensor, which represents the influence of frequency band b on vertex i at time delay τ, and T is the length of the temporal convolution kernel;

[0125] Add nonlinear transformation after convolution: Δv' i (t)=σ(Δv i [t]), where Δv' i (t) is the displacement vector after the activation function, and σ(·) is the nonlinear activation function;

[0126] Add a slight displacement to the base model vertices: Among them, p i (t) is the final position of the i-th vertex at time t, is its initial rest position;

[0127] Specifically, this micro-displacement superposition framework can accurately capture the multi-band dynamics in the sensor vibration signal and convert it into vertex-level micro-deformations through frequency domain feature extraction and time domain convolution mapping; the detailed information of the vibration is retained in Fourier transform and frequency band aggregation, and the mapping network adapts to the response characteristics of different structures through trainable weights, and generates stable and coherent temporal displacements after nonlinear activation; the final displacement superposition does not affect the original skeletal animation, and can display the real vibration deformation of mechanical components such as mechanical gears and bearings at the micro scale; the fusion of signal processing and graph convolution technology provides an efficient and scalable solution for the subtle dynamic performance in high-fidelity mechanical animation.

[0128] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for intelligently generating 3D animation models based on image recognition, characterized in that: include, Step S1, collecting a continuous frame image sequence of the target object through a multi-view visual sensor, and extracting a multimodal feature set including surface texture features, motion trajectory features, and spatial posture features; Step S2: constructing a spatiotemporal convolution kernel module, inputting the multimodal feature set into a neural network including a motion trajectory prediction layer to generate a joint motion vector field, wherein the motion trajectory prediction layer includes an attention weight allocation mechanism based on the kinematic constraints of a mechanical linkage mechanism; Step S3, driving the three-dimensional skeleton model according to the joint motion vector field, and synchronously performing dynamic detail enhancement processing in the rendering pipeline, wherein the dynamic detail enhancement processing includes an anti-aliasing algorithm based on motion speed adaptation and real-time compensation of material reflectivity; Step S4: outputting an animation model integrated with physical constraints, wherein the model surface detail layer and the skeleton motion layer adopt independent resolution control strategies.

2. The method for intelligently generating a 3D animation model based on image recognition according to claim 1, wherein: The image processing of the multi-viewing sensor in step S1 includes: A multi-view feature fusion network is constructed to align the edge features of the mechanical structure and the motion blur features collected from different viewpoints consistently across views, and generate a composite motion trajectory descriptor with artifact removal.

3. The method for intelligently generating a 3D animation model based on image recognition according to claim 1, wherein: The motion trajectory prediction layer in step S2 includes: The inertia compensation submodule calculates the momentum conservation constraints of each joint through inverse kinematics to generate a smooth motion trajectory that conforms to the dynamic characteristics of the rigid body. The momentum conservation constraints include the center of mass offset threshold and the joint torque limit parameters.

4. The method for intelligently generating a 3D animation model based on image recognition according to claim 3, wherein: In step S2, the compatibility score s is introduced i : Among them, s i represents the compatibility score of the i-th link, α is the sensitivity adjustment coefficient, l i is the actual length of the i-th connecting rod, is the expected length of the i-th connecting rod; Map the compatibility scores to attention weights so that the total weights sum to 1: Among them, w i represents the attention weight of the i-th link, k is the link index variable, and N is the total number of links; Map the joint angle offset of the j-th link to the center of mass displacement: Where Δr represents the mass center offset of the jth connecting rod, l j is the length of the jth connecting rod, θ j is the current angle of the j-th joint, is the initial angle of the jth joint; Bonus rigid body dynamics differential equation, integrating the effects of moment of inertia, damping and stiffness on joint motion, the equation is: Among them, I j is the moment of inertia of the j-th joint, is the angular acceleration of the jth joint, b j is the damping coefficient of the jth joint, is the angular velocity of the jth joint, k j is the stiffness coefficient of the jth joint, θ j is the current angle of the jth joint, τ j is the external torque of the j-th joint; Convert the angle equation into a smooth displacement equation through centroid mapping: in, is the acceleration of the center of mass offset, is the speed of mass center displacement, and Δr is the mass center displacement.

5. The method for intelligently generating a 3D animation model based on image recognition according to claim 1, wherein: The anti-aliasing algorithm specifically includes: The edge sampling density is dynamically adjusted according to the joint angular velocity, non-uniform oversampling is implemented in the rotation axis area, and gradient domain filtering is performed in the direction of the motion vector while retaining high-frequency mechanical structure features.

6. The method for intelligently generating a 3D animation model based on image recognition according to claim 5, characterized in that The dynamic detail enhancement process further comprises: Based on the real-time movement speed of each component of the model, a multi-level level of detail (LOD) control strategy is automatically triggered. The model patch subdivision level of high-speed moving components is negatively correlated with the movement speed, and a simplified collision body calculation mode is enabled above the preset speed threshold.

7. The method for intelligently generating a 3D animation model based on image recognition according to claim 6, wherein: The real-time compensation of material reflectivity is specifically as follows: According to the ambient light sensor data and the changes in the normal direction of the moving parts' surfaces, the anisotropic reflection parameters of the metal material are dynamically adjusted, where the highlight intensity is positively correlated with the joint angular acceleration.

8. The method for intelligently generating a 3D animation model based on image recognition according to claim 7, wherein: In step S3, during the process of performing non-uniform supersampling on the rotation axis region and performing gradient domain filtering in the direction of the motion vector: in the rotation axis region, the number of supersampling points is dynamically allocated according to the joint angular velocity, and then gradient domain filtering is applied in the direction of the motion vector, and a filter kernel is constructed according to an anisotropic weight allocation algorithm, and finally, the anisotropic reflection parameters of the metal material are adjusted in real time according to the angular acceleration; In step S3, a piecewise function model of joint angular velocity-sampling point number is constructed: At the jth joint, the angular velocity is defined as ω j ; Number of sample points n j Expressed as: Among them, n j is the number of super sampling points in the j-th joint area, ω j is the angular velocity of the jth joint, ω1, ω2, ω3 are the angular velocity thresholds, n min ,n max are the minimum and maximum number of sampling points, respectively; m1 and m2 are the slope coefficients of the corresponding intervals; c1 and c2 are the intercept coefficients of the corresponding intervals; In step S3, gradient domain filtering and anisotropic weight distribution in the direction of the motion vector are performed: Assume that the unit vector of the motion vector direction at the jth joint is: Define the filter kernel: Where x is the pixel offset vector, v j is the motion vector, σ ||,j ,σ ⊥,j are the filtering standard deviations along the motion direction and the vertical direction, respectively; Introducing the direction weight function w d (θ)=exp(β j cos(θ-φ j )), where θ is the pixel sampling point relative to u j The polar angle, φ j is the main filtering direction angle, which is u j Direction, β j is the anisotropic sensitivity coefficient; In actual filtering, each sample offset x is multiplied by the weight w d (∠(x,u j )), then convolution kernel K j (x) Complete gradient domain filtering; In step S3, the anisotropic reflection parameters of the metal material are dynamically adjusted: The angular acceleration of the jth joint As input, define the highlight intensity S j The sigmoid map: Among them, S j is the dynamic highlight intensity, R0 is the static reference highlight intensity, R max The maximum highlight intensity, κ is the curve steepness coefficient, is the angular acceleration trigger threshold, is the angular acceleration of the jth joint.

9. The method for intelligently generating a 3D animation model based on image recognition according to claim 1, wherein: Also includes: A real-time physical feedback interface is integrated into the rendering pipeline to receive external input device vibration sensor data, and a micro-displacement superposition algorithm is used to present the microscopic deformation characteristics of the mechanical transmission system in the animation model.

10. The method for intelligently generating a 3D animation model based on image recognition according to claim 9, wherein: The rendering pipeline integrates a real-time physical feedback interface, mapping vibration signals into vertex-level micro-deformations through a micro-displacement superposition algorithm. The specific process includes: Convert the external device vibration sensor data d[n] to the spectrum domain: Where D[k] is the complex amplitude of the kth frequency component, d[n] is the vibration amplitude of the nth sampling point, and N FFT is the discrete Fourier transform length, ι is the imaginary unit; Energy is aggregated on the preset B frequency bands to obtain the spectrum feature sequence: Among them, s b [t] is the energy of the b-th frequency band at time t, is the frequency index set of frequency band b, |·| represents the modulus of the complex amplitude; Construct a convolutional mapping network from vibration spectrum to vertex displacement, and calculate the displacement Δv of the i-th vertex i The calculation formula for [t] is: Where Δv i [t] is the displacement vector of the i-th vertex at time t, W i,b,τ is the mapping weight tensor, which represents the influence of frequency band b on vertex i at time delay τ, and T is the length of the temporal convolution kernel; Add nonlinear transformation after convolution: Δv' i (t)=σ(Δv i [t]), where Δv' i (t) is the displacement vector after the activation function, and σ(·) is the nonlinear activation function; Add a slight displacement to the base model vertices: Among them, p i (t) is the final position of the i-th vertex at time t, is its initial resting position.

Images